COMPARATIVE STUDIES IN ROMANIAN SYNTAX
Related Elsevier books BROWN & MILLER (eds.)
Concise Encyclopedia of Grammatical Categories
BROWN & MILLER (eds.)
Concise Encyclopedia of Syntactic Theories
CINQUE & SALVI (eds.)
Current Studies in Italian Syntax Offered to Lorenzo Renzi
MANOLIU-MANEA
Discourse and Pragmatic Constraints on Grammatical Choices
Related Elsevier journals Language and Communication Editors: Roy Harris and Talbot J. Taylor Language Sciences Editor: Nigel Love Lingua Editors: Johan Rooryck and Neil Smith
Free specimen copies available on request
For further information on the NHLS series and for details of how to submit a proposal go to: http://www.elsevier.nl/locate/series/nhls
NORTH-HOLLAND LINGUISTIC SERIES
58
COMPARATIVE STUDIES IN ROMANIAN SYNTAX
Edited by VIRGINIA MOTAPANYANE University of New Brunswick, Canada
2000 ELSEVIER Amsterdam - Lausanne - New York - Oxford - Shannon - Singapore - Tokyo
ELSEVIER SCIENCE Ltd The Boulevard, Langford Lane Kidlington, Oxford OX5 1GB, UK
© 2000 Elsevier Science Ltd. All rights reserved. This work is protected under copyright by Elsevier Science, and the following terms and conditions apply to its use: Photocopying Single photocopies of single chapters may be made for personal use as allowed by national copyright laws. Permission of the Publisher and payment of a fee is required for all other photocopying, including multiple or systematic copying, copying for advertising or promotional purposes, resale, and all forms of document delivery. Special rates are available for educational institutions that wish to make photocopies for non-profit educational classroom use. Permissions may be sought directly from Elsevier Science Global Rights Department, PO Box 800. Oxford OX5 IDX. UK; phone: (+44) 1865 843830. fax: (+44) 1865 853333. e-mail:
[email protected]. You may also contact Global Rights directly through Elsevier's home page (http://www.elsevier.nl), by selecting "Obtaining Permissions'. In the USA. users may clear permissions and make payments through the Copyright Clearance Center. Inc., 222 Rosewood Drive. Danvers. MA 01923. USA; phone: (978) 7508400. fax: (978) 7504744, and in the UK through the Copyright Licensing Agency Rapid Clearance Service (CLARCS). 90 Tottenham Court Road, London WIPOLP, UK; phone: (+44) 171 631 5555; fax: (+44) 171 631 5500. Other countries may have a local reprographic rights agency for payments. Derivative Works Tables of contents may be reproduced for internal circulation, but permission of Elsevier Science is required for external resale or distribution of such material. Permission of the Publisher is required for all other derivative works, including compilations and translations. Electronic Storage or Usage Permission of the Publisher is required to store or use electronically any material contained in this work, including any chapter or part of a chapter. Except as outlined above, no part of this work may be reproduced, stored in a retrieval system or transmitted in any form or by any means, electronic, mechanical, photocopying, recording or otherwise, without prior written permission of the Publisher. Address permissions requests to: Elsevier Global Rights Department, at the mail, fax and e-mail addresses noted above. Notice No responsibility is assumed by the Publisher for any injury and/or damage to persons or property as a matter of products liability, negligence or otherwise, or from any use or operation of any methods, products, instructions or ideas contained in the material herein. Because of rapid advances in the medical sciences, in particular, independent verification of diagnoses and drug dosages should be made.
First edition 2000 Library of Congress Cataloging-in-Publication Data Comparative studies in Romanian syntax / editor, Virginia Motapanyane. p. cm. — (North-Holland linguistic series ; 58) Includes bibliographical references. ISBN 0-08-043871-7 1. Romanian language—Syntax. I. Motapanyane, Virginia. II. Series. PC713 .C642000 459'.5--dc21 00-055141 British Library Cataloguing in Publication Data A catalogue record from the British Library has been applied for. ISBN: 008 043871 7 ISSN: 0078-1592 ©The paper used in this publication meets the requirements of ANSI/NISO Z39.48-1992 (Permanence of Paper). Printed in The Netherlands.
CONTENTS The generative approach to Romanian grammar: An overview Gabriela Alboiu and Virginia Motapanyane
1
Feature checking meets the Criterion Approach: Three ways of saying only in Romance and Germanic
49
Josef Bayer and Alexander Grosu The double subject construction in Romanian Alexandra Cornilescu
83
Last resort strategies in DP: Article reduplication in Romanian and French Yves D'Hulst, Martine Coene and Liliane Tasmowski
135
(In)definiteness spread: From Romanian genitives to Hebrew construct state nominals Carmen Dobrovie-Sorin
177
Optimal Romanian clitics: A cross-linguistic perspective Geraldine Legendre
227
Parameters for Focus in English and Romanian Virginia Motapanyane
265
The syntax and morphology of Romanian imperatives Mihaela Pirvulescu and Yves Roberge
295
Bibliography of generative studies on Romanian
315
Name Index
329
Subject Index
335
This page intentionally left blank
PREFATORY NOTE Three papers in this collection originated as presentations at the workshop on "Comparative issues in Romanian syntax", held on August 30, 1996, at the University of New Brunswick, Saint John. Subsequently, these papers have evolved into more complete studies on the structure of possessive DPs, the clitic system, and focus constructions. Also, more linguists have shown their interest in a comparative study of Romanian syntax, and have submitted their contributions for publication. The book benefited from peer reviewing as a collection, with further reviews for each contribution. For valuable advice and timely reports, I am most grateful to Alexandru Grosu and to the following reviewers: Judy Bernstein, Carmen Dobrovie-Sorin, Donka Farkas, Giuliana Giusti, Julia Horvath, Mihaela Pirvulescu, Georges Rebuschi, Catherine Rudin, Tali Siloni. This project would not have been possible without helpful advice from Yves D'Hulst and Martine Coene, and patient guidance from Johan Rooryck, to whom I express my gratitude. I would also like to thank Gabriela Alboiu (University of Manitoba) and Julie Brittain (Memorial University of Newfoundland) for most helpful editing work, and Rod Hill (University of New Brunswick) for technical assistance. Both the workshop and the publication of this collection have been supported by generous grants from the University of New Brunswick, Saint John, and the Social Sciences and Humanities Research Council of Canada (#410951417). Virginia Motapanyane University of New Brunswick Saint John, New Brunswick
This page intentionally left blank
The generative approach to Romanian syntax
\
THE GENERATIVE APPROACH TO ROMANIAN GRAMMAR: AN OVERVIEW Gabriela Alboiu, University of Manitoba Virginia Motapanyane, University of New Brunswick-Saint John
The introduction to this volume provides an outline of Romanian syntax from the perspective of generative grammar. Selected topics will be introduced through examples and inflectional paradigms, followed by a presentation of the (sometimes highly controversial) analyses they prompted in generative studies. Given space limitations, only a few issues have been singled out for discussion, leaving unmentioned several peculiarities of this language. Also, the summary of analyses focusing on a certain issue could not always be exhaustive, as part of the relevant bibliography remained inaccessible to the authors. We take this opportunity to apologize to the linguists who do not find their studies mentioned hi this overview, and assure them that the omission may only stem from our futile attempts to obtain their publications. The papers included in this volume are presented under the respective topic headings, with the relevant commentaries indicated by [•*"] for the reader's convenience. Although the introduction has a list of references, more about each topic can be found by consulting the Appendix to the volume, which contains a general bibliography of the generative studies on Romanian syntax from 1980 to the present.
i. NOMINAL ELEMENTS Romanian nouns display three genders: masculine, feminine and neuter. The neuter gender consists of mixed masculine (singular forms) and feminine (plural forms) morphology, as shown in Table (1).
2
Comparative studies in Romanian syntax Table 1. singular
plural
un baiat/"a boy
MASC
1
o fata/a girl
FEM NEUTER
doi baieji /"two boys' doua fete /"two girls'
1
un ou/'an egg
doua oua /"two eggs'
Definite articles occur as enclitics to the noun, and display different forms for gender and number, as shown by the boldface endings in Table (2). Furthermore, the morpheme for oblique Case, in italics in Table (2), may attach to the definite article. Thus, Table (2) shows the inflection of the definite article as well as the Case marking for nominative (NOM) and accusative (Ace), that is, structural Cases, contrasting with genitive (GEN) and dative (DAT), that is, morphological Case. Table 2 Noun declension MASC/NEUTER
SG
PL
FEM/NEUTER 1
NOM/ACC
baiatul /'the boy; oul /"the egg
fata /the girl'
GEN/DAT
baiatul ui /"of the boy; oulw/ /"of the egg'
fetei / 'of the girl'
NOM/ACC
baiejii /"the boys'
fetele /'the girls'; ouale/the eggs'
GEN/DAT
baiejilor /"of the boys'
fetelor/'of the girls'; oualor/'of the eggs'
Various classes of adjectives may either precede or follow the noun; in most cases, a complementary distribution applies, as shown in Table (3). Adjectives preceding the noun carry the enclitic article and Case morphology, when applicable:
The generative approach to Romanian syntax Table 3 The position of adjectives adjective + noun
noun + adjective
bietul baiat /"poor-the boy
*baiatul biet / "boy-the poor1
acest baiat / 'this boy
*baiatul acest /"boy-the this1
*acesta baiat / 'this-A boy
baiatul acesta /"boy-the this-A'
saracul baiat/'poor-the boy
baiatul sarac / "boy-the poor1
1.1. Morphology Farkas (1990) approaches the lexical mechanism of underspecification of [+N] elements. She demonstrates that Romanian has a feature specification default (FSD) rule, which sets [-fern] as the default feature hi a system where the lexical specification for gender is [fern], and nouns have [+/fern] features. Similarly, the FDS correlated to the number feature is [-pi], in a system where nouns have [+/-pl] features. Nouns of neuter gender lack a lexical specification and their morphology relates to the number specification; so the default feature [-pi] correlates to the default gender feature [-fern], and leads to the inflectional distribution of neuter nouns mentioned in Table (1). This analysis accounts for facts of inflectional morphology and agreement in Romanian, and it also proves that phonological rules of underspecification may be successfully extended to morphosyntax.
1.2. The Internal Structure of DP Cornilescu (1992) and Grosu (1988, 1994) argue for a functional domain which embeds Romanian noun phrases (NP). The functional category D with determiner [det] features, may be lexical or nonlexical Lexical material compatible with D comprises: articles (and negative articles), as hi (4a, b); demonstrative determiners, as hi (4c); article-like quantifiers, as hi (4d); quantifiers, as hi (4e, f). (4)
a. c. e.
fete/e girls-the aceasta. fata this-FEM girl multe fete many-FEM girls
b. d. f
nici o fata not one-FEM girl fiecarefata each girl trei fete three girls
4
Comparative studies in Romanian syntax
The structure of DP consists of two hierarchical levels in Cornilescu (1992): the highest D has [Case] features, added to [det] of a definite type; the lower D has default [det] features and selects the NP. [Case/definiteness] features on the highest D attracts the noun through a cyclic head-to-head movement, as in (5a, b). When the lower D is filled, the noun surfaces in situ, as in (5c, d). If morphological Case is needed hi a configuration like (5c, d), an expletive determiner is inserted hi the highest D to carry the morpheme for oblique Case, as hi (5e, f). (5) a. DP D t
NP
N t b. cartea copilului book-the kid-the-GEN 'the child's book'
d. trei copii three kids
f cartea celor book-the the-GEN three kids 'the three children's' book'
Although controversies exist with respect to the exact configuration of the DP (cf Grosu 1988), it is generally accepted that enclitization results from N movement to D, and morphological Case is discharged within DP (cf Bernstein 1993, Giusti 1993).
1.3. Case §tefanescu (1997) draws a distinction between structural and morphological Case checking: the nominative and the accusative Case features of D are checked from outside when the entire DP enters into a local relation with an inflectional head (e.g., Tense) or a preposition (P). The oblique Case, on the other hand, undergoes checking hi a local relation within DP. 1.3.1. Genitive. Genitive Case assignment brings into discussion possessive constructions, hi which the possessor checks the genitive Case feature as follows: through a genitive morpheme, as hi (6a); or against a preposition. The latter configuration entails afof, as hi (6b), or another genitive Case assigning preposition as hi (6c). In case of non-adjacency between possessor and possessee, the element al (inflected for gender and number) must precede the noun inflected for genitive, as hi (6d). (6)
cartea copilulw/ book-the child-the-GEN
The generative approach to Romanian syntax b. c.
d.
5
tatal atrei copii father-the of three children in spatele copacuh/; in back-the tree-the-GEN 'behind the tree' fratele eel mare al copiilo/brother-the the big of-the-MASC. SG children-the-GEN 'the elder brother of the children'
The standard assumption is that the definite article -L (with various Spell Out implementations) has the capacity to assign genitive Case (Cornilescu 1992, Grosu 1988), as long as it has a D status in syntax (Grosu 1994). Also, prepositions as in (6b, c) check the Case features in a local relation with the possessor DP. The locality for Case checking is emphasized by a requirement for linear adjacency between the Case bearing functional category and the possessor. However, controversies arise with respect to the element al, inserted in the structure when adjacency is unavailable, as in (6d) and (7a). Dobrovie-Sorin (1987) defines al as an inflectional element, and Cornilescu (1992) follows this path, defining al (and eel 'the one') as expletive determiners. As a non-0-binder, al selects a null NP to which it assigns Case under government. Arguing from a comparative perspective against Cornilescu (1992) and Dobrovie-Sorin (1987), Grosu (1994) posits that al consists of a prepositional stem merged with the definite morpheme -L. Then al is a derivational (vs. inflectional) morpheme which behaves as a regular genitive assigning preposition, on a par with in spatele "behind1, in (6c). §tefanescu (1997) integrates the two approaches to al in a Linear Correspondence Axiom (LCA) analysis of Romanian DPs, following Kayne (1994). In this study, a 'of is a prepositional determiner (D/P), on a par with de 'of; dintre 'of,'among'; din 'from'. Unlike its counterparts, a has Agreement features. Thus, possessive and partitive DPs in Romanian display the configuration proposed in Kayne (1994) for cfe-DPs in Romance, with adjustments arising from the presence of morphological genitive Case in Romanian, and, in a-structures, the activation of Agreement features on the D/P head, as in (7). (7)
a.
o fata (de-) a Mariei a daughter (of) of-FEM.SG. Mary-GEN
Comparative studies in Romanian syntax
In (7), the possessee DP moves to SpecD/PP to check the Agreement features of a. Accordhlgly, (7b) suggests that a does not check genitive Case features, but only [agr] features. Other D/Ps cannot replace a in a possessive construction, although they may co-occur, as de-a in (7a).
"^
The paper by D'Hulst, Coene & Tasmowski included hi this volume resumes the discussion of genitive marking from a Minimalist (Chomsky 1995) perspective. The authors rely on the concept of overt [Case] feature checking, implemented hi a local Spec-head configuration. Crucially, Romanian D selects a functional category PossP, whose head P is associated with [Case]. In a genitive construction such as (6a) the 'possessee' moves from N to D transiting through Poss-head, whereas the 'possessor' DP moves to SpecPossP for Case checking. In al constructions, as in (6d), a- hi N raises cyclically to D and incorporates with the definite article stem -L. The 'possessor' DP moves to SpecPossP for Case checking. Comparative paradigms with French suggest that the definite article may be expletive in Romance. Romanian uses this option in possessive article constructions, such as (6d), as a 'last resort strategy". Thus, this analysis offers a uniform treatment for genitive Case checking hi adjacent and non-adjacent Possessee-Possessor strings.
"^
Carmen Dobrovie-Sorin (present volume) argues that Romanian morphologically marked genitives must be analyzed as arguments of the function (from individuals to individuals) denoted by the main N. The functional analysis of the main N is incompatible with fully semantic determiners on the main N: only the definite article (which can be expletive) is allowed. Thus, the necessary cooccurrence of morphological genitives and definite articles hi Romanian, which has so far been attributed to a language-specific mechanism of Case assignment is now a result of the functional interpretation of the main noun, a phenomenon we may expect to appear hi other languages (e.g. Saxon and Hebrew, discussed hi the paper).
The generative approach to Romanian syntax
1
1.3.2. Dative. In Cornilescu (1992), both genitive and dative Case are associated with the highest D in a configuration like (5), and assigned under government to the determiner hi lower SpecDP. Thus, the indirect object DP hi (8a) and the possessive DP hi (8b) undergo a similar checking procedure for dative and genitive Case, respectively. The latter configuration displays the preposition a in the Case assigning D head: (8)
a.
b.
I-am dat-o um« copil. him-have-lSG given-it a-DAT child 'I gave it to a child' stapan a multor oi owner of many-GEN sheep 'Owner of many sheep'
Grosu (1994) points out that the same Case morpheme serves for dative and genitive Case marking, and attaches to the determiner, not to the noun itself, as shown hi (8). In Stefinescu (1997), the oblique Case morpheme is associated with a functional head hi the D/PP domain. The different GEN/DAT reading depends on whether this functional head belongs to a predicational (genitive) or a non-predicational (dative) structure.
1.4. Word Order in DPs 1.4.1. Post-nominal position. Several studies (Cinque 1990, Grosu 1988) demonstrate that overt N movement to D takes place hi Romanian, yielding the linear order noun-adjective, as shown hi Table (3). The adjective is hi a Specifier (Spec) position, and does not interfere with the head-tohead movement. Giusti (1995) remarks that adjectives and demonstratrves/possessives follow the definite noun, as hi (9). The latter occur hi complementary distribution hi this position: (9)
a. b.
fete/e acestea/mele iste{e grrls-the these/my-FEM.PL smart- FEM.PL *fete/e isteje acestea/mele gkls-the smart- FEM.PL these/my- FEM.PL
The linear order in (9a) arises from N attraction to the head D (which carries the definite article le,
8
Comparative studies in Romanian syntax
through a cyclical movement) transiting through the functional heads whose Spec positions host the adjective and the demonstrative. This movement triggers Agreement features, morphologically realized as the endings -a and -le in the demonstrative and the possessive form, respectively, as well as in the adjectival inflection. As expected from this analysis, the adjective cannot move across the demonstrative/possessive, as shown in (9b), since both constituents occupy Spec positions. 1.4.2. Pre-nominal position. Adjectives occur in pre-nominal positions in two environments: (i) when they carry the article, as in (lOa); or (ii) when they are preceded by the demonstrative, as in (lOb). (10)
a. O frumoasa fata/frumoaso fata ne-a salutat a-FEM.SG. pretty girl/pretty-the-FEM.SG. girl us greeted 'A/the pretty girl greeted us.' b. Aceasta frumoasa fata/*Frumoasa fata ne-a salutat this-FEM.so. pretty girl/pretty girl us greeted 'This/0 pretty girl greeted us.'
In this configuration, the pre-nominal demonstrative lacks the Agreement morpheme -a, mentioned for (9a). This inflectional change coincides with the occurrence of the noun in situ, indicating that the type of structural Agreement arising in (9) is not obtained. Supposedly, in (lOa) the indefinite article in D, as a free morpheme, does not attract overt N to D movement. For (lOb), Giusti (1995) proposes that the demonstrative, merged in the highest Spec of the functional domain, moves to SpecDP, where it is attracted by some features of empty D. The analyses summed up hi this section suggest the following characterization of the Romanian word order: when D is realized as a free morpheme, no movement applies within DP and the noun surfaces in situ, following its modifiers. The noun presents the same placement when D is nonlexical and attracts a phrasal constituent (e.g. demonstrative) to SpecDP. However, a bound morpheme in D may trigger N movement to this head inducing a configuration hi which the noun precedes its modifiers.
1.5. Pronouns Two classes of pronouns occur in Romanian: clitic and non-clitic. The non-clitic pronouns have the distribution of NPs and present different lexical entries according to Person, Number and Case (nominative, dative, accusative). Clitic pronouns also have distinct lexical forms for Person, Number and Case, but their paradigm covers only dative and accusative. Within the latter paradigm, we can
The generative approach to Romanian syntax
9
distinguish two series of clitics: syntactic, versus syntactic/phonological clitics (Dobrovie-Sorin 1994: 70). The clitics that are both syntactic and phonological cliticize to the surrounding constituents, either to the left or to the right. Table (11) sums up the pronominal system of Romanian. Table 11 The pronominal system of Romanian NOM
DAT
- clitic
- clitic
+ clitic
phon.
- clitic
+ clitic
+phon.
ISO: eu
mie
imi
-mi-
mine
ma
-m-
2SG: tu
tie
itf
-tf-
tine
te
-te-
3SG.MASC: el
lui
ii
-i-
el
il
-1-
3SG.FEM: ea
ei
ii
-i-
ea
0
-o-
1PL: noi
noua
ne
-ne-
noi
ne
-ne-
2PL: voi
voua
va
-V-
voi
va
-V-
3PL.MASC: ei
lor
le
-le-
ei
ii
-i-
3PL.FEM.: ele
lor
le
-le-
ele
le
-le-
ACC
1.5.1. Clitic pronouns. In the clausal structure, clitic pronouns surface in a fixed position, preceding the verb or the auxiliary moved to the highest inflectional head, as in (12a). Phonological clitics attach either to this constituent or to the constituent to their left, as hi (12b). Syntactically, pronominal clitics attach to the right (ie. to the verb). In configurations with Long Head Movement (LHM) (Lema & Rivero 1989), only phonological clitics may attach to the raised verb (12c). (12)
a.
b.
c.
lonii invita./Ion i-a invitat. Ion them-ACC invites/ John them-ACC-has invited 'Ion invites/has invited them.' Mi-a spus ca-i aduce/ca i-aduce o carte. me-DAT-has said that-him-DAT brings a book 'He told me he's bringing a book for her.' Placutu-mi-a jocul./*Placutu-imi-a jocul. liked-me-DAT-has dance-the 'Hiked dancing.'
The intriguing aspect of the clitic system concerns the placement of the ACC ch'tic o 'her' in modern
10
Comparative studies in Romanian syntax
Romanian. Although this clitic precedes the inflected verb, on a par with the forms in (12), it is obligatorily encliticized to it in complex tenses, and optionally encliticized to infinitives, as in (13). (13)
a.
b.
o rog/ am rugat-o/ *o am rugat her ask-1 SG/have-1SG asked-her/her have-1SG asked 'I ask/asked her.' dorinja de-a o vedea/ de-a vedea-o wish-the of to her see-INF/of to see-INF her 'the wish to see her'
Dobrovie-Sorin (1994:75) considers that post-verbal o indicates the basic position for clitics in the functional domain. Further clitic movement to the highest inflectional head must be independently motivated.
^*>
Legendre's paper, included in this volume, re-assesses the Romanian clitic system in the framework of Optimality Theory. Using a wide comparative perspective, the author argues for a morphological analysis of clitics in general. The major claim is that clitics, as lexical affixes, instantiate functional features and become subject to alignment constraints. This analysis covers a larger domain of empirical data, accounting not only for the crosslinguistically variable cluster-internal order but also for their variable position in the clause.
1.5.2. Anaphoric pronouns. The morpheme se has the phonological variant -s-, and distinct £§//-§/forms for DAT. All the forms have the distribution of clitic pronouns, although they are A-bound. Dobrovie-Sorin (1994, 1998) defines se as an anaphoric ACC clitic, instantiated in Romanian either as a reflexive, as in (14a), or a middle/passive, as in (14b). So Romanian lacks a counterpart to the Italian NOM subject clitic si (Cinque 1988), as shown in (14c, d).
(14)
a.
b.
c.
Ion s-a ras. Ion se-has shaved 'Ion shaved himself' S-au recitat poezii de Eminescu. *e-have-3PL recited poems of Eminescu 'Poems by Eminescu were recked.' (Dobrovie-Sorin 1994,142) Non si e mai contend not si is more happy
The generative approach to Romanian syntax
d.
11
'No one is happy any more.' *Nu se este niciodata multumit. not se is more happy (Dobrovie-Sorin 1994,140)
The non-clitic counterpart of se is sine, occurring in contexts which ensure anaphoric binding, as in (15a). As de Jong (1996) observes, prepositions (other than ACC pe 'on' in clitic doubling constructions) may interfere with the A-binding of sine and rule out sentences such as (15b). (15)
a.
b.
Ion se iube§te pe sine, (de Jong 1996,131) Ion se loves pe himself 'Ion loves himself.' *Ion §i-a construit casa pentru sine. Ion se-DAT-has built house-the for himself 'Ion has built himself a house.'
A bound anaphoric morpheme -insu§i '-self, inflected for Gender and Number, forms lexical compounds either with sine or with a non-clitic pronoun. The binding conditions for the pronouninsu$i complex and the sine-insu§i complex differ, as indicated by the different distribution in relation to prepositions, as shown in (16). (16)
a.
b.
c.
PetrCj a construit casa pentru elj Peter has build house-the for him 'Peter built a house for himself.' Petrej a construit casa pentru el-insu^ Peter has built house-the for he-self 'Peter built himself a house.' ??Petre; a construit casa pentru sine-/«sM§/, Peter has build house-the for him-self 'Peter has built the house for himself.' (de Jong 1996,128)
1.6. Clitic Doubling hi certain contexts, Romanian presents clitic doubling of object DPs. In particular, nouns with the semantic features [+human], [+specific] induce clitic doubling when they carry the object theta-role, as in (17). Direct object DPs are embedded under the preposition pe (the counterpart of Spanish a),
12
Comparative studies in Romanian syntax
whereas indirect object DPs display morphological Case marking for dative. Current analyses of the clitic doubling mechanism proposed for other languages may extend to Romanian as well (Mahajan 1991): (17)
a.
b.
Ion lja invitat peprietenulj meu. Ion him-ACC has invited/>e-fiiend-the-ACC my 'Ion invited my friend.' Ion ija dat o carte prietenuliUj sau. Ion him-DAT has given a book friend-the-DAT his 'Ion gave a book to his friend.'
Clitic doubling applies to any object DP that undergoes Clitic Left Dislocation (CLLD), as in (18a), unless the heading noun has intrinsic quantificational features, as in (18b). (18)
a.
b.
Carteaj Oj cumparasem t; demult. book-the-ACC her-ACC bought-1 sg long ago "The book, I had bought it long ago.' Nimic; nu (*!)am cumparat t;. nothing-ACC not it-ACC have-lSG bought 'I haven't bought anything.'
The ungrammaticality of clitic doubling in (18b) follows from well-formedness conditions on operator-variable chains, which exclude intervening clitics (see Cinque 1990, Dobrovie-Sorin 1990, 1994). 1.6.1. Object raising. Gierling (1996, 1997) offers a new approach to clitic doubling, based on Romanian data. She demonstrates that direct object DPs may overtly move out of VP to the functional domain (see also Alboiu 1998), in which case they induce clitic doubling. Specificity does not explain the range of DPs which undergo clitic doubling. For example, a specific reading applies to both (19a) and (19b), although clitic doubling is optional. (19)
a. b.
Ion pupa mireasa. Ion kisses bride-the-ACC Ion o pupa pe mireasa. Ion her-ACC kisses pe-bride-the-ACC 'Ion kisses the bride.'
Gierling further notices that the option for movement, and therefore, clitic doubling, coincides with
The generative approach to Romanian syntax
13
the (in)capacity of the DP to project its [Focus] feature within VP. By means of comparative paradigms, the author shows that in Romanian only V is focused when the DP does not undergo clitic doubling, as in (19a). However, when clitic doubling applies, as in (19b), both the DP and the verb must be stressed. This indicates that Focus cannot project from a doubled DP (unless it raises to a pre-verbal position as opposed to SpecAgToP). The restriction on the Focus reading suggests that a counterpart to the [Focus] feature may associate with a functional head, presumably Agr0, attracting the DP to SpecAgToP. This feature is denned as [Topic-hood], and appears with the same effects in scrambling structures. Thus, clitic doubling and scrambling receive a uniform account, confirmed by facts of binding and the distribution of Focus stress. 1.6.2. Wh and quantification. Clitic doubling has consequences for the syntax of wh-phrases. Dobrovie-Sorin (1990, 1994) notices that some wh-phrases create clitic doubling chains, whereas others do not, as in (20). (20)
a.
b.
Pecarej *(li)ai invitattj? pe which-ACC him-ACC have-2so invited 'Which one did you invite?' Pecinej ("%)ai invitattj? pe whom-ACC him-ACC have-2so invited 'Whom did you invite?'
Contrasts as in (20a) and (20b) follow from the properties of wh-phrases, insofar as the quantificational [qu] features may or may not percolate from the head N to the maximal projection. When they fail to percolate, as in care 'which', the wh-element quantifies only over NP, and cannot create operator-variable chains. The lexical properties of wh-phrases interact with the syntactic properties of the position they are assigned to. In particular, SpecCP does not qualify as a position for structural operators in Romanian, showing a negative setting for the structural quantifier parameter. In consequence, systematic variations will oppose Romanian to languages in which SpecCP hosts only operators, as for example, English, in which SpecCP has a positive setting for the structural quantifier parameter. Thus, variation between the two languages arises in constructions with Quantifier Raising and in structures which rely on empty operators.
14
Comparative studies in Romanian syntax
2. AUXILIARIES AND MODALS 2.1. Auxiliaries The auxiliaries used in the present perfect, present conditional and future paradigms are distinct from their lexical verb counterparts, as follows. The present perfect (PR.PERF.) and conditional paradigms (COND) are historically related to - but distinct from - the lexical verb a avea 'to have', as in (21). Table 21 Auxiliary vs. lexical 'have' Pr.Perf. 'have'+ past particip.
Cond. 'have'+ bare infinitive
Lexical 'have' (transitive)
1. am cantat 'I have sung'
a§ canta 'I would sing'
am o carte 'I have a book'
2. ai cantat 'you have sung'
ai canta 'you would sing'
ai
3. a cantat 's/he has sung'
ar canta 's/he would sing'
are o carte 's/he has a book'
1. am cantat 'we have sung'
am canta 'we would sing'
avem o carte 'we have a book'
2. aji cantat 'you have sung'
aji canta 'you would sing'
aveji o carte 'you have a book'
3. au cantat 'they have sung'
ar
au o carte 'they have a book'
canta 'they would sing'
o carte 'you have a book'
Likewise, the Romanian future (PUT) auxiliary is distinct from, but diachronically related to, the lexical verb voi/vrea 'to want', which has two (possibly related) paradigms, as hi (22). Table 22 Auxiliary vs. lexical 'want' PUT 'want'+ bare infinitive
Lexical voi 'want'
Lexical vrea 'want'
1. voi canta 'I shall sing'
voi(esc) 'I want'
vreau 'I want'
2. vei canta 'you'll sing'
voi(e§ti) 'you want'
vrei 'you want'
3. va canta 'he/she'll sing'
voie§te 'he/she wants'
vrea 'he/she wants'
1. vom canta 'we will sing'
voim 'we want'
vrem 'we want'
2. veji canta 'you'll sing'
voiji 'you want'
vreji 'you want'
3. vor canta 'they'll sing'
voiesc 'they want'
vor 'they want'
The generative approach to Romanian syntax
15
The present perfect, conditional, and future auxiliaries in Romanian are lexical items inflected for subject Agreement (i.e. person and number), the use of which is restricted to the present perfect, conditional, and future constructions, respectively. The past paradigm consists of the present perfect, derived as in (22) and (23a), and of the past perfect, derived through a suffix -se, as in (23b), and unlike the complex 'had + past participle' in other Romance languages (23 c). (23)
a.
b.
c.
Nu 1- am vazut demult pe Ion. not him have-PR.PERF. 1 so seen of-long pe Ion 'I haven't seen Ion in a long tune.' fl vazusem peloncand venise in Winnipeg, him ses-se-lSGpe Ion when come-se-3SG in Winnipeg 'I had seen Ion when he had come to Winnipeg.' *fl aveam vazut pe Ion cand avea venit him have-PAST. ISG seenpe Ion when has-PAST.3so come 'I had seen Ion when he had come to Winnipeg.'
For future in the past and the past conditional, the perfective morpheme fi 'be' is used as in (24). (24)
a.
b.
Euvoi fi mancat. I will-FUT. ISG be eaten 'I will have eaten.' Eu a§ fi mancat. I have-coND. ISG be eaten 'I would have eaten.'
The presence of an auxiliary in Romanian does not block verb raising out of the VP, as is shown hi the comparative Romanian-French examples in (25), from Dobrovie-Sorin (1994). (25)
a.
b.
Elevii meivor (*toji)vedea (toti)filme bune students-the my will-FUT. 3PL all see-lNF all movies good 'All my students will see good movies.' Les enfants ont (tous) vu (*tous) de bons films, the children have all seen all of good movies 'The children have all seen good movies.'
On the assumption that floated quantifiers and certain adverbials mark the VP boundary (Pollock
16
Comparative studies in Romanian syntax
1989), the examples in (25a,b) show that the lexical verb raises out of the VP in Romanian even in the presence of the future auxiliary; (25c) shows the reverse effect for French. The Romanian present perfect and conditional auxiliaries behave identically to the future auxiliary in that they require the lexical verb to raise out of the VP. Obligatory verb raising in Romanian auxiliary constructions results in strict adjacency between the auxiliary and the lexical verb. The Romanian auxiliary cannot be separated from the lexical verb by any intervening material, unless clitics. The examples in (26) show that the subject DP cannot intervene between the auxiliary and the main verb, and that Romanian does not have SubjectAuxiliary Inversion (SAI). (26)
*A Razvan/el venit ieri? has-PR.PERF.3so Razvan/he come yesterday? 'Did Razvan come yesterday?'
There are languages (e.g. Dutch, German, Italian, French, etc.), in which the distinction between unaccusative and unergatrve predicates is manifested in the choice of the perfective auxiliary: unaccusative predicates select 'be', while unergatives select 'have' on a par with transitive verbs. In Romanian, the present perfect selects the same auxiliary (i.e., historically derived from 'have') for all verbs. Consider (27a) with a transitive verb, (27b) with an unaccusative, and (27c) with an unergatrve. (27)
a.
Au batut ei copilul. have-3PL beaten they child-the "They have beaten the child.'
b.
Au sosit toji musafirii. have-3PL arrived all guests-the
c.
'All the guests have arrived.' Au zburat toate vrabiile depe crengi have flown all sparrows-the from branches 'All the sparrows have flown from the branches.'
Romanian does, however, allow for structures with 'be' and an inflected past participle, as illustrated in (28), but these structures are semantically distinct from the present perfect constructions. (28)
Copiii sunt plecaji. children-the are left-3PL "The children are gone.' (i.e. are not here)
The generative approach to Romanian syntax
17
2.1.1. Analyses. Based on most of the distinctive features outlined above, Dobrovie-Sorin (1987, 1994) analyzes the Romanian auxiliary as a syntactic clitic. The Romanian clitic auxiliary is assumed to be base-generated under IP and to take as complement the IP to which the lexical verb raises. In view of the strict adjacency requirement, this complement IP will not project a SpecIP. Romanian auxiliaries are then captured under the definition in (29). (29)
Auxiliaries are verbs that: (i) do not assign any thematic role (not even external); (ii) govern a V" that lacks a SpecV"; (where V1 can be any projection headed by V, namely, I, C)
Dobrovie-Sorin (1987, 1994) analyzes the Romanian future construction, together with the perfect and conditional construction, as a monoclausal structure involving an auxiliary as defined in (29). The future auxiliary construction does not license a pronominal clitic in a small clause (sc)(contrary to the French 'futur-proche'), as shown in (30), so a biclausal analysis is excluded.
(30)
Ilj va
(*Uj)
him will-3SG (*him) phone 'He will call him. ' Romanian, however, has a periphrastic future, yielding two biclausal structures, as in (3 1). (31)
a.
Ion o [sa-i ceara ina§ina]. Ion has [SA-him-DAT ask- SUB J. 3 so car-the]. 'Ion will ask him for the car. '
b.
Ion are [sa-i ceara ma$ina]. Ion has [SA-him-DAT ask-SUBJ.3so car-the]. 'Ion will ask him for the car. '
In (3 1), both the reduced auxiliary o and the non-reduced are 'have' select a subjunctive clause (identified by the subjunctive morpheme sa), in which the embedded verb carries Agreement features and which disallows clitic climbing. It is argued (§tefanescu 1997) that the differences between the two structures are merely phonological. Avram (1998) proposes a unified analysis of the auxiliary and the main verb avea 'have'. The author argues that the different values of avea result from the different positions it occupies in the structure. The position of avea is determined by the complexity of the SC with which it merges in the derivation: the more complex the SC, the higher the position the auxiliary occupies in the structure.
18
Comparative studies in Romanian syntax
In particular, it is proposed that avea can be inserted: (i) under VP (then, it is a lexical verb), (ii) under TP (as a PR.PERF auxiliary), and (iii) under MoodjP - a projection higher than TenseP but lower than AgrSP (i.e. as a COND auxiliary). 2.1.2. The properties of fi 'be'. Dobrovie-Sorin (1994) discusses the properties of the perfective particle/? 'be': (i) its incompatibility with the present perfect, and (ii) its invariable nature. In view of its distribution, the author argues that./? is base-generated under I, as in (32).
(32)
In (32),/? forms a discontinuous morpheme with the past participle inflection under I. Since/? has the [perfect] features assigned at a pre-syntactic level, it may combine with other auxiliaries, as illustrated in (24). Moreover, since/? is part of a lexical constituent, it will be inaccessible to any rule of syntax, such as movement to Agreement, which explains why/? is invariant. In a different vein, Motapanyane (1995) proposes that perfective/? is base-generated as a V^ exclusively selected by T(ense) and does not raise to inflection at all in modern Romanian (although it moved to Agr in the older stages of the language). Romanian allows for a series of (mainly) unaccusative verbs to occur in structures of inflected/? + past participle, as in (33). (33)
Sunt mancata, pot sa rezist. am eaten can-1 so SA resist- SUB J. 1 so 'late, I can last.'
Avram (1998) argues that constructions like (33) do not have the value of a present perfect and that they have properties distinct from the structures which have the auxiliary avea 'have'. In her analysis,/? merges with a sc containing the past participle, and the subject DP of/? is assigned two theta-roles, similar to the control-pro structures in Manztni and Roussou (1997).
The generative approach to Romanian syntax
19
2.2. Modals Romanian modal verbs behave like then* Romance counterparts insofar as they display a complete inflectional paradigm. Some peculiarities arise with respect to Agreement marking and Tense restrictions. For example, trebuie 'must' lacks subject verb agreement morphology, untikeputea 'can', as shown in (34a, b). Also, epistemic modals, unlike deontic modals, seem to obey Tense restrictions (to present or imperfect (IMP)), as shown in (35), although they inflect as lexical verbs for all other forms. (34)
a.
Copiii/fata trebuie sa piece. kids-the/girl-the must-3PL/3so SA leave-SUBJ.3PL/3SG "The children/the girl must leave.' b. Copiii pot/ fata poate sa piece. children-the may-3PL / girl-the may-3SG SA leave-SUBJ "The children/the girl may leave.' (35) lonputea/ *a pututfi bolnavde nu s-a prezentat. Ion may-lMPERF/ has-PR.PERF could be ill since not REFL-has presented 'Ion may have been ill since he didn't show up.'
All modal verbs in Romanian subcategorize for subjunctive clauses, as mentioned for (34). However, putea 'can' also takes a bare infinitive complement, from which clitic climbing is obligatory, as in (36). (36)
lonle poate (*le) primi chiar acum. Ion them can them receive-lNF right now 'Ion can see them right now.'
Dobrovie-Sorin (1994) argues that modals, unlike auxiliaries, raise to Infl in Romanian^ because they combine with Tense features, precede adverbial clitics, and host ch'tic pronouns. Crucially, modal verbs in Romanian behave like lexical verbs, in contrast to auxiliaries, which have a clitic-like behaviour. In this framework, epistemic putea 'can' is defined as a raising verb, while the root/deontic modal involves a control configuration, as in other Romance languages. In clitic climbing structures, as (36), the bare infinitive lacks an Infl node and the chtic must climb to the matrix Infl. Motapanyane (1995) shows that putea 'can' undergoes verb restructuring with the "reduced" infinitive form, hi a manner similar to Rizzi's (1982) analysis of this construction in Italian. Avram (1998) argues that putea 'can' is a lexical verb associated with two parallel structures: a VP-
20
Comparative studies in Romanian syntax
complex, when it takes a bare infinitive, and a biclausal one, when it takes a sentential complement. The former is the result ofputea behaving like a "light verb" which merges with a VP, while the latter results from the merging ofputea with a subjunctive MoodPhrase (MP). This analysis argues against the view that deontic modals enter control structures, as opposed to epistemic modals which are raising verbs (cf. Gueron and Hoekstra 1987 and Dobrovie-Sorin 1994 for Romanian), since both deontic and epistemic putea may freely alternate the VP/MP complements.
3. THE CONFIGURATION OF ROOT CLAUSES 3.1. Declaratives with Indicative Verbs Romanian declarative clauses have a relatively free word order that allows for all the sequences in (37), equally translatable as 'John has eaten the apple pie'. (37) a. A mancat Ion placinta cu mere. has eaten Ion pie-the with apple b. Ion a mancat placinta cu mere. Ion has eaten pie-the with apple c. A mancat placinta cu mere Ion. has eaten pie-the with apple Ion d. Ion {placinta cu mere\ a mancat-Oj. Ion pie-the with apple-FOC has eaten- it e. [Placinta cu mere\, Ion a mancat-Oj. pie-the with apple-TOP Ion has eaten- it
VSO SVO VOS SOV OSV
An object DP in pre-verbal position heads a chain with a resumptive pronoun whenever it is definite, as in (37d, e). Quantified DPs can also surface pre-verbally, as Focus, as in (38a), and even as Topic, when they have a strong reading, as hi (38b). These DPs do not relate to resumptive pronouns: (38) a. Ion un film a vazut. (S)OV Ion a movie-FOC has seen 'It is a movie that Ion has seen (not something else).' b. (*Ni§te) filme, Ion a vazut destule. O(S)V some movies-TOP Ion has seen enough. 'As for movies, Ion has seen enoguh.'
The generative approach to Romanian syntax
21
The number of pre-verbal constituents in declarative clauses is theoretically unlimited. Consider the examples in (39), in which pre-verbal sentential adverbs, VP adverbs, prepositional phrases, subjects and objects appear. In (39b) the subject is non-lexical and realized nspro, since Romanian is a null subject language. (39)
a.
b.
Adesea in vacanjele de vara elevii primeau teme de la iscoala. often in holidays-the of summer students-the received homework of at school 'Students would often get homework from school during the summer holidays.' leri in pare copiilor le- a dat cate o inghe{ata. yesterday in park children-the them-has given each an ice-cream 'He/she bought the children each an ice cream in the park yesterday.'
In Romanian declarative sentences, negation takes the form of a free morpheme nu ('not') which obligatorily precedes the finite verb. There is an adjacency requirement between nu and the verb, obligatorily circumverted only by clitic-like elements (auxiliaries, pronominal clitics and adverbial intensifiers such as mai 'more', prea 'too, very', tot 'still', as in (40). (40)
a.
b.
c.
d.
Ion maine nu canta la pian. Ion tomorrow not plays at piano 'Ion won't be playing the piano tomorrow.' Nu-mi mai place (*imimai) culoarea galben. not-me more like-3SG me more colour-the yellow 'I don't like the yellow color any more.' Mama nu iar mai tot certa daca ar fi linistiti. mother-the not them-would more continuously scold if would be well behaved 'Their mother wouldn't be scolding them all the time if they were well behaved.' Profesoara - ar (*mereu) fi mereu in §coala daca n - ar avea de lucru. teacher-the not-would always be always in school if not-would have of work 'The teacher wouldn't always be in the school if she wouldn't have work to do.'
In (40a), the negative morpheme and the verb are adjacent. In (40b), a clitic pronoun and an adverbial intensifier intervene between negation and the verb, on a par with the conditional auxiliary structure in (40c). Clitic-like elements are not allowed post-verbally or hi any other position, as shown in (40b). So clitic adverbs are in complementary distribution with full fledged adverbs like mereu 'always' in (40d). The word order in the sequence between the negative morpheme nu and the lexical verb, shown hi (41a)/(42a), is extremely rigid. Dobrovie-Sorin (1987, 1994) notices that only the presence of an
22
Comparative studies in Romanian syntax
auxiliary modifies this fixed linear order, yielding (41b)/(42b). (41)
a. b. (42) a.
b.
Neg - Pron - Adv (short) - V Neg - Pron - Aux - Adv (short) -fi - V Nu le- o mai prea citesc zilele astea. not them it more very read days-the these 'I do no more read it to them these days.' loana nu i 1- ar mai fi dat azi. loana not him it-would-3SG. more be given today 'Today, loana wouldn't have given it to him.'
3.1.1. The split IP hypothesis. The empirical data presented above have been analyzed hi a number of ways, the general consensus being that the lexical verb always raises out of the VP to a functional head (Cornilescu present volume, Dobrovie-Sorin 1994, Motapanyane 1995, Stefanescu 1997). Support for such an analysis is taken from the position of VP adjoined adverbs and floating quantifiers with respect to the lexical verb, as in (43), taken from Dobrovie-Sorin (1994). (43)
a.
b.
Elevii meivad des filmebune. students-the my see-3PL often films good 'My students often see good movies.' Elevii mei citesc to\j o poezie de Verlaine. students-the my read-3PL all a poem by Verlaine 'All my students read a poem by Verlaine.'
Under the assumption that certain adverbs and floating quantifiers are generated immediately in front of the VP (Pollock 1989, Sportiche 1988), the Romanian data in (43) shows obligatory raising of the lexical verb, above des and to\i. Several proposals have been made with respect to the landing site of the raised verb and the functional projections relevant to the build-up of the Romanian Inflectional Phrase (IP). DobrovieSorin (1987, 1994), Rivero (1994), and Cornilescu (present volume) argue for verb raising to the highest functional node within the root IP, irrespective of whether an auxiliary is present or not. That is, V moves to an Inflectional head (I) specified for Tense/Agreement in Dobrovie-Sorin (1994); to Agreement (Agr) in Rivero (1994), and to a Mood head available in root and embedded clauses in Cornilescu (present volume). Dobrovie-Sorin (1994) claims that Romanian does not present clear evidence for stipulating separated T and Agr projections (TP/AgrP), so that Agr is viewed as an affix, nominal in nature,
The generative approach to Romanian syntax
23
which adjoins to T, under I. Verb raising takes place to I. The Romanian auxiliary is base-generated fully inflected and cliticized to CP/IP, a projection which dominates IP; thus, auxiliaries do not interfere with verb raising, as shown in (44). (44) CP/IP Aux
"^
CP/IP
On the other hand, Comilescu (present volume) argues that there is both morphological and syntactic evidence for a split IP hi Romanian that accommodates at least the verbal categories of Mood, Tense, and Aspect, hierarchically ordered as hi (45). (45) MoodP > AgrSP > TP > AspP The author argues that the indicative and the subjunctive pattern identically vis-a-vis subjects and adjuncts, which leads to the assumption that the verb raises overtly to M hi all Romanian clauses.
Motapanyane (1995) and §tefanescu (1997) argue that a distinction should be kept between structures involving an auxiliary and structures involving simple tenses. In simple tenses, the finite verb is assumed by both authors to raise to the highest functional head. This is the AgreementSubject (Agrs) head in Motapanyane and the Person head hi §tefanescu . In complex tenses, Motapanyane (1995, 28) analyzes the auxiliary avea as devoid of Tense features (in the spirit of Dobrovie-Sorin 1994), while carrying Agreement features. So Agr selects the auxiliary that raises to Agrs to check the Agreement features, while the lexical verb raises only as high as AgreementObject (Agr0) or T. The adjacency requirement between the auxiliary and the lexical verb is captured in this analysis by positing that subjects can never surface hi SpecTP, a constraint attributed to the EPP. Motapanyane's exploded IP is reminiscent of Belletti (1990), with Agrs being the highest functional head hi the Romanian declarative clausal architecture, as shown in (46).
24
Comparative studies in Romanian syntax
(46)AgrsP
Stefinescu (1997) adopts Motapanyane's analysis in a configuration where Agrs is further split into Person and Number heads. On the basis of dialectal and diachronic evidence, she places the auxiliary in the Number head and shows that it moves to the Person head to check the Agreement features. The lexical verb will only raise as far as T. To capture the obligatory auxiliary verb adjacency, she retains Dobrovie-Sorin's (1994) suggestion that not all the functional heads have Specs and stipulates the necessary assumption that the NumberP does not project a Spec position. 3.1.2. Subject positions. VS(O) sentences are unmarked and highly productive in Romanian and, in contrast to Italian and Spanish, they are not restricted to tensed clauses. Post-verbal subjects have been shown to surface in their base-generated position; that is, SpecVP (Dobrovie-Sorin 1987, 1994, Motapanyane 1989, 1995, §tefanescu 1997, Cornilescu present volume). In the preMinimalist framework, in situ subjects are analyzed as acquiring nominative Case under government by the verb that has raised to I (Dobrovie-Sorin 1994, Motapanyane 1989). Later studies adopt a checking analysis in which SpecVP cannot be viewed as a Case position: Motapanyane (1995) argues that nominative Case checking for post-verbal subjects takes place at LF in SpecTP, a position unavailable to subjects in the overt syntax.
Cornilescu (present volume) offers two analyses for in situ lexical subjects: they may either
The generative approach to Romanian syntax
25
raise at LF to a subject-Case position (which for the author is SpecAgrsP in the default Case) or they can acquire Case by coindexation with pro, an expletive licensed by the Romanian Agr head. As exemplified in section 3.1, the Romanian subject can also surface pre-verbally. Some controversy surrounds the landing site as well as the status of the subject position in this context. Dobrovie-Sorin (1994) argues that the pre-verbal subject receives nominative Case in SpecVP and moves to SpecIP which is a Topic position (a non-argumental/non L-related position). In Dobrovie-Sorin's account, this is the position which also hosts topicalized elements such as object DPs and adverbials. Motapanyane (1994, 1995) argues against a non-argumental pre-verbal subject position and keeps distinct the position occupied by the fronted subject DP and those of the topicalized and focused elements. On the basis of tests with bare quantifiers in subject position and NP-movement in raising constructions, the author shows that movement of the subject DP to the pre-verbal position creates A-chains which do not interfere, in the case of bare quantifiers, with the parallel A'-movement to pre-verbal Focus. The pre-verbal subject lands in the highest Spec of the functional projection hosting the raised finite verb; this is an L-related, argumental position, located between Topic and Focus.
"^
Cornilescu (present volume) also discusses the pre-verbal subject position and concludes that it is not L-related (in the spirit of Dobrovie-Sorin 1994). Unlike Dobrovie-Sorin (1994) or Motapanyane (1994), she argues that there are two argumental postverbal subject positions in Romanian, both available at Spell Out; that is, SpecAgrsP and SpecVP. Under the assumption that weak pronouns and clitics have to occupy their Case-checking position before Spell Out (see Cardinaletti 1996), subject clitics will indicate the Agrs and the nominative Case position in a Romanian clause as in (47): the pronouns, which usually behave as full DPs (see section 1, Table 11) receive a clitic definition in a post-verbal position: (47)
a.
(*El)vine(e/)tata.1 he comes he dad
'It must be noted that the unmarked word order in (47) is VS, as in (i). (i) Vine tata/el. comes father/he All the other versions entail specific intonation, with a Focus reading on the verb in (47a), or a Topic reading on tata'dad' and a Focus reading on the verb in (47b). The pre-verbal pronoun in (47a) yields a grammatical sentence with intonational breaks between the verb and the post-verbal tata 'dad' (e.g., El vine, /a/a, a*ar cine-lprime$(e?rDad, he conies, but who's receiving him?')
26
Comparative studies in Romanian syntax
b.
'Dad will be coming!' Tata, vine el. dad comes he
Since the subject clitic in (11) can only occupy SpecAgrsP as opposed to SpecVP, it suggests the existence of two post-verbal argumental subject positions in Romanian. 3.1.3. Topic and Focus. Section 3.1. showed that the left periphery of the Romanian declarative clause allows for a number of word order sequences provided the fronted DPs can acquire the required interpretation (i.e. Topic, Focus, etc.). We briefly mention below some of the analyses available to interpret these empirical facts. Following Cinque (1990), Dobrovie-Sorin (1990, 1994) discusses left dislocation structures in Romanian. She maintains Cinque's distinction between Clitic Left Dislocation (CLLD) in Romance and the English left dislocation or topicalization (ELD), but diverges from Cinque's analysis in two respects: first, she shows that both CLLD and ELD occur in Romanian, as shown in (48a, b); second, she argues that CLLD involves movement, as opposed to base-generation in a SpecCP or a position adjoined to CP/IP. (48)
a.
b.
Pe Ion; lj- am intalnit (*pe el;) anul trecut. pe Ion-TOP him-have-1 so met pe him year last 'I met Ion last year' (Cat despre) lorij, (pe elj) nu \- am vazut de anul trecut. as for Ion-TOP pe him not him-have-1 so seen of year last '(As for) Ion, I haven't seen him since last year'
The author notices the following different syntactic properties: The left dislocated element of CLLD, as hi (48a), can be of any maximal category, and there is no theoretical limit to the number of dislocated elements to this pre-verbal segment. Moreover, in these structures the sentence-internal element can only be a clitic, emphatic pronouns being ruled out, and there is a 'connectivity' effect. On the other hand, ELDs, as in (48b), essentially restrict left-dislocation to DPs, and allow only one constituent at a time. With ELDs, an emphatic pronoun is grammatical and there is no 'connectivity' eflfect. The important conclusion is that, while Romanian ELDs do not rely on movement, CLLD structures do. With respect to the landing she for CLLD, Motapanyane (1994, 1995) argues for a clear hierarchy for Topic and Focus, on the basis of tests with the interrogative morpheme oare, base-generated in C (see Appendix to Motapanyane, present volume). In (49), the constituent preceding oare is in
The generative approach to Romanian syntax
27
Topic, whereas the constituent following oare holds a Focus position. (49)
Scrisorile, oare ieri le- a primit Ion? (sau azi) letters-the-TOP Q yesterday-FOC them has received Ion (or today) 'As for the letters, was it yesterday that Ion received them, or today?'
*&*
Motapanyane (present volume) reinterprets the analyses of earlier studies in a Minimalist framework. The author points out that [Focus], as a non-categorial feature, cannot be recognized by the computational system unless it combines with semantically related formal features, in particular, [wh] and [Tense]. This hypothesis leads to a parametrization of Focus: (i) [Focus/wh] (as in English); (ii) [Focus/Tense] (as in Romanian). Since the syntactic manifestation of [Focus] depends on [Tense] in Romanian, [Focus/Tense] merges in T and leads to the projection of two SpecTP positions: one to check the [D]/[EPP] features of T, and one to check the [Focus] feature of T.
Other authors have argued for the existence of distinct functional projections in the Romanian clausal architecture, triggered by Topic and Focus. Comilescu (present volume - following Rizzi 1997) argues for the existence of a Topic-Focus articulation in the CP segment of Romanian declarative sentences. Constituents endowed with Topic or Focus features end up in a Spec-head configuration with Topic or Focus through movement to the pre-verbal "initial" field. Alboiu (1998) distinguishes between two types of Focus in Romanian which trigger two distinct types of DP raising: (i) Focus which involves sentential scope-taking and is related to the pre-verbal segment (i.e. above IP); and (ii) a rhematic type of Focus associated to the VP, the domain of the Romanian rheme (i.e. of elements that represent new/non-presupposed information). For the first type of Focus, the author follows Kiss (1995) in assuming a Focus head in the left-periphery of the clause which attracts constituents with [+Focus] features. For the second type of Focus, the author argues - primarily based on the reversal of binding interactions - that VOS constructions in Romanian involve object shift across and above the subject DP. In this analysis, the in situ subject, together with whatever else remains within the VP, acquires maximal rhematic prominence as a result of object raising to SpecAgToP; the raised object will acquire a presuppositional reading, irrespective of the initial semantic type. 3.1.4. Long Head Movement. Romanian shares with the Balkan languages and earlier stages of some Romance languages the property of non-finite verb raising (i.e., participial and infinitival forms) to a position across the inflected auxiliary. This type of construction, defined as Long Head Movement by Rivero (1989, etc.), is optionally present in the Romanian perfect, future and conditional auxiliary constructions, as in (50).
28
Comparative studies in Romanian syntax
(50)
a.
Vedea-voi atunci ce voi face! see-INF-will-1 so then what will-1 so do-iNF
b.
*Vedea-voi atunci ce face- voi! see-INF-will-1 so then what do-INF-will-1 so
c.
Voi vedea atunci ce voi face! will-1 so see-INF then what will-1 so do-INF TU see what I'll do then.'
LHM is restricted to main clauses, as shown by its unavailability in the embedded clause hi (50). This observation allows Rivero (1994) to equate LHM with head movement of the non-finite verb to C. However, unlike V-to-C in Germanic, LHM strands pronominal clitics (Dobrovie-Sorin 1994). In contrast to Slavic, where it is obligatory in those languages that license it, LHM is optional in Romanian, and it is not semantically equivalent to the structure in which the auxiliary has not been by-passed. LHM in Romanian is associated with some form of illocutionary force and never with a neutral statement. In view of this, Rivero (1994) argues that Romanian has "residual" LHM, just like English has "residual" verb-second (as discussed hi Rizzi 1991).
3.2. Interrogatives hi Romanian, 'yes-no' interrogatives allow for a relatively free word order, to the exclusion of SAI, as in (51). (51)
(lon)va (*Ion)veni (Ion) maine? Ion will-3sG ion come-ENF Ion tomorrow
Consider next the permissible word order sequences in Romanian wh-interrogatives in (52). (52)
(Ion) cand (*Ion) va (*Ion) veni (Ion)? Ion when Ion will Ion come Ion
Wh-interrogatives display the subject either in a Topic position, above the wh-phrase, or in a postverbal position, disallowing pre-auxiliary subjects and SAI. Dobrovie-Sorin (1994) and Cornilescu (present volume) argue that the finite verb does not raise to C in Romanian interrogatives since there are no word order asymmetries between declarative and interrogative sentences. Hence, the absence of SAI. However, wh-movement to SpecCP prevents pre-verbal subjects. 3.2.1. The structural quantifier hypothesis.
Dobrovie-Sorin's (1990, 1994) analysis of wh-
The generative approach to Romanian syntax
29
interrogatives relies on the formulation of the structural quantifier parameter. Unlike English, Romanian has a negative setting for this parameter, which entails that SpecCP is compatible with [-quantified] constituents as well. Thus, a wh-phrase moving to SpecCP may create two types of chains: an operator-variable chain, as in (53a), or a chain with clitic doubling, as in (53b). (53)
a.
b.
Ca{i
elevij
a
examinat profesorul tj?
how-many students has examined professor-the 'How many students did the teacher examine?' Pe caji elevi; ij - a examinat profesorul tj? pe how-many students them has examined teacher-the
For a presentation of the structural quantifier parameter in relation to clitic doubling and relatives we refer the reader to section 1.6.2. and section 5, respectively. 3.2.2. Multiple wh-interrogatives. Romanian is a language with multiple wh-fronting (see Rudin 1988, PesetsM 1987), as illustrated in (54a); (54b) shows that multiple wh-fronting is compulsory. (54)
a.
b.
Cinej C6j [ 6j spusese 6j ]]? who what said 'Who had said what?' [Cinej (despre C6j) [crezi ca 6; mia povestit e/(*despre ce)]]? who about what think-2SG that me-DAT has told about what 'Who do you think told me about what?'
For (54), Comorovski (1996) considers Romanian to be a language that allows reiteration of CP, yielding the structure in (55).
spusese t^ Rudin (1988) shows that multiple wh-interrogatives in Romanian and Bulgarian share a number of
30
Comparative studies in Romanian syntax
properties which differ systematically from the multiple wh-constructions in Slavic languages. Kraskow (1990) defines these properties as follows: (i) long multiple wh-movement from thatcomplements; (ii) wh-island extraction; (iii) no lexical material (clitics, adverbs, parentheticals or particles) may separate the two fronted wh-phrases; (iv) the fronted wh-phrases observe a fixed order (subject first). With respect to the last property, also known as the Superiority Effect, Comorovski offers a discourse-linked (D-linked) analysis.
3.3. Imperatives A distinction is made in Romanian between "true" and "surrogate/suppletrve" imperatives (see Isac 1994, Motapanyane 1995, Rivero 1994). The true imperative in Romanian is restricted to the affirmative, second person singular and is identifiable by morphology not shared by the same person in any other tense/mood in the system, as shown in (56). Pronominal clitics follow the imperative and phonologically encliticize to it. In true imperatives, an overt subject is ruled out, unless it is phonologically accented and contrastively focused: (56)
Sunama (*tu)/ tu (nu altcineva)! call-lMP.2sG-me you/you-FOC not someone else 'Give me a call/YOU give me a call!'
Surrogate imperatives are not restricted to person and number; they can be negated and resort to borrowed forms, which include: (i) the infinitive, restricted to second person singular negative contexts, as in (57a); (ii) the subjunctive, in affirmative and negative contexts, for all persons, as in (57b); and (iii) the indicative, restricted to second person, as in (57c). (57)
a.
b.
c.
Nute spala cu apa rece! not you wash-IMF with water cold 'Don't wash yourself with cold water!' Sa (nu) piece (copiii)! SA not leave-SUBJ.3PL kids-the 'They/the kids may (not) leave!' Nuva mai certaji (voi)! not you more argue-IND.2PL you 'You two/guys stop arguing!'
In surrogate imperatives, as in (57), the pronominal clitic precedes the lexical verb, irrespective of whether an infinitive, a subjunctive or an indicative form is used. A lexical subject is licit in the
The generative approach to Romanian syntax
31
subjunctive and in indicative forms. hi affirmative surrogate imperatives, the subjunctive mood marker sa can be absent, in which case the pronominal clitic encliticizes to the verb, as in (58). (58)
a. b.
Slaveasca-1 pe Dumnezeu! worship- SUBJ.3PL-himpe God *fl slaveasca pe Dumnezeu! him worship-SUBJ.3PL/?e God 'May they worship God!'
These structures have been analyzed (in Dobrovie-Sorin 1994, Rivero 1994) as involving verb raising or proposing to a position higher than T/Agr, to which the verb usually raises out of the VP. The ungrammatical example in (58) would arise from the fact that the verb did not move above T/Agr. 3.3.1. Internal structure . Rivero (1994) argues that imperative movement is head-to-head raising to C: V moves to T/Agr for phi-features, then to C, where it is attracted by the imperative affix. This analysis is supported by the data in Dobrovie-Sorin (1994) and Isac (1994). The former study also emphasizes that 'Verb proposing" in Romanian imperatives strands the clitics, so it is distinct from verb-second in Germanic, where clitic are taken along with V-to-I-to-C. 3.3.2. (In)transitivity.
*k®
Pirvulescu and Roberge (present volume) offer a detailed analysis of the verbal morphology in Romanian imperatives and consider its relevance for the syntax of imperative constructions. The authors notice that the alternation -e/-i in the ending of a certain class of imperatives (Q.%.,fierbe/fierbi 'boil') correlates with the transitivity value of the verb, as in (59). (59)
a.
b.
Fierbe macaroanele! boil-IMP.2SG pasta-the 'Boil the pasta!' Fierbi acum, c- a§a-ti trebuie! boil-IMP.2sG now that so you-DAT needs 'Boil now (in your own sauce), you deserve it.'
32
Comparative studies in Romanian syntax The contrast between the transitive (59a) and the intransitive (59b) values of the same verb is related to the composition of the functional domain hi imperative clauses: in the absence of Tense, imperative constructions force the projection of a position for the internal argument, which is reflected hi the morphological alternation in (59).
4. SENTENTIAL COMPLEMENTS 4.1. Embedded Indicative CP 4.1.1. Word order. Root and embedded indicative clauses present a similar "free" word order, in which constituents may cooccur in the pre-verbal segment (as Topics or Focus), as in (60a). The lexical subject may surface at three different points in the linear order, as in (60b). The complementizer ca 'that' marks the head C: (60)
a.
b.
Nu cred [ca ieri in pare copiilor tata le- a dat inghejata]. not think-ISG that yesterday in park kids-the father them-has given ice cream 'I don't think that dad bought the kids ice cream hi the park yesterday.' Nu §tiam [ca (Mircea) i- a dat (Mircea) un inel (Mircea)]. not knew-ISG that Mircea her-has given Mircea a ring Mircea 'I didn't know that Mircea gave her a ring.'
Note that ca 'that' is obligatory hi indicative CPs, as shown hi (61), a property which Romanian shares with other Romance languages. (61)
loana §tia [*(ca) va lua examenul]. loana knew-3so that will-3SG take exam-the 'loana knew she would pass the exam.'
The properties of embedded declaratives such as those hi (60a, b) have led researchers to consider that the internal structure of these clauses is identical to the structure of root declaratives. Since a separate discussion would be irrelevant, we refer the reader to section 3 for details. 4.1.2. Extractions. The obligatory presence of the complementizer ca 'that' is likely to interfere with extraction phenomena by triggering the "that-trace" effect discussed hi Chomsky (1986) for English, and hi Rizzi (1990), for English and Romance. However, any element can be extracted across ca 'that' in Romanian: hi (62a) the extraction originates hi the subject positions, hi (62b), hi the object position.
The generative approach to Romanian syntax (62)
a.
b.
33
Cinej crezi [tj' ca [va capata tj postul]]? who think-2so that will-3SG get job-the 'Who do you think will get the job?' Ce postj crezi [tj' ca [va capata Ion tj]? what job think-2so that will-3SG get Ion 'Which job do you think (that) Ion will get?'
The grammaticality of (62) falls under the analyses in Rizzi (1982, 1990), where the lack of "thattrace" effect is related to the null subject parameter in Romance languages. Focused and topicalised elements can also be extracted across ca. This possibility is expected, since movement to Focus and Topic involves a form of wh-movement (realized through operator-variable chains or clitic doubling chains), and observes the same constraints as the structures in (62). For arguments towards a whmovement analysis of fronting to Focus and/or CLLD see Alboiu (1998), Dobrovie-Sorin (1994), Motapanyane (present volume).
4.2. Subjunctive Clauses Subjunctive clauses occur as complements, adjuncts and root clauses (the latter, as surrogate imperatives, see section 3.3). As complements, they have replaced the infinitives almost completely. The subjunctive morphology in Romanian shares properties with both Romance and Balkan languages (see Dobrovie-Sorin 1994, Farkas 1982, 1984, 1985, Kempchinsky 1986, Motapanyane 1995, Terzi 1992). Romance languages mark the subjunctive mood synthetically (i.e., on the verb stem), while Balkan languages mark it analytically (i.e., using a free morpheme). In Romanian, the subjunctive is marked on the verb stem in the third person, while in the first and second persons the verb form is indistinct from that of the indicative. In all cases, the invariable subjunctive morpheme sa precedes the verb, as in (63). (63)
Sper [sa piece maine la Londra]. hope- ISG SA leave-3SG tomorrow at London 'I hope she will leave for London tomorrow'
The word order sequence between the subjunctive particle sa and the verb is rigid: the negation nu 'not' follows the subjunctive morpheme and precedes the pronominal clitics, adverbial intensifiers and the auxiliary fi (when present), as in (64a, b). Beside clitic-like elements, nothing can disrupt this rigid word order: (64)
a.
Ar
fi
dorit [sa nu-i
mai fi dat
ve§tiproaste.]
34
Comparative studies in Romanian syntax
b.
would be wished SA not him more be given news bad 'He wished he hadn't given him any more bad news.' SA - Neg - Pron - Adv (intensifier-type) -fi-V
No lexical material can precede the embedded sa, as shown in (65a), except for heavily stressed focused constituents, as in (65b). (65)
a.
b.
A spus [(*florile, de la piaja, loana) sa le trimita direct la birou.] has said flowers-the from at market loana SA them send-Sso straight to office 'He said, from the market, loana must send the flowers straight to the office.' ?A spus televizor sa nu cumpere. has said tv-set-FOC SA not buys 'It was a tv-set he had said not to buy.'
Within the Balkan group, Romanian presents the peculiarity of having a subjunctive complementizer ca 'that' (different from indicative ca 'that'), so that ca-subjunctives may occur, hi certain contexts, in free alternation with the .sa- subjunctive clauses. In the presence ofca, the word order in the subjunctive complement is similar to the word order in the indicative complements, as shown in (66a). In general, ca may not be adjacent to sa, although some dialects accept this sequence, as hi (66b). In adjunct contexts, the sequence ca sa may occur freely, as hi (66c). (66)
a.
b.
c.
A spus [ca, florile, loana sa le trimita la birou.] has said that flowers-the loana SA them send-3so to office 'He said that, the flowers, Joan should send them to the office' ?Sper [ca sa piece.] hope-lSG that SA leave-3so 'I hope he leaves.' Au intrat incet [(ca) sa nu ne trezeasca.] have-3PL entered slowly that SA not us wake-3PL "They came in slowly not to wake us up.'
There are certain contexts hi which ca is obligatorily excluded, as shown hi (67), namely: relative subjunctives (67a), wh-subjunctives (67b), constructions with raising verbs (67c) and control structures (subject control, hi (67d) and object control in (67e)). (67)
a.
Caut o fata [care (*ca) sa §tie engleze§te]. search-ISG a girl which that SA knows English 'I'm looking for a girl that can speak English.'
The generative approach to Romanian syntax b.
c.
d.
e.
35
§tiu [cum(*ca)sa scriu cu stiloul]. know-1 so how that SA write-1SG with p en-the 'I know how to write with a fountain pen.' Totibaietiij sau nimerit [(*ca) sa fie bolnavitj. all boys-the REFL have-3PL happened that SA were sick 'All the boys happened to be sick.' Mioaraj a inceput [(*ca) sa S6j pregateasca de plecare]. Mioara has started that SA REFL prepare-3SG of departure 'Mioara started to prepare her departure.' Maria lj- a convins pe lonj [(*ca) sa cumpere elj televizorul.] Mary him has persuaded pe-Ion that SAbuys hetv-set-the 'Mary persuaded John that he should buy the tv-set.'
Exclusion of ca in (67a, b) is attributed to the "doubly filled Camp" filter (Kempchinsky 1986). Verb raising constructions, as in (67c), exclude lexical complementizers because of more general wellformedness conditions on argumental chains (Motapanyane 1995). For control structures, as in (67d, e), a question arises concerning the nature of the controlled subject. Dobrovie-Sorin (1994) and Motapanyane (1995) define the null subject as pro, licensed by the verb-subject agreement. However, Terzi (1992) points out that ca is excluded in these structures, a restriction which resembles a PRO theorem violation. Thus, Terzi, following Kempchinsky (1986), proposes a uniform treatment of subjects in subjunctive and infinitive control structures, as PRO categories. To summarize their distribution, ca and ca-less subjunctives may alternate in complement and adjunct positions (66), but not in relatives/wh-subjunctives, control and raising structures (67), which exclude ca. 4.2.1. Internal structure. Discussions that relate to the internal structure of the Romanian subjunctive clause have been numerous and sometimes conflicting. Issues relating to the status of the subjunctive morpheme sa, to the type and number of functional projections, to subject obviation, and to raising out of subjunctive clauses have all been tackled hi extensive studies (especially Dobrovie-Sorin 1994, Farkas 1982, 1984, 1985a, Grosu and Horvath 1984, Kempchinsky 1986, Motapanyane 1995, Rivero 1989, 1994, Terzi 1992, etc.). Dobrovie-Sorin (1991, 1994) defines sa as an ambiguous category, which has both inflectional and complementizer properties. She argues that although sa is base-generated hi C, it is indistinguishable from inflection. This is possible because the adjacent C and I undergo a mechanism of "functional incorporation" in these contexts. Crucially, Dobrovie-Sorin (following Haider 1988) considers the sa-subjunctives as "merged projections" (i.e. CP/IP), which suppress pre-verbal subjects, so all the
36
Comparative studies in Romanian syntax
material to the left of sa undergoes left-dislocation. Dobrovie-Sorin's (1994) view is challenged by a number of authors (e.g. Motapanyane 1995, Terzi 1992, etc.) which argue for an exclusive inflectional status for sa, while still accounting for the Clike properties of the subjunctive morpheme. Terzi (1992) also offers cross-linguistic evidence pointing to the inflectional status of the subjunctive marker sa. Following work by Rivero (1994) on Balkan languages, both authors argue that sa is base-generated in the Mood (M) head, whereas the subjunctive ca, unambiguously a complementizer, is base-generated in C (cf. Cornilescu, present volume). 4.2.2. Subject obviation. Dobrovie-Sorin (1991, 1994), Farkas(1985b), Kempchinsky( 1986), and Terzi (1992), among others, discuss the obviation effect which arises in Romance with certain matrix verbs, in particular, volition verbs, when they select a subjunctive complement: the subject of the subjunctive clause shows obligatory disjoint reference (hence the term Subject Disjoint Reference effect - SDR) from the matrix subject, as shown in (68). (68)
Anaj quiere que [pro],^ venga con nosotros. Ana wants that he come with us
(Spanish)
Kempchinsky (1986) accounts for the SDR effect in Romance as follows: the subjunctive operator, situated in CP, triggers V-to-I-to-C, as a means to check on the matrix verb's selectional properties. This movement entails an extension of the Binding domain to the matrix clause, so that a pronominal in subject position in the subjunctive clause would be wrongly bound if it were coreferent with an argument in the matrix, as in (68). However, this configuration does not arise in Romanian, where the subjunctive operator is identified by the particle sa, versus V-to-I-to-C. Lack of V-to-I-to-C entails separate Binding domains for the matrix and the embedded clause, which explains the lack of SDR effects in subjunctive complements in this language, as illustrated in (69). (69)
Mihaij vrea [(ca maine) pro,^ sa piece la Bra§ov]. Mihai wants (that tomorrow) SA leave-SUBJ.SSG at Brasov' 'Mihai wants to leave for Brasov tomorrow.'
Dobrovie-Sorin (1994), on the other hand, argues against a separate Binding domain for Romanian subjunctive complements: she derives the lack of SDR effects from the fact that sa- subjunctives are accessible to anaphoric binding. In these contexts, pro has [+ /- anaphoric] properties, qualifying as a "contextual anaphor": in complements of control verbs, pro is an anaphor, while in complements of volitional verbs, pro is a pronominal.
The generative approach to Romanian syntax
37
Farkas (1985b) offers a solution that accounts not only for the Romance and Romanian data, but also for the obviation idiosyncrasies in Hungarian and Serbo-Croatian. After observing that subjunctive clauses are obviative only in contexts in which an infinitive is also possible, the author views obviation as one complement form (the infinitive) blocking another (the subjunctive). Romanian does not have two complement types, with only the subjunctive being available; therefore the blocking mechanism is neutralized, which entails lack of SDR effects. Terzi (1992) argues that the absence of subject obviation among Balkan languages (including Romanian) cannot be accounted for by the concurrent lack of infinitival constructions. She relates the absence of SDR to the structural ambiguity characteristic of subjunctives subcategorized for by volitional verbs which involve two structures, as illustrated in (70). (70)
a. b.
loiij vrea [CP [c s& manancej [w [pro]; [M tj...] Ion wants SA eat-sUBJ.3sG Ion; vrea [CP [c [MP PRO; [M sa manance ] Ion wants SA eat-suBJ.3so 'John wants to eat.'
The absence of a lexical complementizer in (70b) legitimizes the presence of PRO which requires coreference of matrix and embedded subject. In languages where a lexical complementizer is present (such as Romance, excluding Romanian), PRO is impossible (since it would be governed, contrary to its definition) and obviation cannot be concealed. 4.2.3. Raising constructions. Grosu and Horvath (1984) show that raising verbs induce NPmovement from subjunctive complements when ca/ca 'that' is deleted, as shown in (71). (71)
a.
b.
Separe [ca totibaietii sunt bolnavi]. it seems-lND.3sG that all boys-the are-!ND.3PL sick 'It seems that all the boys are sick.' Baietii par [(*ca).sa fie toti bolnavi]. boys-the seem-IND.3PL that SA be-IND.3PL all sick 'All the boys seem to be sick.'
NP-movement is grammatical in (71b) because the subject trace is governed across the S(=IP) boundary, when C is deleted. In other Romance languages, the complementizer cannot be deleted in subjunctive clauses so that raising of the (71b) type never applies. Rivero (1989), on the other hand, maintains that subjunctive clauses in Romance are uniformly of
38
Comparative studies in Romanian syntax
the CP-type. In Romanian, in particular, these complements are neither reduced nor restructured. Accessibility to the subject position is obtained through an agreement relation between C/I which voids the CP[C pfDP I VP]] sequence of its potential barrierhood with respect to DP. Motapanyane (1994) shows, however, that subject verb agreement is practically unobtainable across ca in structures like (71b). In her analysis, NP-movement is triggered by the Extended Projection Principle applying to the matrix clause (versus Case checking purposes), and creates an argumental chain of the form [NP( [pro, sa-V tj ]].
4.3. Infinitives Romanian infinitives are analytical, consisting of a mood marker a 'to' and an infinitive stem, as in (72a). In old Romanian, the infinitive morphology presented the mood marker a and an infinitive suffix -re(a) on the stem (72b), such as is presently seen hi French. It seems that the generalization of the mood marker a relates, diachronically, to the loss of -re. The latter is still visible in nominalized infinitives, as in (72c). (72)
a. b.
c.
apleca, a manca,... to leave to eat ...au vrut milostivul Dumnedzau [a nu has wanted good-the Lord to not lasarea acestu pamant far1 de oameni....] leave this land without of people (Ureche 8) "The good Lord did not want to leave this land without people.' Amintirea razboiului ii urmare§te fara incetare. remind-re-the war-the-GEN them follows without stop-re 'The memory of the war follows them incessantly.'
The word order between the mood marker a and the verb resembles the word order hi a subjunctive string, including the use of the auxiliary.// 'be'. Compare (42) with (73). (73)
a. b.
a - Neg - Pron - Adv (intensifier-type) -.//-verb a nu-1 mai vedea/ a nu-1 mai fi vazut to not him more see/ to not him more be seen 'To not see him again/ to not have seen him again'
Unlike the subjunctives, the infinitive clauses do not host lexical constituents hi the left periphery
The generative approach to Romanian syntax
39
segment preceding the mood marker, even when a lexical complementizer is present, as in (74a). All lexical material, including the subject, must follow the verb. The pronominal subject in (74b) shows that lexical subjects are in nominative Case. (74)
a.
b.
Am plecat [inainte de [(* pe Ion Maria) a-1 saruta (Maria pe Ion)]], have-1 so left before of pe-lon Maria to-him kiss-INF Maria pe-lon 'I left before Mary kissed John.' Ii tinea lau§a [pentru a avea tu/(*tine) timp sa strangivasele.] them kept-3SG at door for to have you-NOM/you-ACC tune SA pick dishes-the 'He was keeping them at the door to give you time to pick up the dishes.'
4.3.1. The distribution of infinitive clauses. The loss of the infinitive in the Balkan languages is a striking phenomenon of area! contact among languages that are genetically unrelated. However, this replacement is not total in modern Romanian, where a few contexts may still display an infinitive complement, in free alternation with subjunctive complements; that is, after aspectual, modal and some raising verbs, as in (75a). Infinitives are still productive as adjuncts (75b), sentential subjects (75c) and relatives (75d). The latter present an obligatory complementizer de: (75)
a.
b.
c.
d.
Am inceput [a citi]. have-ISO started to read-INF 'I have started reading.' Am plecat [fara a- 1 revedea.] have-1 so left without to him see-again-INF 'I left without seeing him again.' [A tacea] inseamna [a- i accepta minciunile.] to shut means to-him-DAT accept lies-the 'Keeping silent means accepting his lies.' Ma inspaimanta gandul [* (de) a iubi un monstru]. me scares thought-the of to love-INF a monster 'The thought of loving a monster scares me.'
The general pattern emerging from the distribution of infinitives can be summed up as follows: (i) environments in which the infinitival clauses have been mostly supplanted by subjunctive clauses (i.e., in complement position); (ii) environments in which infinitivals occur as frequently as subjunctive clauses, in free alternation with the latter (i.e., subject position, complements to the modalputea 'can' or to the raising verbparea 'seem'); and (iii) environments where the infinitive is the most frequent option (i.e., adjunct clauses), or the only option (i.e., relative clauses).
40
Comparative studies in Romanian syntax
4.3.2. Internal structure. In establishing the internal structure of infinitive clauses, researchers have to take into account the peculiar syntax of these clauses in restructuring contexts and in indirect interrogatives. In both contexts, modern Romanian excludes the mood marker a 'to', as in (76). (76)
a.
b.
loanapoate (*a)pleca. loana can-3SG to leave-INF 'loana can leave.' Nu ai [de ce (*a) fi necajit]. not have-2so of what to be-lNF sad 'There is no reason for you to be sad.'
Dobrovie-Sorin (1994) deals with the facts by defining a as categorially ambiguous between complementizer and inflectional morpheme, on a par with sa. So a is base-generated in C, and its inflectional properties arise from its adjacency with I and the implicit reanalysis process that applies to adjacent functional elements. With respect to nominative subjects, the author argues that V-to-I assigns Case to SpecVP under government. Since lexical subjects freely alternate with non-lexical subjects in these contexts, Dobrovie-Sorin (1994) assumes that controlled infinitival subjects are pro versus PRO elements; control needs to be viewed as an anaphoric type of relation (following Borer 1989, Bouchard 1984, etc.), rather than being dependent on the presence of PRO. Motapanyane (1995) argues for an exclusive inflectional status for a, deriving the peculiarity of (76) from selectional restrictions, allowing only a VP constituent in complement position. So a is basegenerated in the M head, and selects an AgrP with a hierarchy similar to subjunctive clauses, as in (47). However, the verb moves only to T in infinitives (versus AgrS in subjunctives), which accounts for the impossibility of pre-verbal subjects. This analysis does not exclude PRO in pre-verbal subject position. In a Minimalist framework, Comilescu (present volume) adopts the hierarchy headed by an M head. However, this M carries the [wh] features of C, yielding the feature combinations [+inf/-wh] and [+inf7+wh]. The former is manifested as the mood marker a, merged in M, whereas the latter lacks a lexical realization, and V-to-M applies, as hi (76b). V-to-M destroys the configuration for Case checking on lexical subjects, so infinitives will have either lexical subjects or wh-elements, but not both.
4.4. Supines The supine in Romanian is formally indistinct from the past participle, both having the ending -t/-s.
The generative approach to Romanian syntax
41
Traditional grammars, however, keep the two moods separate, mainly because of their distinct thematic properties: supines are active verbs, as in (77a), while past participles are passives, as in (77b). The supine clause is necessarily preceded by the element de 'of. (77)
a.
b.
M-am apucat [de vopsit casa.] me-have-lSG started of painted-sup house-the 'I started to paint the house.' L-am gasit [spart.] it have-1 so found broken-PART 'I found it broken.'
Many prepositions, such as la 'at' in (78a), subcategorize for supine forms. In this context, the supine behaves as a predicational DP rather than a sentential complement. That is, the supine may take a definite article, as in (78b), causing the change of Case on its complement. The "nominalization" of the supine is not allowed when the selector is de 'of, as in (78c), repeating (77a). (78)
a. b.
c.
M-a ajutat Maria [la spalat vasele.] me has helped Maria at washed-sup dishes-the M-a ajutat Maria [la spalatul vaselor.] me has helped Maria at washed-sup-the dishes-the-GEN 'Maria helped me to wash the dishes.' *M-am apucat [de vopsitul casei.] me have-ISG started of painted-sup-the house-the-GEN
A distinction is thus drawn between de 'of and other prepositions insofar as only the former may head sentential complements. Various verbs (mostly aspectual) and adjectives (mostly of the tough-class) subcategorize for desupines, which usually alternate with subjunctive complements in that context, as in (79). (79)
a.
b.
c.
S- a apucat [de citit]/ [sa citeasca.] REFL has started of read-sup/ SA read-sUBJ.3sG 'He/she started reading.' Victor a terminat [de cantat la pian]/ ??[sa cante la pian.] Victor has finished of played-SUP at piano/ SA play-suBJ.3sG at piano 'Victor has finished playing piano.' E greu [de stabilit cine e primul]/ [[sa stabilim cine e primul].
42
Comparative studies in Romanian syntax is hard of decided-sup who is first-the/ SA decided-SUB J.I PL who is first-the 'It is hard to decide who is first.'
De-supines also occur as relatives, as in (80a), root exclamatives, as in (80b), or in the "recipe syntax", as in (80c). (80)
a.
b.
c.
maisina [de cusut] machine of sewn-SUP 'Sewing machine' De neimaginat! of unimagined-sup 'It's unimaginable!' Detinut la racoare. of kept-sup at cold 'To keep it hi a cold place.'
The word order in a supine clause indicates a strict adjacency in the string fife-Verb-Complement. Neither clitic-like elements (pronouns, negation, short adverbs) nor VP-adjoined adverbs may intervene between the elements of this string, as shown in (8 la), with the grammatical form in (81b). For negation, the supines resort to the derivational suffix ne- 'un-', specific to nominal elements, as shown in (81c). (81)
a. b.
c.
I- am dat [de (*nu-1 mai) citit (*des)] him-have-lso given of not it more read-sup often Nu i 1-am mai dat des [de citit.] not him it have- ISG more given often of read-SUP 'I have not given him to read it often any more.' E [de necrezut]. is ofunbelieved-sup 'It is unbelievable!'
4.4.1. Internal structure. The word order restriction illustrated in (81) led to analyses which postulate the absence of a functional domain in supine clauses (Grosu & Horvath 1987, Grosu 1994, Motapanyane 1998). In the same studies, de 'of is shown to occupy the head C (see also §tefanescu 1997). With respect to the internal structure of the supine clause, Motapanyane (1998) considers that the supine form is selected as such from the lexicon and merged in V. That is, -t/-s is a derivational (versus inflectional) morpheme, and the resulting lexical item is of the category [+V],[+N]. Feature checking is implemented through V movement to a [+V] head, that is "light v"
The generative approach to Romanian syntax
43
in the framework of Chomsky (1995), and then to a [+N] head, that may be either C occupied by de 'of or D. Thus, the supine clause has the structure [CP>VP] as complement to verbs, or the structure [DP>VP] as complement to P or in an unselected context. 4.4.2. Tough-constructions. Adjectives of the tough-chss select exclusively supine clauses (versus subjunctives) when they trigger object deletion, as shown in (83). (83)
a.
b.
Aceste problemej sunt greu [de rezolvat tj.] these problems are difficult of solved- SUP 'These problems are difficult to solve.' *Aceste problemej sunt greu sa le; rezob/i tj. these problems are difficult SA them solve-SUBJ.2SG
Grosu and Horvath (1987) and Grosu (1994) assume that Romanian tough- constructions are derived through Null Operator (NO) chains, as in English (Chomsky 1986). The ungrammaticality of (83b) follows from finiteness restrictions in configurations which license NOs. That is, finite clauses block NOs raising from their domain, as in (84b), contrasting with (84a) in which the NO raises cyclically across non-finite domains. (84)
a.
b.
Aceste alunCj sunt greu [de incetat [de ron|ait tj.] these peanuts are difficult of stopped- sup of crunched- SUP 'It is difficult to stop crunching these peanuts.' *Aceste alone, sunt greu [de incetat [sa kj ronjaim tj.] these peanuts are difficult of stopped-sup SA them crunch- SUB J.I PL
An alternative analysis is proposed in Motapanyane (1998), where (84a) involves object to subject movement, instead of relying on NOs. Support for an NP-movement analysis comes from the possibility of by-phrase insertion, lack of agreement with the predicative adjective, and exclusion of infinitive complements.
s. (NULL) OPERATORS (NOS) Dobrovie-Sorin (1990, 1994) establishes a structural quantifier parameter (presented in sections 1 and 3), which promotes a dissociated analysis of wh-movement and quantification in Romanian. An important prediction of this analysis (which strongly influenced subsequent studies on related subjects in Romanian syntax) is that Romanian lacks NOs: since NOs receive their quantificational feature from the structural position (i.e. SpecCP), they fail to be licensed in Romanian, where
44
Comparative studies in Romanian syntax
SpecCP is not a position for structural operators. The lack of constructions relying on NOs is thus explained: clefts, infinitive relatives, towg/z-constructions. On the basis of comparative data, Grosu (1994) contradicts this prediction by showing that the existence of NOs is independent of the structural quantifier parameter, and depends solely on the structural configuration. He points out that chains which include resumptive pronouns or clitic doubling and do not exhibit a lexical head indicate the presence of NO in SpecCP (cf Bayer and Grosu, present volume). Empirical evidence comes from comparatives, relatives, too/enoughconstructions, as illustrated in (85). (85)
a.
b.
c.
lonj e §i mai puternic decat a fost tatal liuY Ion is and more strong than has been father-the his 'John is even stronger than his father was.' Ni§te carji ce am rnjeles de la dansul ca vreji sa *(le) citiji. some books that have-1 so understood from him that wish-2PL SA them read-2PL 'Some books which I understood from him that you wish to read (them).' Ion e prea periculos pentru a-*(l) angaja./ sa-*(l) angajam. Ion is too dangerous for to him hire-INF/SA him hire-SUBJ-lPL 'John is too dangerous for us to hire .'
All the sentences in (85) present A'-chains with clitic/resumptive pronouns. Pied-piping tests on these constructions yield ungrammatical results (see Grosu 1994, 204-225). Since Pied-Piping of lexical material is possible with overt operators but not with NOs, the structures in (85) must include NOs. Consequently, if NOs are proved to occur in Romanian syntax, the lack of clefts and infinitive relatives must stem from an independent constraint.
*®*
Bayer and Grosu (present volume) approach operator constructions in general by postulating an operator feature whose "spreading" across the boundaries of the left-branch islands is the licensing factor for Pied-Piping and VP-scope effects. Data from three constructions (ie. adnominal adjectives, o«/y-clauses and constructions with left DP-external adjectives) in Romance (especially Romanian) and German provide support for the existence of the operator feature spread and indicate the structural environment in which it may or may not apply.
The generative approach to Romanian syntax
45
6. CONCLUSIONS This summary of generative studies on Romanian syntax presented pioneering work in the field in which more questions were formulated than solved. A few linguists' names occur repeatedly, emphasizing the scarcity of generative approaches to Romanian grammar. As expected, a relatively small number of studies means that a more limited array of topics have been investigated. For example, if nothing has been said about the adverbial system or the syntax of adjunct clauses it is because those aspects of grammar have not yet been explored, rather than because they lack relevance for grammatical theory. However, there is much hope hi this respect, since generative grammar is adopted now in Romanian universities, and doctoral students participate in exchange programs, joining the linguistic programs in North America, and engaging hi intensive research. A second observation following from the present overview concerns the importance of comparative studies for defining the structure of Romanian. Syntactically, Romanian is a hybrid between Romance and Balkan languages, and many of its peculiarities can be understood only with reference to equivalent paradigms hi Romance and Balkan. However, much more work has been done on Romanian as a Romance language than in relation to Balkan or other genetically unrelated languages. The present volume contributes to redressing this comparative balance by proposing cross-linguistic accounts which include Balkan languages (G. Legendre), English (V. Motapanyane), German (J. Bayer and A. Grosu) and Hebrew (C. Dobrovie-Sorin).
REFERENCES Alboiu, G. (1998). Object raising hi Romanian. Paper presented at the Annual Meeting of the Canadian Linguistic Association, Ottawa, June 1. Avram, L. (1998). Comparative remarks on the syntax and semantics ofaputea. Manuscript, University of Bucharest. Belletti, A. (1990). Generalized verb movement: Aspects of verb syntax. Rosenberg & Sellier, Torino. Bernstein, J. (1993). Topics in the syntax of nominal structure across Romance. Doctoral dissertation, City University of New York. Borer, H (1989). Anaphoric Agr. The Null Subject Parameter, edited by O. Jaeggli & K. Safir, 69109. Dordrecht: Kluwer Academic Publishers. Bouchard, D. (1984). On the content of empty categories. Foris, Dordrecht. Cardinaletti, A. (1996). Subjects in clause structure. Manuscript, University of Venice. Chomsky, N. (1986). Barriers. MIT Press, Cambridge, Mass.
46
Comparative studies in Romanian syntax
Chomsky, N. (1995). The Minimalist Program. MIT Press, Cambridge, Mass. Cinque, G. (1988). On si constructions and the theory ofarb. Linguistic Inquiry 19, 521-581. Cinque, G. (1990). Agreement and head-to-head movement in the Romance Noun Phrase. Paper presented at LSRL 20, Ottawa, March 15. Cinque, G. (1990). Types of A '-dependencies. MIT Press, Cambridge, Mass. Comorovski, I. (1996). Interrogative phrases and the syntax-semantics interface. Kluwer Academic Publishers, Dordrecht. Cornilescu, A (1992). Remarks on the Determiner system of Romanian: The demosntratives al and cel.Probus4, 189-260. Dobrovie-Sorin, C. (1987). Syntaxe du roumain.: Chaines thematiques. These de Doctorat d'Etat, Universite de Paris 7. Dobrovie-Sorin, C. (1990). Clitic Doubling, Wh-Movement, and Quantification in Romanian. Linguistic Inquiry 21, 351-397. Dobrovie-Sorin, C. (1994). The syntax of Romanian, comparative studies in Romance. Mouton de Gruyter, New York. Dobrovie-Sorin, C. (1994). On impersonalse/si in Romanian, French and Italian. In: Paths Towards Universal Grammar. Studies in Honor of Richard S. Kayne (G. Cinque, J. Koster, J.-Y. Pollock, L. Rizzi and R. Zanuttini, eds.), pp. 137-153. Georgetown University Press, Washington, D.C. Dobrovie-Sorin, C. (1998). Impersonal si constructions hi Romance and the passivization of intransitives. Linguistic Inquiry 29, 399-437. Farkas, D. (1982). Intensionality and Romance subjunctive relatives. Indiana University Linguistics Club, Bloomington. Farkas, D. (1984). Subjunctive complements in Rumanian. In: Papers from the Xllth Linguistic Symposium on Romance Languages (P. Baldi, ed.), pp. 354-372. John Benjamins, Amsterdam. Farkas, D. (1985a). Intensional descriptions and the Romance subjunctive mood. Garland Publishing Inc., New York. Farkas, D. (1985b). Obligatorily Controlled Subjects hi Romanian. In: Proceedings of the 21st Annual Meeting of the Chicago Linguistics Society (W.H. Eilfort, P.O. Kroeber and K.L. Peterson, eds.), pp. 91-100. CLS, Chicago. Farkas, D. (1990). Two cases of underspecification in morphology. Linguistic Inquiry 21, 539-550. Gierling, D. (1996). Further parallels between clitic doubling and scrambling. In: Proceedings of the 13th Eastern States Conference on Linguistics '96 (A D. Green and V. Motapanyane, eds.), pp. 113-123 . Cornell University, CLC Publications, Ithaca, NY. Gierling, D. (1997). Clitic doubling, specificity and Focus in Romanian. In: Clitics, pronouns and movement (J. Black and V. Motapanyane, eds.), pp. 63-85. John Benjamins, Amsterdam. Giusti, G. (1993). Enclitic articles and double definiteness: A comparative analysis of nominal
The generative approach to Romanian syntax
47
structure in Romance and Germanic. The Linguistic Review 11, 231-255. Giusti, G. (1995). Heads and modifiers among determiners: Evidence from Rumanian. In: Advances in Rumanian linguistics (G. Cinque and G. Giusti, eds.), pp. 103-126. John Benjamins: Amsterdam. Grosu, A. (1988). On the distribution of genitive phrases in Rumanian. Linguistics 26, 931-949. Grosu, A. (1994). Three studies in locality and Case. Routledge, New York. Grosu, A. and J. Horvath. (1984). The GB Theory and Raising in Rumanian. Linguistic Inquiry 15, 345-353. Grosu, A. (1987). On non-finitness in extraction constructions. Natural Language and Linguistic Theory 5, 181-196. Gueron, J. and T. Hoekstra. (1987). T-chains and the constituent structure of auxiliaries. Paper presented at GLOW Conference, Venice. Haider, H. (1988). Matching Projections. In: Constituent Structure (A. Cardinaletti.; G. Cinque & G. Giusti). Annali di ca' Foscari XXVII4, 101-123. Isac,D. (1994). Sentence negation in Romanian. Revue roumaine de linguistique 29 (3/4), 371-393. Jong, J. J. de. (1996). The case of bound pronouns in peripheral Romance. (Groningen Dissertations in Linguistics 16.) Grafisch Bedrijf Ponsen & Loojinen BV, Wageningen. Kayne, R. (1994). The antisymmetry of syntax. MIT Press, Cambridge, Mass. Kempchinsky, P. (1986). Romance subjunctive clauses and Logical Form. Doctoral dissertation, UCLA. Kiss, K.(1995). Introduction. In: Discourse configurational languages (K. Kiss, ed.), pp. 3-27. Oxford University Press, New York. Kraskow, T. (1990). Multiple Wh-movement in Slavic. Paper presented at the Workshop on Scrambling, Tilburg University. Lema, J. and M.-L. Rivero. (1989). Long Head Movement: ECP vs. HMC. Proceedings ofNELS 20, pp. 333-347. University of Massachusetts: GLSA. Mahajan, A. (1991). Clitic doubling, object agreement and specificity. Proceedings ofNELS 21, pp. 263-277. University of Massachusetts: GSLA. Manzini, Rita and E. Roussou. (1997). A Minimalist Theory of A-movement and Control, Manuscript, London University College. Motapanyane, V. (1989). La position du sujet dans une langue a 1'ordre SVO/VSO. Rivista di grammatica generativa 14,75-103. Motapanyane, V. (1994). An A-position for Romanian subjects. Linguistic Inquiry 25, 729-734. Motapanyane, V. (1995). Theoretical implications of complementation in Romanian. Unipress, Padova. Motapanyane, V. (1998). De-supines and object-to-subject raising. Paper presented at LSRL 28, April 18, The Pennsylvania State University. Pesetski, D. (1987). Wh-in-situ: Movement and unselective binding. In: The Representation of
48
Comparative studies in Romanian syntax
(In)dejiniteness (E. Reuland and A. G.B. ter Meulen, eds.), pp. 98-130. MIT Press, Cambridge Mass. Pollock, J.-Y. (1989). Verb movement, UG and the structure of IP. Linguistic Inquiry 20, 365-424. Rivero, M.-L. (1989). Barriers and Rumanian. In: Proceedings of the LSRL 17 (J. de Cesaris and C. Kirschoner, eds.), pp. 289-312. John Benjamins, Amsterdam. Rivero, M.-L. (1994). Clause structure and V- movement in the languages of the Balkans. Natural Language and Linguistic Theory 12, 63-120. Rizzi, L. (1982). Issues in Italian syntax. Foris, Dordrecht. Rizzi, L. (1990). Relativized Minimality. MIT Press, Cambridge, MA. Rizzi, L. (1991). Residual Verb Second and the Wh-Criterion. Technical Reports in Formal and Computational Linguistics 2. University of Geneva. Rizzi, L. (1997). The fine structure of the left periphery. In: Elements of grammar (L. Haegeman, ed.), pp. 281-337. Kluwer Academic Publishers, Dordrecht. Rudin, C. (1988). On multiple questions and multiple wh-fronting. Natural Languages and Linguistic Theory 6, 445-501. Sportiche, D. (1988). A theory of floating quantifiers and its corollaries for constituent structure. Linguistic Inquiry 19, 425-449. $tefanescu, I. (1997). The syntax of Agreement in Romanian. Department of Linguistics, MIT Working Papers in Linguistics 14, Cambridge, Mass. Terzi, A. (1992). PRO infinite clauses: A study of the inflectional heads of the Balkan languages. Doctoral dissertation, City University of New York.
Feature checking meets the Criterion Approach 49
Feature Checking Meets the Criterion Approach: Three Ways of Saying only in Romance and Germanic*
Josef Bayer, Friedrich Schiller University, Jena Alexander Grosu, Tel Aviv University
i. INTRODUCTION 1.1. Background The received wisdom within the version of Generative Grammar known as the Theory of Government and Binding was that a quantified nominal whose "understood" scope properties do not correspond to its overt position may achieve a suitable scope position through covert movement, specifically, through the application of a process known as Quantifier Raising (QR). This position, which was vigorously advocated in May (1977, 1985), came under attack at the end of the eighties from scholars who either argued against LF as a component of the grammar (e.g. Koster 1987, Lappin 1991) or else proposed that the process called QR should be allowed to apply to quantified and non-quantified nominals alike (Reinhart 1991). More recent developments in linguistic theory, in particular, in the version of the Minimalist Program in Chomsky (1995), led to an even more radically modified view of LF. On the one hand, covert phrasal movement, and more generally, covert raising of a category (i.e., covert Move-a) was ruled out in general as uneconomical, Movea, as well as phrasal Pied-Piping, being licensed only in the overt component by the need to ensure convergence at the PF interface. On the other hand, Minimalism allows a new kind of process, the raising of features. Importantly, neither phrases nor features can raise in order to ensure
"We would like to thank the following colleagues who helped us at various stages with theoretical as well as with dataoriented discussion: Anna Cardinaletti, Norbert Corver, Carmen Dobrovie-Sorin, Marc-Ariel Friedemann, David Gaatone, Julia Horvath, Giuseppe Longobardi, Virginia Motapanyane and Georges Rebuschi. Thanks also to the audiences of the 1998 Incontro di Grammatica Generative! in Verona and the 1998 Generative Grammatik im Stiden at Salzburg where parts of this research were presented.
50
Comparative studies in Romanian syntax
interpretability, but only in order to prevent a crash at one of the two interface levels. This view is thus in.sharp conflict with May (1985); as Hornstein (1995) puts it, the minimalistic grammar of movement is "semantically myopic". In short, the primary movement process is the raising of a single feature, and the primary driving force behind this process is the need to check off uninterpretable features. Within Minimalism, there are two ways in which an "interpretable" operator feature can achieve scope by raising: (a) if it happens to be matched with an uninterpretable feature on the target of movement and thus raises in order to check off the latter, or (b) if it is PiedPiped by another feature of the lexical item to which it belongs. Option (b) is licensed by the (unavoidable, in Chomsky's view) stipulation that all the formal features of a lexical item are automatically "dragged along" when one of them is attracted by a target for checking purposes. It needs to be stressed that the checking processes envisaged by Chomsky are different hi spirit from the "criteria" proposed in Rizzi (1991), although the two share an operational configuration, the Spec-Head configuration. Rizzi's criteria involve the matching of two interpretable features, and moreover allow an element endowed with such features to convey them to an underspecified recipient by what he calls "dynamic agreement" (see his treatment of French direct questions). Nonetheless, the chasm between the two views may be less deep than it initially seems. We would like to suggest that strict Minimalism can be modified to incorporate Rizzi's transmission mechanism without fundamentally altering its conceptual orientation, if it is acknowledged that not every type of movement is triggered by a "suicidal" attractor. Assume, for example, that an XP is endowed with a feature which is [+int(erpretable)] but occupies a position hi the phrase marker that does not allow it to be fully interpreted. If such objects reach the LF interface, they will induce a crash. To become fully interpretable, they need to move to a position which fills in the missing specification. Assume for concreteness that an operator (henceforth Op) in situ has a set of [+int] features one of which requires Op to have scope over an appropriate domain. Then we can say that a scope domain "attracts" the features of Op. But attraction here is not exactly what it is in the standard cases, because the domain to which Op needs to raise is contentually defined, and - as already pointed out - this operation ultimately involves the matching of interpretable features. In short, the particular implementation of criteria a la Rizzi that we suggested above constitutes a limited extension of Minimalism which preserves the view that movement is licensed by the need to ensure convergence at the two interfaces, while allowing two types of features to achieve this goal: (a) [hit] features are attracted and seek to "commit suicide", and (b) un(der)specified [+int] features, which seek to attain full interpretability. In short, it seems that a marriage of Chomsky's and Rizzi's views hi the way just outlined makes conceptual sense, if it turns out to be empirically needed. 1.2. Goal In this article, we pursue a number of interrelated theoretical and descriptive goals. On the
Feature checking meets the Criterion Approach 51 descriptive side, we compare and contrast three distinct syntactic constructions with essentially the same semantic import, one of which has not, to the best of our knowledge, been carefully described hi the earlier (at least, generative) literature so far. These are illustrated with synonymous Italian, English and Romanian examples in (la-c); for ease of reference, we will call such constructions types (I), (II) and (III) respectively. (I)
a. b. c.
La sola Maria si e presentata the only Maria REFL is presented Only Mary showed up. Singura. Maria/ Maria singura s- a prezentat. only Mary Mary only REFL has presented
(I) (II) (HI)
What distinguishes these constructions is that the force of only seen hi (II) is conveyed by agreeing adjectives in (la) and (Ic), which are moreover DP-internal in the former and DP-external in the latter (all these items are in italics). Furthermore, (I-HI) are increasingly constrained in their distribution in that order hi ways that will be made explicit below. Finally, while some version of (II) seems to exist hi all the languages with which we are familiar, (I) and (HI) appear to be less common; out of the five languages we address hi this study (English, French, German, Italian and Romanian), only French exhibits all three. On the theoretical side, we will propose a unifying analysis of (I-HI) which derives their distributional properties from then- featural make-up. In particular, we will argue that increasing distributional restrictions correlate with increasing "deficiencies" in the Formal Feature (FF) set of the italicised items hi (1) (and their counterparts hi other languages). A more general theoretical innovation that we will argue for, and which goes hand hi hand with the already proposed (A) generalized version of Chomsky's checking-driven F-raising mechanism, which, recall, incorp orates (A') Rizzi's semantics-oriented criteria, is that the grammar of movement must recognize two types of covert movement, with distinct properties. Bayer (1998) offers abundant empirical evidence for this distinction, and proposes to analyze it hi terms of Move-F, which he views as a purely mechanical process of the Computational System dictated by the "suicidal greed" of uninterpretable FFs, and Move-a, which he views as a semantically-oriented operation that resembles QR, but differs from it in that it is not simply adjunction, but rather movement to the specifier of a (potential) projection headed by an Op feature which is also inherent hi the phrase to be moved. The remainder of this paper is organized as follows. In section 2.1., we illustrate the operation of mechanism (A) hi a construction other than the three which form the focus of this paper, and the ability of operator FFs to spread across the boundaries of left-branch islands. In sections 2.2. and 2.3., we discuss construction (I), showing how interpretable Op features can "spread" within such
52
Comparative studies in Romanian syntax
domains as DP and PP, crossing, in the process, the (language-specific) island boundaries that separate a DP from its immediately containing PP, as well as a left-branch DP from its containing DP. In section 3., we argue that the spread of operator FFs is blocked in constructions of type (II) by the lack of categorial FFs in items like the italicized one in (Ib). We will argue that the only way for such items to achieve scope wider than their overt position is to take part in long distance movement, an operation reflected hi sensitivity to prepositional and left-branch islands. In section 4., we show that constructions of type (HI) are acceptable only if items like the italicized one in (Ic) can acquire appropriate scope on the basis of their overt position. Covert F-spread or long-distance movement is ruled out by the assumption that the kind of items under consideration fail to make their operator FFs visible to the computational system
2. FEATURE-SPREADING AS A LICENSING FACTOR 2.1. Spec-Head Agreement and C-Command We begin by illustrating the workings of mechanism (A), hi particular, of its (A') component, hi an unrelated construction, specifically, one where an Op feature originates within the Spec (of the Spec, etc.) of a DP that undergoes overt Pied-Piping. Consider the English data in (2) and the parallel dialectal German data hi (3). (2) a. {Which student's mother's canary] did you poison? b. [No student's mother's canary] have I ever poisoned. (3) a. [Welchem Studenten seiner Mutter ihren Kanarienvogel] hast du vergiftet? which student-DAT his mother-DAT her canary-ACC have you poisoned 'Which student's mother's canary have you poisoned?' b. [Keinem Studenten seiner Mutter ihren Kanarienvogel] habe ichje vergiftet. no student-DAT his mother-DAT her canary have I ever poisoned "No student's mother's canary have I ever poisoned.' Observe that the features [wh] or [neg] associated with the italicized items occur within the Spec of the Spec of the bracketed DP. These features are demonstrably responsible for a number of properties of these data, in particular, for the fact that the bracketed DPs hi (2) occur hi the leftmost position of a verb-second (V2) configuration (something that is permitted hi English only under special conditions), and the fact that negative polarity items are licensed within the VP, suggesting that the italicized items have scope over VP (see the (b) subcases of (2)-(3)). Importantly, not only are the italicized items necessary for the presence of such effects, but they moreover cannot induce them from just any position within a containing DP, as shown by the deviance of the data hi (4).
Feature checking meets the Criterion Approach 53 (4) a. ?*[A canary that -which student owned] did you poison? b. *[A student with no canary] have I (ever) seen. In short, the bracketed constituents in (2)-(3) induce the same effects as when the maximal DP is the minimal one that includes Ops, as is the case, for example, in which/no canary. In past literature, there have been a variety of attempts to analyze Pied-Piping and scope out of Spec effects, none of them fully satisfactory with respect to data like (2-3). For example, Safir (1986) suggested, with respect to a different Pied-Piping configuration (illustrated by the height of the lettering on the covers of which), that the minimal \vh-DP undergoes (cyclic) extraction out of higher DPs and ultimately adjoins to the top DP. Such an account does not obviously generalize to data like (2a), because which in the example just given can undergo long extraction out of DP, while left-branch items like those in (2a) cannot. Reinhart (1987) simply stipulated that Specs (of Specs) may bind in cases like (2-3). In her account, a DP in SpecDP carries a referential index and an operator index in order to distinguish anaphoric and variable binding. Nothing is assumed, however, about feature percolation and its consequences for Pied-Piping. Chomsky (1995) sketched an account of the obligatoriness of Pied-Piping in cases like (2a) which did not require the italicized hem or its features to raise from their overt position, but his account does not obviously generalize to other instances of Pied-Piping, neither does it offer a solution to the scope effects. What we want to show now is that the (A') component of mechanism (A) provides a straightforward account for both the Pied-Piping and the VP-scope effects in (2-3). To see this, consider the following (partial) representation of the top DP in (2a, b).
which/no student's
mother' s
canary
We have refrained from indicating the internal structure of the minimal DP because we do not wish to take a stand on whether the italicized item is under D or under SpecDP. If it is under D, its Op properties automatically project to its maximal projection, that is, the minimal DP. If it is in SpecDP, we need to assume a null D with an unspecified Op feature that receives specification from the
54
Comparative studies in Romanian syntax
italicized item in SpecDP by mechanism (A'); this specification will automatically spread to DP. Now, if we make the further assumption that the possessive morpheme 's can be endowed with an unspecified Op feature, the transfer mechanism (A') will apply twice, ultimately ensuring that the Op features of the corresponding italicized item become properties of the maximal DP. We note that the sequence of operations we have sketched is applicable regardless of whether the phrases in SpecDP are base-generated in this position, or whether they are raised into it from SpecNP. Comparable operations may be assumed to take place in the derivations of (2b) and (3a-b). The fact that the maximal DP inherits the Op properties of the italicized items enables it to raise to the Spec of a suitable category (CP or NegP), and also ensures that polarity items within VP are ccommanded, and thus licensed, by an appropriate Op feature. Since the raising of DP to SpecCP or SpecNegP has automatic scope effects, it seems reasonable, given the (extended) (A)-type model we are assuming, to attribute this process to ks (A') component. At the same tune, the fact that this type of movement is overt requires an appeal to Chomsky's notion of "featural strength", or some equivalent device that is not semantically oriented. We do not claim that this is the only way in which these results can be achieved (e.g., the theory in Kayne 1994 seems capable of achieving them in different ways), and we certainly do not claim to have a general theory of Pied-Piping, which, to the best of our knowledge, nobody has (construction-specific and language-specific variation, as well as optionality under certain circumstances, remain largely unexplained). We do believe, however, that the Minimalist approach incremented with criteria offers a natural alternative to earlier stipulative accounts. As has often been noted, the constructions in (2-3) are not obviously suitable for direct semantic intepretation, and there have been a variety of proposals to implement "reconstruction". It seems to us that the proposals in Chomsky (1993) offer a rather natural implementation. We remind the reader that the proposal under consideration assumes that A'-chains are generated with full copies of the moved phrase, and that such chains undergo independent deletion operations in both the PF and the LF portions of the derivation. The deletions on the branch of the derivation that end at LF are driven by the need to ensure interpretability by semantic operations. To illustrate, we indicate the material that needs to be deleted hi (6a, b) to achieve a suitable input to semantics for (2a) and (2b) respectively. With respect to (6b), we assume that the item no is split into sentential negation and an existential quantifier, an operation that is needed independently of chain formation, and that may well be accomplished by post-LF purely semantic operations.1
'See Bech (1955/57: 76ff), who seems to have been the first to propose that negative elements like German kein are in fact "cohesive" forms of negation and the indefinite determiner em. On this point, see also Stechow, A. von & W. Gender (1997).
Feature checking meets the Criterion Approach 55 (6) a. [Which student's mother's canary] did you poison [which student's mother's canary] b. [No student's mother's canary] have I ever poisoned [no student's mother's canary]
2.2. Adnominal Adjectival Operators and Feature Spreading within DP Having illustrated the operation of the generalized (A) mechanism, we now turn to a closer examination of construction (I). A number of languages allow agreeing elements with adjectival morphology that are internal to DP and induce a construal similar to the one that results from the combination of DP with only or its counterparts in other languages. We illustrate this state of affairs hi (7) and (8) with data from Italian, French and German, noting that Romanian and English do not allow such constructions. These adjectives are invariably pre-nominal, even in French and Italian (where post-nominal APs are quite common; see Cinque 1994, 25 5f), and moreover do not allow (adverbial) modification of any kind; these are two properties that point to probable functional status. We provide data in which the containing DPs occur in both subject and object position with "malice aforethought"; the precise reasons will become clear in section 4: (7)
(8)
a. La sol -a Maria si e presentata. the alone-AGR Maria REFL is presented 'Only Mary showed up.' b. La seul -e rente peut resoudre ce probleme. the alone-AGR queen can solve this problem 'Only the queen can solve this problem.' c. Die alleinig -e Konigin kann dieses Problem losen the alone -AGR queen can this problem solve 'Only the queen can solve this problem.' a. Ritengo la sol -a regina la rappresentante vera de-1 popolo see-1 so the alone-AGR queen the representative true of-the people 'I view only the queen as the true representative of the people.' b. Je considere la seul -e reine comme la representante I consider the alone-AGR queen as the representative legitime du peuple legitimate of-the people 'I view only the queen as the legitimate representative of the people.' c. Ich halte die alleinig -e Konigin fur die legitim Vertreterin des Volkes I hold the alone -AGR queen for the legitimate representative of-the people
56
Comparative studies in Romanian syntax 'I view only the queen as the legitimate representative of the people.'
To avoid confusion, we note that we have glossed the italicized items as 'alone' simply because they are typically homonymous in the corresponding languages with words that mean 'unaccompanied', 'unaided', or 'only' in phrases like the only girl he ever loved. For completeness, we note that type (I) constructions belong to a somewhat elevated register in Italian, French and German, and also that English and Romanian go one step further in disallowing such constructions altogether. For example, the data in (9-10) can at most have some of the meanings mentioned at the end of the preceding paragraph, but not the one that interests us here. (9) (10)
a. The {sole, *alone} queen can solve this problem. b. I view the (sole, *alone} queen as the legitimate representative of the nation. a. Singur-a reginava rezorva problem-a aceasta. alone -the-AGR queen will resolve problem-the this 'The queen, who is alone, will solve this problem' b. O consider pe singur-a regina drept reprezentant-a her consider-\SGpe alone-the queen as representative-the legitima a poporu-1 -ui. legitimate of people-the-AGR -GEN.AGR 'I view the queen, who is alone, as the legitimate representative of the people.'
We will consider in section 4 how parametric variation of such "adjectival Ops" can best be stated. Let us now consider what the LF of data like those hi (7-8) might look like. To the extent that the adjectives at issue have the essential import of'only', they need to be construed as propositional Ops (see Bayer, 1996 for detailed justification). To achieve this outcome, they must c-command (at least) the clause's Complete Functional Complex (CFC) in the sense of Chomsky (1986), that is to say, the verb and (the traces of) all its arguments (we will return to this point hi subsequent sections). How can this result be achieved? We will argue that the following sequence of operations applies: (i) the Op feature spreads to DP; (ii) DP is covertly reordered to a scope position if it is not already in such a position; (ui) suitable tampering with a chain that contains the Op leaves it in an appropriate scope position. Step (i) is unproblematic, given (the (A') component of) (A). Cinque (1994) proposes that a variety of adnominal (adjectival) modifiers of N occur hi the Spec of a number of well-defined functional projections, while leaving open the possibility that at least some of the items which do not exhibit demonstrably phrasal properties are in functional head positions. With respect to the kind of data in (7-8), we find the latter kind of analysis more plausible than its alternative, and will hi fact assume
Feature checking meets the Criterion Approach 57 it in what follows (note that semantically plausible adverbial modifiers of the adjective, such as French tout and German gam 'completely', are decidedly out). Nonetheless, we outline the featurespreading mechanism under both analyses. Under the head analysis, the DPs in (7-8) have essentially the structure in (11). We eschew the issue of the exact nature of the functional category by labeling it "W(P)", while noting that it must be able to license agreement between the adjective, the noun, and D:
(11)
Grimshaw (1991) proposes that the functional categories of "extended projections" agree with the lexical category in categorial features; for example, the lexical and functional categories of a nominal extended projection are all assumed to carry the features [+N,-VJ. One way of reconstructing this idea within a framework that employs feature checking is to assume that the noun carries an interpretable N(ominal) feature and that all the functional heads of its extended projection carry uninterpretable N features. This will force the N-feature of the noun to raise covertly by successive adjunction operations to all the functional heads above it, checking off their uninterpretable features one by one. Since at each such step, the category that hosted an adjunct on the preceding step raises, too, the Op features of the adjective under W will raise to D, ultimately spreading to DP.2 Under the Spec analysis, the head W is either empty or contains the adjectival suffix. In the spirit of Chomsky (1993, 1995), let us assume the former, which yields the structure in (12).
2
We have constructed our account in terms of the categorial feature of the noun because Chomsky (1995) considers that categorial features are interpretable, and thus undeletable, in contrast to say, Case features, which he views as uninterpretable. The view that categorial features are interpretable is not unchallengeable, but not unreasonable, either (we note that various formal semantic theories posit a tight correlation between syntactic categories and logical types). Note that the covert raising of formal features does not obviously account for morphological agreement within DP, since at least some of the agreeing features are viewed as uninterpretable (e.g., Gender and Case), and if they take part in usual checking operations as the N-feature raises from head to head, the uninterpretable features would be deleted too soon. We leave open the treatment of morphological agreement within DP.
58
Comparative studies in Romanian syntax
(12)
DP
la la die
sola seule alleinige
Konigin
The null head W now carries, in addition to an uninterpretable N-feature, an underspecified Op feature. In virtue of the mechanism (A') (dynamic agreement), the adjective specifies this feature, which, in virtue of head-to-head raising, will reach D and, ultimately, DP, by transitivity. Step (ii) is needed only in the derivation of (8). Provided that the surface subject is "high enough" to have scope over the CFC, the DP in (7) does not have to undergo further raising. Actually, the status of (7) warrants closer consideration because the overt presence of DP in a scope position does not automatically mean that the Op will take scope from that position. Thus, Bayer (1996) proposed that Ops like the one under consideration cannot end up in just any position that guarantees clausal scope, but only hi the Spec of a specific category, which he labeled "Particle Phrase" (PrtP), and which, he argued, includes VP as a complement of its head. This view falls under the theory of quantification developed by Beghelli & Stowell (1995), which proposes to account for relative Op scope in terms of the relative hierarchical height of functional projections, and which has also proven beneficial for the characterization of the linear order of Ops in languages where linear order reflects relative scope, such as Hungarian (Kiss 1991; Szabolcsi 1995). Note that, under the assumption that VP is a complement of Prt, PrtP is the hierarchically lowest of the scopeoriented functional categories. We note that, according to Bayer (1996), the DP in data like (7) is not itself in SpecPrt, but one of the copies that it created in raising from SpecVP is. What is needed, then, in order to achieve an interpretable LF, is the application of step (iii), and in the following way: everything but the Op must be deleted from the copy in SpecPrtP; nothing but the Op must be deleted in the lowest copy, and any additional copies must be entirely deleted.3 3
Szabolsci (1995) points out that in data with subject quantified phrases, the Beghelli & Stowell hypothesis would seem to assume an improper chain, that is, one in which movement to an A"-position is followed by movement to an A-position, on the assumption that the Specs of such phrases as Beghelli & Stowell's Distributive)? or our PrtP probably are A'positions. The force of this objection is unclear to us. First, the notion "A-position" is far harder to characterize on independent grounds within a model that assumes functional categories than in the GB model, since such positions as [SpecTP] are clearly not argument positions in the GB sense of (potential) theta-positions. We thus see no conceptual problems with assuming that SpecPrtP is an A-position. Second, the kind of reasons that exclude "classical" improper
Feature checking meets the Criterion Approach 59 In the derivation of data like those in (8), step (ii) is unquestionably necessary, since DP is clearly not in a clausal scope position in overt representation. Note, however, that step (ii) is inconsistent with the hypothesis advanced in Chomsky (1995) to the effect that all covert movement is Move-F. Consistency with this hypothesis could only be achieved, if the Op features (together with the remaining FFs) raise and adjoin to Pit; although it must be stressed that Move-F is supposed to fall outside the known processes of categorial movement, such a process would perhaps resemble head movement more than phrasal movement (cf. Brody 1998, 382). Bayer (1996, 57-59) presents a number of arguments against scoping of Pit by head movement. To the extent that head movement equals Move-F plus Pied-Piping of the minimal morphological carrier (perhaps a phonological word), these arguments apply to Move-F as well. While Move-F would immediately allow Op features to reach a proper scope position, this implementation of step (iii) seems to be all too easy. We return to the issue of the choice between analyzing QR as Move-a or Move-F hi section 3.
2.3. Feature Spreading from DP into PP We now turn to DPs which, just like those in (8), fail to be in scope position in overt representation, but which, unlike those in (8), are complements of a P. These data provide an indication of the kind of factors that could hi principle adjudicate between feature raising and phrasal raising for scope assignment (see comments on step (ii) hi the preceding section), but in fact, no decision is yet possible, for reasons that will become apparent below. Consider the data in (13a-c), and note that the italicized items have the same force as those in (7-8). For completeness, we add the superficially similar Romanian data hi (13d), which, just like (10), disallows this kind of construal for the italicized adjective. (13)
a. n re ha parlato [pP con [^p la sol-a regjna]] the king has spoken with the alone-AGR queen "The king has spoken only with the queen.' b. On ne peut plus compter [pP sur (j^, la seul -e reine]] one not can more rely
on
the alone-AGR queen
chains, as in (i), do not apply here, (i) a. It was asked by John [who had seen Mary]. b.*Who was asked by John [ t had seen Mary]. In (ib), instead of inserting a dummy in the non-theta position in the matrix subject position, as in (ia), who has raised from SpecCP, an unquestionable A'-position, to the matrix subject position. Since who has "used up" its Case feature within the lower clause, it cannot check the matrix Case feature. This situation does not arise in cases like (6), where movement is confined to a single verbal extended projection. In short, we see nothing obviously improper about the chain that the Beghelli & Stowell hypothesis would assume with respect to the examples in (7).
60
Comparative studies in Romanian syntax 'One can no longer rely only on the queen.' c. Man kann sich nicht [PP auf [DP die alleinig-e Konigin verlassen]]. one can REFL not on the alone -AGR queen rely 'One cannot rely only on the queen.' d. Nu se mai poate conta [PP pe [DP singur-a regina]] not REFL more can rely on alone -the-AGR queen 'It is no longer possible to rely on the queen, who is alone.'
Assuming that the Op features of the italicized adjectives reach the containing DP by the mechanisms discussed in the preceding section, we are now faced with a potentially surprising result: on the one hand, we may expect DP to move to SpecPrtP (or its Op and remaining FFs to move to Pit), just as in the derivation of (8). On the other hand, given the very reasonable assumption made in Chomsky (1993) to the effect that constraints on movement are insensitive to the overt/covert status of movement, we may expect that the data in (13a-c) should be deviant, because overt movement cannot strand Ps in the corresponding languages. That the data at issue are not deviant points to the conclusion that the PP island boundary is circumvented in some way. The solution we wish to propose was suggested to us by a comparable, if weaker, contrast found in relation to left-branch constituents. While the left-branch constraint on overt extraction is weaker in German data like (3) than in English data like (2) (presumably, because neither the stranding of an affix nor the reordering of a non-constituent are involved; see Chomsky 1995, 263), most speakers nevertheless witness a clear effect, as illustrated in (14a-b). But such an effect is absent in comparable type-(I) constructions, as illustrated in (14c). (14)
a. [[Welchem Studenten] seine Ansichten] meinst du, seien which-DAT student his views think you are-SUBJ beriicksichtigt worden? taken-seriously become 'Which student's views do you think were taken seriously?' b. ??[Welchem Studenten] meinst du, [ t seme Ansichten] seien beriicksichtigt worden? c. [[Dem alleinigen Konig] seine Ansichten] sind beriicksichtigt worden. the-DAT alone king his views are taken-seriously become 'Only the KING'S views were taken seriously.'
The contrast between (14b) and (14c) is unsurprising, in view of what was said in section 2.1. about the data in (2-5): the operator FF in (14c) can spread to the largest containing DP by exactly the kind of local steps that were posited for (2)-(3), thus circumventing the need to violate the leftbranch constraints of German. What we want to suggest is that a similar tack is possible with respect
Feature checking meets the Criterion Approach 61 to the data in (13a-c). Thus, a number of writers have provided evidence that Ps function in certain respects very much like alternative realizations of Case. For example, Grosu (1994) shows that affixal Case and Ps function in the same way with respect to a variety of matching and non-matching effects in relative clause constructions of various kinds. Galhnan (1996, 1997) shows that certain German vocabulary items that fail to exhibit overt Case inflection, for example, nichts 'nothing', may not occur hi environments where some oblique Case, in particular, dative, is required (see (15a)). Nonetheless, this item can occur as complement of the P mil, which requires precisely dative Case, as illustrated in (15b). Grosu (1994) provides extensive evidence that in German and various other Indo-European languages, Ps always count as oblique Case for the purpose of the (non)matching effects he studied. All this suggests that in (15b), P satisfies a requirement that is violated in (15a), namely, that items like nichts needs to exhibit some oblique Case morphology in oblique contexts. (15)
a. *Peter hat nichts widersprochen. Peter has nothing objected 'Peter has objected *(to) nothing.' b. Peter ist mit nichts zufiieden. Peter is with nothing satisfied 'Peter is satisfied with nothing.'
Now, if Ps can function as alternative realizations of Case hi situations like those just described, it seems reasonable to assume that they function as realizations of Case in other respects as well, in particular, in further extending an extended nominal projection. If so, we may assume that Ps are, or at least can be, endowed with an uninterpretable N feature, with the result that the N-feature that has raised as far as D(P) will further raise to P, and ultimately spread to PP. The assumption of Nfeature movement to P has a clear advantage. Being independently motivated, it explains why PPs are transparent to certain kinds of processes, a fact that has caused various researchers to propose unconventional accounts.4
3. PARTICLE OPERATORS We now turn to constructions of type (II). As already noted hi section 2.2., items with the semantic force of only need to be construed as prepositional Ops. Bayer (1996) argues that this construal can
4
To illustrate, Ross (1967) simply stipulated that PPs are in fact NPs. Moritz & Valois (1994) proposed covert movement of DP to SpecPP, an operation that is somewhat suspicious for French (the language they studied), since it is never overtly attested. Various other proposals could be adduced.
62
Comparative studies in Romanian syntax
sometimes be straightforwardly derived from the overt representation, in particular, when this particle is base-generated in the head position Prt, as seems to be the case in (16). Bayer also observes that there are other situations where the needed construal is not obviously derivable in this way, because the particle appears to form a constituent with constituents smaller than VP, for example, with subjects, objects or PPs. Data that point to this conclusion are provided hi (17). Observe that the proposed constituent [only XP] occurs in the first position of what is plausibly a V2-construction in (17b), and in the Focus position of a cleft construction in (17c)), two positions that have been widely regarded as reliable diagnostics for constituency. (16) (17)
John can [only [play the piano] a. John can play [only [the piano]]. b. [Only [the piano]] can John play. c. It's [only [by working hard]] that we will ever achieve anything.
Now, under the view that the strings within the more inclusive sets of brackets in (17a-c) form constituents, additional already familiar covert operations will need to apply. In (17a), for example, covert raising is unavoidable.5 Just as in the cases we examined in sections 2.2. and 2.3., we are faced with the dilemma of whether this operation constitutes feature raising to Prt, or phrasal raising to SpecPrt followed by selective deletion in the resulting chain. Consider now (18a-c), which are essentially synonymous with (13a-c) respectively, and compare them with the corresponding subcases of (19); for completeness, we have added Romanian examples hi (18d) and (19d), since Romanian does have the land of construction under consideration here. (18)
a. fl re ha parlato solo con la regina. the king has spoken only with the queen "The king has spoken only with the queen.' b. On ne peut plus compter settlement surla reine. one not can more rely only on the queen 'One can no longer rely only on the queen.' c. Man kann sich nicht nur auf die Konigin verlassen. one can REFL not only on the queen rely 'One cannot rely only on the queen.'
5 Kayne (1998) denies the need for covert movement even in such cases. This result is purchased at the cost of a large number of overt movements. For space reasons, we refrain from discussing the implications of his proposal here.
Feature checking meets the Criterion Approach 63 d. Nu se mai poate conta numai pe regina.6 not REFL more can rely only on queen (19)
'It is no longer possible to rely only on the queen.' a. *fl re ha parlato con solo la regina. b. *On ne peut plus compter sur settlement la reine. c. *Man kann sich nicht auf nur die Konigin verlassen. d. *Nu se mai poate conta pe numai regina.
The only observable difference between (18) and (19) is that only is attached to a PP in the former and to the DP complement of P in the latter, so this difference must be somehow responsible for the contrast in acceptability between the two sets of data. What we want to suggest, following Bayer (1996), is that the Op features of only, in contrast to the Op features of the kind of adjectives we studied in sections 2.2. and 2.3., are unable to spread upwards beyond their overt position for reasons that we will make explicit below. If so, in data like (19), either the complement of P or its Op features must raise directly to SpecPrtP or Pit respectively. Note that the unacceptability of the data in (19) parallel the unacceptability of P-stranding in the corresponding languages. In contrast to Italian, French, German and Romanian, P-stranding is possible in English, and, correlatively, data like (19) are also possible. To be entirely accurate, some speakers of English prefer data of the form shown in (18) to data of the form in (19), but data of the former kind are nonetheless often used in English, while hi German, French, Italian and Romanian, they do not seem to be ever used.7 In support of the claim just made, we offer a sample of attested English data parallel to (19); these have appeared hi print, and thus cannot be dismissed as mere speech errors. (20)
a. George Lakoff has pointed out to me that on the basis [of [only the facts considered so far]], it would be unnecessary to state the Sentential Subject Constraint... (J.R. Ross, 1974) b. ... syntactically appropriate [to [only the non-preferred reading of the ambiguity]] (P. Gorell, 1987) c. It has scope [over [only the matrix-clause element it binds]] (A. Davidson, 1988) d. The computation 'looks [at" [only F and a sublabel of K]] (N. Chomsky, 1995)
6
Owing to a strange and ill-understood requirement of Romanian grammar, regina is, and in fact, must be construed as definite despite the absence of the enclitic definite article. By and large, if the complement of a P consists of a single word, in particular, a noun, the enclitic is morphologically suppressed and semantically "understood." If the complement of P consists of more than one word and the enclitic can in principle occur, a definite reading is obtained, only if the enclitic is morphologically present. This phenomenon is not found with all prepositions; for example, cu 'with' constitutes an exception. 7
See also Ross & Cooper (1979) for relevant judgments.
64
Comparative studies in Romanian syntax
Another set of data where restrictions on scope parallel restrictions on extraction is found with leftbranch constituents. Consider the data in (21-22). (21)
a. Which student's canary did you poison?
(22)
b. *Which student's did you poison canary? a. (?)[[JOHN's canary] only] have I poisoned. b. *[{[JOHN's only], [JOHN onfy]'s } canary] have I poisoned.
While for various speakers (22a) may suffer from the postposition of only (witness that the test cannot be made with Focus-preceding only), the example seems to be clearly better than those in (22b). We will consider the (possible) implications of the facts in (18-22) after making a proposal on the factors which, in our view, block the upwards spread by local steps of the Op features of the italicized items. According to Bayer (1996), particles such as only cannot acquire categorial status unless they appear in a potential scope position, which, in core cases like (13), is a sister to VP.8 The reason for this is that particles, in contrast to functional elements such as D and C, lack inherent syntactic categorial features, an assumption that is motivated by their highly promiscuous behavior, that is, by their ability to combine with virtually any potentially phrasal category. Building on an idea in Rothstein (1991), one may say that only is a "minor" functional category, that is, a category that has no categorial FFs to project, but only semantically-oriented ones. As far as their external syntax goes, phrases of the form [Pit XP] or [XP Pit] have the essential distribution of XP. Within the framework we proposed in section 1.1., we can express this state of affairs by assuming that Prt has a completely unspecified interpretable categorial feature, which needs to derive its specification from the categorial FFs of XP; if we furthermore assume that the unspecified categorial FF of Prt can optionally be strong, we can account for [XP Prt] orders by assuming overt raising of XP to the particle's Spec. Now, just as a [+wh] DP or a [+neg] QP fails to be a question or a negative statement all by itself [Prt DP] does not amount to a Focus construction all by itself. The Op feature inherent in Prt can only play a role in the semantics, and this is only possible if a suitable scope position has been reached. Thus, a PrtP is a phrase headed by an Op feature inherent in Prt. According to this reasoning, [only [the piano]] in (17a,b) is simply a DP with an unlicensed Op feature, while [only \play the piano]] in (16) is categorially a VP and semantically a PrtP in virtue
8
We will not consider here examples in which only takes scope over non-propositional domains, as, for example, in (i) and (ii), where only functions as a scalar Op. (i) John could see only three/few of the soldiers. (ii) The relatives of only three/few soldiers showed up. For discussion and for a proposal on how to unify the quantificational and scalar uses of only, see Bayer (1996, chapter 2)-
Feature checking meets the Criterion Approach 65 of the fact that the Op feature inherent in Pit can now have appropriate scope. Morphologically, particles are invariant elements, that is, elements which lack an inflectional paradigm altogether. From the perspective of Grimshaw (1991), only cannot even abstractly agree in categorial features with N, since it has no categorial features (by assumption). Translated into a feature-checking framework, this generalization takes the following form: only lacks not only inherent (interpretable) categorial features, but also uninterpretable categorial features. When combining with a DP, it cannot attract the interpretable N-feature that has raised to D, so that in cases like (19), D+N can in principle raise to P only directly, without adjoining to only. If so, the Op feature of only is unable to get a "free ride" on an independent raising process, and thus cannot reach P(P). The Op feature of the particle will then reach the conceptual-semantic interface without being able to attain a suitable scope position, and the derivation will crash due to a violation of the Principle of Full Interpretation (Chomsky 1986). Let us now return to the parallelism between restrictions on overt movement and restrictions on the scope of particles that were noted in connection with (18-22), and let us address it in conjunction with the effects reported on in Bayer (1998, 1999) which we already alluded to in section 1.2. Given Chomsky's (1993) hypothesis that constraints on movement apply in both the overt and the covert mode, the parallelism just noted points to the conclusion that (18-20) and (22) involve long distance covert movement. Bayer (1998) observes that some speakers find (16) with focus on the piano more felicitous than (17a). The difference is subtle and seems to be more pronounced in pairs such as They didn 't call any student at home and They called no student at home. As also reported there, preliminary experimental work using speeded grammaticality judgements and the self-paced reading technique suggests that comparable distinctions are reflected in on-line comprehension. This ties in with the conclusion that QR is a marked operation. Reinhart (1993, 1997) and Fox (1994, 1995) urge, on the basis of different data, the even stronger conclusion that QR is permitted just hi case it yields a reading that is not attainable without movement; unless further qualifications are added, this position would make the factually incorrect prediction that data like (17a) are ruled out, since, given that Focus is confined to the phrase the piano, (16) achieves a non-distinct reading without movement. Be this as it may, what matters for our purposes is that data that involve QR have a more or less marked status, and thus stand in contrast with data in which covert operations on features have either no semantic import or no semantic motivation; an illustration of the former is provided by English sentences without auxiliaries, where, according to Chomsky (1995), V-features covertly raise to T and Agr; illustrations of the latter involve feature percolation hi DP and PP as already discussed. Bayer (1998) sought to account for the distinct status of the two kinds of processes by taking the position that processes like V-Raising hi English and percolations of the sort described above are instances of FF-Raising (i.e., Move-F with Pied-Piping of the remaining FFs), while QR is an instance of phrasal raising. Let us point out, however, that with respect to the assumptions
66
Comparative studies in Romanian syntax
about bare phrase structure and covert movement, the term "phrasal movement" may be misleading from the outset. If Greed-driven covert movement is semantically motivated, the element to be moved is an amalgamation of [+int] features, not a morphosyntactic entity such as a DP or a PP. The question is then how much will be moved covertly at a time. Chomsky suggests that it is a single feature that may under certain conditions Pied-Pipe other features. We feel, however, that this issue bas to be resolved empirically. There is, for example evidence that covert movement may undo Principle-C effects.9 If so, a plausible reason may be that the semantic structure to be raised can involve more than simply the feature that heads the QP. Seen from this perspective, covert movement of an Op may be the movement of a [+int] feature that can carry along other [+int] features, essentially giving rise to LF Pied-Piping. The gist of the proposal in section 1.1. was, recall, that both non-semantically and semantically oriented operations on features can be analyzed as driven by the need to ensure convergence at the LF interface, although this goal is achieved hi distinct ways. Now, the data in (2-3), which we analyzed in terms of the semantically oriented mechanism (A'), do not have the marked status of QR constructions. The crucial difference between the two kinds of situation in which mechanism (A') is activated is that hi (2-3), the configuration needed for activating it is achieved hi virtue of Merge, while in data with QR, a suitable configuration can only be achieved by a self-propelled operation of Move. The constructions with marked/unmarked status can then be characterized as follows: Operations dictated by the need to eliminate [-hit] features, or by the need to identify un(der)specified interpretable features hi configurations created by Merge, are unmarked; operations of the latter kind that also appeal to Move are marked. To avoid any possible confusion, we hasten to stress that we do not predict marked status for just any construction hi which phrases have wider scope than then" overt position, but only for those that must resort to QR; for some discussion of wide-scope effects that can circumvent appeals to QR (see Reinhart 1997 and Tsai 1994 among others). In conclusion, we propose the (A) mechanism as an operation which is entirely driven by nonsemantic forces but which enables semantically relevant formal features such as Op features to surface in places which do not count as islands hi spite of their morphophonological locatedness hi the string. We propose hi addition an (A') process of covert movement which is entirely driven by semantic forces. This latter process comes close to the familiar QR-style movement, although it must be seen somewhat different, given the Minimalist reinterpretation of linguistic theory. Two aspects
9 See, for instance, Gueron & May (1984) and May (1985, chapter 4) on extraposed result clauses with so: (i) a. *I told her, that the concert was attended by many people last year who made Mary, nervous b. I told her, that the concert was attended by so many people last year that I made Mary, nervous See also the discussion of antecedent-contained deletion, especially conclusions about the construction's status in the context of Minimalism (see Kennedy, 1997,684ff).
Feature checking meets the Criterion Approach 67 are relevant: (a) it can be naturally integrated into the Spec-head agreement mechanism of the criterion approach; the earlier adjunction approach can perhaps entirely be dispensed with, at least if the Beghelli & Stowell approach proves to be equivalent; (b) self-propelled covert movement may be movement of [+int] features, but there is no necessity of assuming with Chomsky that it is nothing but the movement of a single feature. To the extent that it makes any sense to speak of covert XP-movement, that is, movement of the semantic side of XP, we concede that covert movement may also be covert Move-a. We think this will do as much harm as taking Move-F as covert head movement. As far as we can see, whatever account of language-specific P-stranding options is assumed can be extended to data like (19-20) regardless of whether the displacement process affects phrases or just FFs. As for the deviance of (22b), it can be attributed either to the fact that QR purports to reorder a non-constituent (i.e., to the factor which, according to Chomsky 1995, 263, blocks data like (21b)), or to the fact that only purports to combine with a nonconstituent
4. DP-EXTERNAL ADJECTIVES We now turn to constructions of type (HI). These are found in French and Romanian, but not in Italian, and despite initial appearances to the contrary, they do not exist in English or German either. Descriptively, they exhibit the same kind of agreeing, non-modifiable adjectives with Op import as constructions of type (I), with the notable distinction that these adjectives sit outside DP, rather than within it. Just like the non-agreeing particles discussed in section 3., they can hi principle either precede or follow DP, subject to certain conditions of "relative heaviness" that will be made explicit below. We provide a first illustration of type (HI) constructions in (23), using both French and Romanian data. (23)
a. Seul-e la reine/la reine seul-e peut encore nous aider alone-FEM.SG the queen the queen alone can still us help 'Only the queen can still help us.' b. Singur-a regm -a / regina singur-a ne mai poate ajuta. alone-FEM.SG queen-the-FEM.so queen-the alone us still can help 'Only the queen can still help us.'
While type (ffl) constructions have remarkably parallel properties hi both languages, we will use data from both languages in what follows, and this, because each language sheds light on certain aspects about which the other is silent. Thus, the fact that the adjective agrees with DP in number and gender is only orthographically detectable in French, but in Romanian, it is phonologically detectable as well. At the same tune, French has cleft and complex inversion constructions, which
68
Comparative studies in Romanian syntax
Romanian lacks, and these give a more complete picture of the privileges of occurrence of the items that form the focus of this section. The data in (23) illustrate the fact that the adjectives under consideration agree with their DPs, as well as the feet that they can both precede and follow the latter. The fact that they are DP-external is extremely clear in pre-DP position, since they precede the definite article in French and do not host the enclitic definite article in Romanian, as DP-initial adjectives do. To illustrate the last point, observe that in (24a), which is not a type (III) construction, it is the first of the two pre-nominal adjectives that hosts the enclitic, while in (24b), which is a type (III) construction, the enclitic is hosted by the second adjective. (24)
a.
b.
Interesant-a nouaideea reginei.... interesting-the new idea of queen-the 'The interesting new idea of the queen...' Singura nobil-a noastra regina... alone noble-the our queen 'Only our noble queen ...'
DP-external status can also be demonstrated with post-DP adjectives. Thus, in French and Romanian, DP-internal post-nominal ordinary adjectives may eidier precede or follow a complement of N, depending on its "heaviness". That is, the adjective needs to be quite heavy in order to follow the complement, as illustrated in (25-26). (25)
(26)
a. La secretaire intelligente du roi... the secretary-FEM intelligent-FEM of-the king 'The king's intelligent secretary...' b. *La secretaire du roi intelligente... c. La secretaire du roi si belle et intelligente... the secretary-FEM of-the king so beautiful-FEM and intelligent-FEM a. Secretara inteligenta a regelui... secretary-the-FEM intelligent-FEM of king-the-GEN 'The king's intelligent secretary...' b. * Secretara regelui inteligenta... c. Secretara regelui atat de frumoasa §i inteligenta... secretary-the-FEM king-the-GEN so beautiful-FEM and intelligent-FEM
hi contrast, the kind of adjectives under consideration here can absolutely not occur immediately before a post-nominal N-complement, but can occur after such a complement (subject to certain
Feature checking meets the Criterion Approach 69 conditions; see below). The examples in (27a) and (28a) demonstrate that the only possible reading that emanates from these sentences is the irrelevant non-quantificational reading. (27)
(28)
a. La fille seule du roi peut encore nous aider. the daughter alone of-the king can still us help 'The unaccompanied daughter of the king can still help us.' b. ?La fille du roi seule peut encore nous aider. the daughter of-the king alone can still us help 'Only the king's daughter can still help us.' a. Fiica singura. a regelui ne mai poate ajuta. daughter-the alone of king-the-GEN us still can help 'The unaccompanied daughter of the king can still help us.' b. ?Fiica regelui singura ne mai poate ajuta. daughter-the king-the-GEN alone us still can help 'Only the king's daughter can still help us.'
Having argued for the DP-external position of the adjective and having touched on matters of relative heaviness, we wish to show that comparable considerations of relative heaviness play an important role in determining the acceptability of A+DP and DP+A constructions. This is not immediately apparent in (23), where A and DP are presumably balanced for heaviness, but becomes detectable in data like (27b) and (28b), which are only acceptable if the italicized item is intonationauy separated from the preceding DP and carries relatively heavy stress (two factors that are usually taken to contribute to "heaviness"). In contrast, if the order of DP and A is reversed, acceptability is straightforward and unproblematic. Importantly, if the length and complexity of DP is increased to a sufficient degree, stress and intonational insulation of A are no longer sufficient to salvage the construction, as illustrated below: (29)
a. *La fille aihee de notre noble roi seule peut encore nous sauver. the daughter eldest of our noble king alone-FEM can still us save b. *Fiica cea mare a nobilului nostru rege singura. ne mai poate salva. daughter the eldest of noble-GEN our king alone-FEM us still can save 'Only the eldest daughter of our noble king can still save us.'
Hard to pin down as the notion "heaviness" may be, it has sometimes been noted that non-referential nominals (in particular, indefinite ones) count as heavier than referential ones (Postal 1974). In keeping with this observation, the order A+DP is preferred when DP is indefinite: (30)
a. {Seul un miracle /?*un miracle seul} peut encore nous sauver.
70
Comparative studies in Romanian syntax alone a miracle a miracle alone can still us save b. {Singura o minune /?*o minune singura} ne mai poate salva. alone a miracle a miracle alone us still can save 'Only a miracle can still save us.'
Finally, definite personal pronouns, which generally count as quite light, are usually unacceptable in the order A+DP: (31)
a. {Elle seule l*seule elle} peut encore nous sauver. she alone alone she can still us save b. {Ea singura. /*singura ea} ne mai poate salva. she alone alone she us still can save 'Only she can still save us.'
Nonetheless, as Georges Rebuschi (p.c.) pointed out to us, the deviant versions of (31) become acceptable is the pronoun is heavily stressed and used deictically, that is, accompanied by a pointing gesture (something that arguably turns it into a proper name). In short, to the extent that the notion "heaviness" is understood, the acceptability of A+DP vs. DP+A orders is determined to a significant extent by relative heaviness. A further property of the construction under consideration that takes us closer to proposing an analysis is that the adjective, just like the homonymous items of constructions of type (I), resist modification even by items that make conceivable semantic sense. Thus, the French and Romanian items tout and complet respectively, which function essentially like the English item all in she was all alone in the room, are completely impossible in (23), (27-28) and (30-31). On this basis, we propose that the operator adjectives of type (HI) constructions are functional heads, much like those of type (I) constructions. The similarity between the adjectives of type (I) and type (HI) constructions goes in fact one step further: both belong to the nominal extended projection (although, of course, to different levels of that projection). Thus, in contrast to the categorial "promiscuity" of the particles of type (II) constructions, the adjectives of type (HI) constructions seem to attach to DPs only. We illustrate this contrast between type (II) and type (IE) Ops with respect to predicative nominate, which are NPs, not DPs. (32)
a. Jean Jean b. Ion e
est {seulement l*seul} ouvrier. is only alone-M-SG worker {numai / *singur] muncitor.
Feature checking meets the Criterion Approach 71 Ion is only alone-M-SG 'John is only a manual worker.'
worker
We are now ready to propose a structure for type (III) constructions. We submit that these differ minimally from type (I) constructions in that the functional projection headed by the adjective is below DP in the latter and above DP in the former; this is schematically shown in (11) (reproduced below) and (33) respectively.
(11) WP
x-^\
w a a lie
scla seule alleinige
NP 1 1 Maria reine Konigin
(33)
seule singura
la -a
We have used different labels for the two functional categories because the items that head them exhibit some important differences, despite their many similarities. After noting one further similarity, we turn to the differences. Since both Ws and Zs exhibit the same adjectival morphology and inflectional properties, we see no reason to assume a difference in their interpretable categorial features. Furthermore, in order to capture the fact that both are members of the nominal extended projection, we need to extend to Z the kind of assumption we proposed to make about W in section 2.2.: that Z includes an uninterpretable N-feature (or D-feature) which forces the interpretable categorial feature of N that
72
Comparative studies in Romanian syntax
has covertly reached D to further raise to Z. We will not consider in detail the mechanisms that ensure morphological agreement within the nominal extended projection (see footnote 2), but it seems to us that whatever mechanisms are responsible for agreement in (11) can also be assumed to be operative in (33). Having outlined a number of properties shared by W and Z, we now address the difference alluded to two paragraphs earlier, which concerns the distributional privileges of WPs and ZPs. Whereas the former can occur in overt representations both in positions that do and in positions that do not have scope over the clause's CFC, as can be seen by examining (7), (8) and (13), the latter are only felicitous in positions that do have such scope. We proceed to support this claim. In addition to the pre-verbal subject position, type (III) constructions may occur in cleft-Focus position and in the Topic position of complex inversion constructions, as illustrated in (34-35). In view of the fact that Romanian does not have such constructions, we are forced to limit our illustrations to French:10 (34)
(35)
10
a. C'est {seule la directrice du conseil d'administration, it is alone the directress of-the council of-administration la reine seule} qui est encore en mesure de nous aider, the queen alone who is still in measure of us help 'It is (only the directress of the administrative council, the queen only} who is still in a position to help us.' b. C'est {seul un ami de son fils, son filsseul} que it is alone a friend of his son his son alone that le roi est encore pret a recevoir. the king is still ready to receive 'It is {only a friend of his son, his son only} that the king is still willing to see.' {Seule la reine, la reine seule} est-elle encore capable de nous aider? alone the queen the queen alone is-she still capable of us help 'Is it only the queen who can still help us?'
We have used heavier DPs with the order A+DP because data like (i) are unacceptable. (i) *C'est5ew/e la reine qui peut encore nous sauver. it is only the queen who can still us save We conjecture that the Focus position of clefts, which is an automatic stress position, is more sensitive to relative heaviness than normal argument positions, and also that seule is presumably somewhat heavier than la reine. This distinction seems not to be significant enough to block the A+DP order in data like those in (23).
Feature checking meets the Criterion Approach 73 In contrast, type (III) constructions may not function as post-verbal subjects, direct objects, Pobjects/verbal PP complements, and foci of pseudo-cleft constructions, as illustrated in (36-39) respectively. (36)
(37)
(38)
(39)
a. Le palais ou reside {(*seule) la reine, la reine (*seule)} the palace where lives alone the queen the queen alone b. Palatul in care locuie§te {(*singura) regina, regina (*singura)}... palace-the in which lives alone queen-the queen-the alone 'The palace in which only the queen lives...' a. Le roi est pret a recevoir {(*seul) un ami de son fils, son fils (*seul)} the king is ready to receive alone a friend of his son his son alone b. Regele e gata sa primeasca {(*singur) un prieten al king-the is ready SA receive alone a friend of fiului sau, pe fiul lui (*singur)}. son-the-GEN pe son-the his alone 'The king is willing to see {only a friend of his son's, his son only}.' a. J'ai envoye cette lettre {(*seule) a notre reine, I have sent this letter alone to our queen a (*seule) notre reine, a notre reine (*seule)}. to alone our queen to our queen alone 'I sent this letter (only) to the queen.' b. M-am uitat {(*singura) la regina noastra, me-have-lso looked alone at queen-the our la (*singura) regina noastra, la regina noastra (*singura)}. at alone queen-the our at queen-the our alone 'I looked (only) at our queen.' a. Ce que je reussis a voir est({seulement, *seule}) la surface du lac (*seule)}. that which I manage to see is only alone the surface of-the lake alone b. Ceea ce reu§esc sa vad G({numai, *singura}) that-which manage-1SGSA see-suBJ.lSG is only alone suprafaja lacului (*singura). surface-the lake-the-GEN alone 'What I manage to see is just the surface of the lake.'
Particularly significant are the facts that (34b), which is the cleft version of the unacceptable full versions of (37a), is acceptable, and that ZPs in pseudo-cleft Focus position (which, unlike the cleft Focus position, is not a CFC-scope position) are unacceptable. In short, the contrast between (23) and (34-35) on the one hand and (36-39) on the other unmistakably points to the generalization that
74
Comparative studies in Romanian syntax
ZPs need to have scope over the clause's CFC in overt representation. More carefully put, ZP's need to occur in positions where tampering with their chain (see step (iii) of section 2.2.), if necessary, enables Z to take CFC-scope.11 The italicized qualification is motivated by the existence of acceptable constructions in which Z and DP are discontinuous in overt representation, and in which Z has CFC-scope:12 (40)
(41)
(42)
a. La reine a seule 1' autorite de dissoudre le parlement. the queen has alone the authority of dissolve the parliament 'Only the queen has the right to dissolve the parliament.' b. ?La reine peut seule nous aider dans cette situation difficile. the queen can alone us help in this situation difficult 'Only the queen can still help us in this difficult situation.' a. Regina are singura. dreptul sa dizoh/e parlamentul. queen-the has alone right-the SA dissolve-SUBJ.SSG parliament 'Only the queen has the right to dissolve the parliament.' b. ?Regina mai poate singura sa ne ajute. queen-the still can alone SA us help-SUBJ. 3 so 'Only the queen can still help us.' a. Seule peut encore nous sauver notre grande et noble reine. alone can still us save our great and noble queen
11 The kind of tampering that takes place in cleft constructions like (34) depends on one's analysis of clefts. If the Focus phrase is reordered from within IP, as proposed, for example, in Kayne (1994), we have a straightforward chain, and tampering has its usual sense. If the Focus phrase is base-generated in its surface position, as proposed, for example, in Chomsky (1977), some extension of the notion "chain" appears to be needed. In any event, some form of reconstruction is certainly needed in clefts in general and in data like (i) in particular. (i) C' est seule une photo de lui-meme it is only a picture of himself que le roi serait pret a accepter that the king would-be ready to accept To satisfy Condition A of the Binding Theory, une photo de lui-meme must be reconstructed in the position of object of accepter, while seule needs to retain its surface position under a base-generated analysis, or undergo reconstruction to a position that has CFC-scope under a movement analysis. 12 The difference between the (a) and (b) examples in (40) and (41) seems to rest on the fact that meaningful verbal elements need to be within the scope of the Op in overt representation. Thus, main verbs can certainly raise overtly in French and Romanian (Pollock 1989), but if they raise to the left of the Op in constructions like (23-24), the result is degraded, as shown below: (i) ?*La reine boit seule du Dom Perignon. the queen drinks alone some Dom Perignon 'Only the queen drinks Dom Perignon.' (ii) ?* Regina bea singura sampanie de calitate. queen-the drinks alone champaign of quality 'Only the queen drinks high quality champaign.' The (b) sentences in (40-41) are thus arguably somewhat harder to accept than the (a) sentences because modals have more content than the verbs be/have.
Feature checking meets the Criterion Approach 75 b. Singuranz mai poate sah/a marea §i nobila noastra regina. alone us still can save great-the and noble-the our queen 'Only our great and noble queen can still save us.' The discontinuities in (40-41) arise due to the combined effect of overt V-Raising out of VP and DP-Raising out of ZP, stranding Z in SpecVP. The discontinuity in (42) is arguably due to a stylistic process that post-poses DP out of ZP. We note that (42) is distinct from the more usual French/Romanian data with post-verbal subjects, such as (36), because, in contrast to the latter, (42a-b) belong to a very elevated stylistic register, and the post-verbal subject must moreover be quite heavy and intonationally insulated from the preceding context. The distributional restrictions on ZPs we have just noted are straightforwardly accounted for is we assume that the Op features of ZP, in contrast to those of constructions of type (I) and (II), are, for some reason, unable to undergo QR. On the background of our account of data of type (la) and (Ib) there is a natural solution. Recall that (la), La sola Maria si e presentata, is licit because FF(so/-) can travel to the top node of DP, and since DP is presumably in a proper scope position by virtue ofbeing in subject position, the Op feature corresponding to sola is licensed. Were the DP in question in a non-subject position, it would have to undergo covert movement in the sense of QR. In (Ib), Only Mary showed up, the particle was said to be a head which turns the whole phrase into a certain semantic entity (essentially, a quantified phrase), while being syntactically inert. Since, according to our assumptions, only does not have a syntactic feature that could project, it is the DP which projects. Again, if this DP is in subject position, the particle has scope over a CFC. If it is not, [xp only XP] has to undergo covert movement to SpecPrtP essentially by identifying the head of PrtP. Consider now (Ic), {SinguraMaria, Maria singura} s-aprezentat, and assume the DP in it is a ZP of the kind shown in (33). What is this ZP like? It consists of a head, singura, which is an inflected adjective, and a DP which agrees with it at least in gender and number. In terms of bare phrase structure, it differs hi one important way from [DP only DP]: It is headed by Agr. With respect to checking, it also differs in one important way from [DP D [sol-Agr NP]: The DP-external agreeing head must have a D-feature which attracts and is checked either overtly by the complement DP or covertly by FF(DP). These two options, which we will not explore in more detail, are attested by the two basic orderings between the external Op and the DP. The DP-internal agreeing [+Op] adjective (or its Agr-part), on the other hand, must have an N-feature which attracts FF(NP). Consider now (11) and (33) which we repeat here for convenience.
76
Comparative studies in Romanian syntax
(11)
DP
D
WP W
la la die
sola seule alleinige
NP Maria reine Konigin
(33)
In both cases, we view Move-F as a semantically blind morphological process. In (11), FF(NP) raise to W, and FF(WP) raise to D. Both processes are driven by the need to check the respective head's N-feature as encoded in the Agr of D. In the second process, the [+Op] feature of sol- gets a free ride to D, and will, as a consequence, be present at the DP-node. In (33), FF(DP) raise to Z, unless DP raises overtly, which also gives an attested result as we have seen. In either case, Z has a Dfeature which is checked offby FF(DP). Assume now that in order to undergo QR, XP must be such that its head bears the [+Op] feature. The two familiar cases are the type (II) construction and the type (I) construction: the relevant head hi (n) is Prt; the relevant head in (I) is a derived feature complex that appears hi the functional categories D or P. Construction type (in), as shown in (33), differs from both of these cases because Z, seule/singura, is strictly speaking not a functional head. Its functional part is Agr. Checking this featural complex by raising FF(DP) will not have the required semantic effect because now Agr will disappear without [+Op] becoming a feature of the formal head of the phrase. We think that it is precisely this lack of formal marking which prevents ZP to undergo semantically motivated movement to a pre-VP scope position. Notice now that seule/singuradoQS contain [+Op] albeit not as its formal head. If ZP has moved to the canonical preVP subject position as in (23), (30) and (31) or if Z has been stranded in such a position as hi (4042), and if this position is sufficiently high to enable [+Op] to take scope over the CFC VP, then QR becomes unnecessary. The scope requirement is accidentally met because of ZP's appearance in
Feature checking meets the Criterion Approach 77 subject position. What about clefting and complex inversion as exemplified in (34) and (35) respectively? All we can say at this moment is that any account of these constructions hi French will have to guarantee that [+Op] is ultimately copied into a position higher than VP. To summarize, our account of the pre-/post-V asymmetry observed hi connection with the type (IE) construction rests on the idea that QR presupposes formal marking of XP with [+Op], and that under close scrutiny such formal marking is absent hi ZP.'J Then, those ZPs which enable [+Op] to take scope must have moved to a scope position by a process that is independent of the motivations underlying QR. Before concluding, we will take a look at some data of English and German that we briefly alluded to at the beginning of this section. The data, which are illustrated hi (44), share superficial properties with both type (II) and type (IE) constructions. (44)
a. (The queen alone, *alone the queen} can still help us.14 b. {Die Konigin allein, allein die Konigin} kann uns noch helfen. the queen alone alone the queen can us still help
Just like the operator elements of type (HI) constructions, the italicized elements hi (44) are homonymous with items that mean 'unaccompanied' or 'unaided.' On the other hand, these elements are uninflected, just like the particles of type (n) constructions. The latter property is not significant for English, where adjectives are always uninflected, but it is potentially significant for German, where adnominal adjectives are inflected. To see whether the constructions hi (44) are of type (II) or (HI), we need to check then" distributional privileges hi post-verbal contexts. The following data show that they are allowed hi such contexts, and thus point to the conclusion that they are of type (H): (45) (46)
a. This proposal concerns the queen alone. b. This remark was meant for you alone. a. Dies interessiert {die Konigin allein, allein die Konigin}. this concerns the queen alone alone the queen b. Ich habe dieses Buch {der Konigin allein, allein der Konigin}gezeigt I have this book the-DAT queen alone alone the-DAT queen shown
13 Claiming that +Op is missing in the formal syntactic representation of (33) and that this fact prevents ZP from undergoing QR does not imply that +Op is the only trigger for QR. As we pointed out already, there are reasons to believe that QR can also apply to XPs which are not quantified in the narrow sense. What is at issue here, however, is the fact that ZP will fail to activate a PrtP because formally it does not amount to more than a DP. 14 In section 3., we proposed that the unspecified categorial FFs of operator particles that can both precede and follow XP need to be allowed to be either strong or weak. The English operator particle alone has necessarily strong unspecified features, since it is only allowed in post-XP position.
78
Comparative studies in Romanian syntax 'I have shown this book to the queen alone.' c. Dies war {fur die Konigin allein, allein fur die Konigin} bestimmt. this was for the queen alone alone for the queen meant 'This was meant for the queen alone.'
5. SUMMARY OF RESULTS In this paper, we have addressed three syntactic constructions with comparable semantic import, but with distinct distributional properties and distinct cross-linguistic privileges of occurrence. The constructions (1-13), which were discussed at some length hi Bayer (1996), have been partly reanalyzed here; construction (HI) has been, to the best of our knowledge, neither described nor analyzed in the earlier generative literature. We have proposed to analyze their intra-linguistic and cross-linguistic distributional properties in terms of differences in the featural make-up of lexical items with operator import that individual languages may or may not have. In particular, we proposed that in constructions of type (I) the Op's features may percolate up the phrase marker as a consequence of purely formal and non-semantic checking operations. Constructions of type (II) have only Op features, and that these are forced to undergo movement to a scope position unless such movement is blocked by known constraints on Move-a. Type (HI) constructions reveal an asymmetry between pre- and post-verbal position by virtue of the fact that their Op feature is not formally represented in a way that would permit QR-style movement to a scope position. A semantically well-formed structure can only be achieved in those cases where the relevant phrase is already hi a potential scope position for reasons that have nothing to do with semantics proper. In analyzing these various constructions, we pointed to the need to recognize two types of covert movement operations, one purely formal and one semantics-oriented. The present results support conclusions about two types of covert movement that were reached in Bayer (1996, 1998, 1999), but they clearly go beyond that in presenting a more fine-grained picture of the interaction between the syntax of inflectional morphology and semantic interpretation. In doing so we have shown that important insights from both the Criterion approach by Rizzi (1991,1997) and the feature-checking approach of Chomsky (1995) can be successfully integrated into a more unified account.
Feature checking meets the Criterion Approach 79
REFERENCES Bayer, J. (1996). Directionality and Logical Form. On the scope of Focusing particles and Wh-insitu. Kluwer, Dordrecht. Bayer, J. (1998). Two types of covert movement. Paper presented at the Workshop on acquisition and variation in syntax and semantics, Trieste, SISSA. Bayer, J. (1999). Bound Focus or: How can association with Focus be achieved without going semantically astray. In: The Grammar of Focus (G. Rebuschi and L. Tuller, eds.). John Benjamins, Amsterdam. Bech, G. (1955/57). Studien Ueber das deutsche Verbun infmitum. Copenhagen. Reprinted in 1981 by Niemeyer, Tubingen. BeghelU, F. and T. Stowell. (1995). Distributivity and negation. In: Ways of Scope Taking (A. Szabolsci, ed.), pp. 71-107. Kluwer, Dordrecht. Brody, M. (1998). Projection and phrase structure. Linguistic Inquiry 29, 367-398. Chomsky, N. (1977). On wh-movement. In: Formal Syntax (P. Culicover, T. Wasow and A. Akmajian, eds.), pp. 71-132. Academic Press, New York. Chomsky, N. (1986). Knowledge of Language. Praeger, New York. Chomsky, N. (1993). A Minimalist Program for linguistic theory. In: The View from Building 20 (K. Hale and S. J. Keyser, eds.), pp. 1-52. MIT Press, Cambridge, Mass. Chomsky, N. (1995). The Minimalist Program. MIT Press, Cambridge, Mass. Cinque, G. (1994). On the evidence for partial N-Movement hi the Romance. In: Path towards Universal Grammar (G. Cinque; J. Koster; J-Y. Pollock; L. Rizzi and R. Zanuttini, eds.), pp. 85-110. Georgetown University Press, Washington, DC. Davison, A. (1988). Operator binding, gaps and pronouns. Linguistics 26, 181-214. Fox, D. (1994). Quantifier scope and VP-ellipsis. Manuscript, Cambridge: MIT. Fox, D. (1995). Economy and scope. Natural Language Semantics 3, 283-341. Galhnann, (1996). Die Steuerung der Flexion in der DP, Linguistische Berichte 164, 283-314. Gallmann, (1997). Zur Morphosyntax und Lexik der w-W6rter. Sprachtheoretische Grundlagen fur die Computerlinguistik. 340. Stuttgart: Arbeitspapier #107 des SFB. Gorell, P. (1987). Studies of human syntactic processing: ranked parallel versus serial models. Doctoral dissertation, University of Connecticut, Storrs. Grimshaw, J. (1991). Extended Projection. Manuscript, Brandeis University. Grosu, A. (1994). Three studies in locality and Case. Routledge, London. Gueron, J. and R. May. (1984). Extraposition and Logical Form, Linguistic Inquiry 15, 1-31. Haider, H. (1993). Deutsche Syntax - generativ: Vorstudien zur Theorie einer Projektiven
80
Comparative studies in Romanian syntax
Grammatik. Gunter Narr Verlag, Tubingen. Horastein, N. (1995). Logical Form. From GB to Minimalism. Blackwell, Oxford. Kayne, R. (1994). The antisymmetry of syntax. MIT Press, Cambridge, Mass. Kayne, R. (1998). Overt versus covert movement. Syntax 1, 128-191. Kennedy, C. (1997). Antecedent-contained deletion and the syntax of quantification. Linguistic Inquiry 28, 662-688. Kiss, K. E. (1991). Logical structure in syntactic structure: The case of Hungarian. In: Logical structure and linguistic structure (C.-T.J. Huang and R. May, eds.), pp. 117-147. Kluwer, Dordrecht. Koster, J. (1987). Domains and dynasties: The radical autonomy of syntax. Foris, Dordrecht. Lappin, S. (1991). Concepts of logical form in linguistics and philosophy. In: The Chomskyan turn (A. Kasher, ed.), pp. 300-333. Blackwell, Oxford. Larson, R. (1988). On the double object construction. Linguistic Inquiry 19, 335-391. May, R. (1977). The grammar of quantification. Doctoral dissertation, Cambridge: MIT. May, R (1985). Logical Form: Its structure and derivation. MIT Press, Cambridge, Mass. Moritz, L. and D. Valois. (1994). Pied-Piping and Specifier-Head agremeent. Linguistic Inquiry 25, 667-707. Pollock, J.-Y. (1989). Verb movement, Universal Grammar, and the structure of IP. Linguistic Inquiry 20, 365-424. Postal, P. (1974). On Raising. MIT Press, Cambridge, Mass. Reinhart, T. (1991). Elliptic Conjunctions: Non-Quantificational LF. In: The Chomskyan Turn (A. Kasher, ed.), pp. 360-384. Blackwell, Oxford. Reinhart, T. (1993). WH-in-situ in the framework of the Minimalist Program. OTS Working Papers, Utrecht University (published hi 1998 in Natural Language Semantics 6, 29-56). Reinhart, T. (1997). Quantifier scope: How labor is divided between QR and choice functions. Linguistics and Philosophy 20, 335-397. Reinhart, T. (1998). Wh-in-situ in the framework of the Minimalist Program. Natural Language Semantics 6, 29-56. Rizzi, L. (1991). Residual Verb Second and the Wh-Criterion. Technical Reports in Formal and Computational Lingustics 2, University of Geneva. Rizzi, L. (1997). The fine structure of the left periphery. In: Elements of Grammar (L. Haegemann, ed.), pp. 281-337. Kluwer, Dordrecht. Rooth, M. (1985). Association with Focus. Doctoral dissertation, University of Massachusetts at Amherst. Ross, J. R. (1967). Constraints on variables in syntax. Doctoral dissertation. Cambridge: MIT. Ross, J. R. (1974). Excerpts from constraints on variables hi syntax. In: On Noam Chomsky: Critical essays (G. Harman, ed.), pp. 165-200. Anchor Press-Garden City, New York. Ross, J. R. and W. Cooper. (1979). Like Syntax. In: Sentence Processing (W.E. Cooper and
Feature checking meets the Criterion Approach 81 C.T.Walker, eds.), pp. 343-418. Lawrence Erlbaum, Hillsdale, NJ. Rothstein, S. (1991). Heads, projections, and category determination. In: Views on Phrase Structure (K. Leffel and D. Bouchard, eds.), pp. 97-112. Kluwer, Dordrecht. Safir, K. (1986). Relative clauses in a theory of binding and levels. Linguistic Inquiry 17, 663-690. Stechow, A. von and W. Geuder. (1997). Sind manche Sprachen prdziser als andere? Uber morphologische Kategorien und ihre Interpretation. Manuscript, Seminar fur Sprachwissenschaft, Universitat Tubingen. Szabolsci, A. (1995). Ways of Scope Taking(A. Szabolsci, ed.), pp. 109-154. Kluwer: Dordrecht. Tsai, Wei-tien D. (1994). On economizing the theory of A -bar dependencies. Doctoral dissertation, Cambridge, Mass: MIT.
This page intentionally left blank
The double subject construction in Romanian
83
THE DOUBLE SUBJECT CONSTRUCTION IN ROMANIAN Alexandra Cornilescu, University of Bucharest
i. INTRODUCTION In the pre-Minimalist descriptions of Romanian, one finds the following presentations of the Romanian subject: a) The subject is projected VP-internally and may remain there, since nominative Case is assigned under government by the verb raising to Inflection. Alternatively, the subject, which has already been assigned Case in SpecVP, may move to SpecIP, which is a Topic position in Romanian, rather than an argumental, or L-related one. This is the analysis in Dobrovie-Sorin (1987, 1994). b) A different view is defended in Motapanyane (1989). Employing a more richly articulated functional structure, she considers Romanian clauses to be AgrSPs and claims that there are two argumental subject positions, SpecVP and SpecAgrSP. Nominative Case is either assigned in SpecVP under government by the verb in Tense, or in SpecAgrSP by Spec-head agreement. Evidence is provided that SpecAgrS cannot be an A1 position and must be argumental (A-position). Therefore, both authors concur that nominative Case is assigned post-verbally under government by the verb that has raised to I or T, but differ in the interpretation of the pre-verbal subject position, considered either a position of topicalization or an argumental, Case-related position. Several important studies on Romance (Cardinaletti and Roberts 1991; Roberts 1991), have argued for the existence of two AgrSPs, both of them pre-verbal, at various stages in the evolution of various Romance languages (e.g. Old French). Recently Cardinaletti (1996) argues that in UG there may be two AgrSPs, both pre-verbal in Italian. The two positions are specialized. The lower Agr2SP * I would like to thank Ileana Baciu, Larisa Avram Dana Isac and Carmen Dobrovie-for helping me with suggestions and clarifications in writing this paper. Special thanks are also due to the editors and reviewers, who helped me in every way, in matters of content and style. Remaining errors are all mine.
84
Comparative studies in Romanian syntax
is a Case position, as well as the position where ^-features are checked, securing subject-verb agreement. In pro-drop languages like Italian, this is also the position which licenses pro. The higher pre-verbal subject position Ag^SP is also argumental and, in Italian or French, it accommodates strong pronouns and strong DPs, while the weak nominative pronouns remain in the lower Agr2SP. It is also the position which hosts prominent nominals like the dative Experiencer of "psych" verbs. Cardinaletti concludes that this is the position for the semantic "subject of experience" or "subject of predication". Against this background for Romance, we will try to establish the properties of the AgrS projection(s) in Romanian, as well as its/then" relative position with respect to the verb. This amounts to defining the syntactic properties of the subject, as it occurs pre-verbally or postverbally. Evidence will be presented supporting the following claims: a) There is only one AgrSP in Romanian, which is nevertheless post-verbal. Consequently, both SpecVP and SpecAgrSP are post-verbal positions in Romanian. b) With very few exceptions, nominative Case is regularly checked hi SpecAgrSP, which is the nominative Case position of Romanian c) The pre-verbal subject positions) can roughly be described as a non-argumental or non-Lrelated, position(s). Our analysis confirms the findings of Dobrovie-Sorin (1994), although our argumentation differs. More specifically, the pre-verbal subject is left dislocated or focalized. d) This description of the Romanian subject is supported by various morphological and syntactic facts. In section 2., we briefly inspect the morphology of the Romanian verb, outlining the functional structure of the Romanian finite clause. Romanian sentences will be described as Mood Phrases and it will be seen that hi Romanian the verb raises to the Mood head (M), and not higher to C. In section 3., we indicate that there are two post-verbal subject positions: SpecVP, a thematic position, and AgrSP, a post-verbal position to the left of SpecVP. In section 4., we describe a double subject construction, which can be taken to provide crucial evidence not only that AgrSP is postverbal hi Romanian but also that SpecAgrSP is the major nominative checking position hi Romanian. In section 5., we prove that the pre-verbal subject occupies an A1 position (Topic or Focus), so that there is only one AgrS position in Romanian.
2. ON INFLECTION AND VERB MOVEMENT IN ROMANIAN 2.1. Functional Categories of the Romanian Verb This investigation cannot proceed without first defining the skeleton of the Romanian clause,
The double subject construction in Romanian
85
in as much as it is relevant for the syntax of the subject. Like other researchers on Romance and on Balkan languages (Alexiadou 1994; Cinque 1997; Rivero 1994) we accept the Split Inflection hypothesis (Chomsky 1991, Pollock 1988), postulating at least the verbal categories of Mood, Tense, Aspect for the Romanian clause. In her study on verb movement, Rivero (1994), drawing principally on evidence from Balkan languages, provides syntactic evidence in favour of a Mood position. While Rivero (1994) chiefly relies on syntactic facts for assuming the existence of a Mood Phrase (MP) and Tense Phrase (TP), the morphology of the Romanian verb also provides evidence for the categories of Mood, Tense and Aspect, even if one ignores the Agreement features of person and number. It is tempting to initially consider only finite and non-finite inflectional forms, ignoring the analytical tenses and moods. In Romanian, these are as shown in (1) for the verb merge 'go', hi 1st person singular. (1)
Indicative - Present Imperfect Simple Past Past Perfect Subjunctive - Present Infinitive Gerund Past Participle
MERG MERGEAM MERSEI MERSESEM SA MERG A MERGE MERGAND MERS
/'I go' /'I was going' /'I went' /'I had gone' /'(that) I go' /'to go' /'going' /'gone'
Like other Balkan languages, Romanian possesses Mood markers; namely, the particles SB for the subjuncive and a for the infinitive. Examination of the forms listed above reveals the existence of two different stems in the paradigm of this verb: merg- and mers. Since this variation is systematic for regular verbs of the third conjugation, and since it does not seem to be a case of phonological conditioning, (e.g., both merg and mers occur hi front of/E/, as in el merge 'he goes'and el merse 'he went') it is natural to assume that this opposition is the overt manifestation of one of the functional categories hi the lower part of the functional system of the clause, presumably a manifestation of Aspect. We will say that, for a verb like merge 'go', the Aspect system opposes an imperfective verbal stem: merg-, which, for example, appears hi the imperfect, to a perfective verbal stem : mers-, which occurs, for instance, hi the simple perfect. Just as in other Romance languages (e.g. French), the
86
Comparative studies in Romanian syntax
substantive content of this formal contrast is a difference between the [+ durative] imperfective forms, and the [-durative] perfective forms. An intuitively satisfactory result is that the imperfective stem merg- is found in all the tenses/moods which are understood as durative and imperfective: present, imperfect, gerund and infinitive, while the other stem, mers- is found in all those tense/moods which are clearly non-durative and perfective. This contrast is further illustrated in (2a) for the stem merg- and in (2b) for the stem mers-. (2)
a.
b.
tu mergi - PR.2SG /'you go' tu mergeai - IMP.2SG /'you were going' a mergQ - INF /'to go' mergwd - GER /'going' tu mersQi - S.PAST.2SG /'you went' tu mersese$L - PAST.PF.2sG /'you had gone' mers - PAST.PART. /'gone'
The TP will accommodate the features [+/-present, +/-past (anteriority) ]; specifically:Tense —> +/Present, and -Present --> + /- Past (Anteriority). The present and the imperfect will differ in terms of +/- present, while the simple perfect and the past perfect will differ in terms of +/-past (anteriority), an opposition morphologically marked by -se-: el veni 'he came' vs. el venise 'he had come'.Thus, even if the overt realization of Tense and Agreement is normally synchretic, the Tense projection is independently filled by -se-. Ignoring subject agreement features, one obtains the table of the four indicative forms as in (3). Table 3 Tense morphology present
simple past
imperfect
past perfect
merg 'I go'
mersei 'I went'
mergeam 'I was going'
mersesem 'I had gone'
+ present
- present
- present
- present
+ durative
- durative
+ durative
- durative
The imperfect contrasts both with the present, through the feature [+/- present] and with the past perfect, through the feature [+/-anteriority]. Agreement features of [person] and [number] differentiate between any of these forms and non-finite ones. The order of these functional projections, indicated in (4), is assumed to hold cross-linguistically, pending evidence to the contrary, though in some languages some of these functional categories may not project at all (see Alexiadou 1994; Cardinaletti 1996; Cinque 1993a, 1997; cf Ouhalla 1992):
The double subject construction in Romanian
(4)
87
MoodP > TenseP > AspP
Baker's (1985) Mirror Principle is observed, bringing further support for the proposed order. Thus in a form like mersesem (PAST.PERF. Iso) the order of the affixes, and, therefore, of the functional heads, in a derivational rather than checking analysis, is as shown in (5).
(5)
[[[[ mer v] + s(e) Mp] + se T] m^s]
Moreover, Romanian verb forms should always be taken as positively specified for a particular Mood, even in the absence of a Mood particle. For instance, the present subjunctive is formed with the Mood particle sa followed by an inflected form, which is different from that of the indicative, at least in the 3rd person, so that the indicative/subjunctive opposition is overtly marked even when the Mood marker sa is absent. The Mood feature is overtly checked, since the verb raises to M. (6)
a.
Ion citeste cartea. Ion reads book-the 'Ion reads the book.'
(indicative)
b.
(7)
(8)
Ion sa citeasca cartea! (subjunctive) Ion SA read-suBJ.3so book-the 'Ion should read the book!' a. Pre§edintele traie§te. (indicative) president-the lives "The president is living.' b. Traiasca pre§edintele! (subjunctive) hve-suBj.3SG president-the 'Long live the president!' traie§te 'lives': traiasca 'live': +indicative
+subjunctive
+present +durative +3rd person
+present +durative +3rd person
Rivero (1994) proves that in the subjunctive, the mood head must always be lexically filled, either by the particle sa, or by the inflected verb which skips the clitic phrase and fills the empty M head by Long Head Movement (LHM), as in (9).
(9)
a.
Cre§tinii
sa-1
slaveasca
pe Dumnezeu.
88
Comparative studies in Romanian syntax Christians-the SA him glorify-SUBJ.SPLpe God b.
'Let the Christians glorify God!' Cre§tinii slaveasca1 pe Dumnezeu! Christians-the glorify-SUBJ.3PL \ampe God 'Let the Christians glorify God!'
The existence of LHM suggests that Romanian verbs must overtly raise to MP to check their Mood feature. Further evidence that verbs always raise to M in overt syntax will be provided below. We provisionally accept that Case is checked in AgrPs, rather than in the specifiers (Specs) of verbal inflectional categories (TP or AspPs), following Alexiadou (1994), Cardinaletti (1996) and Koizumi (1995), among others. The inflectional system of the Romanina finite verb presents rich person and number marking, providing evidence for an AgrSP. At the same tune, if Baker's Mirror Principle is observed, the morphology of the Romanian subjunctive, in examples like (5), (6) and (9) above, indicates that the position of AgrSP is below MP: (10)
MP > AgrS> TP > AspP.
hi conclusion, the morphology of the Romanian verb indicates that the AgrSP projection is lower than MP, although it precedes TP and AspP. Since the verb raises to MP, AgrSP is post-verbal. The Clitic Phrase (C1P) is between MP and AgrS (i.e., immediately above AgrS), but lower than negation, as suggested by forms like (9), analysed in (11) following Sportiche (1992). (11)
[saM] + [L cl] + [slaveasca ^.... SA him glorify
The clitics are the Case features of the verb (Borer 1984). As it raises, passing through C1P, the verb will take them along and checks its Case features. The clitics (pronominal or adverbial) and the verb form a cluster which behaves like one syntactic X-head unit (see Dobrovie-Sorin 1987, 1994). Negation and the Mood particle sa in the subjunctive will become elements of the same complex head as the verb raises to the relevant projections (i.e. NegP and, finally MP), as illustrated in (12). (12)
a.
Ei [sa nu le
mai citeasca\.
they SA not them more read-sueJ.3PL 'They shouldn't read them any more.' b
[sa J + [ nu N J + [le CJ + [ mai Adv] + [citeasca v] SA not them more read
The double subject construction in Romanian
89
The distribution of a complex head such as [mood particle +neg + clitics+ inflected form] is the same as that of a simple verbal form as far as the subject and the adjuncts are concerned. As shown by examples (13) and (14), adverbs and subjects either precede or follow such complex heads. This indicates that both the complex head and the simple verb form occupy the same position, M. (13)
a.
b.
(14)
a.
b.
Ion maine sanu-l mai trimita acolo pe copil. Ion tomorrow SA not him more send-SUBJ.3SG there pe child 'Tomorrow Ion should not send the child there again.' Ion maine trimite copilul acolo. Ion tomorrow sends child-the there 'Tomorrow Ion sends the child there.' Maine sanu-l mai trimita Ion pe copil acolo. tomorrow SA not him more send-8UBJ.3SG loupe child there 'Ion should not send the child there again tomorrow.' Maine trimite Ion copilul acolo. tomorrow sends Ion child-the there 'Tomorrow Ion sends the child there.'
All these very general considerations support the view that the verb raises overtly and obligatorily to the MP, which is always projected.
2.2. Verb Raising to M: Evidence from Adverb Positions Further evidence regarding verb movement hi Romanian comes from examining the placement of adverbs. Since Pollock (1988), adverb positions have been considered important in establishing how far various heads raise. The general idea is that adverbs check their inherent features either in the specifier position of an appropriate head or in a position of adjunction to a semantically suitable phrase (see Cinque 1993b). In important work on adverbs, Alexiadou (1994) and Cinque (1997) opt for the first solution and state the general principle that (at least some) adverbs are projected as Specs (rather than adjuncts), obeying the following licensing condition: (15) Generalized Licensing Criterion (i) An [ +F] head must be in a Spec-head relation with a [+F] XP. (ii) An [ +F ] XP must be in a Spec-head relation with a [+ F ] head. If [+F] corresponds to ^-features (gender, number, person), then the Spec is an A-position; when the [+F] corresponds to Op features, (+whl +neg/ +Topic), the Spec is an A'-position.
90
Comparative studies in Romanian syntax
We wish to remain neutral on whether adverbs are Specs or adjoined phrases, although we provisionally adopt the first position (i.e., that adverbs are Specs). In either case, however (if multiple adjunction/ specification is disallowed), adverbs may be used to ascertain the positions to which the verb raises. The assumption is that adverb positions are fixed (at LF), since adverbs must agree with appropriate heads / projections which are projected in a rigid word order themselves. We first examine aspectual adverbs, such as regulat 'regularly', zilnic 'daily', mereu 'always', iar 'again', and time adverbs, such as: acum 'now', atunci 'then', curdnd 'soon'. Both series may merge lower in the VP or directly in the Specs of the AspP and TP. What matters is that in Romanian both aspectual and time adverbs may check their features overtly, appearing in the appropriate Specs. The examples below suggest that when both types of adverbs are present the (slightly) preferred order is indeed TP > AspP, as shown by the word order in (16). (16)
a. b. c.
Ion vine
acum la mine zilnic.
Ion comes now to me daily Ion vine acum zilnic la mine Ion comes now daily to me. ?Ion vine zilnic acum la mine. Ion comes daily now to me.
The examples in (17) below show that the time and aspectual adverbs may both precede subcategorized prepositional objects, which are surely in the VP and have no reason to raise. This word order pattern is expected if the time and aspectual adverbs are in the Specs of the respective functional projections and the verb has raised past them to MP, as hypothesized above. (17)
a.
Ion apealeaza acum zilnic la parintii lui.
b.
Ion resorts now daily to parents his Ion se gandeste acum rareori la parintii lui. Ion REFL thinks now seldom of parents his
Negative adverbs (e.g. de loc, defel 'at all') will be licensed in the Spec of the NegP, directly below the MP, while adverbs of manner are apparently licensed in the Spec of a Voice Phrase, lower than the AspP (cf. Alexiadou 1994; Cinque 1997). Provisionally accepting a biclausal analysis of compound tenses (cf Alexiadou 1994; DobrovieSorin 1987, 1994), let us examine the position of the past participle, which contrasts French paradigms, as shown in (18a), with their Italian and Romanian counterparts, as illustrated in (18b) and (18c), respectively.
The double subject construction in Romanian (18)
91
a. Les enfants riont [ pas du tout bien repondu a la maitresse ]. b. I bambini non hanno \detto mica piu bene alia maestra]. c. Copiii n-au [raspuns de loc bine profesoarei] 'Children have not responded at all well to the teacher.'
The examples show that in Italian and Romanian the past participle verb in its own small clause raises past the adverb of manner and the negative adverb, to a position which can only be M, while in French the verb apparently raises only to the VoiceP (cf. Alexiadou 1994), being preceded by the two adverbs. On the strength of the morphological and syntactic evidence presented so far, it may be concluded that the verb always raises overtly to the highest inflectional projection hi the area of the morphosyntactic features, the position M. At the same time, the data suggest that in Romanian, although the verb always raises to M, it never continues to C; there is no I-to-C. A similar remark was made for Spanish in Suner (1994). The following may be regarded as arguments against raising to C: a) In Romanian there are no main clause/subordinate clause asymmetries (see Isac 1996). b) There is a class of pre-verbal adverbs, presumably in SpecMP or higher, which always precede the finite verb: (19)
a.
(20)
b.
lonabia fl a§teapta pe Petru. Ion hardly him awaits pe Petru. 'Ion can hardly wait for Petru.' *Ionffl a§teapta ad/a pe Petru. Ion him waits hardly pe Petru 'Ion can hardly wait for Petru.'
What is of interest for us is the position the V occupies when a w/z-phrase undergoes vf/z-movement, assuming that Wi-phrases move to SpecCP. The examples hi (21) show that even in interrogative sentences, the respective adverbs still precede the verb. This proves that the verb has not changed its position. (21)
a.
b.
Pecine abia a§teaptalon? pe whom hardly awaits Ion? 'Who can John hardly wait for ?' *Pe cine a§teapta abia Ion ? pe whom awaits hardly Ion ?
92
Comparative studies in Romanian syntax
Of course, if the verb does not raise past the pre-verbal subject to C, one will have to find a different account of the often-noticed impossibility of a pre-verbal subject in questions (see Baciu 1996): * Ce Ion cite^tel 'What Ion reads?' hi conclusion, the verb always raises to M, but not further, so that the Romanian finite (possibly also non finite) sentence is (at least) an MP.
3. POST-VERBAL SUBJECT POSITIONS In this section, I attempt to establish the claim that there is a post-verbal AgrS position in Romanian, using syntactic evidence. As already seen, verb morphology hints that AgrSP is lower than MP in Romanian, but Baker's Mirror Priniciple, on which our hypothesis has so far rested, is too strong and has been shown not to work in many instances (see, for example, Sportiche 1992). Therefore, distributional evidence is necessary to strengthen the claim. The examination of postverbal subject sentences will prove that there are two post-verbal subject positions: SpecVP and a second argumental position to the left of SpecVP, but still post-verbal, a position which has the properties ofSpecAgrSP.
3.1. The Post-verbal Subject in SpecVP It has often been convincingly shown that in Romanian the subject may remain in SpecVP where it is Case-marked under government by the verb raising to Inflection (see Dobrovie-Sorin 1987,1994). hi the Minimalist framework, where a distinction is made between thematic positions, in the Spec of lexical categories, and Case positions, in the Spec of functional categories, SpecVP cannot be a Case position. Alternative analyses are available. One possibility is that the subject in SpecVP checks Case covertly by raising to a subject Case position at LF. Since, even at LF, movement is costly and must be motivated, a different analysis is preferable. Being a pro-drop language, Romanian has the inexpensive possibility of expletive pro insertion. The subject might check its Case through coindexation with pro, an expletive licensed by the AgrS head. When the subject stays in SpecVP, the surface order is V(X)SO, with the verb raising to the higher functional projections past the subject in SpecVP. The subject may precede subcategorized constituents, the direct object or a prepositional object: (22)
Numai rareori §tiau toateelevele toatalectia. only seldom knew all pupils-the all lesson-the 'Only seldon knew all the pupils the whole lesson.'
The double subject construction in Romanian (23)
93
Adesea apelau multe dintre ele la bunicii lor dupa ajutor. often resorted many of them to grandparents-the their for help 'Many of them often resorted to their grandparents for assistance.'
There is no strict adjacency requirement between the verb and the subject, so that adverbs can intervene, as in (24). (24)
Vara dormeau deseoritoate fetele peterasa summer-the slept-3PL often all girls-the on terrace 'In summer, all the girls slept on the terrace.'
In this respect, Romanian is different from Celtic languages. Significantly for our discussion, Roberts (1994) notices that in Welsh no adverb can intervene between the verb and the post-verbal subject, and concludes that in Welsh the post-verbal subject must always be in Spec AgrS, while the verb is presumably in some higher intermediate position between C and AgrS.
3.2. The Post-verbal Subject in SpecAgrSP Apparently, in Romanian there is a second post-verbal subject position higher than SpecVP. Thus, time and aspectual adverbs may intervene between the post-verbal subject and the direct object, as in (25), or the prepositional object, as in (26), which are clearly in VP (or even higher). (25)
(26)
a. Vara, faceau toate elevele zUnic probleme la matematica. summer-the did-3PL all pupils-the daily problems at maths 'In summer, all the pupils were solving math problems.' b. Aproape fara gre§ invatau copiii pe atunci cate o limba straina. almost without fault learned chilrdren-the on then each a langugage foreign 'In those times, all the children learned a foreign language almost without exception.' c. Din fericire, primeau to£ copiii in vremea aceea of fortune received-3PL all children-the at time-the that regulat cadouri frumoase de Craciun. regularly presents nice for Christmas 'Fortunately, in those times all the children were receiving nice Christmas presents on a regular basis.' In tinere|e apelau toate fetele cu regularitate la unchiul lor bogat dupa ajutor. in youth resorted-3PL all girls-the with regularity to uncle-the their rich for assistance 'In their youth, all the girls appealed to their rich uncle for assistance, on a regular basis.'
94
Comparative studies in Romanian syntax
There are good reasons to believe that this post-verbal subject position in Romanian is SpecAgrSP. A comparison of Romanian and Italian is helpful because certain properties characteristic only of pre-verbal subjects in SpecAgrSP in Italian appear to be true of post-verbal subjects in Romanian as welL One such property is Quantifier Floating (QF). Cardinaletti (1996) establishes that in Italian, QF is compatible only with pre-verbal subjects, as in (27b): the pre-verbal subject in SpecAgrSP licenses the quantifier left behind in SpecVP (or adjoined to the VP in other analyses). An inverted subject in Italian is not compatible with QF (c£(28b)). hi Romanian, not only the pre-verbal, but also the post-verbal, subject may launch quantifiers, as the correctness of (30c) shows. Even though the subject is post-verbal in (30c), it has to be to the left of the quantifier, like in Italian, in a ccommanding position, which can be assumed to be AgrSP: (27)
(28) (29)
a. b. a. b. a. b.
(30)
a. b. c.
Tutti i soldati sono andati via. 'All the soldiers have gone away.' /soldati'sono tutti'andativia. Sono andati via tutti i soldati. *Sono tutti andati via / soldati. To]f soldati au plecat de aici all soldiers-the have left from here Soldati au plecat to\j de aici soldiers-the have left all from here Au plecat de aici to\j solda\ji have left from here all soldiers-the *Au plecat to\f de aici soldati. have left all from here soldiers-the Au plecat soldati de aici foj/. have left soldiers-the from here all
Here are more examples, where the post-verbal subject is separated from the floated quantifier by adverbs of time, aspect and manner. Thus, in (31b), the aspectual adjunct infiecare vara 'every summer' intervenes between the subject and the quantifier. In (32b), the verb is unergative (i.e. rode 'laugh') and the post-verbal subject precedes the tune adverb atunci 'then' in SpecTP, while the floated quantifier amdndoi 'both' precedes a subcategorized PP (de cele intdmplate 'at what had happened'). In (33b), the verb is transitive; the post-verbal subject, ei 'they' is followed by the aspectual adverb niciodaft 'never', preceding the floating quantifier to\i 'all' and the direct object: (31)
a. Pe vremea aceea plecau in fiecare vara on time-the that left-3PL in each summer amandoi parintii lui in strainatate.
The double subject construction in Romanian
95
both parents-the his in abroad 'At that time his parents both went abroad.' b. Pe vremea aceea plecau paring lui on time-the that leftSPL parents-the his in fiecare vara amandoi in strainatate. in each summer both in abroad (32)
(33)
'At that time his parents both went abroad.' a. Grozav mai rasesera atunci ei amandoi de cele intamplate. terribly more laughed then they both of those happened-PAST.PART 'Then, they both laughed terribly at what had happened.' b. Grozav mai rasesera ei atunci amandoi de cele intamplate. terribly more laughed they then both of those happened-PAST.PART "Then, they both laughed terribly at what had happened.' a. N- au uitat niciodata ei toji toata lecjia. not have forgotten never they all all lesson-the "They all have never forgotten the whole lesson.' b. N- au uitat ei niciodata toji toata lecjia. not have forgotten they never all all lesson-the 'They all have never forgotten the whole lesson.'
We have assumed that the verb raises to MP in Romanian. Since the subject appears in post-verbal position, yet it precedes the adverbs in Spec TP and SpecAspP and the floated quantifier, it can only be in some AgrP, say AgrSP. This distributional fact agrees with the morphology of the Romanian verb, which indicates the same functional order MP > Agr SP > TP >AspP. In section 4. below, we shall prove that the post-verbal SpecAgrSP position is a/the nominative Case position in Romanian.
3.3. Interpreting the VOS Order in Romanian Before turning to that problem, we should mention what might look like another post-verbal subject position in Romanian. In the sentences considered so far, the subject has consistently preceded the object(s), in VXSO or VSXO word orders. However, the VOS structure is also available in Romanian: (34)
a. Au
ca§tigat premii prestigioase urmatorii
studenti.
have won prizes prestigious following-the students 'The following students have won prestigious prizes.' b. Si- au cumparat vile hixoase toji noii demnitari.
96
Comparative studies in Romanian syntax
(35)
REFL have bought villas lavish all new-the officials 'All the new officials have bought themselves lavish villas.' a. A spart u§a bancii un hoj foarte priceput. has broken door-the bank-the a burglar very skilled 'A very skilled burglar has broken the bank door.' b. A telefonat Mariei Ion, nu Petre. has telephoned Mary-DAT lon-FOC not Petre 'It was Ion who called Maria, not Petre.'
The examples look like the inversion structure of Italian or Spanish. The traditional analysis of VOS in these languages involves right adjunction of the Subject to the VP or base-generation in this position (Burzio 1986; Rizzi 1982). As first remarked by Dobrovie-Sorin (1994) and Motapanyane (1989), the post verbal subject is not VP adjoined in these Romanian structures. In fact, there is evidence suggesting that the subject cannot be adjoined to the VP. This hypothesis would predict dial the subject can appear after an object clause. But this position is awkward or even completely unacceptable in Romanian: (36)
a. ?*I- a spus lui Ion ca vremea va fi frumoasa Petru. him has said to Ion that weather-the will be nice Petru b. Petru i- a spus lui Ion ca vremea va fi frumoasa Petru him has said to Ion that weather-the will be nice 'Petru told Ion that the weather will be nice.'
If one considers the interpretation of the VOS sentences above, it becomes clear that in this order the V and the O are part of the presupposition of the sentence. The subject, on the other hand, is focalized, introducing new information, and may even be a contrastive Focus, as in (35b). A more plausible analysis of the VOS order in Romanian would involve object movement (scrambling); that is, movement of the object past the subject in SpecVP, to SpecAgrO. As is known, scrambling may create binding relations. In support of such an analysis, notice that the subject can be bound by an object quantifier only in the (V)OS order, as in sentences (37a), (38a), (39a) below, though not in the SO order, as in the sentences (37b), (38b), (39b) below. Compare: (37)
a. Ce i- a cumparat fiecarui copil tatal lui de Crachm? what him has bought each-DAT child father-the his for Christmas 'What did his father buy for every child for Christmas?' b. *Tatal lui i- a cumparat fiecarui copil bomboane de Craciun. father-the his him has bought each-DAT child sweets for Christmas
The double subject construction in Romanian
(38)
(39)
97
'For each child, his father bought sweets for Christmas.' a. Unde 1- a chemat pe fiecare copil profescrul lui ? where him has called pe each child teacher-the his 'Where did his teacher call every child ?' b *Profesorul lui (1)- a chemat pe fiecare elev la cancelarie. teacher-the his him has called pe each student to staff-room 'His teacher called every child to the staff room.' a. Ce - a invajat-o pe fiecare fetija mama ei ? what has taught her pe each girl mother-the her 'What has her mother taught each little girl ?' b. *Mama ei a rnvatat-o bunele maniere pe fiecare fetija. mother-the her has taught her good-the manners pe each girl 'Her mother has taught each little girl the good manners.'
Since in the scrambled VOS order the subject is the most deeply embedded constituent of the whole IP, it will be maximally stressed and focalized as desired (see Cinque 1993a). In Romanian, object movement may affect any DP, not only strong or specific DPs as in Dutch or German (see examples in (34) above). The Romanian counterpart of object movement in Germanic is clitic doubling. Object movement and clitic doubling are independent structures. Movement of the object is apparently triggered by the need for the subject to get maximal stress b the sentence, that is, to become focal (a point also defended in Zubizaretta 1994). In such cases, movement continues to be driven by the need to check certain features of constituents, but, against the requirements of Greed, one constituent a (the direct object) moves solely to satisfy some property of another constituent b (in this case, the [+Focus] feature of the subject). A principle along the following lines seems to be in operation (see Alexiadou 1994): (40)
Move a although a has weak features to satisfy a property of P. Otherwise a violation at PF will occur.
Lasnik (1992) has discussed similar cases, where movement, though still morphologically driven, benefits the target of movement, rather than the moved element itself (the "Principle of Enlightend Self-Interest"). To the extent that these analyses are well-founded, they suggest that a weakening of Greed is called for. It may be concluded then that in the SOV order, the subject continues to be in SpecVP, so that there are only two post-verbal subject positions: SpecVP and SpecAgrSP. In the next section, we prove that AgrSP is a/the nominative Case position, as well as the position which licenses pro in null-
98
Comparative studies in Romanian syntax
subject sentences in Romanian.
4. THE DOUBLE SUBJECT CONSTRUCTION 4.1. The Data and the Traditional View Romanian is not usually described as a clitic subject language. However, there is one particular double subject construction, where a pronoun doubles a lexical DP. It can be argued that the pronominal subject in this construction is in fact a weak pronoun and a phonological chtic. This pronoun will be referred to as the "clitic subject". Romanian thus becomes comparable with chtic subject languages, like French or certain dialects of Italian. It is generally agreed (see Cardinaletti and Starke 1994) that weak pronouns and clitics must occupy their Case-checking position before Spell Out. For instance, weak pronoun subjects necessarily occupy Spec AgrSP before Spell Out. The interest of the double subject construction (DSC) for an investigation of the syntax of the Romanian subject is that it contains a weak pronoun subject (the chtic subject), and then, since elites and weak pronouns occupy their Case-checking position before Spell Out, the position of this weak pronoun subject simultaneously indicates the place of AgrSP and the nominative Case position(s) in the Romanian clause. It will be noticed that Romanian, because its chtic subject is always postverbal, contrasts sharply with French, as well as all of the northern Italian dialects, which have preverbal clitics. The DSC under analysis belongs to the "langage familier et populaire"(Byck 1937) and marks certain illocutionary attitudes of the speaker: threat, promise, reassurance. Linguists have spoken of an "emploi affectif du pronom personnel en roumain" (Byck 1937). The chtic subject must immediately follow the verb. The lexical subject is usually post-verbal, as hi (41), but it may as well be pre-verbal, as hi (42). Consider the following attested literary examples: (41)
a. Vazand el dracul ca n- are ce face.... seeing he devil-the that not has what do-MF 'The Devil seeing that there-s nothing to do for him...' b. Dar o sa vie ea o vreme cand o sa se but will SA come she a tune when will SA REFL gaseasca cineva sa scrie despre vitejiile romanilor. find someone SA write about deeds-the Romanians-the 'But there will come a time when someone will write about Romanians' heroic deeds.' d. Sa trece ea §/' asta.
The double subject construction in Romanian
(42)
99
REFL will pass she and this 'This too will pass.' e. Ne cam codeam noifetele. REFL quite hesitated we girls-the 'We, the girls, were quite hesitant.' a. Mama §tie ea ce face. mother knows she what does 'Mother knows what she is doing.' b. Ion vine el mai tarziu. Ion comes he more late 'Ion will come later.'
An interesting empirical fact is the existence of two "dialects" which exhibit distinct agreement patterns. In one of them, that of standard familiar Romania^ which is illustrated in all the examples above, the pronoun agrees in gender, person and number with the lexical subject. In the second, non-standard variety, the pronominal subject is invariably the 3rd person singular masculine form el, irrespective of the gender and number of the lexical subject. The verb naturally agrees with the clitic, not with the lexical subject. Here are attested literary examples, hi which the singular masculine pronoun el 'he' doubles feminine nouns (43a, b) or plural nouns (43c) in subject position. (43)
a. Are sa-1 certe elpreoteasa pe popa will SA him scold he priest-wife-the/?e priest "The priest's wife is surely going to scold her husband.' b. Ne-a venit el apa la moara. US-DAT has come he the water-the to mill 'The water has come to our mill.' c. Las' c- a pajit-o el ho\ji. leave-lMPER that has got it he thieves-the 'You may be sure that the thieves have got it!"
The construction has always caught the attention of reputed Romanian linguists, who have been intrigued by the possibility of doubling the subject, an uncommon possibility for Romanian. Indeed, this language may or must double the post-verbal direct, as in (44), and indirect object, as hi (45), but not the post-verbal or pre-verbal subject, as in (46). (44)
a.
L-amvazatpe copil. him have seeupe child 'I have seen the child.'
100
Comparative studies in Romanian syntax b.
(45)
(46)
Am vazut copilul have seen child-the 'I have seen the child.' (/-) am dat copilului bomboanele. him-DAT have given child-the-DAT sweets-the 'I have given the child sweets.' a. In curte se jucau copiii lui Petre. in yard REFL played children-the of Petre 'Petre's children were playing in the yard.' b. Copiii lui Petre se jucau in curte. children-the of Petre REFL played in yard 'Petre's children were playing in the yard.'
Moreover, even in left dislocated structures, such as (47) and (48), where doubling of either object is obligatory, the subject is not doubled. For instance, in (49) the subject precedes other left dislocated constituents without being doubled. (47)
(48)
(49)
a.
Pe Ion, *(/)- am vazut in curtea §colii. pe Ion him have seen in yard-the school-the-GEN 'Ion, I have seen him hi the schoolyard.' b. Copilul, *(/)- am vazut in curtea §colii. The child him have seen in yard-the school-the-GEN 'The child, I have seen him in the schoolyard.' Copilului, *(i)- am dat bomboanele. child-the-DAT him have given sweets-the "The child, I have given him sweets.' Ion, de la Paris numai fleacuri a cumparat. Ion from of Paris only trifles has bought 'Ion, in Paris, he only bought trifles.'
Several important properties of the DSC have been noticed by Romanian linguists. Philippide (1929) considers that the pronoun is pleonastic, and that the pronoun and the lexical subject make up one constituent, the lexical subject being an apposition. He also correctly points out that in this construction the pronoun is always post-verbal: "On emploie parfois le pronom personnel a cote du substantif, en apposition. Pour le sujet, cet emploi pleonastique existe settlement la ou le verbe precede". Against Philippide (1929), Byck (1937) claims that, although initially the noun may have been an
The double subject construction in Romanian 101 apposition to the pronoun, at the current stage the pronoun forms one constituent with the verb, not with the noun. He emphasizes that the pronoun is unstressed and forms a phonetic unit with the preceding verb. Moreover, the pronoun always agrees with the verb, but it may fail to agree with the noun, as in the dialectal examples hi (43). It may certainly be added that the lexical subject may be separated from the pronoun when it is pre-verbal, as in (50a), or when there is an adverb between the pronoun and the lexical subject, as in (5 Ob). Therefore, they cannot be said to form one constituent: (50)
a.
b.
Fratele meu vine el mai tarziu. brother-the my comes he more late 'My brother, he will come later.' Las' ca vin ei maine copiii, n- ai grija. leave that come they tomorrow children-the not have worry 'The children, they'll come tomorrow, don't worry.'
Most linguists agree that, originating in an appositional structure, the repetition of the subject has become a means of rendering the speaker's afFectivity and emotional intensity. Traditional grammarians have thus established that the pronoun forms a phonological and perhaps also syntactic unit with the verb, and that the pronoun is an expletive. It is important that the DSC should not be mixed up with other situations where there is a pronominal copy of the subject, accompanied by the presuppositional adverb §/ 'also', as in §/ el 'he too' or by the emphatic reflexive pronoun el insu§i 'he himself. There are important distributional differences between these structures: a) The clitic pronoun in the DSC must be adjacent to the verb, as in (5 la) and (52a) versus (51b) and (52b)), while the other two combinations may be separated from the verb by an intervening constituent, such as the adverb mdine 'tomorrow' in (53b) and (54b). In the DSC, the adverb may not appear between the clitic and the verb: (51)
a.
b. c. (52)
a.
Vine el tata maine. comes he father tomorrow 'Father will come tomorrow.' *Vine maine el tata. comes tomorrow he father Vine el maine tata. comes he tomorrow father Tata vine el maine. father comes he tomorrow
102
Comparative studies in Romanian syntax b.
(53)
a.
b.
(54)
a.
b.
* Tata vine maine el. father comes tomorrow he Tata vine §/ el maine. father comes he too tomorrow 'Father too will come tomorrow.' Tata vine maine §/e/. father comes tomorrow he too 'Father too will come tomorrow.' Tata vine el insu§i maine. father comes he himself tomorrow 'Father himself will come tomorrow.' Tata vine maine el insu§i. father comes tomorrow he himself 'Father himself will come tomorrow.'
b) In the DSC the subject is either post-verbal or pre-verbal. In contrast, hi the other two combinations the subject is always pre-verbal, a post-verbal subject causes severe ungrammaticality. There is a clear difference between the example (55a) and the examples (55b, c). (55)
a.
b. c.
Vine el tata mai tarziu. comes he father more late 'Father will come later.' *Vine§/ el tata mai tarziu. comes too he father more late *Vine el insu$i tata mai tarziu. comes he himself father more late
c) The other two combinations may appear with any type of lexical subject, and they are not sensitive to quantifiers, as in (57) and (58)). On the other hand, the DSC construction excludes bare quantifier subjects, whether they are pre-verbal or post-verbal, as in (56). (56)
a.
b. c.
Vine el Ion / unchiul Ion /stapdnul comes he Ion / uncle Ion/ master-the 'Ion/uncle Ion/the master, he'll come.' *Vine elfiecare. comes he everyone *Vine eloricine. comes he anyone
The double subject construction in Romanian 103 (57)
a.
b.
(58)
a.
b.
Fiecare vine §/ el cu speranja. everyone comes too he with hope 'Everyone comes, he too, full of hopes.' Oricine vine §/' el cu speran^a ca va fi ajutat. anyone comes too he with hope-the that will be helped 'Anyone comes, he too, with the hope that he'll get help.' M/»emnuvine el insu$i la tine sa ceara ajutor. nobody nu comes he himself to you SA ask help 'Nobody comes in person to you to ask for help.' Orice copil vine el insu§i la profesor sa ceara ajutor. any child comes he himself to teacher SA ask help 'Any child comes in person to see the teacher for help.'
d) Finally, a distributional contrast is also apparent in the fact that either of the two phrases may itself co-occur with a clitic subject, while the phrases themselves cannot co-occur: (59)
(60)
a.
Ion vine el maine §/ el, n -ai grija. Ion comes he tomorrow too he not have worry 'Ion will come tomorrow too, don't worry.' b. Ion vine el negre§it el lnsu$i, n -ai grija. Ion comes he unfailingly he himself not have worry 'Ion will unfailingly come himself, don't worry.' *llon vine (el) negre§it §/ el el insu§i. Ion comes (he) unfailingly too he he himself
Whatever the correct analysis of these constructions may be, the combinations §/ el 'he too', el insu$i 'he himself are clearly phrasal units independent from the verb, while the pronoun in the DSC depends on the verb. These phrasal units are probably distributed in some adjoined positions, perhaps adjoined to SpecVP.
4.2. The Pronominal Subject in the DSC Two issues regarding the status of the pronoun have yet to be settled: (a) its argumental (or nonargumental) status and (b) its phrasal status. In order to examine the latter, it must first be determined whether the clitic subject is and X head or an XP category (i.e., whether it is a syntactic or merely a phonological clitic).
104
Comparative studies in Romanian syntax
As to the first question, the Romanian pronoun is a non-referential expletive. This is why it may cooccur with a lexical subject, in the same way in which an object clitic occurs with a lexical object in clitic doubling constructions. In languages such as French, where the clitic is argumental, it does not normally appear with a lexical subject, since one of the two nominals will be devoid of a theta-role. From this point of view, the Romanian clitic differs from its French counterpart and is similar to subject clitics in certain Italian dialects discussed in Poletto (1991), Rizzi (1986b) and Safir (1985), where the clitic subject may be accompanied by a lexical subject: (61)
French Trentino Romanian
(62)
French
(63)
Trentino
(64)
Romanian
fl
mange. he eats El magna he eats Mananca el. (with appropriate intonation on V) eats he *Jean il mange//Jean mange. Jean he eats Jean eats El Gianni el magna.// *E1 Gianni magna. he Gianni he eats he Gianni eats Vine el tata.//Vine tata.//Tata vine el. comes he dad cames dad dad comes he
Moreover, in non-standard dialects, the clitic has the invariable 3rd person masculine form el 'he', and in this case it must be followed by the lexical subject, if the intended meaning is to be expressed. This is clearly an indication that the pronoun is expletive and cannot be viewed as some anaphoric copy of the subject: (65)
Nea venit el apa la moara. us-DAT has come he water-NOMFEM.SG-the to mill "The water has come to our mill.'
As to the second question raised above, on the phrasal status of this pronoun, there is evidence that the pronominal subject is still a DP, not a D. In the typology of Cardinaletti and Starke (1994), it should better be viewed as a weak pronoun rather than a clitic, comparable to the French //, or the Italian egli /esso. A comparison with other Romance language is again helpful. Rizzi (1986b) has established the
The double subject construction in Romanian 105 existence of two types of subject clitics in Romance. On the one hand, there are head clitics, placed in the verbal agreement head AgrS and adjoined to the verbal morphological features in a configurations like (66a) below. There are also DP clitics, or weak pronouns, which occupy Spec positions (SpecVP, SpecAgrSP), as in (66b) below, cliticizing to the verb in the phonological component: (66) a.
AgrSP Spec
b.
AgrSP
AgrS' AgrS Clitic
TP
A head clitic part of AgrS0 is often a sort of morphological specification , (always) expressed on the head of AgrSP, independently of the element that is realized in SpecAgrSF, which can be a null subject or a phonetically realized DP. The clitic performs morphosyntactic functions and is often not involved in referential relations. Thus, there are dialects where the clitic is the assignor of the nominative Case to the pre-verbal subject, by means of Spec-Head Agreement. Poletto (1991) discusses dialects like Trentino/ Florentine where the clitic is an obligatory Spell Out of Inflection, always accompanying the preverbal subject: (67)
La Maria la magna.//*La Maria magna. the Maria she eats the Maria eats
It may be that the pre-verbal and post-verbal subject check Case with different heads. In some dialects, the subject clitic is obligatory with the pre-verbal subject, which checks Case with Agr, but cannot appear with the post-verbal subject, which checks Case in a different manner. This is the case of the Modern Trentino variety (see Poletto 1991, 43), illustrated below: (68) (69)
La Maria la riva.//*La Maria riva. the Maria she arrives the Maria arrives *La riva la Maria.// Riva la Maria. she arrives the Maria arrives the Maria
In many northern Italian dialects (e.g. Trentino, Fiorentino, Veneto), the clitic subject is also the head which licenses and assigns Case to pro, performing the morphosyntactic functions of strong
106
Comparative studies in Romanian syntax
Agr. Pro is licensed by the clitic head in Agr through Spec-head agreement, in contrast to French, where the clitic is in SpecAgrSP. These Italian dialects are, therefore, pro-drop languages, unlike French. The only difference between these dialects and standard Italian is that the strong Agr is realized in phonetic form not only in concrete verbal morphology, but also in a syntactic position, by a sort of reduplication of Agreement: (70)
pro el magna // *magna he eats eats
Binding and quantification facts may also help hi distinguishing between clitics and full DPs, if we accept a uniformity condition regarding the phrasal status of chain members (see Dobrovie-Sorin 1998 and Progovac 1993). French instructively contrasts with certain Italian dialects along these lines. French has subject clitics in SpecAgrSP in syntax and cliticization to V in the phonological component. For some speakers a lexical subject can occur left dislocated with a subject clitic, even without any perceptible intonation break, but a bare quantifier (BQ) is impossible in this environment: (71)
(72)
Jeann' a rien dit.//Jeanil n' a rien dit. Jean not has nothing said Jean he not has nothing said 'Jean did not say anything.' Personne n1 a rien dit.// *Personne il n1 a rien dit. nobody not has nothing said nobody he not has nothing said 'Nobody said anything.'
Rizzi (1986b) explains that in the correctly formed quantifier sentences, the pre-verbal subject must be in SpecAgrS. It cannot be a left dislocated element, since BQs cannot be left dislocated, leaving (clitic) pronouns behind. This restriction on BQs is expressed below (see Rizzi 1986b, 395): (73)
A pronoun cannot be lexically bound by a non-lexically restricted (bare) quantifer.
As Rizzi explains, this statement essentially amounts to chinning that pronouns cannot function as primary variables. They sometimes acquire variable status only parasrtically, through binding from licit primary variables. When a pronoun is directly bound by a BQ an ill-formed chain will result. As is known, QPs must move to an A1 position at LF, leaving variables as their trace. On the assumption that in French the clitic occupies the subject position, as in (66b), the ungramrnaticality of Personne il n'a rien dit follows at once. At LF, the EQpersonne must be in
The double subject construction in Romanian 107 an A1 Op position. On the assumption that the BQ is already in an A'-position, it binds the pronoun in subject position in direct violation of constraint (73). If the BQpersonne were to further raise by QR from its left dislocated position, it would leave behind a variable in an A'-position instead of in an A-position. Italian dialects like Florentine or Torinese sharply differ from French. In these dialects, the subject clitic is obligatory, even when the lexical subject is a BQ. This is significant since, as shown by the example of French, BQs can only appear in an argumental subject position (SpecAgrS), and not in a left dislocated position. Examples(74) are due to Rizzi (1986b, 356). (74)
Torinese Florentine:
a.Gnun 1' a dit gnent. nobody he has said nothing b.Nessuno 1' ha detto nulla. nobody he has said nothing
The Italian examples are well-formed, since the clitic is adjoined to Agr under the Agr head, hi configuration (66a), allowing the BQ to occupy the argumental subject position, SpecAgrSP. Intuitively, in this case, the clitic is treated as merely an element of verb morphology. The BQ undergoes QR, leaving behind a (licit) variable hi the argumental Case-marked subject position. As to the clitic, one might follow Rizzi and say that the clitic is bound by a licit primary variable, acquiring variable status; a (descriptively) better solution is to assume that chains should be uniform as to phrase structure status (the uniformity condition on chains (see Dobrovie-Sorin 1998; Progovac 1993), so that only the elite trace (i.e. the subject position itself) enters a well formed chain of the form (QP^ vblj) However, the situation is not so clear cut; Poletto (1991, 42) cites the Modem Trentino variety where the clitics are, demonstrably, heads, but where doubling of the clitic subject by a BQ is, nevertheless, prohibited: (75) Trentino *Nisun el vien qua.// ?Nisun vien qua. nobody he comes here nobody comes here Poletto's explanation relies on the fact that in this dialect, the clitic is a nominative Case assigner. As such it will be coindexed with the pre-verbal subject so that it will have to participate in all chains invoking the subject position. When the quantifier undergoes QR, the clitic will be coindexed with the subject trace and will count as the member of a chain having the form (QP;, vblj, clj, in violation of the uniformity condition on chains. The clitic enters a chain having the form (QPi5 vb}, 4), distinct from the chain corresponding to examples (74) above, where only the clitic trace position
108
Comparative studies in Romanian syntax
(Le., the subject position itself) enters the chain. Moreover, the head clitic will count as A-binding the variable left by the raising quantifier, a variable which should be A1 - bound, hi conclusion, the possibility of doubling a BQ by a subject clitic unambiguously shows that the BQ is in subject position and that the clitic is a head (as in Torinese and Florentine), but, as suggested by the Trentino data, the feet that BQ subjects may not be doubled by clitic subjects does not always mean that the clitic is a DP in Spec position. A second reliable diagnostic which may be used to determine the D versus DP status of a pronoun is that of coordination. Clitic subjects must be repeated when there is verb coordination, as shown by the Trentino examples below (from Rizzi 1986b, 402-403). In contrast, in French, the verbal heads are coordinated leaving the subject in SpecAgrSP: (76)
French Trentino
a. Elle chante et danse she sings and dances b. c.
*La canta e balla. she sings and dances La canta e la balla. she sings and she dances
Let us examine the Romanian facts now: to begin with, the clitic surely plays no part in Case checking. It appears equally felicitously with pre-verbal or post-verbal subjects. This is not unexpected, since the DSC represents a feature of familiar, affective speech, which could hardly have assumed any morphosyntactic role in the economy of Romanian. Romanian subject clitics are optional, and as such, appear to behave like full DPs, not heads. Notice first that subject ch'tics do not have to be repeated under verb coordination, as in (77a), although they may be as in (77b, c). In contrast, object ch'tics, which are adjoined to the verbal head (in an Agreement projection), being syntactic clitics, must be repeated under verb coordination: (77)
(78)
a. Las' ca vine el tata §i te invaja minte. leave that comes he father and you teaches mind 'Be sure that father will come and teach you a lesson.' b. Las ca vine el tata §i te invaja el minte. leave that comes he father and you teaches he mind 'Be sure that father will come and teach you a lesson.' c. Tata vine el §i te invaja (el) minte. father conies he and you teaches he mind a.*Z- am vazut§i am placut.
The double subject construction in Romanian 109 him have seen and have liked b. L- am vazut §i /- am placut. him have seen and him have liked 'I have seen him and I have liked him.' This is an indication that the pronominal subject is in SpecAgrSP. Applied to Romanian, the binding and quantification facts mentioned above strengthen this result. Like its counterparts in French or Trentino, the Romanian pronoun is sensitive to the type of lexical subject which it doubles. All the attested examples we found have a definite referential postverbal subject, as in (41) and (43) above. What is theoretically significant is that, like in Modern Trentino or French, BQs and more generally quantified subjects are excluded in post-verbal as well as, even more strongly, in pre-verbal position in the DSC: (79)
(80)
a. *Are sa se mire el flecare de una ca asta. will SA REFL wonder he everyone of one as this 'Everyone is going to wonder about such a thing.' b. *Nu vine el nimeni pe vremea asta, n -ai grija. not comes he nobody/>e time-the this not have worry 'Nobody will be coming in such a weather." c. *Cumpara el un elev cartea pana la urma. buys he a student book-the until at end 'Some student will buy the book in the end.' a. *Fiecare are sa se mire el de una ca asta. everyone will SA REFL wonder he of one like this 'Everyone is going to wonder about a thing like this.' b. *Nimeni nu vine el pe vremea asta. nobody not comes he in weather-the this 'Nobody will be coming in such a weather.' c. *Un elev cumpara el cartea, n -ai grija. a pupil buys he book-the not have worry 'Some student is going to buy the book, don't worry.'
Principle (73) above is operative in Romanian as well, in examples like (80), which are similar to the French examples in (72b). In examples (79), the post-verbal BQ subject will eventually raise to an A1- position , giving rise to a chain of type (BQj, c\, var^), where the variable is again wrongly Abound by the clitic and the members have unequal phrasal status. Thus, the results of the coordination diagnostic, together with impossibility of doubling a BQ (or any quantified subject) by a clitic, leads one to the conclusion that the clitic subject holds the position SpecAgrSP.
110
Comparative studies in Romanian syntax
But now comparison of the Romanian examples with any of the examples from French or Italian immediately shows a striking difference. The Romanian clitic is systematically post-verbal. In fact, the clitic can never be pre-verbal; that is, in the DSC the pronoun is always post-verbal: (81)
a. £7 vine tata, n -ai grija. he comes father not have worry b. Vine el tata, n -ai grja. comes he father not have worry c. Tata vine el, n -ai grija. father comes he not have worry
This paradigm actually provides strong evidence for a post-verbal AgrS position in Romanian. The conclusion that Romanian has a post-verbal AgrSP seems hard to disprove. The adjacency of the pronoun to the verb is a direct effect of its structural position directly below the verb cluster, a position that makes possible phonological cliticization. At the same time, when the verb raises by LHM, the subject pronoun, as well as the object clitics, are left behind: (82)
Slaveasca-/ ei, preojii cat or vrea. glorify-SUBJ-him they priests-the how will want 'Let the priests glorify him as much as they want.'
A second tentative result is suggested by the similarity of the French examples with Left Dislocation (72b) and the Romanian DSC with pre-verbal subjects, as in (80). Namely, the Romanian pre-verbal subject (sometimes) behaves like a left dislocated constituent. It appears that the pre-verbal subject position is not an L-related argument position in Romanian. The pre-verbal subject is a Topic, rather than an argument.
4.3. Problems of Case If the pronoun is a specifier, that is to say, if it is a full DP, then there is the problerm of how nominative Case is assigned to the two DPs. Neither of them can get inherent partitive Case, since both of them are definite and partitive Case is restricted to indefinites (Belletti 1988). Also, since both DPs are phonetically realized, Case cannot be transmitted through chain formation, as it is in (expletive pro, DP) chains. The two nominative Cases are presumably checked in different positions. In the discussion of Case, it is profitable to remember that there exists a subdialect of Romanian where the clitic is invariably marked for the 3rd person singular, masculine, possibly having different
The double subject construction in Romanian 111 4>-features from the following lexical subject, as in (43), repeated as (83). When this happens, the veib agrees with the clitic, not with the lexical subject.The lack of agreement between the fall lexical subject and the verb shows that Agr is not involved in checking the Case of the lexical subject. It is likely that the latter checks its nonunative feature against a different head: (83)
a. Are sa-1 certe elpreoteasa pe popa will SA him scold he priest-wife-the/?e priest 'The priest's wife is surely going to scold her husband.' b. Ne-a venit el apa la moara. us-DAT has come he the water-the to mill "The water has come to our mill.' c. Las' c- a pa{it-o el ho\ji. leave-IMPER that has got it he thieves-the 'You may be sure that the thieves have got it!"
The Romannian facts are parallel to those of Renaissance Veneto(RV), discussed in Poletto (1991), except that in RV the clitic is pre-verbal. In this dialect, a lexical subject that has undergone free inversion co-occurs with a subject clitic in pre-verbal position. The verb agrees with the clitic, not with the post-verbal subject: (84)
(85)
a. RV El viene quel suofraelo. he comes that his brother "That brother of his will come.' b. R Vine el fratele acela al lui comes he brother-the that of his 'That brother of his will come' Z/e sta suspesso le prediche al Sior Geronimo. it is stay suspended the sermons to Mr. Geronimo 'Mr. Geronimo's sermons are being suspended.'
The subject clitic was, at the time, a full DP (see Poletto 1991) and obtained nominative Case by agreement, as the verbal morphology indicates. It is less clear how the full DP subject obtains Case. To solve similar problems, Roberts (1991) assumes two means of assigning nominative Case in UG, in two distinct configurations: (86)
a. Agr assigns Case through Spec-Head Agreement. b.T assigns Case through Government.
112
Comparative studies in Romanian syntax
Such a parameter of nominative Case assignment could, according to Roberts, explain the difference between languages like French and Welsh. In French, the subject is pre-verbal and triggers morphological agreement in person and number with the verb. French exploits the option expressed in (86a). In Welsh the subject is post-verbal and does not trigger morphological agreement in person and number with the verb. Welsh avails itself of option (86b). Roberts further assumes that in Romance languages, the subject can be pre-verbal, as well as postverbal, because both options can be selected. Nevertheless, languages like standard Italian always show morphological agreement of person and number, even when the subject is post-verbal, while Welsh never does. To account for this difference in the properties of the post-verbal subject, Roberts (1991) formulates a rule of co-superscripting between the heads of AgrP and TP, which applies in Romance, but not in Welsh. The cos-uperscripting rule determines the sharing of the morphological agreement features of person and number, so that the verb and the post-verbal subject always agree in person and number in Romance: (87)
Coindex Agr and T.
A rule like (87) is then active hi Romance, because both Agr and T are able to assign nominative but it will fail to apply hi Welsh. Adapting Roberts' ideas to current theory, one might claim that nominative Case may be checked either in SpecAgrSP, or in SpecTP, with independent properties of the language determining which option is used. Let us consider the facts of RV and dialectal Romanian, where the post-verbal subject does not agree with the verb, while the clitic does. The configuration of the Romanian sentences under discussion is as follows: (88) -Ursp clitic subject [ V+Agr [ ^DP, t,... [w t2 [V.t,...]]]] Apparently, in RV and in dialectal Romanian, just as in other Romance languages, both mechanisms of Case assignment can be exploited in principle, and must be exploited in certain situations, as is the case in DSC. The clitic subject necessarily occupies the position SpecAgrSP, since a weak pronoun or a clitic must reach its Case checking position before Spell Out. A second reason for the clitic merge in SpecAgrSP rather than in a lower position is that the clitic is expletive and SpecAgrSP is a nonthematic position. Consequently, the clitic will check its Case by agreement with the verbal
The double subject construction in Romanian 113 agreement head. The post-verbal subject checks Case by overt or covert movement to SpecTP. Cosuperscripting of Agr and T does not apply, so the post-verbal subject does not agree with the verb. In standard Italian / Romanian, unlike Welsh or the Romanian dialect discussed above, there is always agreement with a post-verbal subject. This may be interpreted as showing that standard Romanian / Italian possesses only one Nominative checking position, namely, SpecAgrSP. As mentioned above, the post-verbal subject in SpecVP may either check Case covertly at LF, (see Cardianletti 1996) or by coindexation with an expletive pro in SpecAgrSP, which forms a chain with the post-verbal subject (see Alexiadou 1994). If the latter solution is adopted, nominative Case will uniformly be checked before Spell Out and there will standardly be a unique nominative checking position, namely SpecAgrSP. As for the DSC, we may claim that, exceptionally, through co-superscripting of Agreement with Tense, the Case-checking potential of Tense is "activated"; that is, the Case/ person/ number/ gender features of Agreement are transmitted to its complement, Tense, allowing a second nominal, the lexical subject in the DSC, to check its nominative feature overtly or covertly. Through cosuperscripting, the two subjects share
AgrSP> TP postulated in our analysis. Alternatively, the lexical subject may stay in SpecVP and check Case at LF; in this case, there may be intervening constituents between the pronominal subject and the lexical subject: (89)
(90)
Nu vin ei copiii maine. not come they children-the tomorrow 'They are not going to come tomorrow, the children.' Vin ei maine copiii, n -ai grija! come they tomorrow children-the not have worry 'The children, they'll come tomorrow, don't worry.'
4. CONCLUSIONS 1. The pronoun in the DSC is an expletive weak one, cliticizing to the verb at phonlogical form, so there is a subject clitic in Romanian as well. The clitic always agrees with the verb, even when it does not agree with the full subject. This indicates that the clitic checks Case by Spec-head agreement with the verb in AgrS. 2. In the DSC there are two full nominative DPs, which check Case in different positions. The lexical post-verbal subject checks Case in SpecTP, moving into this position overtly or covertly. In
114
Comparative studies in Romanian syntax
the present analysis, the SpecTP position is made use of only in the DSC. Outside the DSC, when the clitic is not present, the Case of the post-verbal subject is uniformly checked in SpecAgrSP, overtly or covertly (see also Cardinaletti 1996). In contrast, other researchers (e.g. Motapanyane 1989) have argued that all post-verbal subjects get Case from the verb in Tense under government, while pre-verbal subjects check Case in a pre-verbal AgrS position. 3. Since weak forms and clitics occupy then" Case-checking position before Spell Out, the position of the subject clitic must be SpecAgrS. In fact the clitic is projected in that position, since it does not need any theta-role. Given that the subject clitic is always post-verbal, it follows that AgrSP is post-verbal in Romanian. The verb always raises past the subject Case-checking position toward some higher functional projection, the MoodP in our analysis.
5. ON THE NATURE OF THE PRE-VERBAL SUBJECT POSITION IN ROMANIAN In the previous sections, it has been proved that in Romanian there is a post-verbal AgrSP below MP, whose Spec is the nominative checking postion, as well as the position where pro is licensed. In this section, we will argue that this is the only AgrSP position in Romanian. At stake is the nature of the pre-verbal subject position. Two views have been expressed on this matter, as previously stated: a) The pre-verbal subject position is a Topic, left dislocation position (see Dobrovie-Sorin 1987,1994). b) The pre-verbal subject position is argumental and it is the nominative checking position for pre-verbal subjects. This second view is expressed in Motapanyane (1989).
5.1. Previous Results The arguments adduced to defend the non-argumental or the argumental nature of the pre-verbal position are briefly summarized in this paragraph. Dobrovie-Sorin (1987) argues that the subject, which receives Case from the governing Inflection, scrambles to SpecIP, a topicalization position in Romanian, more exactly a position of left dislocation. The main argument is that several XPs may be topicalized in Romanian, and that they may precede or follow the topicalized subject: (91)
a. leri Ion drept in mijlocul gradinii faceaplaja yesterday Ion right in middle-the garden-the-GEN made beach 'Yesterday, Ion was sunning himself right in the middle of the garden.'
The double subject construction in Romanian 115 b. leri, drept in mijlocul gradinii, Ion facea plaja. yesterday right in middle-the garden-the-GEN Ion made beach c. Ion ieri, drept in mijlocul gradinii facea plaja. Ion yesterday right in middle-the garden-the-GEN made beach 'Yesterday, Ion was getting a tan right in the middle of the garden.' (92)
a. Ion tot flori i -a adus Mariei. Ion also flowers her has brought Maria-DAT 'Ion has brought flowers too for Maria.' b. Mariei Ion tot flori i -a adus. Maria-DAT Ion also flowers her has brought 'Ion has brought flowers too for Maria.' c. Flori Mariei ION i -a adus (nu Petre). flowers Maria-DAT Ion-Foe her has brought not Petre 'Ion is the one who brought flowers for Maria, not Petre.'
On the other hand, Motapanyane (1989) claims that SpecAgrSP is an argumental pre-verbal position in Romanian. Briefly, her argument is that there are types of pre-verbal constituents which cannot be considered to be left dislocated, but which nevertheless may function as pre-verbal subjects in Romanian. One example is that of BQs. She holds that BQs must occupy A-positions at S-structure, so as to form well-formed chains when they raise at LF. Since BQs may be pre-verbal subjects, it follows that at least for BQs, there has to be a pre-verbal subject A-position. To prove her point, Motapanyane (1989) compares the behaviour of objects and subjects. She notices that emphatic strong direct objects may scramble to a position between the subject and the predicate, or even to a position to the left of the subject, as in (93). In contrast, BQ direct objects cannot appear in either of these positions, as in (94b, c). (93)
a.
b.
c.
(94)
a.
lonpetojiii invitase la masa. Ion pe all them invited at dinner 'Ion has invited them all for dinner.' Ion pe soldaji ii admira, nu pe profesori. lonpe soldiers them admired notpe teachers 'Ion admired the soldiers, not the teachers.' Pe solda^i Ion ii admira. pe soldiers Ion them admired 'As for the soldiers, Ion admired them.' Profesorul a invitat pe cineva la masa professor-the has invited pe someone at dinner "The professor has invited someone for dinner.'
116
Comparative studies in Romanian syntax b. c.
*Profesorul pe cineva a invitat la masa professor-the pe someone has invited at dinner *Pe cineva profesorul a invitat la masa pe someone professor-the has invited at dinner
Motapanyane (1989, 22-23) accounts for the ungrammaticality of (94a, b) in the following terms: a BQ cannot scramble out of an argumental position, because it would end up in an A'-position at S-structure (cf Rizzi 1986b). Given that, at LF, the BQ undergoes QR, if the position out of which it raises is A1, an ill-formed chain will result, where the variable bound by the BQ is in an A'-position, instead of an A-position. To avoid this undesirable result, BQs in argumental positions must not move until LF (and, therefore, cannot scramble). However, as Motapanyane observes, in Romanian BQs function as both pre-verbal and post-verbal subjects: (95)
a.
b.
Cineva a batut la u§5. someone has knocked on door 'Someone knocked at the door.' A batut la u§a cineva. has knocked on door someone 'Someone knocked at the door.'
Along the same line of reasoning, the BQ subject should also be in an argumental position at Sstructure; this leads one to the conclusion that the two subject positions hosting the BQs in the examples above must both be argumental Case positions. In this analysis, SpecAgrSP is a Case position: "Par consequent, le sujet ne monte pas en scrambling en SpecAgr, mais pour recevoir le cas assigne a cette position" (Motapanyane 1989, 25). This elegant argument cannot really stand. In the first place, empirically, the examples presented above are not convincing, since, in spite of (94b, c), there surely are instances where an emphatic (focalized) direct object BQ felicitously appears between the subject and the predicate. Examples of type (94b) become acceptable if cineva 'someone', harder to emphasize, but still a possible Focus, as (96c) shows, is replaced by nimem 'nobody', as in (96a), or oricine 'anybody', as in (96b). An emphatic quantified direct object may appear between the subject and the predicate, in the same position as in (94b). Moreover, examples like (97) show that a BQ object may even felicitously precede the pre-verbal subject, occupying the same position as it does in ( 94c). (96)
a.
Profesorul pe nimeni
n -a
invitat la el.
The double subject construction in Romanian 117
b.
c.
(97)
a.
professor-the nobody-FOC not has invited at him "The professor has invited NOBODY to his place.' Profesorul pe oricine ajuta cum putea. professor-the pe anybody-FOC helped how could 'The professor helped ANYBODY as he could.' Laurmaurmei chiar§i Maria pe cineva at end end-the-GEN even and Maria pe somebody-Foe tot va trebui sa invite still will need SA invite-SUBJ. 3 so 'After all, even Maria will have to invite SOMEONE.' Ceva chiar §i Maria §tie la matematica. something-FOC even and Maria knows at mathematics 'Even Maria knows SOME THING in maths.'
Secondly, more recent research on clause structure has proved that in fact, the occurrence of a BQ in an A'-position before Spell Out is not theoretically impossible. Our claim is that in Romanian the unique SpecAgrSP position is post-verbal and that it is also the standard nominative checking position, at least hi finite clauses. The pre-verbal subject position is not L-related, and the pre-verbal subject appears to have (operator) A' properties.
5.2. The Left Periphery A brief review of some more recent results regarding the structure of the left periphery of the clause is needed at this point, since we believe that the Romanian pre-verbal subject always occupies positions in that area. In a synthetic presentation of this problem, Rizzi (1997) proposes that the C system, actually the CP > IP(operator) area, may be more highly articulated into a number of positions, which check different non L-related features of sentences, denning a "force-finiteness system" with the following configuration: (98)
(Force)... (Topic)... (Focus)...( Fin IP)
The Fin(iteness) system is the old IP, therefore the MP for Romanian. The highest projection is the old CP, supplying differences of illocutionary force, through features like [imperative], [declarative], etc. For the present analysis, what matters most is the Topic-Focus articulation of the sentence. The Top(ic) system interprets the Topic/ Comment structure of the sentence, while the Foc(us) system deals with its Focus / Presupposition articulation. Although Topic and Focus may be formally
118
Comparative studies in Romanian syntax
similar in many languages, there are clear interpretational differences between them, sometimes accompanied by distributional differences, allowing one to distinguish between Top/ Foc.(Rizzi 1997;Zubizaretta 1994). The basic syntactic difference between the higher Op projections and the lower morphosyntactic projections is that the Spec of the former is an A'-position, checking an Op feature, while the Spec of the latter is an L-related A position. A constituent endowed with Topic or Focus features must end up in a Spec-head configuration with a Top or Foe head, containing the respecive feature, to satisfy theTop / Foe Criteria. In Romance, Topics may involve resumptive clitics. A typical topicalization structure in Romance, also available hi Romanian, is the Clitic Left Dislocation Structure (CLLD), whose syntax is discussed in. Cinque (1990) for Romance, and Dobrovie Sorin (1994) for Romanian. The following are the most salient Topic/Focus differences: a) Focus is quantificational, Topic is not. This is the main difference between them, dividing the class of A dependencies into those that involve a quantifier that binds a variable, and those that involve non-quantificational binding, that is, binding of a null constant, a DP trace, as with CLLD: (99)
a. b.
(100) a. b.
// tuo librot ho comprato tj. (FOCUS) 'It is your book I've bought.' *// tuo libro{ toj ho comprato t;. the your book it have-1 so bought // tuo libro, k>j ho comprato tj. 'As for your book, I've bought it.' *// tuo librot ho comprato tj. the your book have-1 so bought
(TOPIC)
The Focus phrase legitimately binds a variable in (99a), but not in (99b), where neither the ch'tic nor its trace is a variable. As to (lOOb), under classical assumptions on the typology of empty categories, it is also ruled out right away: The topicalized phrase in (lOOb) is not quantificational. Therefore the empty category in object position has no legitimate status; it cannot be a variable, since there is no quantifier to bind it. It is in fact a clitic trace, so that structure (lOOa), where the empty category is bound by a clitic, is well-formed. Consider now some Romanian examples: (101)
a.
b
Cartea tat am cumparat-Oj tj. book-the your-TOP have- 1SG bought it 'Your book, I bought it.' *Cartea ta am cumparattj.
The double subject construction in Romanian 119
c.
d. (102)
a.
b. (103) a. b.
book-the your-TOP have-lSG bought Cartea ta, am cumparat-Oj tb nu pe a lui. book-the your-FOC have- ISO bought it not pe of his 'It is your book I've bought, not his.' *7Cartea ta{ am cumparat t(, nu pe a lui. book-the your-FOC have-1 so bought not pe of his M/n/Cj n -am cumparat tj. nothing-Foe not have-1 so bought 'I bought NOTHING.' *Nimici nu l^am cumparat tj. nothing-Foe not it have-1 so bought Ni§te car\f^ am cumparat ti; nu caiete. some books-FOC have-1SG bought not notebooks *Ni§te cdr]ii \Q, -am cumparat tj. some books-FOC them have-1 SG bought
In Italian, Focus and Topic are formally associated not only with distinct intonation, but also with different types of chains, involving variables in the case of Focus, but DP-traces k the case of Topic. In other words, Focus is always quantificational. This is not always the case in Romanian, where the form of the chain, specifically doubling by a clitic, depends more on the inherent properties of the moved DP, and less on the Topic / Focus interpretation. Thus fronted definite direct / indirect objects must, or at least, may be doubled, even when they are contrastively stressed (focalized), as shown by example (lOlc), in contrast with the Italian (99b). What matters, however, for the present discussion, is that a BQ and generally quantified subjects in a Focus configuration will always be marked by the absence of the resumptive clitic in Romanian as well Examples (102) and (103) involve quantificational Focus, behaving like the Italian examples in (99). Thus, the focalized elements in (102) and (103) are genuine quantifiers and must bind syntactic variables. This happens in (102a) and (103a) but not in (102b) and (103b), in which potential bindees are the clitic and its trace, neither of which qualifies as a syntactic variable (i.e., a non-pronominal empty X-max category in an A- position). In conclusion, in Romanian, BQs clearly distinguish between Topic and Focus: they are excluded in the CLLD structure, and cannot be Topics, but they easily allow focalization. b) Weak crossover phenomena offer a second test differentiating Topic and Focus. A Topic never gives rise to any weak crossover effects, as shown in (104). However, such effects may be detectable with a Focus, as in (105).
120
Comparative studies in Romanian syntax
(104)
a. Un copilj nu- \ abandoneaza maica-sa; a child not him abandons mother-bus 'His mother does not abandon a child.' b. *Un copilj nu abandoneaza maica-sa; a child-FOC not abandons mother-his (105) a. Pe oricare copilj ilj rasfaja maica-sa;/ mama luij. pe any child-TOP him spoils mother his mother his 'His mother spoils any child.' b. *Pe oricinet rasfap maica-saj. / mama luij pe anybody-FOC spoils mother-his mother his A further illustration of this difference is the contrast between questions and appositive relative clauses. Questions, which (may) involve genuine quantifiers, may be sensitive to weak crossover effects, as in (106) and (107), in contrast with appositive relative clauses, as in (108), which merely involve anaphoric pronouns. (106) a.
?*Pe cinej iube§temama luijCu adevarat? pe whom loves mother his with true b ?*WhOi does hiSj mother really like ? (107) *Ce copilj ajuta mama lui; latraduceri? what child, helps mother his; at translations 'What child does his mother help with his translations?' (108) Ion, pe carej maica-sa; fl iube§tecu adevarat, este un copil talentat. Ion pe which mother-his him loves with true is a child gifted 'Ion, whom his mother truly loves, is a gifted child.' c) Uniqueness further distinguishes between Focus and Topic. A clause may contain as many Topics as are consistent with its topicalizable arguments, as hi example (109a); on the other hand, there is a unique Focus. A Focus and one or more Topics can be combined in the same structure. In that case, the focal constituent can be both preceded and followed by Topic constituents, as in example (109b). (109) a. Mariei cu banii ace§tia florile nu-i le pojicumpara. Maria-DAT with money-the this flowers-the not her them can buy 'You cannot buy the flowers for Mary with this money.' b. Lui Ion gre$eala asta la proces nu i- o iarta nimeni, sunt sigur. to Ion this mistake-FOC at trial not him it forgive nobody am sure "This mistake, nobody's going to forgive Ion for it at the trial, I am sure.'
The double subject construction in Romanian 121
5.3. Our View We may now return to the pre-verbal subject position in Romanian. Since Case and (f>-features are checked in AgrS or lower hi TP, it is difficult to see why another L-related position would be needed. There are however several logical possibilities which would warrant one more pre-verbal AgrSP. First, it is still conceivable that certain subjects may not check Case in the lower AgrS position simply because this position is empty, as is the case in non-finite clauses. We will not deal with nonfinite clauses in this paper(see Cornilescu 1997; Dobrovie-Sorin 1987, 1994). In the same line of thought, a very strong case for a pre-verbal AgrSP structure would be the existence of situations where nominative Case could only be checked in this pre-verbal position. For instance, the preverbal subject might get Case in that position from some Case assignor in a higher functional projection, such as C. A situation of this type is discussed in Rizzi (1982, chapters 3 and 4), who shows that in a class of gerundial complements, nominative Case is checked by an auxiliary that has raised to C (see Rizzi's Aux -to Comp rule). In Italian examples like (109a ) a full subject can in fact be found in a position internal to the sentence (i.e., internal to the IP), comparable to the position of the subject in English. Notice that a BQ subject (qualcuno 'someone') is grammatical in (110), although BQs cannot be topicalized and have to occupy argumental positions. (110) Avendo Gianni/eghV qualcuno telefonato a Maria... having Gianni/ he / someone telephoned to Maria (from Cardinaletti 1996, 10) Although identifying a nominative position as a position governed by C is surely correct, this line of reasoning cannot apply to Romanian, since in Romanian there is no V-to-C movement, so there will be no way of identifying a nominative position as one governed by the V in C. Second, it may be, as has been argued, that, even in finite clauses, a pre-verbal SpecAgrSP position is needed to accommodate those pre-verbal elements (for instance BQs), which must be in Aposhions before Spell Out, in order to avoid the formation of ill-formed chains. This is the argument offered by Motapanyane (1989), outlined above. However, the mere presence of BQs as pre-verbal subjects is not sufficient reason to speak of a pre-verbal AgrSP. The Op area of the clause does contain a position that accommodates quantificational elements and out of which quantifiers need not raise at LF; this is the Focus position. BQ subjects might thus occupy the Focus position when they are pre-verbal, and perhaps they always do. In theory, following Cardinaletti (1996), we might hypothesize that there are two AgrSPs, one of them post-verbal, and one pre-verbal. But this option
122
Comparative studies in Romanian syntax
would imply finding evidence that the pre-verbal position is argumental. Only then could it be accepted that there is a second pre-verbal AgrSP in Romanian. The present analysis has in fact, been unable to find evidence for a pre-verbal Case/ argumental subject position in Romanian. On the contrary, the facts we have examined suggest that there is no pre-verbal argumental subject position in Romanian. The pre-verbal subject occupies one of the positions in the left periphery of the sentence, namely it is either Topic or Focus . Two types of arguments are presented: a) At least in certain constructions, like the DSC, the pre-verbal subject can only be described as a Topic, because it has only Topic properties. Significantly, constituents which cannot be topicalized cannot be pre-verbal subjects in these constructions either, clearly because there is no pre-verbal AgrSP to accommodate them. b) Since there is no pre-verbal AgrSP, Romanian has only one pre-verbal position for quantifiers, the Focus position. Consequently there are distributional differences between languages like English, French, Italian, etc., which have two pre-verbal positions for BQs, namely SpecFocP and SpecAgrSP, and Romanian where only one position, Focus, is accessible to BQs.
5.4. The Pre-verbal Position in the DSC The DSC construction has the advantage of throwing light on the properties of the pre-verbal subject position hi contrast with the postverbal AgrSP position, because both positions are lexicalized. The characteristic chain of the DSC, that is, pre-verbal subject +weak / clitic pronominal subject + DP-trace, could, in theory, be either an A or an A1 chain, depending on its head. If the pre-verbal subject position were argumental, that is, a pre-verbal Ag^SP were available, any subject could appear in this construction, including BQs. A BQ in SpecAgTjSP before Spell Out would raise at LF, say to check a [+Focus] feature, leaving behind, as predictable, a variable hi the higher pre-verbal SpecAgrSP, as shown hi (111). This variable is licit, since it is A'-bound by the raised BQ, and it cannot be A-bound by the lower pronoun in Spec Agr2SP. In fact, it will ccommand, and licitby bind, the pronoun in the lower Agr2SP, a pronoun which becomes a secondary variable (see Rizzi 1986a). At LF, the chain would be as hi (111). (111) [FocP BQ, Foc [+Focus] L^sp t, [MP V +Agr..[Agr2SP clitic subject
[w t, [ ^
On the other hand, if there is no pre-verbal AgrS, a BQ should be excluded from the DSC. Before Spell Out, the BQ would already be in an A1 position where it could check its Op feature. But then, in the DSC, the BQ cannot bind a variable, since hi the chain hi (112) below neither the clitic hi SpecAgrSP, nor its trace hi SpecVP, qualifies as a variable.
The double subject construction in Romanian 123
(112) [FocP BQ, [Foc +[Focus] [MP V+Agr..[^SP clitic subject [w t, [ t. The (non)-acceptability of pre-verbal subject BQs is a crucial test for deciding between (111) and (112). But, as we have already mentioned, BQs are completely ungrammatical in the DSC. This proves that there is no pre-verbal AgrSP to host them. The examples in (113) have (112) as the underlying structure. (113) a. *Nimeninu vine el la seminar pe wemea asta. nobody not comes he to seminar on weather-the this 'Nobody will come to the seminar in such a weather.' b. *Cineva nu spune el a§a o prostie. somebody not says he such a nonsense 'Somebody will not say such nonsense.' The pre-verbal subject position in the DSC is not argumental; it is a non-quantificational A'-position, and therefore, a Topic, left dislocation position. The chain (pre-verbal subject + weak /clitic pronominal subject + DP-trace) has to be interpreted as an instance of the CLLD construction. I would, incidentally, decline to express an opinion on the issue of whether the left dislocated element is base-generated in a peripheral position (see Cinque 1990; Dekydtspotter 1994) or involves nonquantificational w/z-movement (see Dobrovie-Sorin 1994). On either account, the expletive pronoun in SpecAgrSP, with which the left dislocated constituent is coindexed, licenses a DP in the argumental SpecVP position. Its trace in SpecVP leads to the formation of chains of the type (Left dislocated DP + weak /clitic pronominal subject + DP trace). Several properties of the construction confirm that the DSC with pre-verbal subject illustrates the Topic / Comment articulation of the sentence, the Topic being left-dislocated. a) First, the lexical subject in the DSC does not allow contrastive stress pre-verbally or postverbally: This means that, in this construction, the lexical subject cannot be interpreted as a (nonquantificational) Focus: (114) a. b.
*Tata vine el, n- ai grija, nu mama. father-FOC comes he not have worry not mother *Vine el, tata, nu mama, comes he father-FOC not mother
b) Second, since the DSC is not quantificational, but involves topicalization, one consequence that can be detected in this construction is that there are no weak cross over effects:
124
Comparative studies in Romanian syntax
(115) a. Copilulj
nu prea este e^ iubit demama luij.
child-the; not really is he j loved by mother his ( 'The child is not really loved by his mother.' b. Copilul; chiar de maica- sa^ii prea esteelj iubit child-the even by mother-bis not really is he loved. c. Copilulj nici pe maica- sa^- o prea iube§teelj. child-the j neither pe mother-his j not her really loves he { 'The child does not even love his mother.' c) The behavior of questions is also revealing. As mentioned above the questioned word is an interrogative quantifier and it is naturally assumed to move to SpecFocP, the quantifier position. Consequently, a questioned subject is excluded in the DSC, because its pre-verbal subject is a Topic, not a Focus: (116) a. b.
*Cinevine el mai tarziu? who comes he more late ?Ion, care vine el mai tarziu, are Ion who comes he more late will sa- §i ia banii de la secretariat. SA REFL take-8UBJ.3SG money-the from the office 'Ion, who comes later, will pick up his money from the office.'
On the other hand the DSC is possible if a constituent different from the subject is questioned: the topicalized subject of the DSC must be placed above the w/z-Focus position of the interrogative word, as in examples (117). Also as expected, in the non-quantificational relative clause construction, the subject itself may be relativized hi the DSC. The relative pronoun is probably placed hi SpecF (the old SpecC), above the left dislocated position, as hi (118). (117) a. Ion, cand o veni el acasa, cine §tie. Ion when will come he home who knows 'Who knows when Ion will come home.' b. Femeia, unde s- o fi dus ea dupa ajutor, cine poate afla ? woman-the where REFL will have gone she for help who can find out 'Who can find out where has the poor woman gone for help?' (118) Fratele
meu, care§tie
el ce
face, te
va ajuta.
brother-the my who knows he what does you will help
The double subject construction in Romanian 125 'My brother, who knows what he's doing, will help you.' d) In a more recent analysis of CLLD, Dekydtspotter (1994) claims that CLLD has mixed argumental and non-argumental properties. For instance, unlike wft-movement and (English) topicalization, CLLD allows floated quantifiers(FQs). (119) French
AUK enfants je leur ai to children I them have all spoken 'I have spoken to all the children.'
This is an argumental property, which should characterize pre-verbal subjects (see Cardinaletti 1996). Pre-verbal subjects in the Romanian DSC are surely compatible with FQs. But of course in the examples below, illustrating DSCs with FQs, there is no need to assign any role to the dislocated phrase in licensing the FQ. We may simply say that the FQ has been licensed by the pronoun in SpecAgrSP, rather than by the lexical subject in LD position: (120) Copiii vin ei maine toji,n -ai grija. children-the come they tomorrow all not have worry 'The children, they will all come tomorrow, don't worry.' In (120), copiii 'the children' sits hi a LD position, ei 'they' is in SpecAgrSP, to\j is in Spec VP. In conclusion, it appears that the lexical subject in the DSC offers a clear example of pre-verbal subject which must be interpreted as a Topic. Furthermore, the non- occurrence of subject BQs in the DSC proves that there is no pre-verbal AgrSP in Romanian.
5.5. Weak Pronouns Cannot Be Topicalized Arguing against the view that pre-verbal subjects are left- dislocated in Italian, Cardinaletti (1996) mentions the existence of weak subject pronouns which cannot be left-dislocated, but which nevertheless occur as pre-verbal subjects SpecAgrSP: the Italian egli lesso are cases in point, as shown in (121a). She remarks (1996, 10) that "weak pronouns cannot precede another leftdislocated element", as in (121b). (121) a. b.
Egli e molto simpatico. 'He is very nice.' *Egli a Maria [pro non le ha parlato ancora].
126
Comparative studies in Romanian syntax he
to Maria
not her has talked yet
Comparable examples are not available in Romanian. Ah1 types of pronouns may appear in subject position even followed by topicalized constituents or by parentheticals. In Cardinaletti's analysis, only strong pronouns in the higher AgrSP can be followed by parentheticals, while weak pronouns like French, //, Italian egli/esso, cannot: (122) a.
b.
c.
El Mariei nu i- a- vorbitinca he Maria-DAT not her has talked yet 'As for Maria, he hasn't yet talked to her.' Asta, dupa cate vad, e o prostie. this as how-much see^lso is a nonsense 'This, as I see it, is just nonsense.' El / Acesta, dupa cate vad
este un geniu.
he this-one as how-much see-ISG is 'As I see it, he/this one is a genius.'
a genius
Significantly, the only clear instance of a weak subject pronoun in Romanian, namely, the pronoun occurring in the DSC, cannot appear in pre-verbal position, because weak pronouns cannot be dislocated. Again, there is no pre-verbal AgrSP to accommodate the weak pronominal subject: (123) a. b.
*Elvine tata. he comes father Vine el tata. comes he father
The impossibility of a pre-verbal weak pronominal subject in the DSC is an important piece of evidence proving the correctness of our analysis. In conclusion, the facts examined regarding the DSC testify that the pre-verbal subject must be a Topic in this construction: constituents that may not be topicalized (BQs, weak pronouns) cannot appear as pre-verbal subjects in this construction because there is no pre-verbal argumental position (SpecAgrSP) to host them.
5.6. Bare Quantifier Subjects As a matter of principle, the discussion of the left periphery of the clause has indicated that there is
The double subject construction in Romanian 127 an A1 position which can accommodate BQs at S-structure, namely Focus. This is or may be a quantificational position, out of which the BQ does not need to further raise at LF, and from which the quantifier may bind the subject variable in SpecAgrSP. We will hypothesize then that pre-verbal BQ subjects occupy the Focus position and will try to substantiate this claim. The type of argument we propose relies on the uniqueness of the Focus position. Under this assumption, it follows that in clauses where there already exists one focalized constituent in the left periphery, a pre-verbal BQ subject is impossible, either before or after this Focus, since a Topic position is not accessible to BQs and Romanian does not possess any preverbal AgrSP. In the Romanian clause, there is a unique pre-verbal quantifier position, the Focus position. This is not true in English, or in any other languages that have a pre-verbal AgrSP, where one quantifier may be in SpecFocP and the other in SpecAgrSP: (124) a. B.
Whom does no one love? What did someone say ?
Several facts regarding the syntax of questions suggest that there is no pre-verbal AgrSP in Romanian. It is generally accepted that the questioned constituent is a (quantificational) Focus. Notice that, since the questioned constituent occupies the Focus position, a second Focus cannot appear before or after the questioned constituent: (125) a.
b.
*La Paris cine pleaca maine ? to Paris-Foe who leaves tomorrow 'Who leaves for Paris tomorrow?' *Cine la Paris pleaca maine ? who to Paris-FOC leaves tomorrow 'Who leaves for Paris tomorrow?'
Appositive relative pronouns, probably hi SpecC/Force, which are not quantifiers but resumptive pronouns, are, hi contrast, compatible with focalized constituents: (126) Omul
acesta, care in via]a lui
n -a
invatatnimic
man-thethis who in life-the his-FOC not has learned nothing ne invata pe noi ce este lingvistica. us teaches pe us what is linguistics-the "This man, who has not learned anything in his life, is teaching us what linguistics is.'
128
Comparative studies in Romanian syntax
One may then take for granted the uniqueness of the Focus position, accommodating focalized constituents, including quantifiers and w/i-constituents. As mentioned above, questioned constituents in Focus may be preceded by a Topic constituent, in the same Topic + Focus order found in declaratives, in examples like (96) above: (127) a.
b.
Mariei ce- i cumpara tata ? Maria-DAT what her buys father 'What will father buy for Mary?' La Paris cine pleaca ? to Paris who leaves 'Who leaves for Paris?'
In particular, the focalized/questioned constituent must be preceded by a topicalized subject: (128) Profesorul pe cine prefera? professor-thepe whom-FOC prefers 'Whom does the professor prefer?' The same sequence topicalized subject + focalized direct object was apparent hi our earlier examples (96), one of which is repeated as (129). (129) Profesorul pe nimeni n -a invitat la el. professor-the nobody-FOC not has invited to him "The professor has invited NOBODY to his place.' It is essential that the subject should be a Topic in such examples, because there are again restrictions on the possible types of subjects. Predictably, BQs, which cannot be Topics, cannot appear in such structures either, as shown in (131) and (133). Since there is only one Focus position, already occupied by the w/i-word, we expect, and get, contrasts between acceptable cases where a Topic (LD) subject constituent precedes an interrogative quantifier, as in (130a) and (132a), and unacceptable cases where a non-interrogative quantifier subject precedes an interrogative one, as in (13 la) and (133a). In every case, the subject may appear post-verbally (the b. examples of each pair below): (130) a.
I on pe cine prefera? Ion pe whom prefers 'Who does Ion prefer?'
The double subject construction in Romanian 129 b.
(131) a. b.
(132) a.
b. (133) a. b.
Pe cine prefera Ion? pe whom prefers Ion 'Whom does Ion prefer?' *Cineva pe cine prefera in asemenea cazuri? someone pe whom prefers in such cases Pe cine prefera cineva in asemenea cazuri? pe whom prefers someone in such cases 'Who does someone prefer in such cases?' loncui nu-i ajuta? Ion who-DAT not him help 'Whom doesn't Ion help ?' Cui nu- i ajuta Ion? whoDAT not him help Ion *Nimeni cui nu-i ajuta? nobody who-DAT not him help Cui nu- i ajuta nimeni? whoDAT not him help nobody 'Whom does nobody help?'
Where two quantifiers compete for the same position in order to check their [+Focus] feature, the sentence results in ungrammaticality. In conclusion, the subject position to the left of the Focus position is not an Agreement position, but a Topic position. At the same time, questions also prove that there is no pre-verbal AgrS position below the FocusP either. Cardinaletti (1996) mentions that in Italian a full subject, in SpecAgrSP, is sometimes allowed to occur between the w/z-constituent and the verb. This position is, however, not available for left dislocated items such as (134b), which precede the w/z-phrase in Italian, as shown in (134c). (134) a. Achi Gianni/nessuno ha connsegnato 1' invito? to whom Gianni/ nobody has given the invitation? 'To whom did Gianni give the invitation?' 'Did Gianni give the invitation to anybody?' b??Achi I1 invito Gianni 1' ha connsegnato? to whom the invitatiion Gianni it has given c. L1 invito, a chi Gianni I1 ha connsegnato? the invitatiion to whom Gianni it has given 'As for the invitation, who did Gianni give it to?'
130
Comparative studies in Romanian syntax
(134c) can be represented as in (135) showing the order of the three projections in Italian. (135)
[Topicp rinvito[FocusP [ a chi [^p Gianni Ilia consegnato]]]
It is highly significant that no possibility of a pre-verbal subject is available in the Romanian equivalents of these examples, obviously because there is no SpecAgrSP below Focus where the subject could remain. Moreover, there is no Topic position below a w/z-Focus either, as confirmed by example (136b). The only pre-verbal position accessible to the subject is the Topic position to the left of the Wi-Focus, as in (137). (136)
a. b.
(137)
a.
b.
*Cui loni -a dat invitatia? who-DAT Ion him has given invitation-the *Cui invita$ia i- a dat- o Ion? who-DAT invitation-the him has given it Ion Ion cui i- a dat invitatia? Ion who-DAT him has given invitation-the 'Who did Ion give the invitation to? Invitatia cui i- a dat- o Ion? invitation-the who-DAT him has given it Ion 'The invitation, who did Ion give it to?'
hi sum, the facts we have examined prove that Romanian possesses only one pre-verbal quantifier position, namely, Focus, whereas, two positions ought to be available if there were a pre-verbal argumental subject position. We conlude that there is no AgrSP before the verb in Romanian. Preverbal BQ subjects occupy the Focus position.
6. CONCLUSIONS We have argued on the basis of morphological and syntactic facts that Romanian clauses are (at least) Mps. The functional structure of the MP includes (at least) the following ordered categories: MP > C1P> AgrSP> TP > AspP.... VP There are two argumental subject positions hi the Romanian clause, both of them post-verbal: a) the thematic position which is SpecVP; b) the Case and/wo-licensing position which is SpecAgrSP. Case is uniformly checked hi SpecAgrSP, except for the DSC, where SpecTP is also involved. Depending on the intrinsic semantic properties of the DP, it will check Case overtly or covertly. The pre-verbal subject is in the non-argumental position of left dislocated Topic or Focus, depending
The double subject construction in Romanian 131 on the particular [+Topic] or [+Focus] feature it must check. The position that we have reached is similar to that expressed in Alexiadou (1994) on Modern Greek. She also claims that in SVO the subject cannot be in SpecAgrSP and that to license a pre-verbal subject an additional TopicP and, we should say, FocusP, is needed. The Agr verbal features do not include the Topic/Focus feature, so such features are checked in special operator projections. Thus, although we use different arguments, we have confirmed the analysis hi Dobrovie-Sorin (1994). Romanian gives more prominence to the Topic Comment and Presupposition Focus articulation of the sentence than other Romance languages. Romanian again turns out to be a more Topic oriented than Subject-oriented language, a feature which is hi line with the higher degree of orality often discussed hi the literature.
REFERENCES Alexiadou, A. (1994). Issues in the syntax of adverbs. Doctoral dissertation, Potsdam University. Baker, M. (1988). Incorporation. University of Chicago Press, Chicago. Belletti, A. (1988). The Case of unaccusatives. Linguistic inquiry 19, 1-34. Beletti, A. (1990). Generalized verb movement. Rosenberg and Sellier, Torino. Borer, H. (1984). Parametric syntax. Foris, Dordrecht. Burzio, L. (1986). Italian syntax. Reidel, Dordrecht. Burzio, L. ( 1994). Case uniformity. Manuscript, Johns Hopkins University. Byck, J. (1937). L'emploi affectifdu pronom personnel en roumain. Buletin de linguistique, 15-33. Cardinaletti, A. (1996). Subjects and clause structure. University of Venice Working Papers in Linguistics 6(1), 55-95. Cardinaletti, A and I. Roberts (1991). Clause structure and X-Second. In: Levels of representation (W. Chao and G. Harrocks, eds.), pp. 34-73. Foris, Dordrecht. Cardinaletti, A and M. Starke (1994). The typology of structural deficiency. On the three grammatical classes. University of Venice Working Papers in Linguistics 4(2), 41-109. Cinque, G. (1990). Types ofA-Bar dependencies. MIT Press, Cambridge, Mass. Cinque, G. (1993a). A null theory of phrase and compound stress. Linguistic Inquiry 24, 239-297. Cinque, G. (1993b). Functional structure in the Romance sentence and the DP. Course notes, Girona summer school of linguistics. Cinque, G. (1997). Adverbs and functional heads. A cross-linguistic perspective. Oxford University Press, New York. Chomsky, N. (1981). Lectures on Government and Binding. Foris, Dordrecht. Chomsky, N. (1991). Some notes on Economy of derivation and representation. In: Principles and parameters in comparative grammar (R. Freidin, ed.), pp. 417-454. MIT Press, Cambridge,
132
Comparative studies in Romanian syntax
Mass. Chomsky, N. (1995). The Minimalist Program. MIT Press, Cambridge, Mass. Cornilescu, A. (1996). Concepts of modern grammar. Bucharest University Press, Bucharest. Comilescu, A. (1997). Some HAVE curiosities. Manuscript, University of Bucharest. Dekydtspotter, L. (1994). The Left Dislocated construction. Manuscript. University of Paris 7. Deprez, V. (1989). On the typology of syntactic positions and the nature of chains. Doctoral dissertation, Massachusetts Institute of Techonology. Diesing, M. (1992). Indefinites. MIT Press, Cambridge, Mass. Dobrovie-Sorin, C. (1987). Syntaxe du roumain. Doctoral dissertation, University of Paris 7. Dobrovie-Sorin, C. (1994). The syntax of Romanian. Comparative studies in Romanian. Mouton de Gruyter, Berlin. Dobrovie-Sorin, C. (1998). Impersonal se construction in Romance and the Passivization of unergatives. Linguistic Inquiry 29, 399-437. Gueron, J. and T. Hoekstra, ( 1995). The temporal interpretation of Predication. In: Small Clauses (A. Cardinaletti and M-T. Guasti, eds.), pp. 77-109. Academic Press, New York. Hoop, H. de (1992). Case configurationality and Noun Phrase interpretation. Doctoral dissertation, University of Groningen. Isac, D. (1996). Sentence structure and verb movement in Romanian. Cahiers de linguistique theorique et appliquee 29, 99-107. Kayne, R (1989). Null subjects and clitic climbing. In: The Null Subject Parameter (O. Jaeggli and K.Safir, eds.), pp. 239-261. Kluwer, Dordrecht. Kayne, R.(1993). Towards a modular theory of auxiliary selection. Studio Linguistica 47, 3-31. Kayne, R (994). The antisymmetry of syntax. MIT Press, Cambridge, Mass. Koizumi, M. (1995). Phrase structure in Minimalist syntax. Doctoral dissertation, Massachusetts Institute of Technology. Lasnik, H. (1992). Case and Epletives. Notes toward a parametric account. Linguistic Inquiry 23, 381-406. Mahajan, A. (1990). The A / A -bar distinction and movement theory. Doctoral dissertation. Massachusetts Institute of Technology. Motapanyane, V. (1989). La position du sujet dans une langue a 1'ordre SVO / VSO. Rivista di grammaticagenerativa 14, 75-103. Ouahalla, J. (1992). Functional categories and parametric variation. Routledge, London. Pana- Dindelgan, G. (1974). Sintaxa transforma^onala a grupului verbal in limba romdna. Editura Academiei, Bucharest. Pana-Dindelgan, G. (1976). Sintaxa limbii romdne. Bucharest University Press, Bucharest. Poletto, C. (1991). The diachronic developments of subject clitics in North Eastern Italian dialects. University of Venice Working Papers in Linguistics 1, 15-32. Pollock, J,Y. (1988). Verb movement, UG and the structure of IP. Linguistic inquiry 20, 365-
The double subject construction in Romanian 133
424. Progovac, L. (1993). Long-distance reflexives. Linguistic inquiry 24, 755 - 722. Ramchand, G.C. (1996). Two subject positions in Scottish Gaelic: The syntax semantics interface. Natural Language Semantics 4, 165-191. Raposo E. and Uriagereka J. (1995). Two types of Small Clauses. Towards a syntax of Theme / Rheme relations. In: Small Clauses (A. Cardinaletti and M-T. Guasti, eds.), pp. 179198. Academic Press, New York. Rivero, .ML. (1994). Clause structure and V-movement in the languages of the Balkans. Natural Language and Linguistic Theory 12, 63-120. Rizzi, L. (1982). Issues in Italian Syntax. Foris, Dordrecht. Rizzi, L. (1986a). Null objects in Italian and the Theory of pro. Linguistic Inquiry 17, 501-557. Rizzi, L. (1986b). On the status of subject clitics in Romance. In: Studies in Romance linguistics (O. Jaeggli and C. Silva-Corvalan, eds.), pp. 391-420. Foris, Dordrecht. Rizzi, L. (1990). Relativized Minimality. MIT Press, Cambridge, Mass. Rizzi, L. (1997). The fine structure of the left periphery. In: Elements of grammar. Handbook of generative grammar (L. Haegeman, ed.), pp. 281-337. Kluwer, Dordrecht. Roberts, I. (1991). Verbs and diachronic syntax. Kluwer, Dordrecht. Roberts, I. (1994). Long Head Movement, Case and Agreement in Romance. In: Verb Movement (N. Hornstein and D. Lighfoot, eds.), pp. 207-242. Cambridge University Press, Cambridge. Safir, K. (1985). Syntactic chains. Cambridge University Press, Cambridge. Sportiche, D. (1992). Clitic constructions. Manuscript, University of California Los Angeles. Starke, M. (1995). On the format for Small Clauses. In: Small Clauses (A. Cardinaletti and M-T. Guasti, eds.), pp. 237-260. Academic Press, New York. Suner, M. (1994). V-Movement and the licensing of argumental wh-phrases in Spanish. Natural Language andLinguisic Theory 12, 336-378. Uriagereka, J. (1994). Aspects of clitic placement in Western Romance. Linguistic Inquiry 25, 79-123. Zubizaretta, M.L. (1994). The grammatical representation of Topic and Focus. Implications for the structure of the clause. University of Venice Working Papers in Linguistics 4(1), 97132.
This page intentionally left blank
Last resort strategies in DP
135
LAST RESORT STRATEGIES IN DP: ARTICLE REDUPLICATION IN ROMANIAN AND FRENCH Yves D 'Hulst, Leiden University Mar tine Coene, Antwerp University Liliane Tasmowski, Antwerp University
i. INTRODUCTION 1.1. Background In a number of specific syntactic contexts, like (1), Romanian inserts before a nominal phrase (vecinilor 'neighbors') which bears genitive Case and acts as the possessor of another noun (copilul 'the child'), a special kind of determiner, al, traditionally referred to as "the possessive article". The exact nature of this possessive article, as well as its origins, are rather unclear. (1)
copilul inteligent child-the-MASC.SG intelligent-MASC.SG al vecinilor OT/-MASC.SG neighbours-the-MASC.PL.GEN/DAT 'the neighbours' intelligent child'
The general purpose of this paper is to give an account of the syntax of the possessive article and
"This paper is an extended version of a talk which was first presented at the Workshop on Romanian Syntax (Saint John, NB, August 30,1996), and subsequently at Going Romance (Utrecht, December 12-14,1996) and the Possession and Dispossession conference (Bucharest, June 2-4, 1997). We would like to thank the audiences of these conferences, as well as Larisa Awam, Alexandra Comilescu, Carmen Dobrovie-Sorin, Corina Lascu, Ileana Oancea, Dorin Perie, Sanda Reinheimer, Johan Rooryck and two anonymous reviewers for their helpful suggestions and comments. This research was supported by grant G 0342.97 of FWO-Vlaanderen.
136
Comparative studies in Romanian syntax
to explain how it relates to the licensing of genitive Case. Our analysis will be mainly based on two assumptions within Chomsky's (1995) Minimalist Program: (a) that movement is driven by the need to check features and (b) that movement in overt syntax follows from the fact that the features to be checked are strong. We will show that the possessive article al is used in Romanian whenever the genitive DP denoting the possessor (vecinilor 'neighbors'), henceforth DPgen, cannot raise to the position where its strong Case-feature can be checked; that is, in contexts where we assume that overt movement of DPgen is required by virtue of the second assumption above, but blocked for independent reasons (here the presence of the adjective inteligent 'intelligent'). We will argue that al heads a projection which contains another position where the strong Casefeature of DPgen can be checked. Structures containing al, therefore, offer an alternative way to check strong Case-features, hence an escape hatch against a violation of obligatory overt movement. Structures containing al are more complex, hence more costly, than those without. Therefore, they should be available only if they are the only means to satisfy the grammatical requirement of genitive Case checking. In this sense, structures with al constitute a last resort (in the sense of Chomsky 1995, Marantz 1995 among others). This paper is organized as follows. In section 1. we will briefly summarize the Minimalist stance on movement. Section 2. introduces some basic facts about the properties of nominal elements (nouns, adjectives and determiners) in Romanian and the structure of DPs. In section 3., we will address the issue of the internal structure of a/-phrases. After discussing the two major proposals that have been offered in the literature, we will sketch an alternative hypothesis. The basic idea of our proposal will be that the Romanian a/-phrase is a DP of its own, and as such offers a position for the checking of genitive Case. The problem of licensing genitive Case in Romanian will be addressed in section 4. Adopting the Minimalist hypothesis that Case features are licensed through checking in Spec-head configurations, we will show that the hypothesis sketched in section 3., allows genitive Case to be uniformly checked both in DPs without a/-phrases and in DPs that contain al. Independently of the genitive phrase they contain, ^/-phrases have properties of then- own, hi particular Case features. We will argue, hi section 5., that nouns have the possibility of licensing Case features. Finally, in section 6., we will show that the last resort strategy Romanian exploits for licensing genitive phrases is to be found hi other Romance languages as well, albeit in other domains and for other reasons. More specifically, we will show that reduplication of the determiner hi post-nominal superlatives hi French is quite similar to the syntax of Romanian a/-phrases.
Last resort strategies in DP
13 7
1.2. Overt Movement and Feature-checking It follows from long-standing research in the theory of Principles and Parameters that specific functional projections are endowed with the property to check specific features. Typical preMinimalist examples include Agr or C which check a Case and a w/z-feature, respectively: hi (2a) the subject moves to SpecAgrP in order to have its Case feature checked by Agr; similarly, the w/z-phrase of (2b) moves to SpecCP where its w/z-feature is checked by C. (2)
a b
\j^f Jerry Agr [yp tj fooled Tom]] [CP Who C^ did Jerry fool tj] ?
Chomsky (1995) reinterprets the relationship between feature checking and movement as the consequence of an operation of feature attraction: the strong nominal (D) features on Agr hi (2a) and on C in (2b) attract the subject and the w/i-phrase, respectively, to their specifiers (Spec) positions. Furthermore, movement obligatorily applies in overt syntax, as in (2), whenever the attracting feature is strong; otherwise, movement will apply after Spell Out. Consider a structure like (3), with a being a head bearing a strong feature F, and P being the closest element with a [+F] specification. Since the feature of a is strong, movement of p should apply in overt syntax. If movement of p"13" (=ZP) to SpecXP or adjunction of p to a is barred for independent reasons (conditions on movement) (3) will be ruled out.
(3)
[xp[ *XP /
Spec
X'
This conclusion is empirically motivated for derivations like those in (4). In (4a), fooled is unable
138
Comparative studies in Romanian syntax
to check the strong [V] feature on C as lexical verbs, for whatever reason, are not allowed to move out of VP. Similarly, in (4b) the strong D (or wh) feature on C cannot be checked as where is not able to move out of the w/z-island. (4) a.*[Cp Wh0i [c, fooled, [^ Jerry [w b.*[CP Wherei [c, did [^ I wonder [CP if [^p Jerry ft [VP fooled Tom]]]]]]]? Usually, languages exploit a number of (last resort) strategies to avoid non- convergent derivations like (4). English cto-insertion, as in (5), is a classical case and we will argue that the Romanian possessive article is another, albeit a more complex one. (5)
[CP Who; [c, did, [^p Jerry t, [w fool tj]]]?
2. ASPECTS OF THE SYNTAX OF ROMANIAN DPs In the nominal sphere, Romanian differs from other modern Romance languages in at least two major respects: (a) the syntax of the definite article and (b) morphological Case. We will give a rough sketch of these issues in 2.1. and 2.2. In 2.3., we will address the basics of DPs with morphological genitive Case (DPgen).
2.1. The Definite Article Unlike its equivalents in other Romance languages, the Romanian definite article is an enclitic: it is either preceded by a noun, as hi (6), or by an adjective, as in (7). Furthermore, it agrees in gender and number with the noun as can be seen (for gender) hi the contrast between the (a) and (b) examples and (for number) in the contrast between the examples (a) versus (a') and (b) versus (b')-1 (6)
a.
a'.
copilul frumos child—the- MASC.SG beautiful-MASC.SG 'the beautiful child' copiii frumo$i
Romanian has three gender classes: masculine, feminine and neuter. "Neuter" is somewhat misleading: in the singular, neuter nouns pattern in all respects with masculine nouns, taking masculine determiners and occurring with adjectives that agree as if neuters were masculine; in the plural, they fully pattern with feminine nouns. Since neuters do not interfere with the facts we will be considering, we will disregard them for the rest of this article.
Last resort strategies in DP
139
child-the- MASC.PL beautiful- MASC.PL 'the beautiful children' b.
copila frumoasa child-the-FEM. so beautiful- FEM.SG
b'.
copilele
'the beautiful child' frumoase
child-the- FEM.PL beautiful- FEM.PL 'the beautiful children' (7)
a.
frumosul
copil
beautiful-the-MASC.SG child-MASC.SG a'.
'the beautiful child' frumo§ii
copii
beautiful-the- MASC.PL children- MASC.PL 'the beautiful children' b.
frumoasa copila beautiful-the-FEM.SG child-FEM.so
b.
frumoasele
'the beautiful child' copile
beautifiil-the-FEM.PL children-FEM.PL 'the beautiful children' The enclitic nature of the definite article has been analyzed by Dobrovie-Sorin (1987) as the consequence of raising the noun or adjective to the left-side of the article. Building on this insight and Abney's (1987) DP-hypothesis, Grosu (1988), Giusti (1993), Bernstein (1993) and Cornilescu (1994) among others reinterpret the position of the article as the result of head movement of the noun to D, in the case of N-D configurations like (6), or as the result of movement of the adjective phrase to SpecDP, in the case of A-D configurations like (7). Their proposal can be readily restated in Minimalist terms, assuming the definite article has a strong Nfeature attracting a [+N] category, a noun as in (8a), or an adjective as in (8b) (see Longobardi 1994). (8)
a
copilul
frumos
child-the-MASC.so beautiful-MASC.SG 'the beautiful child' b
frumosul
copil
beautiful-the-MASC.SG child-MASC.SG 'the beautiful child'
140
Comparative studies in Romanian syntax DP
a'. b'.
copiluj-lfrumos frumosUj-l tt
copilj
2.2. Morphological Case Unlike other Romance languages, Romanian has a productive opposition between nominative/accusative and genitive/dative Case. Although virtually nonexistent for both nouns and adjectives, this Case distinction is fully articulated for determiners, including the definite article.2
2
Only feminine singular forms may be said to be morphologically Case-marked since they have a distinct form for nominative/accusative.
Last resort strategies in DP
141
Table 9 masculine singular
plural NOM /ACC
GEN /DAT
NOM /ACC
GEN /DAT
N
copil3
copil
copii
copii
D
-1
-lui
-i
-lor
N-D
copilul
copilului
copiii
copiilor
feminine singular
plural NOM /ACC
GEN /DAT
NOM /ACC
GEN /DAT
N
copil
copile
copile
copile
D
-a
-i
-le
-lor
copilei
copilele
copilelor
4
N-D
copila
2.3. Distribution of DPgens Turning to DPgens, we see from the contrast in (10) that they must be adjacent to the definite article: whenever the adjacency is broken down by an adjective as, for example, in (lOb), the structure is ruled out. Nevertheless, structures like (lOb) can be rescued if DPgen is preceded, as in (1 la), by what Romanian grammars call the "possessive article". Note, furthermore, that the possessive article is barred when adjacent to the definite article, as in (lib). In sum, configurations with adjacency between definite article and DPgen on the one hand, and configurations with adjacency between the possessive article and Dpgen, on the other hand, are in complementary distribution:
3
Masculine singular nouns which end in a consonant are underlyingly -Cu#. Final -u deletes unless it is needed for syllabification as in (i). (i) [li.tru] [co.pi.lul] [ta.bii] litre child-the taboo
4
Copila is the result of the obligatory phonological deletion rule in (i) which is pervasive throughout Romanian morphology (see Tasmowski 1989, Van Eeden 1994):
(0
a-»0/_v
(ii)
cdnta-am —»cantam sing-lPL
fdtS-o —*fato girl-voc
masS-a -+ masa table-the
142
Comparative studies in Romanian syntax
(10) a. copilul vecinilor child-the-NOM/ACC.MASC.SG neighbors-the-GEN/DAT. MASC.PL 'the neighbors' child' b. *copilul
inteligent
vecinilor
child-the-NOM/ACC intelh'gent-MASC.SG neighbors-the-GEN/DAT 'the neighbors' intelligent child' (11)
a. copilul inteligent child-the-NOM/ACC. MASC . so intelligent-MASC. so al
vecinilor
O/-MASC.SG neighbors-the-GEN/DAT.MASC.PL 'the neighbors' intelligent child' b. *copilul al vecinilor child-the-MASC.SG a/-MASC.SG neighbors-the-GEN/DAT.MASC.PL 'the neighbors' child' The contrasts in (10) and (11) highlight the issues any analysis of DPgens in Romanian should address: (12)
(i). How are DPgens licensed in Romanian? (ii). What is the relationship between DPgens and the possessive article?
The issue raised in (12.ii), at least partly, concerns the internal syntax of a/-phrases and will be addressed in section 3. where, on the basis of morphological properties of possessive articles, a specific proposal about the syntactic structure of a/-phrases will be developed. Discussion of question (12.i) will be postponed until section 4. where we will show that the proposal in section 3. allows a unitary account of the licensing of DPgens, whether they occur in a/-phrases or not.
3. INTERNAL SYNTAX OF THE POSSESSIVE ARTICLE 3.1. The Morphological Paradigm The possessive article agrees in gender and number with the possessee, not with the possessor (Le. DPgen): al in (13a) is masculine singular because fiu >son= is masculine singular. The other examples illustrate the rest of the agreement pattern of the possessive article. (13) a. bietul
fiu
al
Jaranului
poor-the-NOM/ACC.MASC.SG son O/-MASC.SG farmer-the-GEN/DAT.MASC.SG
Last resort strategies in DP
143
'the farmer's poor son' b. bie ii fii poor-the-NOM/ACC.MASC.PL sons ai ranului a/-MASC.PLfarmer-the-GEN/DAT.MASC.SG 'the farmer's poor sons' c. biata file poor-the-NOM/ACC.FEM.SGdaughter-FEM.SG a ranului a/-FEM.SGfarmer-the-GEN/DAT.MASC.SG 'the farmer's poor daughter' d. bietele fiice poor-the-NOM/ACC.FEM.PLdaughters-FEM.PL ale ranului a/-FEM.PLfarmer-the-GEN/DAT.MASC.so 'the farmer's poor daughters' As can be seen from these examples and the table in (14), the morphological paradigm of the possessive article closely resembles that of the definite article, a resemblance which has a historical ground: it is generally accepted that the possessive article, or at least part of it, goes back to Latin ille, just like the definite article. Table 14 article
masculine
feminine
singular
plural
singular
plural
definite
-1
-i
-a
-le
possessive
al
ai
a
ale
Comilescu (1993, 1994, 1995) argues that the possessive article is a D, thereby explaining both its morphological resemblance and etymological relatedness to the definite article. Grosu (1988, 1994) goes further, assuming the possessive article contains the definite article: under his hypothesis, possessive articles are lexical compounds of the preposition a and the definite article. If the resemblance between possessive and definite articles were the result of more deeply shared properties than those proposed here, one would expect that, besides number and gender features, the possessive article would also bear a Case feature just like that of the definite article. This expectation seems to be justified given the existence of a genitive/dative form like a/or, as can be seen in table (15). We should note however, that the genitive/dative paradigm is defective and
144
Comparative studies in Romanian syntax
that alor has a distribution which is rather different from the regular al, ai, a, ale forms.5 Table 15 masculine
article
singular
feminine plural
singular
plural
NOM/ACC
definite
-1
-i
-a
-le
possessive
al
ai
a
ale
-i
-lor
GEN/DAT
definite
(u)-hii
possessive
-lor
alor
alor
On the basis of these distributional facts we may complement the issues any analysis of genitives in Romanian should address: (16)
(i) How are genitive phrases licensed hi Romanian? (ii) What is the relationship between genitive phrases and possessive articles? (iii) What is the agreement rule of the possessive article with the possessee? (iv) Assuming the possessive article bears a Case feature, why is its regular, semantically void use restricted to nominative/accusative?
3.2. The Construct State Hypothesis Cornilescu's (1994) answer to (16.ii) is a specific proposal concerning the internal structure of a/-phrases: she analyzes the possessive article as a D taking a DP/QP complement in a configuration mirroring the internal structure of gen-assigning construct state nouns.6
5
Unlike Nom/Acc. forms, alor can surface immediately after the definite article; alor also has a pronoun-like referential interpretation (as Span, los de Juan 'those of Juan') contrasting with the semantically void al, ai, a, ale. (i) casa alor mei
house-the-FEM.so O/-GEN/DAT.PL mine-MASc.pL 'my parents' house' 6
"Construct state" refers to a construction found in Semitic languages where DPs containing a possessor may yield two derivations: a "free state", as in (i.a) or a construct state, as in (i.b). In (i.a) the possessor is introduced by a preposition and the possessee is (normally) introduced by the article. The construct state lacks both these properties. (i) a. ha-bayit sel ha-'is Hebrew the-house of the-man 'the man's house'
Last resort strategies in DP
145
Construct state nouns have been analysed as nouns which move to D as shown in the analysis of the Hebrew example in (17a).7 Cornilescu assumes that movement of the noun to D allows the noun to assign Case under government to a possessor in adjacent position. To the extent that a government relationship and adjacency also obtains between the possessive article and the possessor (i.e. DPgen) in Romanian, Cornilescu assumes a construct-state-like analysis for these constructions as well. Her analysis is given in (17b). (17)
a.
beytha-'is house the-man 'the man's house' beyt,.
b.
al fetei flr/-MASC.SG girl-the-GEN.FEM.SG
'the one of the girl' DP D' D^~^ DP al D' D
NP
feteri
N' N
'/ Note, however, that there are some striking differences between construct state constructions like (17a) and cr/-phrases which cast doubt on analysing a/-phrases as in (17b): (i) In construct state constructions the complement of D is an NP; however, the possessive article has a DP as complement, since an NP complement would yield ungrammaticality, as shown in (18). Given (18), it follows that only ^-constructions have a structure with recursive
b.
7
beyt ha-'is house the-man 'the man's house'
Hebrew
Within the Generative framework, the N-to-D raising hypothesis for construct state nouns has been advanced and motivated among others by Borer (1989,1996), Dobrovie-Sorin (present volume), Longobardi( 1994,1996),Ouhalla (1991), Ritter (1987, 1988), Rouveret (1994), and Siloni (1997a, 1997b). Several of these authors point out that construct state constructions, while more pervasive in Semitic, are present in other languages as well, including Romance languages.
146
Comparative studies in Romanian syntax
DP, as in (17b), whereas other construct state constructions may also have a D-NP structure. This contrast is difficult to explain on the basis of (17). (18)
*alfete 0/-MASC.SG gkl-GEN.FEM.SG
'the one of girl'
(ii) The possessor appears in a Spec position with construct state nouns and in a complement position in al- constructions. As a consequence, licensing of the genitive on the complement in (17b) is less straightforward within a framework where Case is licensed through feature checking in a Spec-head configuration and no longer through assignment under government.8 (iii) Construct state nouns raise and substitute for D; no raising and substitution takes place hi possessive article constructions: the possessive article is taken to be a lexical item without further internal structure, inserted hi D. Cornilescu's answer to (16.iii) seems less problematic: she assumes that the agreement between the a/-phrase and the possessee derives from a Spec-head agreement relation in a lower functional projection of the possessee-noun (labelled AgrGenP), as hi (19). (19)
8
elevele bune ale students-the-FEM good-FEM.PL a/-FEM.PL acestui profesor this-GEN.MASC.so teacher-MASC.SG 'this teacher's good students'
Note that (17a) is neutral between these positions: a government relation obtains between D and the specifier of its complement and Case features can be checked in a Spec-head configuration in the course of the derivation.
Last resort strategies in DP
147
Note, however, that deriving the features of al from a Spec-head relationship is not consistent with the fact that determiners in Romanian, and in Romance in general, draw their gender and number features from their complement, as shown in (20). (20)
a.
[or1111 [NP student]] a-MASC.SG student-MASC.SG
b-
[DP<>[NP student ]] a-FEM.SG student-FEM.se [DPni te [NP studen i]] INDEF.PL students-MASC.PL
c.
3.3. The P-D Compound Hypothesis Grosu (1988,1994) offers an alternative answer for question (16.ii), analyzing a/-forms as lexical compounds of a preposition (a) and the definite article (-/). This analysis does not call for specific complementation rules as k Comilescu's construct state analysis: prepositions take DP complements. Furthermore, Grosu's hypothesis makes it possible to simplify Comilescu's basic
9
The existence of Num as a functional head of the noun has been suggested independently by Ritter (1991) and Valois (1991).
148
Comparative studies in Romanian syntax
insight that definite articles assign genitive Case: taking a/-forms to contain the definite article, it is possible to reduce the property to assign genitive Case to a single lexical item: the definite article. A number of diachronic, comparative and synchronic arguments support Grosu's hypothesis. First, one of the etymological hypotheses traces a/-forms back to Latin preposition ad + *illu. Second, preposition and determiner compounding is not so uncommon in Romance languages; French has the prepositions a and de compounding with the definite determiner in au, aux and du, des and similar cases exist in Italian, Spanish, Portuguese, Catalan and Occitan. Third, Romanian prepositions followed by a genitive complement always bear some form of the definite enclitic article, as is illustrated below in (21). (21)
[PP [P' [P in spate-le] [DP casei ]]] in back.-the-MASC.SG. house-the-GEM.FEM.SG 'behind the house'
However, the parallel between a morphologically complex preposition as in spatele 'behind' and the possessive article breaks down in a number of important respects: (i) Genitive-assigning prepositions can be parsed into a stem and a definite enclitic article with idiosyncratically fixed gender and number features, whereas the possessive article, or the definite article it contains, dynamically agrees with the possessee.10 (ii) Genitive-assigning prepositions all have semantic content (mostly locative), whereas the possessive article is inserted for pure syntactic reasons and does not have any semantic content. (iii) Article incorporation in Romance is restricted to monosyllabic prepositions (e.g. French de 'of, a 'to', etc.). Putting aside possessive al, Romanian monosyllabic prepositions typically lack this option: compare Romanian (22a) with Italian (22b). (22)
a. b.
*cul prim tren with-the first train col primo treno
In the next section, we will explore another possibility which is in line with Grosu's original insight that a/-forms have an internal structure and that they contain the normal definite article.
10 The number and gender features of complex prepositions in Romanian either depend on the N-stem they contain (e.g. In spatele * behind' with -le agreeing with neuter singular spate 'back') or they are idiosyncratically fixed (e.g. dedesvbtul 'under' with masculine singular -/, contra 'against' with feminine singular -a). The observation in the text is not affected by this specification: genitive assigning prepositions are inherently specified for gender and number, while al draws its features from the outside.
Last resort strategies in DP
149
3.4. The N-D Compound Hypothesis In 2.1. we reinterpreted Cornilescu's (1994) analysis of the enclitic position of the Romanian definite article as the result of the attraction of a +N category to D or SpecDP by the strong nominal feature on the definite article. This results in regular N-D sequences, like (7a), repeated in (23). If we follow Grosu (1988, 1994) and take the possessive article to contain the definite article, we are compelled to conclude that a bears the categorial feature +N. We will assume that it is a noun. As a consequence, the internal structure of a/-phrases is equivalent to the internal structure of ordinary DP's, as indicated in (24). (23)
copilul child-the- MASC.SG 'the child'
(24)
Since a combines with all nominative/accusative forms of the definite article regardless of gender and number, we may further assume that it is unspecified for number and gender features.n This assumption seems to be independently supported by the behavior of pronominal demonstratives: whenever the demonstrative is not followed by a lexical N, it takes an invariant a ending. Compare pre-nominal acest in (25a) with pronominal acesta in (25b).12 The pattern in (25)
11 An anonymous reviewer notes that the absence of number and gender features on a might argue in favor of a prepositional status, as in Grosu's (1988) analysis. As we argued in section 3.3., such an analysis leaves several problems unanswered. Furthermore, if we take Harris' (1991, 44) view that the gender feature is privative, that is, it is only present for the marked value(s) (feminine in Spanish, and arguably in Romanian as well), approximately half of the nouns (traditionally referred to as "masculine" nouns) have to be analyzed as lacking a gender feature. The number feature can be analyzed in exactly the same way. This means that a is not different from, for example, lup 'wolf: both lack gender and number specifications. 12 The same form acesta is used post-nominally as in (i). We assume that the internal structure of the post-nominal form in (i) equals the internal structure of pronominal acesta in (25b). This implies that post-nominal demonstratives are generated in some specifier position between D and N (see Giusti 1997).
(i)
copilul acesta child-the-MASC.SG this-MASC.SG-a 'this child'
150
Comparative studies in Romanian syntax
suggests that Romanian demonstratives do not license empty Ns and that a is a placeholder for N, similar to English one in this one.13 (25)
a.
acest copil this-MASC.SG child-MASC.SG
a'.
'this child' aceast fat this-FEM.SG girl-FEM.SG
b.
b.'
'this girl' acesta this-MASC.SG-a 'this one' aceasta this-FEM.SG-a 'this one'
If we indeed take the internal structure of a/-phrases to equal the internal structure of ordinary DPs, as in (24), the null hypothesis is that the relationship between a-D and DPgen equals the relationship between N-D and genitives as well. As a consequence, the answer to (16.ii) must follow from the answer to (16.i), the peculiarity of ^/-phrases reducing to their external distribution, not to their internal syntax.
Although we take demonstratives always to have the same internal structure (acest + N, where N might be lexical or a), we do not assume that the existence of post-nominal demonstratives necessarily indicates that pre-nommal demonstratives are generated in the same position and raised to D or SpecDP as proposed in Bernstein (1993) and Giusti (1997), respectively. In the remainder of the text we will turn to the more traditional view, assuming that demonstratives are determiners generated in D. 13 As was pointed out by an anonymous reviewer, Romanian demonstratives seem to contrast with demonstratives in Italian and Spanish, which can apparently license empty Ns. Following Bernstein (1993b), however, the final vowel of Italian and Spanish demonstratives are generated independently as word markers in one of the functional projections between N and D. Interestingly, the final vowel of the demonstrative can drop only if it is followed by a noun, as shown in (i). According to Bernstein, this suggests that it is the word marker that licenses the empty N. Her analysis provides a unified account for Romanian, Spanish and Italian. (i) a quel libro Italian
this-MASC.SG book-MASC.SG
b.
quello this-MASC.SG 'this one'
Last resort strategies in DP
151
4. GENITIVE CASE CHECKING 4.1. The Internal Structure of DPs Recall from section 2.3. that genitive phrases are adjacent to either the definite article or to the possessive article. Within a pre-Minimalist framework, both Cornilescu (1994, 1995) and Grosu (1988, 1994) take these distributional facts as positive evidence for the hypothesis that the definite article assigns genitive Case under strict government. In keeping with the standard Minimalist assumption that Case assignment/checking obtains in Spec-head configurations, we have to separate the Case assignment/checking property from the adjacency requirement and restate the latter in terms of configurational properties of Dps. Ever since Abney's (1987) DP-hypothesis, several hypotheses have been put forward to define a more articulated internal structure for nominal phrases. In the literature, evidence has been given for, at least, a Possessive Phrase (Delfitto & D'Hulst 1995 ), a Number Phrase (Ritter 1991) and a Gender Phrase (Picallo 1991), each of which is the locus for particular features of the noun, as is illustrated in (26). Cinque (1993) also proposes a more highly articulated internal structure for nominal phrases, but on different grounds: he argues that different classes of adjectives are generated in the specifier positions of different functional projections: (26) [op Y [Possp Y [NumP Y [0enP Y U Y ]]]]] Within this more articulated internal structure for nominal phrases, there is a straightforward way to implement the adjacency requirement between D and the genitive phrase if we take PossP to be the complement of D (see Delfitto & D'Hulst 1995) and if we assume that genitive phrases mcve from their base position (which we assume to be SpecNP) to SpecPossP: in a language where Poss requires its specifier to be filled in overt syntax, the adjacency requirement will automatically follow. We assume Romanian to be such a language and analyse overt movement of the genitive phrase to SpecPossP as the consequence of a strong D-feature on Poss attracting the genitive DP. The relevant derivations are given in (27). (27)
copilul vecinilor child-the-masc. sg neighbor-the-gen.masc.pl the neighbors' child'
152
Comparative studies in Romanian syntax
DP vecinilor
N' N copil(u)
The consequence of the proposal in (27) is that it is possible to analyse genitive Case assignment/checking as a Spec-head relation between the genitive phrase and Poss, endowed with the feature "assign genitive Case".14 The ungrammaticality of (28) shows that movement of the genitive phrase to SpecPossP is dependent on N-movement. We therefore conclude that Poss also bears a strong +N feature attracting the possessee.15 (28)
*acest vecinilor inteligent copil this-NM.sc.SG neighbors-the-QEN.MASC.PL intelh'gent-MASC.SG child-MASC.SG 'this intelligent child of the neighbours'
14 If the Case relationship between possessor phrases and Poss is to be analyzed in a uniform way cross-linguistically, we must assume that the feature "assign Case" is subject to parametric variation in order to account for nominative Case on the possessor in languages like Hungarian (see Szabolcsi 1987, 171): (i) az en kalap-om thelhat-poss.lso 'my hat' 13
Note that movement of N to Poss in (28) without further overt raising, as in (i), is ungrammatical as well, suggesting that Romanian has a PF output constraint similar to the Doubly Filled Comp Filter (call it the Doubly Filled PossP Filter). (0 *U [D- acest [p^p [up vecinilorj [Poss' copil; this-MASC.so. neighbor-the-GEN.MASC.PL [XP inteligent [NPtj [N' tj]]]]]] intelligent-MASC.so child-MASC.so 'this intelligent child of the neighbors'
Last resort strategies in DP
15 3
DP D'
PossP acest Spec
Poss' Poss inteligent X vecinilor
NP N' N copil(u)
To summarize, we have come to the conclusion that Poss has the features in (29): (29a) is the feature attracting the possessee (N), prior to subsequent movement to D; (29b) attracts the possessor (DPgen) to SpecPossP and (29c) checks the Case feature of the possessor (DPgen). The relevant operations are illustrated in (30):
(29)
a. {+N} b. {+D} c{assign genitive Case}
(30)
16 Following Cinque (1993) we will assume that adjectives appear in the specifier position of one of the functional projections of the noun. Following Cinque, we will label these projections XP, YP etc.
154
Comparative studies in Romanian syntax
4.2. The Template DP Hypothesis Since we assume that the internal structure of a/-phrases equals the internal structure of "regular" DPs (see section 3.4.), we may take (29) to exhaust the issues raised in (16.i) and (16.ii): a/phrases are copies of "regular" DPs, hosting a PossP projection. We call this the Template DP hypothesis. The relationships between N-D (a-/), Poss and DPgen (vecinilor) in (3 Ib) thus match the relationships in (27), repeated in (3 la), as illustrated in (32). (31)
a.
b.
copilul vecinilor child-the-MASC. so neighbor-the-GEN.MASC.PL 'the neighbors' child' ... al vecinilor ... a/-the-MASC.SGneighbors-the-GEN.MASC.PL '... of the neighbours...'
The peculiarity of a/-phrases being reduced to their external distribution, we will now address the problem of their relationship to the DP of the possessee. Since Poss has strong +N and +D features, it will be inserted in overt syntax. However, if checking of one of its strong features is blocked, no convergent derivation is available. This is the case in (33a) where the adjective has moved from a lower Spec position to SpecDP (through SpecPossP): with SpecPossP filled by the trace of the adjective, movement of the possessor is blocked and the strong +D feature on Poss will remain unchecked. The alternative without PossP in (33b) overcomes this problem but leaves vecinilor with an unchecked Case feature. (33)
*inteligentul copil vecinilor intelligent-the-MASC.SG child neighbor-the-GEN.MASC.PL 'the neighbors' intelligent child'
Last resort strategies in DP
15 5
b. *DP Spec inteligentUj
Structures like (33) constitute the core contexts for insertion of ^/-phrases. Considering (a) that no convergent derivation is available if PossP is generated under the determiner of the possessee (33a), (b) that possessors require their Case feature to be checked as in (33b) and (c) that genitive Case is checked/assigned by PossP as in (29c), the only option left is generating PossP lower in the structure. ,4/-phrases, with the internal structure of (31b) are the only means to realize this option since complementation rules require PossP to be in the complement position of D (see (26)): (34)
inteligentul copil al vecinilor intelligent-the-MASC.SG child neighbors-the-GEN.MASC.PL 'the neighbors' intelligent child'
156
Comparative studies in Romanian syntax DP
Insertion of al thus appears to be a last resort strategy for cases where no PossP can be generated in the complement of the D of the possessee. Considering ^/-phrases as last resorts, we have a straightforward way to account for the fact that, as we have noted before, they have no semantic content. If the DP headed by al merely serves a grammatical purpose, we must assume that its determiner is expletive and that a is a pure placeholder for N. In section 3.4., we have shown that it is reasonable to assume such status for a. As for the expletive status of the determiner, we may note that such an option is available in Romanian since, under standard accounts, a lexical determiner in DPs with a generic (subspecies) reading, as hi (35a), or a possessive distributive reading, as in (36a), only obtains in languages where the lexical determiner may have expletive status (cf. Vergnaud & Zubizarreta 1992). The (b) examples below show that this is indeed the case hi Romanian: (35) a. Les baleines grises sont en train de disparaitreien Californie 'the grey wales are becoming extinct in California' - (French) b. Balenele gri incep s dispar in California. wales-the grey begin to disappear in California - (Romanian) (36) a. Les enfants se sont casse la jambe. the children REFL been broken the leg - (French) 'The children have broken their legs.' b. Copiii i-au fracturat piciorul. children-the REFL-have broken leg-the 'The children have broken then* legs.' Summarizing, we have been led to assume that the internal structure of a/-phrases equals that of "regular" DPs, an assumption which makes a uniform account of genitive phrases possible: that a/-phrases are a last resort, and that the determiner which heads such phrases is expletive.
Last resort strategies in DP
157
5. FEATURE-CHECKING OF/IL-PHRASES The Template DP hypothesis adequately answers the issues raised in (16.i) and (16.ii), but still needs to be made explicit on the issues of
5.1. Licensing of <J>-features and Case
As noted in section 3.1., the possessive article agrees in gender and number with the possessee. Since a/-phrases occur in a specifier position of the (lower part of the) functional structure of the noun, we may view the (|>-features on al as the reflex of Spec-head agreement, exactly as in Cornilescu' s (1994) proposal.17 If gender and number features cannot be positively checked, the derivation will crash as is illustrated in (37a), compared to (37b). (37) a.*saracul copila/ ai/ ale vecinilor poor-the-MASC.SG child a/-FEM.SG/a/-MASC.PL/«/-FEM.PL neighbors-the-GEN.MASC.PL b. saracul copil al vecinilor poo-the-MASC.so child al-M.SG. neighbors-the-GEN.MASC.PL 'the neighbors' poor child' The Case feature of al (nominative/accusative) allows us to formulate a specific proposal with regard to where this Spec-head relation obtains: since Case features call for a Case-assigning head and a functional projection where Case is checked, we postulate the existence of a Case agreement projection in the lower part of the functional structure of the noun and endow the noun with the (optional) feature of assigning Case. This proposal may be reduced to the assumption that the functional structure of nouns even more closely parallels the functional projections of verbs than is generally assumed: both appear to have the possibility of containing two functional projections where Case can be checked.18,19
17 Note that this Spec-head agreement relation strengthens the assumption that a is a pure placeholder for N and is thus either unspecified for 4>-features or lacks these features altogether. 11
Cinque (1993) has convincingly argued that the linearization of adjectives follows from them being generated in specific functional projections of the noun. A similar linearization principle seems to hold for adverbs as well (Cinque 1997). If the linearization mechanisms that operate for adjectives at the nominal level and for adverbs at the verbal level match, as indeed appears to be the case, the proposal of a second Case checking projection in the nominal domain (mirroring the accusative Case checking position in the verbal domain) seems to make sense. At the theoretical level, then, the conclusion that must be drawn from these parallelisms is that the functional structure of
158
Comparative studies in Romanian syntax
(38) a. b-
[ DP Y [posspY^pY [^ Y U Y]]]]] [ C pY [^pY [ W Y UpYt
The hypothesis in (38) makes two specific predictions: (a) if (Romanian) nouns can have the feature "assign nominative/accusative", er/-phrases in SpecAgrP may only come with a nominative/accusative Case feature; (b) a/-phrases can never raise to SpecPossP even if an accessible PossP is present in the main DP. Both predictions are borne out. (39) shows that a/phrases, even when occurring within a genitive DP (inv\ U\torului Y), cannot bear genitive Case (39a); the nominative/accusative form is ruled in (39b). The construction in (40), on the other hand, is ruled out since Poss will fail to check off its feature "assign genitive".20 (39) a.*casa invajatorului simpatic house-the-FEM.SG teacher-the-GEN..MASC.SG nice-MASC.SG alui copilului nostru a/-GEN.MASC.SG child-the-GEN.MASC.SG our b
casa
invajatorului
simpatic
house-the-FEM.SG teacher-the-GEN..MASC.SG nice-MASC.SG al
copilului
nostru
(7/-NOM/ACC.MASC.SG child-the-GEN.MASC.SG OUT 'the house of our child's nice teacher'
nouns and verbs is rather independent of considerations of categorial nature, variation patterns and parametnzation reducing to mere morphological matters. 19
The parallelism between nominal and verbal functional structures suggests that if different heads are involved in the assignment and the checking of Case, PossP should project an AgrP projection as well. See Chomsky (1995, 349ff.) for a different proposal on the status of Agr phrases. 20
Structures like (40) are ruled out even if no PossP is generated in the main DP as in the example below. The ungrammatically of (i) strengthens the Template DP hypothesis as a last resort strategy, although it raises intricate questions concerning the status of expletives with respect to numeration. (I)*[DP ID= cava L^ [DP a invat&toruluij] [^ tj U tj [». tj ]]]]]] house-THE-FEM.SG O/-FEM.SG teacher-the-GEN.MASC.SG
'the teacher's house'
Last resort strategies in DP
invafStoruk-lui
a'.
*alui copilului nostru
b'.
aI copilului nostru
(40)
*casa a inva^atorului house-the-FEM.SG
159
160
Comparative studies in Romanian syntax
5.2. Feature Compatibility and Incorporation In the preceding sections, we have suggested that the functional structure of the noun provides a second position for Case- checking (AgrP) and that the noun may be a Case-assigning head. Furthermore, we have adopted the standard assumption that the -features and the Case feature are checked under Spec-head agreement. The Spec-head agreement hypothesis implies that the features on al match the features on the noun (or its trace). The contrast in (37) confirms this conclusion for a/'s -features. Things are slightly different for the Case feature: recall from section 2.2. that only feminine singular nominals are morphologically marked for nominative/accusative. For all other nouns, morphological Case has been lost completely, meaning that they will be marked in the lexicon with an { Case} feature. If so, Spec-head relations may still obtain between a fully specified a/phrase in SpecAgrP and a partially underspecified N (or its trace) hi Agr. To put it differently: Spec-head relations require feature compatibility, not necessarily feature identity. (41) a. frumosul fiu al regelui cute-the-MASC.so son-MASC.so OT/-MASC.SG king-the-GEN.MASC.so b.
'the king's cute son' frumoasele fete ale regelui pretty-the-FEM.PL girls-FEM.PL or/-FEM.PL king-the-GEN.MASC.SG 'the king's beautiful daughters'
Last resort strategies in DP
\ 61
frumosUj-l frumoasej-le
al{nom/acc} regelui ale{nomjacc) regelui
ijaCaseJ
Feminine singular nouns, on the other hand, are morphologically marked in nominative/accusative. Therefore, their fully specified Case feature is always identical to that of the possessive article. (42)
fhunoasa-a fata a regelui pretty-the-FEM.SG girl-FEM.SG a/-FEM.SG king-the-GEN.MASC.SG 'the king's pretty girl'
162
Comparative studies in Romanian syntax
The morphological distinction between feminine singular nouns in nominative/accusative Case and all other nouns also has a syntactic reflex: while all possessees, irrespective of noun class, allow a/-phrases to check their features hi a Spec-head configuration like (41-42), only feminine singular nominative/accusative possessees offer an alternative to (42):21 (43)
frumoasa fata regelui pretty-the-FEM.so girl-X-FEM.SG king-the-GEN.MASC.SG 'the king's pretty girl'
There are two striking differences between (42) and (43): the possessor in (43) is not preceded by the feminine singular possessive article and the possessee takes a special ending (glossed as X). There are at least three possible ways to account for this special ending: as the result of (a) the mere application of the obligatory phonological deletion rule a —* 0 / _ V; (b) the reduplication of the definite article of the adjective on N, and (c) the incorporation of the possessive article into the noun.22 The first option must be excluded on the basis of similar phonological contexts in which the reduction of a-a to a does not yield a grammatical result, as in (44). (44) a. fata frumoasa a regelui girl-the-FEM.SG pretty-FEM.SG a/-FEM.SG king-the-GEN.MASC.SG 'the king's pretty girl'
21 There is a certain amount of variation among speakers with respect to the acceptability of structures like (43). This variation relates to the categorial nature of the possessor (possessive pronouns vs. full DP) and its thematic role (possessor, agent vs. theme). 22
See fn. 4. Note that this rule is operative in accounts (b) and (c) as well.
Last resort strategies in DP
163
b *fata frumoasa regelui girl-the-FEM.SG pretty-X-FEM.so king-the-GEN.MASC.SG 'the king's pretty girl' A second argument against (a) is the fact that the phonological deletion rule is not optional in standard Romanian: the union of the -| ending of the feminine singular N with the definite article a for example, does not allow (45a). The account based on the mere application of the phonological reduction rule would thus leave unanswered why (46a) exists alongside (46b): (45)
a. b.
(46)
a. b.
*fata-a girl-the-FEM.SG fata girl-the-FEM.SG 'the girl' frumoasa fata a regelui 'the king's beautiful girl' frumoasa fata regelui 'the king's beautiful girl'
We therefore conclude that the special ending on the noun in (43) cannot merely be a matter of phonological rules, but must be sensitive to syntactic structure. As for the article reduplication hypothesis, there seems to be no possible way to account for the fact that it only operates on feminine singular nouns, as is shown by the examples in (47): (47) a. *frumosul fiul regelui cute-the-MASC.SG son-the-MASC.SG king-the-GEN.MASC.SG 'the king's cute son' b. *frumo§ii
fiii
regelui
CUte-the-MASC.PL SOn-THE-MASC.PL king-the-GEN.MASC.SG
'the king's cute sons' c. *frumoasele fetele regelui pretty-the-FEM.PL girl-the-FEM.PL king-the-GEN.MASC.SG 'the king's pretty girls' Therefore, we are led to the conclusion that incorporation accounts for (43): the possessive article adjoins to the raised noun (fat\) and incorporates into it. This hypothesis gives a straightforward explanation for the fact that the so-called special ending only appears on nouns
164
Comparative studies in Romanian syntax
(seethe contrast in (44)), since it subsumes the head-movement constraint. It furthermore allows us to explain why phonological rules that are sensitive to word boundaries are fully productive in structures like (43). One such rule is the phonological reduction rule a —*0 /_Villustrated above, another one is illustrated hi (48).
(48)
a. a -> ua / a _ b. frumoasa basmaua ta pretty-the-FEM.SG scarf-FEM.SG-cr/-FEM.SG your-FEM.SG 'your pretty scarf
Since structures like (43) are restricted to the context of feminine singular nouns in nominative/accusative we expect phonological reduction rules that are sensitive to word boundaries to operate only with these nouns. This is indeed the case: feminine cartea in (49a) is subject to syllabic reduction while frate al in (49b) is not. (49)a. frumoasa cartea regelui pretty-the-FEM.SG book-FEM.SG-a/-FEM.SG king-the-GEN.MASC.SG 'the king's beautiful book' b. frumosul frate al regelui cute-the-MASC.SG brother-MASC.se a/-MASC.so king-the-GEN.MASC.SG 'the king's cute brother' Incorporation of the possessive article seems a likely hypothesis since (a) on independent grounds, incorporation has been shown to provide an alternative way for checking of Case features (see Ferguson 1996) and (b) the distinction between feminine singular nouns that allow structures like (43) and other noun classes that do not, lies in the domain of then- Case features. Considering that incorporation usually constitutes a severely restricted process, we may assume that Case checking through incorporation is too and formulate the restriction on incorporation of al as follows:23 (50)
if overtly incorporates into , then every unchecked feature of is checked by an overtly spelled-out feature of Since we assume that incorporation is an alternative way to check the Case feature of the
23
If restructuring is to be considered an instance of incorporation (possibly at LF), the examples of specific triggers restricting the application of incorporation are numerous; for example, causative verbs (Baker 1988 and subsequent literature), object bare plurals (Delfitto & Schroten 1991), empty possessors (Delfitto & D'Hulst 1995).
Last resort strategies in DP
165
possessive article, (50) will require that the unchecked Case feature of al be checked by an overtly spelled-out feature on the possessee: incorporation is thus available only to those nouns that have overt Case marking; that is, feminine singular nouns used in nominative/accusative as shown hi (51). (51) frumoas -a fat -a regelui beautiful-the-FEM.SG girl-a/-FEM.SG king-the-GEN.MASC.SG 'the king's beautiful girl'
Jrumoasaj-a
5.3. Summary The fact that possessive articles have and Case features to be checked forces us to assume (a) that the internal structure of noun phrases have projections where these features can be checked and (b) that N can be a Case assignor. The fact that Case features can be checked in either of two ways, hi a Spec-head configuration or through incorporation, confirms the latter assumption.
6. FRENCH SUPERLATIVES AND ARTICLE REDUPLICATION The preceding analysis of Romanian a/-phrases crucially depends on the fact that Romanian allows the determiner to function as an expletive. Therefore, we expect that languages which allow the definite determiner to have expletive status may, under specific circumstances, exploit the last resort strategy of template DP's. More specifically, we expect Romance languages, which have been shown to allow their definite determiners to have expletive status (see Delfitto & Schroten 1991 and Vergnaud & Zubizarreta 1992), to exploit the possibility of template DP's. In this section, we will only focus on French, which shows an interesting discrepancy between
166
Comparative studies in Romanian syntax
pre-nominal and post-nominal superlatives.24 Pre-nominal superlatives are formed by introducing the sole comparative marker plus before the pre-nominal adjective as illustrated in (52a). Postnominal superlatives, on the other hand, require the comparative marker to be preceded by an additional definite article as shown in (53). (52)
a.
b. (53)
a. b.
le plus grand ecrivain the more great writer 'the greatest writer' *le le plus grand ecrivain the the more great writer Tecrivain plus populaire the writer more popular 1'ecrivain le plus populaire the writer the more popular 'the most popular writer'
Article reduplication in French superlatives mimics the distribution of possessive articles in Romanian as can be seen comparing the (a) and (b) examples above with the Romanian data in (9) and (10). The parallelism is even more striking if we abstract from independent differences in the syntax of nominals in French and Romanian related to the parameter of N-movement: the crucial factor determining article reduplication appears to be related to word order phenomena. In Romanian the relevant word order phenomenon regards the distance between genitive phrase and determiner, in French it regards the ordering of noun and superlative adjective. An analysis of French superlatives similar to the one we offered for Romanian genitives thus seems in order, a conclusion which appears justified on the basis of the fact that, in predicative position, both Romanian genitives (54) and French superlatives (55) (in the intended reading) require the presence of the determiner. (54) a. Aceasta ma§ina e a regelui. this car isa/-FEM.SG king-the-GEN.MASC.so 'This is the king's car' b. * Aceasta ma§ina e regelui. this car is king-the-GEN.MASC.SG (55) a. Cet ecrivain est le plus grand
24
We are aware of the fact that Romanian offers other contexts which require article reduplication, for example, the demonstrative article eel (see a.o. Giusti 1993a and Coene 1994). On the relationship between eel and al, see Coene (in press).
Last resort strategies in DP
\ 67
this writer is the more great 'This writer is the greatest.' b. Get ecrivain est plus grand this writer is more great 'This writer is greater/*greatest.'
6.1. Applying the Template DP Hypothesis Assuming that adjectives invariantly appear in specifier positions as in (56a) (see Cinque 1993) and that comparative particles head a functional projection of the adjective as in (56b) (see Corver 1997), we may take (56c) to be the rough syntactic structure of DP's containing superlative adjectives. (56) a. bc-
[opDt [D^P plus/mains [^ A]] [DP D [xp [DegP plus/mains [^ A]] X [^ N]]] DP D'
.— D
XP
Degp" Deg'
Deg
X' X:
~AP
plus/mains
A' i A
NP N' 1 N
There appears to be no evidence that French DPs containing pre-nominal superlatives are more complex than (56c). As a consequence, we may take phrases like (52a) to have the internal structure illustrated in (57a), where, as argued by Cinque, the noun moves to an intermediate position between D and N and the superlative plus grand has the internal structure indicated in (56b). Now, if French actually exploits the strategy of template DPs, we expect post-nominal superlatives like le plus populaire in (53b) to be DPs with an internal structure similar to (57a). Under this account, phrases like (53b) receive the structure hi (57b), where the internal DP containing the superlative adjective has the structure of (56c). (57) a.
b.
tp [D, le [xp [^ plus grand] [*, ecrivain; [^ [N, ts ]]]]]]
[DP [D- 1' [XP tx- ecrivahii [w [DP le plus populaire] [Y, t; [^ [N, t; ]]]]]]]]
168
Comparative studies in Romanian syntax
This account for postnominal superlatives presupposes that the template DP of post-nominal superlatives contains an NP headed by an empty N. (58) [up [D, le [^ [Dt%P [Ds%, plus [^ ^ populaire]]]] [*, X [w [N, e]]]]]]
6.2. Superlatives and the Restrictive Clause Applying the Template DP hypothesis to French post-nominal superlatives seems a fair move once it is accepted that the presence of a definite article signals the presence of a DP-shell. But what underlying need justifies the use of such a last resort strategy in this case? Unlike Romanian possessor phrases, where the data wear their explanation on their sleeve, French superlatives do not seem to have any particular morphological properties to justify the use of a template DP. One possibility which comes to mind is that superlatives, (i.e., DegPs headed by plus or mains) need to move from their base position (i.e., from the position where regular adjectives occur), to the specifier of a higher functional projection of the noun. If, for whatever reason, such movement is obligatory in overt syntax, post-nominal superlatives will exploit the last resort strategy of template DPs in order to occur in such a higher Spec position, before LF. This hypothesis would close the gap between Romanian possessor phrases and French superlatives. Unfortunately, it is virtually impossible to give this hypothesis an empirical underpinning, as DPs containing superlatives fail to tolerate the presence of other pre-nominal material (adjectives) which might signal a different distribution of superlatives with respect to their non-modified counterparts. Therefore, we will not pursue this line of argument although it might well be on the right track. The approach we will adopt, will be more interpretative and is based on the analysis of semantic properties of superlatives and the comparative marker they contain. First of all, consider that the comparative marker plus that enters superlatives (59a) or comparatives (59b), also shows up in other contexts such as (60a) with a temporal-like interpretation, or (60b), with a nominal-like interpretation. (59) a.
Mariette est \aplus jolie
fille
queje connaisse.
Mariette is the more beautiful girl that I know 'Mariette is the most beautiful girl I know.' b.
As-tu jamais rencontre une plus jolie fille? 'Have-you ever met a more beautiful girl?'
Last resort strategies in DP (60) a.
b.
169
Apres cette malheureuse rencontre, after this unfortunate encounter Emile n'a/7/ws revu Mariette. Emile NEC has more seen-again Mariette 'After this unfortunate encounter, Emile has not seen Mariette anymore.' Emile a achete plus de hVres. Emile has bought more of books 'Emile has bought more books.'
Although the variation in interpretation might suggest that the different occurrences of plus represent different lexical entries, such an analysis seems untenable considering that there is an overall cross-linguistic tendency to use in all four contexts in (59-60) the same stem or, at least, derivationally linked stems. The table in (61) illustrates this both for Romance and Germanic languages.25 Table 61. Interpretation Languages Romance Rumanian French Italian Spanish Superlative plus piu mas mai Comparative plus piu mas mai Temporal plus piu mas mai Nominal mas mai mult plus piu Germanic English German Dutch Swedish meest mest Superlative most meist meer mer mehr Comparative more meer mer Temporal mehr anymore meer mer / fler mehr Nominal more If superlative, comparative, temporal and nominal meanings of plus are the various interpretations of a single lexical item, we expect this item to have unique and constant semantic properties encoded in the lexicon and meaning differences to be the result of distributional or configurational properties. The constant meaning of plus in (60) seems to be its quantificational force, with the particle quantifying over events hi (60a) and over entities in (60b). In its comparative use in 25
There is an interesting diachronic evidence for considering the four interpretations in (61) as being related to a single lexical stem. Latin had two entries for more: plus and magis. The Romance languages appear to have chosen systematically one or the other stem in all four interpretations, French and Italian utilizing the former, Spanish and Romanian selecting the latter.
170
Comparative studies in Romanian syntax
(59b), plus has quantificational force as well, quantifying over measures, and if comparatives and superlatives are underlyingly similar to one another, the superlative use of plus is quantificational as well (see Gawron 1992, 1995). Assuming that plus is a mere quantifier and, as a consequence, that superlative and comparative interpretations derive from distributional or configurational properties (e.g., definite versus nondefinite DPs), we may narrow down the problem of article reduplication in French superlatives as follows: why do post-nominal superlatives require the definite article (see the contrast in (53)), while post-nominal comparatives do not? The relevant data are given in (62). - definite article superlative reading: (62) a.
Pecrivain le plus populaire the writer the more popular 'the most popular writer' - indefinite article comparative reading: b. un ecrivain plus populaire a writer more popular 'a more popular writer' The crucial difference between comparative and superlative readings is that while the former only compares measures, the latter also compares properties and entities. In predicative contexts like (63a), the comparative contrasts measures of different properties (length and width) associated with different entities (a table and a room). The corresponding superlative in (63b) is quite different, comparing different measures of a single property (weight) associated with entities of the same order (pieces of furniture).26 What the contrast in (63) shows, is that it is part of the inherent semantics of superlatives, and not comparatives, that the measure quantification is restricted to a set of entities (see Gawron 1992, 1995). (63) a. Cette table est plus longue que la chambre n'est large. this table is more long than the room not is large 'This table is longer than the room is large.' b. De tous ces meubles, 1' armoire of all these furnitures-FEM. the-FEM.SG cupboard-FEM.SG
26
To restate this observation in an informal way, what (63b) conveys is that no other piece of furniture is heavier than the cupboard. The construction in (63a), on the other hand, does not imply that the room be less long than the table.
Last resort strategies in DP
171
27
est le plus lourd. is the-MASC.so more heavy-MASC.SG 'Of ah1 this furniture, the cupboard is the heaviest.' Now, recall that the superlative reading hi predicative contexts is available only when the superlative is preceded by the definite article; witness the contrast hi (55), repeated below, and that the presence of the definite article signals a DP template like (56c), repeated hi (65). Therefore, we may conclude that, hi order for the superlative reading to obtain and the set restricting the measure quantification to be properly construed, the comparative particle needs to have an N hi its scope-domain. - superlative reading: (64) a. Get ecrivain est le plus grand. this writer is the more great 'This writer is the greatest.' - comparative reading: b. Get ecrivain est plus grand, this writer is more great 'This writer is greater.' (65) [DP D [xp [D£RP plus/mains [^ A]] X [^ N]]] If so, reduplication of the article with post-nominal superlatives hi French falls into place: with article reduplication, as hi (66b), the DP template will offer the comparative particle an empty N and the restrictive clause will be properly construed. Taking the trace of the noun not to qualify for this purpose, (66a) will be correctly ruled out, since, without a restrictive clause, the superlative semantics will crash. (66) a. b.
27
*[DP 1'ecrivainj [j,^ plus populaire] tj [DP 1'ecrivainj [pp le [D(%P plus populaire] e] tj
Note that in this example, the determiner and the adjective of the nominal predicate agree with the entities being compared (meubles) not the entity which is singled out by the superlative (armoire). Attested examples of this agreement pattern are given below: (i) a. Et, de toutes les saisons, le printemps, s'il est la plus seyante, est aussi la plus difficile a porter. (Raymond Radiguet. Le bal du comte d'Orgel. Paris: B. Grasset. 1923, 25) b. De tous les luxes, la femme est le plus rare et le plus distingue, elle est celui qui coute le plus cher, et qu'on nous envie le plus; elle est done aussi celui que nous devons aimer le mieux a etaler sous les yeux jaloux du public. (Guy de Maupassant. Contes et nouveUes. 1885. Edition Albert-Marie Schmidt & Gerard Delaisement. Paris: Albin Michel. 1959,711.)
172
Comparative studies in Romanian syntax
7. CONCLUSION The two instances of article reduplication we have considered in this article both appear to exploit the same last resort strategy of template DPs, albeit for different reasons. In the case we discuss more extensively (Romanian possessor phrases introduced by at), the template DP offers the Poss both an escape hatch to check its strong nominal feature and a way for the possessor to check its genitive Case in overt syntax. French exploits this very same strategy for interpretative reasons: the possibility of a restrictive clause of superlative quantification asks for an explanation; more specifically, the comparative particle needs to have the noun in its scope so that the restrictive clause can be construed at the conceptual-interpretative interface. Although the use of template DPs appears to be determined by language specific rules, the Template DP hypothesis is based on minimal assumptions. The assumption that empty Ns can be licensed, is independently needed for other cases. Since Romanian and French template DPs occur within other DPs or, as in the case of the predicative use of genitives and superlatives, are locally related to other DPs, the licensing of an empty N in template DPs should receive an account similar to the one in elliptical constructions (the red and the blue car), which, by the way, constitute the most widespread case of empty N licensing. The other basic assumption of the Template DP hypothesis is that the definite determiner heading these DPs is expletive. This assumption places an upper bound to the cross-linguistic distribution of template DPs: only languages that independently allow the definite determiner to function as an expletive can exploit template DPs as a last resort. Both Romanian and French appear to be such languages. Undoubtedly, further research is needed to check whether the above assumptions are all that is needed to license template DPs and whether these assumptions account for the relatively restricted cross-linguistic distribution of template DPs. An interesting test case is continental Scandinavian article reduplication in adjective plus noun constructions.
REFERENCES Abney, S. P. (1987). The English noun phrase in its sentential aspect. Doctoral dissertation, Massachusetts Institute of Technology. Baker, M. (1988). Incorporation. A theory of grammatical function changing. The University of Chicago Press, Chicago.
Last resort strategies in DP
173
Benveniste, E (1966). Problemes de linguistique generate. Gallimard, Paris. Bernstein, J. B. (1993a). Topics in the syntax of nominal structure across Romance. Doctoral dissertation, City University of New York. Bernstein, J. B. (1993b). The syntactic role of word markers in null nominal constructions. Probus 5(1), 5-38. Borer, H. (1989). On the morphological parallelism between compounds and constructs. In: Yearbook of morphology 1 (G. Booij and J. van Marie, eds.), pp. 45-65. Foris, Dordrecht. Borer, H. (1996). The construct hi review. In: Studies in Afroasiatic grammar (J. Lecarme; J. Lowenstamm and Ur Shlonsky, eds.). Academic Graphics, Den Haag. Chomsky, N. (1995). The Minimalist Program. MIT Press, Cambridge, Mass. Cinque, G. (1993). On the evidence for partial N-movement in the Romance DP. University of Venice Working Papers in Linguistics 3(2), 21-40. Cinque, G. (1997). Subject and (finite) verb positions. Paper presented at Incontro di grammatica generativa, Pisa, 22-24 February. Coene, M. (1994). Some reflections on the position of the adjective hi Romanian. Cahiers de linguistique theorique et appliquee 31, 9-26. Coene, M. (hi press). Homo ellum bonum ou existe-t-il un demonstratif provenant du latin ellwiri} Studii $? cercetari lingvistice. Cornilescu, A. (1993). Notes on the structure of Romanian DP and the assignment of the genitive Case. University of Venice Working Papers in Linguistics 3(2), 107-133. Cornilescu, A. (1994). Remarks on the Romanian ordinal numeral. Towards a unitary description of phrases headed by AL. Revue roumaine de linguistique 39, 303-337. Cornilescu, A. (1995). Rumanian genitive constructions. In: Advances in Rumanian linguistics (G. Cinque and G. Giusti, eds.), pp. 1-54. John Benjamins, Amsterdam. Corver, N. (1997). A/wc/z-support as a Last Resort. Linguistic Inquiry 28, 119-164. Delfitto, D. and Y. D'Hulst (1995). On the syntax of null possessive anaphors. Recherches de linguistique franqaise et romane d'Utrecht 14, 13-38. Delfitto, D. and J. Schroten. (1991). Bare plurals and the number affix hi DP. Probus 3(2), 155185. Dobrovie-Sorin, C. (1987). A propos de la structure du groupe nominal en roumain. Rivista di grammatica generativa 12, 123-152. Dobrovie-Sorin, C. (1992). The syntax of Rumanian. Comparative studies in Romance. Mouton de Gruyter, Berlin. Dobrovie-Sorin, C. (present volume). (In)definiteness spread: From Romanian genitives to Hebrew construct state nominals. Ferguson, K. S. (1996). Shortest move and object Case checking. In: Minimal ideas. Syntactic structures in the Minimalist framework (W. Abraham et aliis, eds.), pp. 97-111. John
174
Comparative studies in Romanian syntax
Benjamins, Amsterdam. Gawron, J. M. (1992). Focus and ellipsis in comparatives and superlatives: A case study. Ohio Working Papers in Linguistics 40, 79-98. Gawron, J. M. (1995). Comparatives, superlatives, and resolution. Linguistics and philosophy 18(4), 333-380. Giusti, G. (1993 a). Lasintassi del determinant!. Unipress, Padova. Giusti, G. (1993b). Enclitic articles and double definiteness. A comparative analysis of nominal structure in Romance and Germanic. The Linguistic Review 11, 231-255. Giusti, G. (1995). Heads and modifiers among Determiners: Evidence from Romanian. In: Advances in Rumanian linguistics (G. Cinque and G. Giusti, eds.), pp. 103-126. John Benjamins, Amsterdam. Giusti, G. (1997). The categorial status of Determiners. In: The new comparative syntax (L. Haegeman, ed.), pp. 95-123. Longman, London. Grosu, A. (1988). On the distribution of genitive phra ses in Rumanian. Linguistics 26, 931-949. Grosu, A.(1994). Three studies in locality and Case. Routledge, London. Harris, J. W. (1991). The exponence of gender in Spanish. Linguistic Inquiry 22, 27-62. Longobardi, G. (1994). Reference and proper names: A theory of N-movement in syntax and Logical Form. Linguistic Inquiry 25, 609-665. Longobardi, G: (1996). Chez and the problem of etymology. Manuscript. University of Venice. Marantz, A. (1995). The Minimalist Program. In: Government and Binding Theory and the Minimalist Program (G. Webelhuth, ed.), pp. 351-382. Blackwell, Cambridge. Ouhalla, J. (1991). Functional categories and parametric variation. CroomHelm, London. Picallo, M. C. (1991). Nominals and nominalizations in Catalan. Probus 3(3), 279-316. Ritter, E. (1987). NSO noun phrases in Modern Hebrew. Proceedings ofNELS 17, 521-537. Ritter, E. (1988). A head movement approach to construct state noun phrases. Linguistics 26(6), 909-929. Ritter, E. (1991). Two functional categories in noun phrases: Evidence from Modern Hebrew. In: Syntax and semantics 25. Perspectives on phrase structure: heads and licensing (S. D. Rothstein, ed.), pp. 37-62. Academic Press, San Diego, Ca. Rouveret, A.(1994). Syntaxe dugallois. Principesgenerauxettypologie. Editions CNRS, Paris. Siioni, T. (1997a). Noun phrases and nominalizations. The syntax ofDPs. Kluwer, Dordrecht. Siloni, T. (1997b). Event nominals and the construct state. In: The new comparative syntax (L. Haegeman, ed.), pp. 165-188. Longman, London. Szabolcsi, A. (1987). Functional categories in the noun phrase. In: Approaches to Hungarian 2. Theories and analyses (I. Kenesei, ed.), pp. 167-189. Jate Szeged, Budapest. Tasmowski-De Ryck, L (1989). Cours degrammaire roumaine. Communication and Cognition, Ghent. Valois, D. (1991). The internal syntax of DP. Doctoral dissertation, University of California, Los
Last resort strategies in DP
175
Angeles. Van Eeden, W. (1994). Grammatica van het Roemeens. Grammaticale morfologie. Grammar Publications, Amsterdam. Vergnaud, J.-R. And M. L. Zubizarreta (1992). The definite determiner and the inalienable constructions in French and English. Linguistic Inquiry 23, 657-720.
This page intentionally left blank
(In) defmitness spread
111
(IN)DEFINITENESS SPREAD: FROM ROMANIAN GENITIVES TO HEBREW CONSTRUCT STATE NOMINALS
Carmen Dobrovie-Sorin, Centre National de Recherche, Paris
This paper aims to establish the existence of two universal syntactic constraints imposed by the SpecDP position: (a) SpecDP can be lexically filled only if (a) D is empty or (b) filled with a definite article. (b) SpecDP cannot be filled with bare NPs. These constraints will be derived as consequences of the semantic composition that underlies DPs that contain a SpecDP constituent: (c) The head N denotes a function of type (e,e), which applies to the individual denoted by the DP in SpecDP and yields the individual denoted by the overall possessive DP. I will then show that the "(in)definiteness spread" phenomenon found hi Saxon genitives and Hebrew Construct State nominate (CSNs)) is due to the universal semantic rule in (c), rather than to some marked, construction-specific mechanism of Spec-head agreement in ±def features between the genitive DP and the D of the head N. Relying on the minimalist assumptions of Bare Phrase Structure, I will argue against the X'-theoretical formal universal according to which all nominal projections are projections of D.
* I would like to thank Patricia Cabredo Hofherr, Alex Grosu and Tali Siloni for helpful comments, and Hagit Borer for the patience with which she answered my questions on Hebrew.
178 Comparative studies in Romanian syntax
i. STRUCTURAL GENITIVE CASE is ASSIGNED TO SPECDP Romanian overtly distinguishes between two distinct types of genitive DPs: (1)
a.
b.
(2)
a.
b.
c.
d.
casa vecinului/ vecinilor house-theneighbor-the-GEN/neighbors-the-GEN 'the neighbor's/the neighbors' house' casa unei vecine/ unor vecine house-the a-GEN.FEM neighbor-GEN.FEM /some-GEN.FEM neighbors-GEN.FEM 'a neighbor's/some neighbors' house' o casa a vecinului/ a unei vecine a house a neighbor-the-GEN/a a-GEN.FEM neighbor-GEN.FEM 'a house of the neighbor's/of a neighbor's1 acest obicei al vecinului/ al unei vecinei this habit al neighbor-the-GEN.FEM/a/ a-GEN.FEM neighbor-GEN.FEM 'this habit of the neighbor's/of a neighbor's1 casele de piatra ale vecinului/ ale unei vecinei houses-the in stone ale neighbor-the-GEN.FEM/a/e a-GEN.FEMneighbor-GEN.FEM 'the houses in stone of the neighbor's/of a neighbor's' primii studenji ai mamei mele first-the students ai mother-the-GEN my 'the first students of my mother's'
In all these examples, genitive Case is morphologically marked on the article: -/M/-MASC. so, -eiFEM. so, -/or-MASC/FEM.PL are the genitive forms (which are identical to the dative forms) of the suffixal definite article and unui, unei, unor are the corresponding forms of the indefinite article. The examples in (2) differ from those in (1) by the presence of a pre-genitive particle, made up of an invariable part, a-, followed by the nominative forms of the definite article, which are inflected for number and gender: al-MASC.SG/ a-FEM.SG/ a/'-MASC.PL/ a/e-FEM.PL. This particle agrees with the head noun. The first aim of this paper is to account for the alternation between "bare (morphological) genitives" and "a/-genitives". The account to be proposed below follows Grosu's (1988, 1994) line of analysis, insofar as it attempts to treat Romanian genitives on a par with Saxon genitives, but diverges from Grosu's analysis itself According to Abney (1986), Saxon genitives such as those hi (3a, b, c) are to be represented as hi (3d): (3)
a.
John's brother
(In) defmitness spread b. c.
179
my sister's friend a neighbor's daughter
John my sister a neighbor
i neighbor
brother friend daughter
In these configurations, the possessor DP moves from SpecNP to SpecDP in order to be assigned Case by a D element, namely the 's morpheme. Adapting this analysis, Grosu (1988, 1994) proposes that in Romanian, the assigner of genitive Case is the definite article. He furthermore assumes an adjacency constraint on the assignment of genitive Case, which is meant to account for the paradigm in (4), which shows that the Caseassigner, that is, the definite article, must be adjacent to the assignee, the genitive-marked DP. (4)
a.
b. c.
cartea profesoru/w/ book-the teacher-the-GEN 'the teacher's book' *cartea noua profesoru/w/ book-the new teacher-the-GEN cartea noua a profesoru/w/ book-the new a teacher-the-GEN 'the teacher's new book'
Grosu's proposal successfully accounts for the examples at hand. In (4a) genitive Case can be assigned, because the possessor DP is adjacent to (more precisely, it immediately follows) the suffixal definite article -a. In (4b) the intervening adjective blocks Case-assignment, and the configuration can only be saved by the insertion of al, an element that Grosu analyzes as a dummy preposition carrying a dummy definite article. Although the categorial status of al/a/ai/ale is debatable (see section 2. below), the fact remains that this element contains a
180 Comparative studies in Romanian syntax definite article, which, given Grosu's proposal, may assign genitive Case under adjacency to profesorului, as in (4c). According to Grosu, the assignment of Genitive Case in Romanian and Saxon is alike insofar as: (5)
Genitive Case is assigned (i) under adjacency (ii) by a genitive Case assignor of category D (the definite article and 's, respectively).1
The Abney's and Grosu's analysis summarized here is not completely satisfactory insofar as it relies on language-specific elements, the definite article in Romanian and 's in English, which do not seem to be characterizable in a unitary manner. We may indeed agree that the Romanian definite article is a D element, but such a categorial label is hard to accept for 's, and as a matter of fact this hypothesis is dropped hi Abney (1987), who assumes that D dominates a null Agr rather than 's, as hi (3e). (3)
SpecDP
[DPgen-John]'s [Agr-0][DPgen-t] [DPgen-my sisterj's [Agr-0] [DPgen-a neighbor]'s [Agr-0]
brother friend daughter
Under this alternative analysis, 's is a phrasal affix (see also Grosu 1994, section 7.4.2.) marking genitive Case rather than a Case assignor. No adjacency constraint needs to be postulated, since the raising of N to D cannot put the N-head in front of's. The problem, however, is that we must postulate the existence of a null determiner/Agreement morpheme that does not independently exist in the language, but is restricted to co-occur with a Saxon genitive. Besides, the general proposal is confronted with the following challenges: (6)
a.
What kind of D elements are universally predicted to be Case-assigners? More
1 According to Grosu (1988, 1994), the ungranunaticality of examples such as *the child doll's provides evidence in favor of the hypothesis that's assigns genitive Case under adjacency : sequences of this type can be generated by Head-to Head movement of the N-head doll to the D position occupied by 's. To rule them out one would have to postulate an adjacency constraint.
(In)definitness spread
b. c.
181
precisely, do we expect indefinite or demonstrative articles to be able to assign genitive Case? Do we expect definite articles that are not suffixes on N to assign genitive Case? Do we expect D elements to assign Case under adjacency to any structural position?
In what follows, I will argue instead that the assignment of genitive Case hi Romanian and in English Saxon genitives is not due to a particular Case-assigner, but rather to the structural position occupied by the DP that needs Case: (7)
Structural genitive Case is assigned to SpecDP/SpecNmax.
Saxon genitives undoubtedly occupy a specific structural position inside nominal projections. Although the existence of such a position has been firmly established since the early 70s, its definition has changed depending on the changes of the syntactic model itself. Pre-GB and early GB studies referred to "the subject of NP", or to "SpecNP". The label SpecDP is related to the extension of X'-theory to functional categories (Chomsky 1986), and more precisely to Abney"s (1986, 1987) hypothesis according to which the maximal nominal projection is a projection of the functional category Det rather than a projection of the N-head itself. Because Abneys DPhypothesis is still widely assumed in the current literature on nominal projections, I will use the DP-terminology essentially for readability reasons. This does not mean, however, that I subscribe to the DP-hypothesis as it stands. More precisely, I agree with the D-as-head hypothesis, but not with the idea that all the maximal nominal projections occupying argument-positions are projections of D. The analysis to be proposed below will rely on the following basic assumptions, which are Minimalist in spirit, although they have not as yet been explicitly postulated in Minimalist accounts of DP-structure: (8)
a. b.
Bare NPs (i.e., nominal projections that lack the D-layer) may occupy argument positions, No movement from SpecNP to SpecDP applies.
The proposition stated in (8a) essentially says that we do not want to postulate null D elements in the lexicon, which is a sound methodological option. If two competing analyses cover the same range of data, preference should be given to the analysis that dispenses with null lexical elements over one that does not. The main objection that may be raised against (8a) is the presumed incompatibility between the semantic type of bare NPs and the fact that they are allowed to appear in argument positions.
182 Comparative studies in Romanian syntax Theorists agree that bare NPs are property-denoting expressions (type (e,t)), as indicated by their occurrence in predicate positions, for example, Jean est medecin 'John is physician1; Jean et Marie sont medecins 'John and Mary are physicians'). Only those nominal projections that have the D-layer can have the semantic type of an argument (either individual-type (type e) or generalized quantifier (type ((e,t)t)). Theorists disagree, however, as to whether propertydenoting expressions, in particular bare NPs, may or may not occupy argument positions. Thus, in examples such as John met friends last night, one may assume that friends is a projection of a null D (Longobardi 1994, 1996), which as such is allowed to occupy an argument position, on a par with any other type of DP. Alternatively, one may assume that despite being a mere property-denoting bare NP, friends is allowed to occupy an argument position, due to particular rules of semantic composition (Dobrovie-Sorin 1997, McNally 1995, van Geehoven 1996). The generalization in (8a) can thus be understood as meaning (8c). (8)
c.
Property-denoting expressions are allowed to occupy argument positions.
I will assume (8c) to be correct, which will become relevant only in later sections (starting with section 8.). There is yet another possible interpretation of (8a), shown in (8d). (8)
d.
Those nominal projections that lack D, but instead have a lexically-filled Spec position, end up having the semantic type of arguments (denote individuals).
As we cannot anticipate the evidence in favor of (8d), we ask the reader to set aside (until section 10. below) his questions regarding the referential/argumental status of nominal projections lacking the D-layer and simply assume that (8a) is a possibility allowed by the syntax. Principle (8b) takes seriously one of the central claims of the Minimalist model, the elimination of D-structure (and of its representational correlate, NP-chains). (8b) allows two possibilities: a given DP is either merged in SpecNP and stays there or it is directly merged in SpecDP and inherits a th-feature directly from the N-head itself (for a similar proposal in the V-domain see Manzini & Roussou 1997). Given (8a,b), the position occupied by Saxon genitives is not SpecDP, but rather SpecNmax. I will nevertheless continue to use the labels DP and SpecDP, somewhat improperly, to refer not only to projections of D, but also to maximal nominal projections (Nmax) and their Spec positions (SpecNmax). The label NP will refer to nominal projections embedded under a D element, and SpecNP will be the associated Spec position:
(In)definitness spread (3)
f.
\ 83
DP/Nmax
SpecDP/Nmax
N-nonmax
John's
house
Compare French cfe-phrases: (9)
a.
la maison de Jean
b.
DP
la
N-nonmax
SpecNP
maison
de Jean
We are not interested here in knowing whether the structure in (9b) is the correct representation oflefrere de Jean (we will come back to this issue shortly). I simply want to show that it is possible to distinguish between SpecDP/Nmax and SpecNP. Notice that, following Giorgi and Longobardi (1991), I assume that SpecNP can be on the right side of N (contra Kayne 1994). In what follows, a similar departure from Kayne (1994) will be assumed for the SpecDP position. Coming back to Romanian genitives, I will argue that they occupy the SpecDP rather than the SpecNP position (contra Dobrovie-Sorin 1987 and Grosu 1994): (10)
a.
fata vecinului daughter-the neighbor-the-GEN
SpecDP/SpecNmax vecinului
2
The nghtward position of SpecDP/SpecNmax may be a marked option, made available by language-specific properties (e.g. the suffixal nature of the definite article). I leave this issue open for further research.
184 Comparative studies in Romanian syntax Against the general background of the Bare Phrase Structure (BPS) approach to constituent structure (Chomsky 1994,1995), I assume that bound functional morphemes, the definite suffixal article in particular, do not occupy syntactic positions, but instead merge with their hosts, here the N-head, prior to syntactic Merge. Now, a constituent such as fata 'daughter-the', made up of an N and a D head, can only be labeled D1 or DP (labels such as N, N1 or NP would ignore the fact that the determiner itself belongs to this term). And if fata is assigned the label D1, it can merge with a SpecDP constituent, as in (lOb). In this configuration, fata 'daughter1 is a minimalmaximal category (in other words, both an N and an NP), which is allowed by the BPS format. Alternative analyses of definite possessives can, of course, be imagined: even if made up ofN and D,fata 'daughter' could be labeled N, and as such could merge with a complement and project NP, which would then merge with a null D element, which would atir&ctfata 'daughter' at a later stage in the derivation (see the N-to-D rule assumed by Cornilescu 1993, 1994, Dobrovie-Sorin 1987 , among others). Although maintained in Minimalist studies, this type of derivation is not forced upon us by Minimalist principles, rather it seems to be an undesirable survival of the GBtype of analysis. It is in any case uneconomical compared to (10b), which is the minim?! amount of structure that must be assumed for Romanian definite possessives. hi sum, the basic assumption here is that Romanian genitives occupy the SpecDP position. The preceding discussion was meant to show that this fact could be derived as a consequence of the suffixal nature of the definite article: because the definite article is a suffix in Romanian, an adnominal DP cannot be merged as a complement of N, but only as SpecDP.3 Let me observe, however, that even if the correlation between the suffixal nature of the definite article and the projection of SpecDP does not hold on general grounds, Romanian genitives may stillbe assumed to be in SpecDP, and this is the only hypothesis relevant for our present purposes. It allows us to propose a unified analysis, namely that structural genitive Case is assigned in SpecDP, as stated in (7), for both Romanian and Saxon genitives, a proposal that will be extended to the Hebrew CSNs in the sections that follow. It is important to stress that unlike Grosu's rule stated hi (5), the rule in (7) proposed here does not mention the definite article per se. How can we then explain the contrast in (1) and (2), repeated below, which shows that "bare" genitive Case can only be assigned if the head N carries the definite article. With any other determiner, the genitive DP is obligatorily preceded by al/a/ai/ale (a/-genitives, henceforth):
'Adjunction to DP is another option, which is available if the N+D term is labelled DP (see Section 3 below). As will become clear below, the alternation between simple genitives and a/-genitives indicates that Spec,DPs cannot be identified wjth DP-adjuncts (contra Kayne 1994).
(In)defmitness spread (1)
a.
b.
(2)
a.
b.
185
casa vecinului/ vecinilor house-theneighbor-the-GEN/neighbors-the-GEN 'the neighbor's/the neighbors' house' casa unei vecine/ unor vecine house-the a-GEN.FEM neighbor-GEN.FEM /some-GEN.FEM neighbors-GEN.FEM 'a neighbor's/some neighbors' house' o casa a vecinului/ a unei vecine a house a neighbor-the-GEN/a a-GEN.FEM neighbor-GEN.FEM 'a house of the neighbor's/of a neighbor's' acest obicei al vecinului/ al unei vecinei this habit al neighbor-the-GEN.FEM/a/ a-GEN.FEM neighbor-GEN.FEM 'this habit of the neighbor's/of a neighbor's1
The obligatory co-occurrence of the definite article with structural genitive Case will be analyzed here as a consequence of the language-specific constraint stated in (11). (11)
In Romanian, SpecDP can be lexically filled only if D is filled with a definite article.
Since structural genitive Case can only be assigned to SpecDP, and since SpecDP is subject to the constraint in (11), it follows that genitive Case must co-occur with the definite article. If we leave aside examples like John's every student, the analysis of which is far from clear, Saxon genitives offer an instantiation of another language-specific constraint on SpecDP: (12)
In English, SpecDP/SpecNmax can be lexically filled only if D is empty.
At this point in our investigation, the propositions in (11) and (12) are mere empirical generalizations. An explanation as to why they should hold will be provided in section 10. below, where (11) and (12) will be shown to be the only possible options compatible with the rule of semantic composition that characterizes the elements that occupy SpecDP positions. The choice between the two options is language specific, depending on the particular structure that DPs take in a certain language.
2.,41-GENrnVES Since the constraint in (11) rules out the SpecDP position in examples such as (2), headed by indefinite or demonstrative determiners, the two remaining options are to generate the adnominal
186 Comparative studies in Romanian syntax DP (a) as a complement to N or (b) as an adjunct to DP, as shown in (2c, d), representing (2b).
acest this
acest this
obicei habit
a/unei vecine al a-GEN.FEM neighbor-GEN.FEM
obicei habit
al unei vecine al a-GEN.FEM neighbor-GEN.FEM
I want to remain neutral regarding the choice between (2c) and (2d), which correspond to the two classical hypotheses regarding relative clauses, as either DP-adjoined or as N-modifiers. For concreteness, and also because in certain environments a/-genitives can only be analyzed as being DP-adjoined, I will assume (2d). Now, the only thing we know regarding those adnominal DPs that cannot go to SpecDP but are instead adjoined to DP is that they cannot be assigned structural genitive Case, since this assignment can only apply to SpecDP (see rule 7). This correctly rules out genitive DPs without al/a (*un coleg vecinei 'a colleague neighbor-GEN.FEM.SG1), and one could simply stipulate that al/a. is a default Case assigner. Note however that the internal make-up of <7/-genitives is remarkable: a/-genitrves are made up of a genitive-marked DP, which must be assumed to get its Case as defined in rule (7). This analysis is indeed possible, as proposed by d'Hulst, Coene et Tasmowski (1997): al/a/.. can be analyzed as D'-constituents containing apro-N element (the invariable particle or) followed by the
(In)definitness spread
187
definite article4 Along these lines, (13c) is the representation of (13a, b). (13)
a.
b.
un
un baiat al vecinului a son al neighbor-the-GEN 'a son of the neighbor1 o fata a vecinului a daughter a neighbor-the-GEN 'a daughter of the neighbor1
baiat [Na]+[D/] vecinului son a+the-MASC neighbor-the-GEN. MASC fata [Na]+[Da] vecinului 5 daughter a+the-FEM neighbor-the-GEN.MASC
In this configuration, DPgen occupies the SpecDP position of DP2. Hence, structural genitive Case can be assigned by virtue of rule (7). This analysis of a/-genitives is supported by the fact that they can occur in isolation, as shown in (14a, b), with the representation hi (14c). (14)
a.
b.
4
/4/Mariei numia placut. al Maria-GEN not me-DAT has pleased 'I did not like Maria's.' N- am vazut-o inca pe a vecinului. not-have seen- it yet pe a neighbor-the-GEN 'I haven't yet seen the neighbor's.1
The configurations proposed here are simplified ("Minimalist"?) versions of the more complex structures proposed by d'Hulst, Coene et Tasmowski (present volume). Functional layers such as AgrP, PossP or GenP (see Longobardi 1996, Ritter 1991, Siloni 1994, 1997 among many others) do not appear to be necessary for an account of the data considered here. 3 The definite feminine article does not form a separate segment here, presumably due to a reduction of a hiatus: a (invariable particle) +a (definite article) => a.
188 Comparative studies in Romanian syntax c.
DP
SpecDP DPGen [Na]+[D/]
Mariei 6
a+the-MASC.SG
Maria-GEN
[No]+[Da] a+the-FEM. so
vecinului neighbor-the-GEN. MASC . so
In sum, ^/-genitives are analyzed here as definite possessive DPs headed by a/vo-N element.7
6
As explained in section 8 below, morphological genitive Case must be marked, in the canonical case at least, on the determiner of the genitive DP. Although proper names do not generally carry definite articles in Romanian, the -a ending of feminine singular proper names counts as a definite article on which genitive Case is marked. 7 The exact status of al is not directly relevant for our analysis, which is mainly concerned with a cross-linguistic analysis of Romanian genitives lacking al, Saxon genitives and Hebrew CSNs. The central syntactic proposal made here is that these genitives all occupy the SpecDP/SpecNmax position, in contrast to Romanian a/-genitives, English o/-genitives or Hebrew -SW-genitives, which sit either under NP, as complements or modifiers of the head N, or are adjoined to DP. This analysis of a/-genitives is compatible with the prepositional status of al (Grosu 1988,1994). For lack of space I cannot review here the various arguments in favor of each of the two hypotheses regarding al. If Grosu's view turned out to be correct, we would need to assume that genitive Case is assigned not only in the SpecDP position, but also in a position governed by a preposition carrying the definite article. And indeed, elements that are unquestionably prepositions assign genitive Case in Romanian whenever a definite article is suffixed to them: tnd&atul cosei 'in behind the house-the-GEN/behind the house', deosupra capului 'of-over-the-head-the-GEN/over the head'. Note, however, that some of these prepositions are nominal stems (e.g. in spatele casei 'in back-the housethe-GEN/behind the house'). Even in the absence of a relation to nouns, one may assume that those prepositions that take a definite article acquire nominal status (see numerals in a number of languages), and so, genitive Case assignment in Romanian can be uniformly assigned by rule (7). Some adjectives allow possessive adjectives, not nominal genitives: (i) Vitejii tai fii brave-the your sons 'your brave sons' (ii) *Vitejii {arii soldaji brave-the country-the-GEN soldiers 'the country's brave soldiers' According to Grosu (1994: 170-178), genitive Case is assigned in (i), although not in (ii), due to an adjacency constraint that is too complex to be reviewed here. Some doubt regarding Grosu's hypothesis is shed by the ungrammaticality of the following examples, which indicates that the assignment of genitive Case by adjectival heads to pronouns is not fully productive: (iii) Tampitul tau copil stupid-the your child 'your stupid child' (iv) *bolnavul lui copil ill-the his child 'his suffering child' In sum, the core case, and arguably the only one, of genitive Case assignment in Romanian is covered by the rule in (7).
(In)defmitness spread
189
This type of constituent can occur both as an independent DP and as a DP adjoined to another DP (that is, in (13a, b), a/-can be analyzed as a kind of relative pronoun: the gloss of the overall DP would be 'a son/daughter [that is] one of the neighbor1).8
3. THE ADJACENCY CONSTRAINT ON THE ASSIGNMENT OF GENITIVE CASE Let us now examine the adjacency constraint on the assignment of genitive Case: (15)
a.
b. c.
cartea profesorului book-the teacher-the-GEN 'the teacher's book" *cartea noua profesorului book-the new teacher-the-GEN cartea noua a profesorului book-the new a teacher-the-GEN 'the teacher's new book1
I would like to argue that the adjacency effect need not be stated as such, but is an indirect consequence of the assignment of genitive Case to SpecDP. The first step in showing this is to explain the idea that a post-nominal adjective can merge with a definite noun only by adjoining to it:
(16)
a.
cartea noua book-the new
cartea book-the
noua new
'According to Kihm (to appear), genitives of the construct state type in Semitic (Arabic in particular) are to be analyzed as a certain kind of relative. Although I agree that this analysis is indeed correct for certain genitives, in particular for o/-genitives, I do not believe that it is adequate for the Hebrew CSNs, nor for the Romanian genitives that lack al.
190 Comparative studies in Romanian syntax As already argued above, a suffixal article merges with N prior to the merging of the adjective, and the term N+D can be labeled D or DP, but not N. Since N+D is not labeled N, the adjective cannot be embedded under NP. Nor can adjectives merge with D1 as SpecDP constituents (this position is reserved for maximal nominal projections, as argued below) or undergo adjunction (general principles prohibit an X to adjoin to a X'-constituent). The only remaining option is DPadjunction, as illustrated in (16b).9 Given this configuration for adjectives following definite nouns, the SpecDP position is not available for adnominal DPs. And since genitive Case can only be assigned to SpecDP, we explain why a possessor DP following a post-nominal adjective cannot be marked with structural genitive Case. The only option for merging a possessor is DP-adjunction, and in this position al is necessary. The analysis of adjectives sketched here is only suggestive and it may well turn out to be on the wrong track. Note, however, that its incorrectness would not affect the core of the analysis of genitives proposed here, but only the attempt to derive the adjacency constraint on the assignment of genitive Case from the structural position of adjectives.
4. FRENCH ADNOMINAL DPs It can be argued that certain languages do not have a structural SpecDP position, and correspondingly no structural genitive Case. French is a case in point.
9
1 take the DP-adjunction of post-nominal adjectives to be a language-particular rule, related to the suffixal nature of the definite article. Although I want to leave aside a detailed analysis of the distribution of adjectives in Romanian, let me provide a somewhat broader picture of the various distributional possibilities: (i) cartea (cea) noua (a Mariei) book-the (that-the) new a Maria-GEN 'Maria's new book' (ii) cartea Mariei ??(cea) noua book-the Maria-GEN ??(that-the) new (iii) cartea Mariei (cea) noua§i scumpa book-the Maria-GEN (that-the) new and expensive These examples show that post-nominal adjectives may, but need not, be preceded by cel/cea/cei/cele, a complex element made up of an invariable part, ce (formally identical to the w/r-phrase ce 'what') followed by the definite article. According to Romanian traditional grammars, eel is an "adjectival article", a label that points to the fact that it (almost) always appears in front of an adjective. I take the possible insertion of eel in between the N-head and the adjective to indicate that Romanian adjectives (may) adjoin to the DP containing the N-head. Similarly, I tend to believe that in Hebrew, the obligatory insertion of the definite article ha in between an N preceded by another ha and the post-nominal adjective indicates that ha+Adj is adjoined to \DPha+N\. As observed by Tali Siloni (p.c.), this hypothesis may seem problematic, because the adjective precedes the internal argument of event nominals. For a solution to this problem see footnote 31. below.
(In)definitness spread
191
In clear contrast to Romanian, French shows no overt difference in the realization of genitive DPs: (17)
(18)
la maison du voisin/ de la voisine the house of-the-MASC.SG neighbor/of the-FEM. so neigbor 'the neighbor's house' a. une maison du voisin/ dela voisine a house of-the-MASC. SG neighbor/of the-FEM. so neigbor b. cette maison du voisin/ de la voisine this house of-the-MASC.SG neighbor/of the-FEM. so neigbor c. la maison en pierre du voisin/ de la voisine the house in stone of-the-MASC.SG neighbor/of the-FEM. SG neigbor
Milner (1982, 1995) argues, however, that the categorial status of de varies from (17) to (18): de would be a genitive Case-marker and a preposition, respectively. Although I agree that a distinction is needed in order to account for an important number of empirical phenomena, I believe that lexical ambiguity should be avoided. On methodological grounds, the best possible analysis is one that assumes the lexical identity ofde and views whatever differences might exist between (17) and (18) as being triggered by structural distinctions that are neutralized by the subcategorization properties of fife specified in the lexicon, which can be stated as (19). (19)
Insert de in front of adnominal N-projections.
This lexical definition of de allows it to be inserted in (17) and (18) alike, regardless of whether the structural positions occupied by the fife-phrases are distinct or not. We could thus assume, following Milner (1982), that fife-phrases occupy distinct structural positions depending on the type of D: (20e) would be correct for (20b, c, d), but not for (20a), which would instead be analyzed as hi (20f).10 (20)
a. b. c.
10
1'etudiant de Jean the student of John un etudiant de Jean a student of John cet etudiant de Jean
These structures constitute an updating of those proposed by Milner (1982), see Dobrovie-Sorin (forthcoming) for discussion.
192 Comparative studies in Romanian syntax
d.
(20)
e.
this student of John deux etudiants de Jean two students of John DP,
de Jean
fils
de Jean
Note, however, that there is no decisive evidence in favor of this difference in structure (20)(see Dobrovie-Sorin, forthcoming, and Miller 1992). We may therefore choose to assume that all French genitive DPs occupy the same structural position, which means that in French, a SpecDP position cannot be defined on purely structural grounds. Correlatively, there can be no distinct morphology corresponding to such a position. Hence, the uniform treatment of genitive DPs in French. Although a structural SpecDP position does not seem to be well-motivated hi French, certain genitive DPs differ from others by their subject-like properties. However, the subject status of such genitives need not be assumed to be due to the structural position they occupy, but may instead be directly related to the type of D.11 That is, hi the context of the definite article, dephrases would count as DP-subjects, whereas in the context of any other determiner, fife-phrases 11 As argued in Dobrovie-Sorin (forthcoming), this is a reversal of Milner's (1982,1995) proposal, which postulates an intrinsic difference between a prepositional de and another de which counts as a genitive Case marker. This categorial difference would correlate with a structural difference, which in turn would trigger a different choice of determiner on the head N.
(In)defmitness spread
193
would count either as nominal complements or as DP-adjuncts.12 We may now try to understand why Romanian, but not French, has alternating genitives. What I take to be crucial is the fact that Romanian, but not French, has a SpecDP position that can be structurally identified, and associated to it, a particular rule of genitive Case assignment. Precisely because this rule is structure-dependent, it cannot apply to all possessor DPs, hence a different mechanism of genitive Case assignment is needed. Since French does not have a SpecDP position that can be structurally identified, it does not have a structural rule of genitive Case assignment: adnominal DPs are uniformly Case-marked by de.
s. ROMANIAN GENITIVES AS CONSTRUCT STATE NOMBVALS (CSNS) Romanian genitive constructions exhibit certain similarities to the Hebrew CSNs.13 According to the terminology used in Semitic grammars, the term "noun in the construct state" refers to the head of the overall DP; the DP corresponding to the genitive DP is called "associate": (21)
a.
b. c.
The determiner of the head N is severely constrained: Hebrew CSNs allow only an empty D (or a D filled with a null determiner) and Romanian genitives allow only the definite article (which is a suffix). Both CSNs and Romanian genitives show an adjacency constraint between the head N and the genitive/associate DP. In those contexts hi which (2la, b) are not satisfied, genitive DPs cannot be "bare", but must be preceded by a particle (al/ai/a/ale in Romanian and invariable Set in Hebrew).
These similarities are all the more interesting as Romanian genitives do not have the morphophonological properties characteristic of Semitic CSN*s: a noun "in the construct state" is phonologically reduced (bayit >beyt 'house'), and correlatively forms with its associate a phonological word (Borer 1988).14 Note furthermore that CSNs hi Hebrew and Arabic can be headed not only by nouns, but also by adjectives, gerunds or quantifiers (Borer 1988, 1996; Danon 1996; Fassi-Fehri 1997; Siloni 1994, 1997, 1998). Romanian bare genitives are
12 In Dobrovie-Sorin (forthcoming), the distinction between subject-like and complement-like de-phrases was formulated, somewhat misleadingly, in terms of structural and morphological genitive Case, respectively. 13 Most of the relevant properties of this Hebrew construction are also found in the other Semitic languages. 14 As observed by Martine Coene (p.c.), Romanian possessive adjectives may give rise to a pattern that has the morphophonological properties of the Hebrew construct state: maica-sa, taicO-tu, soru-mea 'mother-her/his, father-your, sister-my1. This construction is lexically restricted, it can only be used with certain terms of relation.
194 Comparative studies in Romanian syntax disallowed in these various constructions; they can only be constructed with definite nouns or with prepositions followed by a suffixed definite article, a case that I leave aside here (see Grosu 1994 and footnote 7. above). Although we have already dealt with the Romanian paradigms in previous sections, the relevant examples are repeated here hi order to show the parallelism with Hebrew as clearly as possible. The examples in (23) and (24) illustrate the constraint on determiners. The incompatibility between genitives and indefinite articles is overt in Romanian, as in (23b), whereas in Hebrew it only appears at the interpretative level.15 Since Hebrew has a null indefinite article (a characteristic to which we will return below), the example in (24)a could have one on the head N. It is the impossibility of the reading "a house of the man" that indicates that CSNs with definite associate DPs are (semantically) incompatible with an indefinite article on the head N. The incompatibility with the definite article ha, which is phonologically realized, can be directly observed, as in (24b).16 (23)
(24)
a.
casa omului house-the-FEM.SG man-the-GEN.MASC.SG 'the man's house'
b.
*o/*aceasta casa omului a/this house man-the-GEN.MASC.SG
a.
beyt ha-iS house the-man 'the man's house' *ha-beyt ha-iS the-house the-man
b.
The examples in (25) illustrate the adjacency constraint. (25)
a.
*casa mare omului house-the-FEMSG big man-the-GEN.MASC.so
b.
*beyt house
ha-gadol ha-iS the-big the-man
The examples in (26) and (27) show the presence of al and Sel in those configurations in which 15
If the "tanwin" (realized as an -n suffix) is analyzed as an indefinite article (for a refinement of this traditional view, see Ayoub 1991), standard Arabic offers an overt manifestation of the incompatibility between construct state and indefinite articles. "Most of the Hebrew examples in this paper are taken from Borer (1988, 1996).
(In)defmitness spread
195
a CSN/genitive DP without al is prohibited: (26)
a.
b.
(27)
a.
b.
c.
o/aceasta casa a omului a/this house aman-the-GEN.MASC.SG 'a/this house of the man' casa mare a omului house-the-FEM.SG big a man-the-GEN.MASC.SG 'the man's big house' bayit Se/ha-iS house Se/the-man 'a house of the man1 ha-bayit Sel ha-iS the-house Se/the-man 'the house of the man' ha-bayit ha-gadol Se/ha-iS 'the house the-big of the-man'
It is worth stressing that French fife-genitives (and English o^genitives) are not constrained by (21a,b), and correlatively no genitive alternation exists in French. The similarities between Hebrew and Romanian illustrated above have led Longobardi (1996) to suggest that Romanian genitives constitute a particular type of CSN, a hypothesis that was favored by the fact that the two constructions had both been independently analyzed as relying on the raising of the head N to D (see Cornilescu 1993, 1994, Dobrovie-Sorin 1987 for Romanian; Borer 1996, Fassi-Fehri 1989,Ritter 1987, Siloni 1994,1997 for Hebrew; Rouveret 1994 extends N-to-D to Welsh). Building on this quasi-unanimous analysis, Longobardi argues that the rule of N-to-D is the defining characteristic of CSNs across languages. This rule would be allowed only if the determiner is empty or a suffix (hence the restriction in 2 la), and would be subject to an adjacency constraint (hence 2 Ib).17 In case N-to-D is blocked, a "prepositional"
17 Against the background of Bare Phrase Structure assumed here, the N-to-D rule can be restated as Head-to-Head Merge, projecting a D' constituent, which subsequently merges with SpecDP, as proposed in section 1 above for Romanian genitives. Since SpecDP can be assigned structural genitive Case, we can account for the lack of Sel. When N-to-D is blocked, for example, by the presence of an overt D element, the SpecDP position is not available and correlatively the possessor DP cannot be assigned structural genitive Case, hence the necessary insertion of Sel. Although it is compatible with my main claims, this analysis is somewhat problematic on both theoretical and empirical grounds. Note first that according to Bare Phrase Structure (Chomsky (1994)), constituent structure is not projected by phrase-structure rules, but is instead gradually built by the Merge of items selected from the lexicon. Under this view, each and every syntactic node must dominate lexical material. Thus, if we want to assume that CSNs are D-projections, we cannot assume that D is an empty syntactic position, but rather that it dominates a null determiner (or Agreement) element listed in the lexicon. The problem is that two distinct null determiners would have
196 Comparative studies in Romanian syntax genitive must be used (see 21c). Note that according to Longobardi, CSNs (a label which, in his analysis, covers Romanian genitives) are allowed by enclitic articles, regardless of whether the latter are definite, as hi Romanian, or indefinite. The tanwin of standard Arabic (see footnotes 15 and 28) offers interesting counter-evidence against the hypothesis that enclisisper se is relevant: the head N of CS nominals in standard Arabic cannot be marked with the indefinite suffix -n, even if the associate of the CS is indefinite (this specification is needed in order to avoid a violation of the so-called "(in)definiteness spread" requirement). According to the generalization in (21 a), a/-less Romanian genitives are allowed, not because the head N is marked with a suffixal/enclitic article, but rather because the article is both enclitic (or rather suffixal) and definite. An understanding of why definiteness is crucial for the phenomena at hand will be provided in later sections (Section 10. and subsequent). Within the analysis proposed in Section 1 above, the suffixal nature of the article is relevant insofar as it allows (maybe forces) first Merge with the head N, resulting in a constituent labelled D1, which in turn allows/forces the genitive DP to merge in SpecDP. As will become clear below, the common properties of Romanian genitives and Hebrew CSNs are precisely due to the fact that (in the relevant cases) the genitive/associate DP merges in SpecDP.
6. ROMANIAN GENITIVES ARE NOT CSNs Longobardi's hypothesis that Romanian genitives constitute a particular case of CSNs cannot explain important contrasts between Hebrew and Romanian. First, Romanian genitives without al are incompatible with numerals on the head N, whereas Hebrew CSNs allow numerals, as shown in (28a), as opposed to (28b). In this context, only or/-genitives are allowed in Romanian, as hi (28c). (28)
a.
b. c.
'arba'a bney melex four sons king Tour sons of a king' *patru fii unui rege four sons a-GEN.MASC.SG king patru fii ai unui rege four sons ai a-GEN.MASC.SG king
to be postulated, one corresponding to an indefinite article (which is allowed in Hebrew with both CSNs and free state nominals), and another one co-occurring with a noun in the construct state. Further empirical problems will arise as we proceed, which will lead us to assume that CS's are not D-projections, but rather bare NPs.
(In)defmitness spread
\ 97
'four sons of a king' Crucially, the Hebrew example in (28a) has a semantic interpretation that is clearly different from that of the corresponding Saxon genitive: (28a) means 'four sons of a king1, whereas a king's four sons means 'the four sons of a king1. Correlative to the contrast between (28a) and (28b) is the one between (29a, b) and (30a,b). (29)
a.
b.
(30)
a. b.
ben melex son king 'a son of a king1 bney melaxim sons kings 'some sons of some kings' *un fiu unui rege a son a-GEN.MASC.SG king *fii unor regi sons some-GEN kings
In this context an al genitive is required in Romanian: (31)
a.
b.
unfiua/ unui rege a son O/-MASC.SG a-GEN.MASC.so king 'a son of a king1 fii ai unor regi sons a/-MASC.PL some-GEN kings 'sons of some kings'
Finally, construct state is productively used to form so-called "CSN compounds" (Borer 1988): (32)
a.
b.
c.
ben melex son king 'a prince' beyt xolim house sicks 'a hospital' gan yeladim garden children
198 Comparative studies in Romanian syntax
(33)
a.
b.
c.
'a kindergarden' beyt ec house wood 'a house of wood' gan peyrot tropyim garden fruit tropical 'a garden of tropical fruit' kos kafe cup coffee 'a cup of coffee'
The examples in (32) are true compounds, those in (33) illustrate a productive construction in which the associate is a modifier of the head N. To refer to examples of both kinds I will use Borer's term of "CSN compound", although compounds constitute just the frozen version of the productive modification pattern. In contrast to CSNs, Romanian genitives cannot be used as modifiers of the head N. In this case, even a/-genitives are prohibited, the only possibility being the use of the preposition de, which can never be used as a genitive marker in Romanian (e.g. it never marks possession): (34)
(35)
a. *o casa (a) lemn(ului) a house (a-FEM.SG) wood(-the-GEN.MASC.so) b. *o gradina (a) fructe(lor) tropicale a garden (a-FEM.SG) fruit(-the-GEN.PL) tropical c. *o cea§ca (a) cafelei a cup (fl-FEM.SG) coffee(-the-GEN.FEM.SG) a. o casa de lemn a house of wood b. o gradina de fructe tropicale a garden of fruit tropical c. o cea§ca de cafea a cup of coffee
In sum, CSNs contrast with Romanian genitives in allowing numerals, as well as an indefinite article interpretation on the head N, and in allowing the associate to be interpreted as a modifier. In all these respects, CSNs behave on a par with French de-genitives and English q/^genitives: (36)
a.
quatrefils
d'unroi
(In)definitness spread
(37)
b. a.
(38)
b. a. b.
199
four sons of a king four sons of a king ???un fils d'un roi/?des/quelques fils d'un roi a son of a king/efos/some sons of a king ???a son of a king/ ?some/several sons of a king la/une tasse de cafe the/a cup of coffee the/a cup of coffee
Although the examples hi (37) are highly marginal, and maybe even unacceptable (for some speakers), they are not syntactically ill-formed in the way the Romanian examples hi (30) are. Note that the examples discussed in this section, which give rise to contrasts between Romanian and Hebrew, are all built with indefinite possessors, whereas the examples of the preceding section, which are parallel hi Romanian and Hebrew, are all built with definite possessors.
7. CSNs WITH DEFINITE POSSESSORS In what follows I will attempt to provide an analysis of the cross-linguistic variation shown by genitive constructions based on the idea that the rule of genitive Case assignment (stated in (7), and repeated as (39) below), proposed in earlier sections for Romanian and Saxon genitives, also holds hi Hebrew. (39)
Structural genitive Case is assigned to SpecDP/SpecNmax.
I will furthermore assume that SpecDP imposes severe constraints on the nature of the determiner on the head N. The only options are those described hi (11) and (12), repeated here as (40i. and ii.), which hold for Romanian and Saxon genitives, respectively (see Section 1. above). (40)
SpecDP can be lexically filled only if (i) D is empty or (ii) filled with a definite article.
Definite possessors (marked with ha in Hebrew and a suffix hi Romanian) are DP-projections,
200 Comparative studies in Romanian syntax which are necessarily arguments (type e or type ((e,t)t), and as such need Case.18 Because they need Case, and because they are not marked with inherent Case, the DP-associates of CSNs must occupy the only position to which structural Case can be assigned, the SpecDP position. They can appear neither as adjuncts nor as complements of N. 5e/-genitives differ from C SN-associates insofar as they are inherently marked with genitive Case, which allows them to occupy any syntactic position; hi particular they may occur as DP-adjuncts. The data seems to indicate that ^/-genitives can only be DP-adjuncts, but I will leave this stronger generalization open for further research. Let us now show that the rule hi (39) and the constraint hi (40) correctly account for the genitive configurations found hi Hebrew, French and Romanian. Due to (40), French examples of the type in (41), where the head N carries the indefinite article, cannot be analyzed as having the fife-phrase in SpecDP; the genitive may either be adjoined to the DP-projection of the head N, or it can be a modifier embedded under the NP projection of the head N. Both possibilities are allowed, because ofe-genitives are inherently Case-marked. (41)
une maison de lliomme a house of the-man
Neither adjunction nor embedding under NP is available for DP-associates of CSNs, which, due to Case requirements, can only be merged hi SpecDP. But this position is not available hi (42a), due to (40). Hence the ungrammaticality of (42a), the only relevant analysis being the one hi which the head N is interpreted as indefinite. The only possibility then is the use of a /Se/-genitive (correlated with the free state), which is like Frenchfife-genitivesinsofar as it is inherently marked with genitive Case. (42)
a. b.
beyt ha-iS 'a house of the-man/the house of the man' bayit Sel ha-iS 'a house of the-man1
Our analysis also accounts for the parallelism between the Hebrew examples hi (42a, b) and the Romanian examples hi (43 a, b). (43)
a.
*un elev
vecinei
mele
'"Recall Chomskys (1981) attempt to derive the Case-filter from the theta-Criterion.
(In)definitness spread
b.
201
a student neighbor-GEN.FEM.SG my un elev aI vecinei mele a student O/-MASC.SG neighbor-GEN.FEM.SG my 'a student of my neighbor1
(43a) is ungrammatical because Romanian genitives lacking al need structural genitive Case which can only be assigned to the SpecDP position, but this position is unavailable here because the head N takes an indefinite article. The example in (43b) is allowed, since a/-genitives can be embedded under NP or adjoined to DP (see section 2. above). In sum, the distribution of adnominal DPs is parallel in Romanian and Hebrew CSNs, and differs from that of ate-DPs or of-DPs in English: the DP-associates of CSNs and the Romanian genitives without al must be assigned structural Case, and this forces them to occupy the SpecDP position; in all other contexts, the genitive/associate DP must be preceded by al and Sel, respectively.
s. INDEFINITE POSSESSORS Let us now come back to indefinite possessors, which give rise to important constrasts between Hebrew and Romanian, see (29) and (30), repeated here. (29)
a.
b.
(30)
a. b.
ben melex son king 'a son of a king' bney rnelaxim sons kings 'some sons of some kings' *un fiu unui rege a son a-GEN.MASC.SG king *fii unor regi sons some-GEN kings
The ungrammaticality of the Romanian example in (30a) is parallel to the ungrammaticality of (42a, b): the genitive DP cannot be in SpecDP, and thus cannot take structural genitive Case; only a/-genitives are allowed, as in (3 la, b), repeated here. (31)
a.
unfiuar/ unui rege a son a/-MASC.SG a-GEN.MASC.SG king
202 Comparative studies in Romanian syntax
b.
'a son of a king1 fii ai unor regi sons OT/-MASC.PL some-GEN kings 'sons of some kings'
The grammaticality of the Hebrew examples in (29) is unexpected if we assume the associate noun, melex, to be a DP: as such it should occupy the SpecDP position, but this is incompatible with the main N being (interpreted as) indefinite (recall again the generalizations in (11) and (12)).» Let us now observe that the indefinite possessors in the Hebrew examples above lack an overt article. It is a general property of Hebrew indefinite nouns, both singular and plural, that they do not take an overt indefinite article, even when they occupy argument positions. The lack of an overt indefinite article is clearly not related to the construct state configuration. First, free state nominals also lack an indefinite article. Second, there is no prohibition against an overt indefinite article on the associate of a CSN, as evidenced by Standard Arabic: although the tanwin is disallowed on the head of a CSN, it is allowed on the associate (e.g. bayt-u rajul-i-n "house-Norn man-Gen-a/a man's house'). The absence of an overt article is compatible with two distinct analyses. According to Longobardi (1996), a null indefinite article fills the D position (Hebrew would be an "article-drop" language in the sense of Crisma (1995)). We will instead assume that indefinites are bare NPs (BNPs) rather than DP-projections.20 Since BNPs do not have the semantic type of arguments (they are property-denoting expressions), we may assume that they do not need Case, and consequently they need not appear hi SpecDP. There is, however, an even stronger requirement on the distribution of BNPs: (44)
BNPs cannot occupy the SpecDP position.
It is important to stress that we do not assume a general ban against BNPs in argument positions: although they are property-denoting expressions, BNPs are allowed to occupy argument-
19
Alex Grosu (p.c.) observes that indefinite construct states are highly marginal, and that they are only acceptable if an adequate context is provided. As will become clearly below, the markedness of indefinite CSNs is expected under our analysis. 2 *There is a third possibility, namely that the D position is empty at D-structure (Longobardi 1996) and filled by N-tcD raising at S-structure.
(In)definitness spread
203
positions of the verb or of the noun (see below).21 The prohibition in (44) follows from the requirement stated in (45), which is quite well established for the verbal domain: in a wide variety of languages, BNPs are allowed to appear as direct objects, and more rarely as subjects, but only in positions that are arguably c-commanded by the verb. (45)
BNPs must be lexically-governed.
The analysis of Hebrew CSNs to be proposed below will be based on the assumption that the constraint in (45) also holds hi the nominal domain: because they must be lexically-governed, BNP-associates of CSNs cannot appear in SpecDP, nor can they adjoin to DP. They can only be merged as sisters of the head N, as hi (29c), which is representation of (29a, b).
(29)
c.
melex melaxim This representation may be simplified, as in (29d), by assuming that even the head N lacks the DP-level.
(29)
d.
melex
2I
There is a growing literature arguing that existential bare plurals are BNP-projections, even when they occupy argument positions (Dobrovie-Sorin 1997; Dobrovie-Sorin & Laca 1998; McNally 1995; van Geehoven 1996). This analysis can be extended to bare nouns in general, in particular to bare singulars in languages such as Hebrew. 22 This syntactic requirement is probably correlated with the particular rules of semantic composition that account for bare NPs in argument positions, see van Geehoven's (1996) "semantic incorporation" or Dobrovie-Sorin's (1997) proposal that bare NPs in argument-positions are property-denoting expressions that restrict the range of an existentially bound argument-variable. Crucially, this argument-variable does not arise as an LF translation of the BMP itself, but instead corresponds to an argument in the thematic grid of the verbal predicate.
204 Comparative studies in Romanian syntax The ban on BNPs in SpecDP can be easily observed in Saxon genitives : (46)
a. some friends' children b. *friends' children 23
Although the relevant Romanian data is more complex, a careful analysis reveals that the constraint stated in (44) also holds in this language. Thus, although in Romanian, bare plurals can be used in subject and object argument positions, they cannot take genitive Case. One could argue that this is so simply because genitive Case can only be morphologically marked on the article. This is however not true, because with masculine proper names, which do not take an article in Romanian, a genitive form is constructed by prefixing lui, which indeed corresponds to a genitive definite article. It is worth mentioning that this type of genitive marking is completely productive, as indicated by the fact that it extends to datives (recall that genitive and dative markings are identical in Romanian): (47)
a.
b.
cartea lui Mihai book-the the-GEN.MASC.SG Mihai 'Mihai's book' I-am serfs lui Ion him-DAT-have-1 so written the-DAT.MASC.so Ion 'I/we wrote to Ion1
This form is completely general for masculine proper names, and we could expect the existence of a similar genitive prefix for Romanian bare plurals.24 Quite interestingly, such an element does "Examples such as those below, brought up by Alexandru Grosu (p.c.), may be taken to provide counter-evidence against (47), as shown in (i). (i) Close friends' children ought to be treated like your own (ii) Paupers' wives rarely wear diamonds, rich men's mistresses often do so. DPs of this type may be analyzed as relying on a modification pattern (see Woisetschlaeger 1983, among others), in which case the Saxon genitive is not in SpecJDP. The corresponding French examples are built with the "mass particle" de and are to be analyzed as relying on modification: (iii) Les enfants d'amis proches devraient etre traites comme les tiens. 'Children of the close friends should be treated as your own.' (iv) Les femmes de pauvres portent rarement des diamands, les maitresses de riches le font souvent. 'Poor women rarely wear diamonds, rich men's lovers often do it.' 2 *The plural counterpart of lui is lor (which is the normal form of a GEN/DAT plural definite article, e.g. fata vecinilor 'daughter-the neigbours-the-GEN.PL'), but this form is not used as a prefix in front of bare plurals, neither as a genitive nor as a dative marker: (i) *fetele lor vecini girls-the lor neighbors (ii) * Am sens lor vecini have written lor neighbors The reason for this restriction is quite easy to understand: lui and lor are forms of the definite article, which are
(In)definitness spread
205
exist, in the form of an uninflected preposition, a for genitives and la 'to' for datives, which are used with cardinal indefinites (the cardinal itself cannot be morphologically marked for Case): (48)
a.
b.
fetele a trei vecini daughters-the a three neighbors 'three neighbors' daughters' Amscris la trei vecini. have written to three neighbors '[I/we] wrote to three neighbors'
In sharp contrast with (48a), the genitive preposition a cannot be used with bare plural indefinites:25 (49)
*fetele a vecini daughters-the a neighbors
The ungrammaticality of (49) can be analyzed as showing that bare plurals are not treated as DPs, but rather as "bare" NPs, which as such cannot appear hi the SpecDP position. Those indefinite possessors which carry indefinite articles are DP projections, and as such they are freely allowed to take either genitive or dative Case: (50)
a.
b.
copiii unor prieteni children-the some-GEN friends 'some friends' children' Am scris unor prieteni. have-lso written some-DAT friends
compatible (although not normally used in Romanian) with proper names, but not with bare plurals, which are indefinite. In other words, examples such as (i>(ii), if grammatical at all (they become so if lor is attached as a suffix to the possessor DP) would be interpreted as constructed with a definite possessor. "The dative preposition la apparently can appear with BNPs: (i) Am scris la vecini. have written to neighbors 'I wrote to my neighbors.' The grammaticality of (i) could be analyzed as indicating that BNPs are allowed in the argument position of indirect objects. Note however that vecini 'neighbors' is only apparently a BNP. It is indeed a general property of Romanian definite DPs that they drop the definite article when they are preceded by a preposition (in curte 'in courtyard', pe masa'on table' etc.; the definite article is obligatory if the noun is modified: in curtea ta 'in courtyar-the your', pe masa rope "on table-the red'). The example in (i) indeed has a definite interpretation. It thus appears that Dative arguments cannot be expressed by BNPs either, on a par with genitives and in contrast with subject and objects. This generalization seems to be cross-linguistically widespread.
206 Comparative studies in Romanian syntax 'I/we wrote to some friends' Returning to Hebrew, it is reasonable to assume that the constraint in (44) also holds in this language. BNPs are illegitimate in the SpecDP position, on a par with Romanian, but they are legitimate if embedded under NP, as in (29d).26 Note now that according to our analysis, CSNs constructed with indefinite possessors, as in (29), come out structurally identical to indefinite "CS compounds", as in (32) and (33). More precisely, we are led to assume that the two readings of ben melex, that is, 'prince' or 'a son of a (particular) king', rely on the same syntactic representation, as in (29d), in which the associate of the CSN functions as a modifier of N rather than as an argument. Although this outcome of our analysis may seem problematic, French data confirms the empirical generalization: (51)
a.
b.
Jean est en train de jouer avec des enfants d'amis (qui habitent a cote). 'Jean is playing with children of friends (who live next door)' Jean est en train de recevoir des parents d'eleves (de sa classe). Jean is receivinb parents of students (of his class).
One must first observe that the indefinite possessor is preceded by de and not by des (contracted form ofde+les), the normal form of indefinite plurals in French (see J'ai rencontre des amis 'I met des friends', meaning 'I met friends'). Since des indefinites are the closest French analogue of Romanian or Spanish bare plurals, the ungrammaticality of possessor fifes-indefinites is yet another instantiation of the constraint stated in (44).27 The uninflected preposition de that appears in (5 la, b) is on the other hand currently used with modifying adnominal complements, some of which have become true compounds:
26
If we extend our analysis of the distribution of indefinite/BNPs to event nominals, we are confronted with the following problem, raised by Tali Siloni (p.c.): assuming that the complement position is occupied by the internal argument, where does the indefinite subject sit? The following options seem to me available: (a) following Borer (1996), we may assume that event nominals involve more structure than object-referring nominals; for example, a verbal projection, in which case the indefinite subject could sit in Spec.VP; or (b) we may assume that internal arguments (marked with Sel) do not sit under N1 but are instead DP-adjoined (see footnote 31). 27 DPs such as des enfants des amis 'children of friends' are grammatical in French, but they can only have the meaning 'children of the friends': des is again the contracted form of de + les, but in possessive DPs, de is interpreted as a genitive marker, and correspondingly, le(s) takes its standard, definite meaning. When des, du, de la are used with an indefinite meaning, de has the role of some sort of "mass particle": (i) Jai examine des enfants malades. I have examined des children sick 'I examined sick children.' (ii) Jai mange du beurre/de la farine/des epinards. I have eaten of butter/of the flour/of the spinach
(In)definitness spread (52)
a. b.
207
tasse de cafe, hvre de classe, fusil de chasse, permis de travail cup of coffee, book of class, gun of hunting, permit of work pomme de terre apple of ground 'potato'
It thus appears that the French examples in (5 la, b) are like the Hebrew examples in (29) insofar as they rely on adnominal modifiers of the category NP rather than on genitive-marked DPs. The French examples also show that, despite the fact that in these configurations the de-complement is structurally a BNP modifier (it is headed by the invariable preposition de rather than by the plural indefinite article des), it can be interpreted as referring to a particular (plural) individual due to the larger linguistic context. Thus, the relative clause and the adnominal complement surrounded by brackets in (5 la, b) trigger a referential reading for d'amis and d'eleves. In sum, Hebrew indefinite CSNs are like French plural "genitives" headed by de insofar as thenassociate is a BNP that is structurally a modifier that may be forced to take a referential/argumental status due to properties of the linguistic context. This analysis explains why indefinite CSNs are completely productive if the associate is interpreted as a modifier, but highly marginal if the associate is interpreted as an argument. The following examples are due to Tali Siloni (p.c.), who confirmed Alexandra Grosu's observation quoted in footnote 19: (53)
?(?) 'eSet pakid nimcet ba-xeder wife clerk is-found (=is) in-the-room 'A clerk's wife is in the room.'
Tali Siloni further observes that indefinite CSNs become more acceptable if the associate is modified (this is parallel to the French examples in (51), in which the referential interpretation of the cfe-phrase is brought about by the relative clause and modifier of the class) or if it takes 'exad'one', which, according to some scholars, is developing into an optional indefinite article: (54)
'eSet pakid 'exad nimcet ba-xeder wife clerk one is-found (=is) in-the-room
The complete acceptability of (54) is expected under our analysis: an indefinite in 'exad is not a BNP, but a DP-constituent, and, as such, it occupies the SpecDP position, and correlatively
208 Comparative studies in Romanian syntax behaves as a true argument rather than as a modifier.28 The difference between 'exad-indefinite associates and BNP-associates is left unexplained under current accounts, according to which indefinite associates lacking an overt determiner are DP-projections headed by a null indefinite article. Returning to indefinite CSNs with BNP-associates, our analysis also explains why they are compatible with numerals: (28)
a.
(28)
b.
'arba'a bney melex four sons king 'four sons of a king'
(28b) is a legitimate structure because melex 'king' does not occupy SpecDP. In (28c, d), compare Romanian genitives without al, which necessarily occupy SpecDP and correlatively disallow numerals. (28)
c. *patru fii unui rege four sons a-GEN.MASC.SG king d. patru fii ai unui rege four sons CK-MASC.PL a-GEN,MASC.so king
Note that the structure in (28b) contradicts the widely-assumed generalization according to which CSNs are incompatible with their D-position being overtly realized (see also (2la), which was assumed above as a starting point of the discussion).29 Against the background of the analysis 28
Although discussion of standard Arabic is beyond the scope of this paper, I expect a tanwin-marked associate to function as a full DP-projection (on a par with 'exad-marked Hebrew indefinites) rather than as a BNP. Correlatively, tanwin-marked and 'exad-maiked associates are expected to contrast with Hebrew indefinite (BNP) associates and behave on a par with Romanian indefinite genitives. I leave this issue open for further research. 29 Note that the representation in (28') in which numerals are hosted by D might turn out to be incorrect. One may instead adopt the analysis of numerals proposed in Danon (1996).! leave a careful analysis of Hebrew numerals open for further research.
(In)defmitness spread
209
proposed here, the above-mentioned generalization need not be maintained. As argued in section 10., the correct part of (2la), which corresponds to the prohibition in (40), follows as a consequence of the rule of semantic composition that characterizes the SpecDP position. Since BNP-associates do not occupy SpecDP, they are not expected to fall under (40). Let us now briefly summarize the overall picture of the syntax of genitive constructions in the various languages under discussion here. I assume that the rule of genitive. Case assignment and the categorial prohibition against BNPs in SpecDP hold universally. The latter was presented in (39) and (40), and is repeated under (55a, b). (55)
a. b.
Structural genitive Case is assigned to SpecDP/SpecNmax. SpecDP can be lexically filled only if (i) D is empty or (ii) D is filled with a definite article.
Languages vary depending on whether they have possessor expressions that are marked with inherent/prepositional Case (see de, of, Sel, al in French, English, Hebrew and Romanian, respectively) or instead require structural genitive Case (see Semitic associates of CSNs and Romanian genitives without al).30 Hebrew and Romanian show that those languages that have possessors with structural genitive Case also have possessors marked with inherent genitive Case. This generalization follows from the fact that, as stated in (55a), structural genitive Case can only be assigned to SpecDP. Since this position is not available in all possessive constructions, due to (55b), inherent genitive Case is also needed in languages with structural genitives. The common properties of Hebrew CSNs and Romanian genitives without al are all due to the fact that structural genitive Case, which can only be assigned in SpecDP is needed. Compare inherently-marked genitives, which are not constrained to appear in particular structural positions. The diverging properties of Hebrew CSNs and Romanian genitives follow from the difference in categorial status between Hebrew and Romanian indefinite possessors: the former are BNPs, whereas the latter are DP-projections, on a par with definite possessors. As a consequence of the requirement in (45), Hebrew indefinite associates cannot appear in SpecDP, but must appear as sisters of N, where they are lexically governed, as required by BNPs, as in (45). In sum, CSNs exhibit a strict correlation between the categorial status of CSN-associates and the
30
Note that morphological Case marking may either count as inherent Case marking (see German genitives) or as structural Case (see Romanian simple genitives, as well as the associates of Arabic CSNs).
210 Comparative studies in Romanian syntax position they occupy: (56)
The associates of CSNs appear in SpecDP (or rather, SpecNmax) or are embedded under NP (as either complements or modifiers) depending on whether they are DPs or BNPs.
Neither DP-associates nor BMP-associates of CSNs can be DP-adjuncts: DP-associates need structural Case, which cannot be assigned to DP-adjuncts, and BNP-associates must be lexically governed by N, a requirement that cannot be satisfied in a DP-adjoined position. Inherentlymarked genitives, ^/-genitives in particular, can be DP-adjuncts.31
9. OVERVIEW OF CSNs AND (IN)DEFBVITENESS SPREAD According to the analysis proposed above, CSNs with definite associates are DP/Nmax constituents with the definite associate in SpecDP/Nmax: (57)
a.
ben ha-melex son the-king
b.
DP/Nmax
N
SpecDP/Nmax
ben
ha melex
The structure in (57b), which is essentially that assumed by Borer (1988), can be slightly modified in order to be rendered compatible with the rule of N-to-D movement (or N-to-D Merge), which is currently assumed for the analysis of CSNs since Ritter (1987). We simply need
31
The core evidence seems to suggest an even stronger generalization, according to which 5e/-genitives are necessarily DP-adjuncts, which seems to conflict with the fact that Se/-genitives can be internal arguments of nouns (see event nouns in particular). Any Hebrew noun can appear either in the CS or in the free state, regardless of whether it selects internal arguments or not. Within GB-type analyses, according to which internal arguments are generated as sisters of the head that selects them (see also the "first merge" requirement, which is a Minimalist restatement of the same hypothesis): internal arguments of nouns (of event nouns in particular) are generated under N1, as sisters of the head N, but they can only remain inside the NP/DP projection of the main N if the latter is in the CS. A noun in the free state pushes its internal argument into an adjunction position: under the analysis proposed here, SW-phrases are DP-adjoined at S-structure/Spell Out, even if they are internal arguments, as in, for example, ha-harisa Sel ha-'ir be-'axzariyut 'the-destruction of the-city in-cruelty". Although the reason for the displacement is different, the obligatory displacement of internal arguments that characterizes free state nominals brings to mind the obligatory displacement of sentential complements.
(In)definitness spread
211
to postulate a null D element to which N moves (or which directly merges with N). Following Borer (1996), I assume that Hebrew non-process NP-projections are N-initial:
SpecNP 0
bevt
ha iS
It thus appears that although our analysis of CSNs with definite associates does not depend on the rule of N-to-D raising (or N-to-D first/direct Merge), it is compatible with such a rule.32 Note, however, that CSNs constructed with indefinite associates are not compatible with N-to-D: by assuming a null D for indefinite NPs, we lose the account proposed in the preceding section, which crucially relied on the idea that Hebrew BNPs lack the functional level and as such are bound to be merged as complements/modifiers of N. This means that the rule of N-to-D raising cannot be assumed to characterize all the instantiations of CSNs. Let us now observe that although the definite morpheme ha appears prefixed on the associate, both the associate and the head noun are interpreted as definite. Borer (1988) accounts for this interpretation by postulating a percolation mechanism by which the +def feature percolates from the complement DP to the head noun. As a result, both the complement DP and the head noun, as well as all intermediate nouns (see examples such as madafsifrey ha-yalda "shelf books theguT, meaning 'the girl's shelf of the books') are marked as definite. Borer is aware that the sharing of +def features between complements and heads is not normally allowed, and she attributes the existence of this marked mechanism to the word-status of CSNs: the percolation of+def features is allowed inside phonological words. More recently, Borer (1996) tries to argue that the percolation of ±def features is mediated by the rule of N-to-D raising.33 The necessity of N-to-D would thus be derivable from the percolation of ±def features, which in Borer's system is the result of the incorporation of the head of the complement into the head of the CSN. Other authors analyze the sharing of ±def features between the associate and the head of CSNs as 32
One could further complicate the structure in (5 7') by adding various functional projections (NumberP, GenP, etc.). Although such abstract layers are not needed for an account of the data at hand, they could be made compatible with my main claims. 33 Note that Borer (1988) discusses only the percolation of+def features. The absence of an explicit analysis of-def readings could be interpreted as indicating that the author believed that no percolation of features was needed in this case (as I explicitly argue here). Borer (1996) adheres to the dominant trend of thinking, which treats CSNs constructed with BNP associates on a par with CSNs constructed with definite DP associates.
212 Comparative studies in Romanian syntax relying on Agreement with either SpecDP or Spec, Agrgen (Fassi-Fehri 1989,1993,; Longobardi 1996; and Siloni 1994, 1997, respectively). This account is problematic: how can it be that the ±def features, although realized only once, are interpreted twice? Agreement phenomena are exactly of the opposite kind: an f-feature (gender, number, person, Case) is interpreted only once, although it shows up on several elements.34 In sum, the current account of ±def spread is a marked, construction-specific mechanism, as acknowledged by Borer (1988). In section 10.1 will argue that this kind of mechanism can be dispensed with. In examples of the type in (57), ha marks as definite only the complement DP, the definite interpretation of the entire DP being due to the particular rule of semantic interpretation that needs to be postulated for DPs that contain a SpecDP constituent. This analysis will be shown to extend to Saxon genitives, Romanian genitives without al, as well as to definite possessive DPs in French and English. Note now that under our analysis of CSNs with indefinite associates, no percolation of-def features is needed, since both the complement and the overall possessive are BNPs, and as such can only be interpreted as indefinite. Thus, although a special rule of semantic composition is needed for CSNs with definite genitives as well as for (both definite and indefinite) Saxon genitives, no such rule is needed for CSNs with indefinite genitives. Consider finally definite compound CSNs: (58)
a.
b.
c.
d.
e.
benha-melex son the king 'the son of king1/ 'the prince' beyt ha-xolim house the-sicks 'the hospital' gan ha-yeladim garden the-children 'the kindergarden' beyt ha-ec house the-wood 'the house of wood' gan ha-peyrot tropyim garden the-fruit tropical
34 Compare also the (static or dynamic) Agreement relation underlying Rizzi's (1991) w/j-criterion (later extended to other "criteria" configurations): arguably, only the element in SpecCP has an interpretable W/-feature. The whfeature on C is a formal feature, which accounts for the position of V+I, but carries no particular meaning. In sum, the dynamic Agreement postulated by the w/j-Criterion does not generate two wA-phrases in the place of one.
(In)definitness spread
f.
213
'the garden of tropical fruit1 kos kafe cup the-cofFee 'the cup of coffee'
These examples clearly show that although the position of ha is fixed at S-structure (or at Spell Out), it can be interpreted in two distinct positions at LF: it may either determine the complement, as in (57), or the head N, as in (58). Any analysis of examples such as (58) is bound to postulate some kind of percolation procedure (Borer 1988), which may be implemented in different ways. We may assume that ha is initially merged with the head noun, and subsequently lowered to the embedded N by some phonological rule. Or conversely, ha is merged with the embedded N, and its +def feature percolates/raises at LF:
(58)
g.
ha melex
The LF raising of ha to D is presumably incompatible with the S-structure rule of N-to-D. Examples of the type in (58) thus constitute another type of CSN for which the rule of N-to-D raising is inadequate. In sum, we are led to cast doubt on the current account of CSNs based on the rule of N-to-D. Note furthermore that our analysis of CSNs with indefinite associates allows us to dispense with construction-specific syntactic mechanisms that ensure that the -def feature of the genitive DP is shared by the head N. We have instead argued that no "indefiniteness spread" need be assumed: both the associate and the overall BCSN are BNPs, which are independently attested in the language. Definiteness spread is not needed for the examples in (58) either: ha raises to D at LF (this is the only construction-specific interpretive rule of Hebrew CSNs), but crucially ha is not interpreted twice as filling both D positions. We are thus left with only one type of CSN, the examples in (57), which seem to need some mechanism by which to force the +def feature of the associate DP to percolate up to the head noun. In the remaining sections, I will argue that this phenomenon, which will be identified with
214 Comparative studies in Romanian syntax the (in)definiteness spread characteristic of Saxon genitives, is due to a particular rule of semantic composition, rather than to some Agreement relation in ±def features between the genitive DP and the D of the head noun.
10. SPECDP AND THE SEMANTIC COMPOSITION OF DPS The remaining sections of the paper will be devoted to the semantic composition of possessive DPs in which the possessor is in SpecDP. The proposal must account both for the (in)definiteness spread phenomenon and for the syntactic constraints stated in (55b) and repeated as (59). (59)
SpecDP can be lexically filled only if D is (i) empty or (ii) filled with a definite article.
It is clear that such a restriction is completely unexpected if we assume that possessive DPs rely on the standard rules of semantic composition, according to which the determiner applies to the property denoted by the NP. I will argue that those DPs which contain a SpecDP constituent cannot be composed in this way: the D1 constituent obtained by merging D (type ((e,t), ((e,t),t)) with NP (type (e,t)) already has the semantic type of an argument (type ((e,t),t)), and as such it can only merge with a DP-adjunct. The generalization obtained so far is that in possessive DPs constructed with canonical D-elements, the possessor can occupy an adjunct position (or a complement position under NP)), but not the SpecDP position. The semantic incompatibility between SpecDP and determiners such as demonstrative or indefinite articles thus matches the negative part of the syntactic universal stated in (59), according to which SpecDP is incompatible with determiners other than the definite article. Run-of-the-mill determiners only allow possessor DPs to merge as complements of N or as DP-adjuncts: (60)
a.
(60)
b.
o fata a vecinei a daughter CT-FEM.SG neighbor-FEM 'a daughter of the neighbor1 DP! (et,t) Dx (et,t)
D (et, (et,t))
NP (et)
o
fata
DP2 (et,t)
(In)defmitness spread a
daughter
215
a-FEM.SGneighbor-FEM
From the point of view of the Merge operation, a DP which contains SpecDP and one that relies on DP-adjunction are comparable: D merges with NP, and then the resulting term, notated Dx in (60b), merges with another DP. The difference resides only in the label of Dx: if Dx is a D'constituent, its sister is in Spec,DP, and if Dx is a DP, its sister is adjoined to it. No typemismatch arises with adjunction configurations: by applying D to NP we get a constituent of type ((e,t),t), which is an adequate type for the host of a DP-adjunct.35 At this point, one might conclude that a DP-adjunct cannot be distinguished from a SpecDP constituent. However, Romanian and Saxon genitives, as well as Hebrew CSNs provide clear syntactic evidence hi favor of the existence of adnominal DPs which are not adjuncts: adjuncts must be preceded by al and by Sel, in contrast to genitives without al and CSN- associates. We have also seen that those adnominal DPs that are not adjuncts impose severe constraints on the type of determiner. This syntactic generalization matches with the semantic incompatibility discussed here. Thus, on both syntactic and semantic grounds we are led to the negative conclusion in (61). (61)
SpecDP is incompatible with a D-element of type ((e,t), ((e,t),t)).
Indefinite, cardinal or demonstrative determiners are of type ((e,t), ((e,t),t)), and indeed they are incompatible with genitive DPs in SpecDP. We may now use the positive syntactic generalizations stated hi (59), namely the fact that SpecDP is compatible with either an empty D or with a definite article, hi order to find out what the rule of semantic composition could be. Since in certain languages SpecDP necessarily co-occurs with an empty D, we know that the rule of semantic composition we are trying to find can dispense with determiners. For the purposes of semantic composition, it does not matter whether D is empty or filled with a null article: under both analyses, D does not count for semantic interpretation. Since D does not count for semantic 33
One problem, however, remains: is it possible to combine a constituent of type ((e,t),t), see Dx, which we now identify with DP, with another constituent of type ((e,t),t), the DP-adjunct? We may assume that the predicatemodification rule, which is currently allowed to combine sets of individuals (type (e,t)), can be extended to the composition of sets of properties (type ((e,t),t)): the denotation afofatSa vecinei 'a daughter of the neighbor's1 is obtained by intersecting the set of properties of an individual y (such that y is a daughter of an individual x) with the set of properties of an individual z (such that z is related to the neighbor). Another possibility is to assume that a/genitives are not DP-adjoined, but rather complements of the N-head (see (2)), in which case we first compose the genitive with the head N, yielding a property/set of individuals (the set of the individuals that are daughters of the neighbor), and then we apply the denotation of D to this set of individuals. Note, however, that this analysis cannot be adopted for 5e/-complements, which, as argued above, can only be analyzed as being DP-adjoined at Spell Out. We may assume LF reconstruction into NP, which is available either due to the presence of an NP-trace in the complement position of the N-head (if we assume first Merge into that position) or due to the presence of a thematic feature on the N-head.
216 Comparative studies in Romanian syntax composition, the denotation of the overall nominal projection is obtained by applying the denotation of the head N to the denotation of the DP in SpecDP. Now, if we assume that both SpecDP and the overall DP denote generalized quantifiers (i.e., are of type (et,t)), we can only relate them by assuming a rule of predicate-modification, with D1 itself denoting a generalized quantifier. But then, we are back to the semantic composition just discussed, which characterizes DP-adjunction. The only possibility left open is to assume an individual-type denotation (type e) for both SpecDP and the overall DP. Correlatively, D', or more precisely the NP dominated by D1, denotes a function that maps individuals to individuals: (62)
The sister of SpecDP denotes a function from individuals to individuals.
To make this proposal explicit, consider the examples in (63a, b, c), represented as hi (63d). (63)
a. b. c.
Mary's sister Mary's youngest sister a neighbor's house
(63)
d.
DP/Nmax
SpecDP/Nmax Marys a neighbor's
D'/Nnonmax sister/youngest sister house
The denotation of the overall DP in (63a) is calculated by applying the sister of function to the individual denoted by Mary. Similarly, (63b) denotes the individual obtained by applying the function youngest sister of to the individual denoted by Mary. Finally, (63 c) denotes the individual obtained by applying the function house of to the individual variable contributed by the indefinite DP a neighbor.36 The free variable in (63c) is bound by existential closure (the exact implementation of which is irrelevant here):
36 Our analysis is consistent with DRT-type approaches (according to which indefinites contribute individual variables to the LF representation, see Heim 1982, Kamp 1981). Within Generalized Quantifier Theory, type-shifting operations (Partee & Rooth 1983) are available, which allow indefinites to denote individuals, in certain contexts at least. Rather more complex are examples such as: (i) Every student's house was carefully observed, where the genitive phrase is "essentially quantificational"; that is, irreducible to individual-type denotation. The correct interpretation 'for every x, x is a student, the house of x was carefully observed' can be obtained by first applying Quantifier Raising to every student; the house-o/fiinction will then apply to the variable bound by the raised quantifier.
(In)defmitness spread (63)
e. f. g.
217
Mary's sister => y = f(x), where f= the sister of A x = Mary Mary's youngest sister => y = f(x), where f = the youngest sister of A x = Mary a neighbor's house => y = f(x), where f = the house of A x=//a neighbor// (//= denotation of)
The analysis proposed here constitutes an extension of Frege's (1891) analysis of functional nouns; for example, the capital of the German empire. Unlike Frege, we do not assume that the functional denotation stated in (62) obtains only if the head N intrinsically denotes an (e,e) function. In our view, the functional denotation is contextually determined: the sister of SpecDP denotes a function from individuals to individuals regardless of the lexical properties of the head N (or of the NP), which may be functional (mother, youngest sister, etc.), relational (brother, sister, friend, etc.), or simply property-denoting (house, pencil, dress, etc.). Note also that the functional analysis proposed here is distinct from the relational analysis that Barker (1995) assumes for the head N associated to a genitive DP: the relational analysis is compatible with any kind of D (see Barker (1995)), whereas the functional analysis is only compatible with an empty D (or with a definite article, as shown below). The semantic analysis proposed here for Saxon genitives extends to the Hebrew CSNs that contain a definite DP-associate: (64)
a. b.
beyt ha-iS house the-man galgaley ha-mexonit wheels the-car
In contrast to Saxon genitives, the DP-associate of CSNs is located on the right of the head N, as in (64c), representing (64a, b). (64)
c.
DP/Nmax
DYNnonmax beyt galgaley
SpecDP/Nmax ha-iS ha-mexonit
In addition to the different location of SpecDP, the head N of CSN cannot be modified by an adjective nor can it take a DP-complement as a sister. Given these limitations, D' can only dominate bare N-elements. Beyond these syntactic peculiarities, Hebrew CSNs with definite DP associates can be assumed to rely on the rule of semantic composition proposed above for Saxon
218 Comparative studies in Romanian syntax genitives: (64)
d.
beyt
ha iS => y = f(x), where f = house of and x = //the man//
If the head N is plural (e.g. galgaley 'wheels'), the overall DP denotes the maximal group (plural individual) associated to the individual denoted by ha mexonit 'the car' by the function wheels of. (64)
e.
galgaley ha mexonit => y = f(x), where f= wheels of and x=//the car//
Finally, the same rules of semantic composition can be assumed for Romanian DPs constructed with genitive DPs without al: (65)
a. casa vecinului house-the neighbor-the-GEN.MASC.SG b. rojile ma§inii wheels car-the-GEN.FEM.SG
Although it is not necessary for the functional analysis of the head N, the definite article is compatible with such an analysis: we may analyze it either as being expletive (Milner 1982,1995; Vergnaud & Zubizarreta 1992) or as indicating that the main N denotes a function from individuals to individuals (unlike Lsbner 1985,1 do not assume that this characterization of the definite article should be extended to all its uses). The semantic analysis proposed here thus completely matches the cross-linguistic syntactic generalizations, according to which SpecDP can be lexicalized only if D is empty or filled with a definite article. The choice between these two possibilities, namely the obligatory presence of the definite article in certain languages (e.g. Romanian) versus its obligatory absence in other languages (English or Hebrew), is not due to semantic composition, but rather to morphosyntactic cross-linguistic differences. One can speculate and say that SpecDP may co-occur with a definite article only if the latter is a suffix, as in Romanian. This is however only a tentative generalization, which must be further investigated.
11. (IN)DEFINITENESS SPREAD Saxon genitives are known to be characterized by "(in)definiteness spread": the overall DP is interpreted as either definite or indefinite depending on whether the genitive DP itself is definite or indefinite (Jackendoff 1974). This generalization is supported by the syntactic tests of
(In)definitness spread
219
indefiniteness: (66)
a. b.
There is a man/*the man/??John in the garden. There is a man's dog/*the man's dog/??John's dog in the garden.
Since current analyses of Saxon genitives assume the canonical kind of semantic composition, based on the necessary presence of a determiner, they are forced to assume some (copying, agreement or percolation) mechanism that provides the main N with the lacking Det-features by borrowing them from the determiner of the genitive DP. Such construction-specific mechanisms are problematic for reasons that we have already discussed above in relation to similar proposals regarding CSNs. The rule of semantic composition proposed above allows us to provide a natural, non stipulative explanation for the (in)definiteness spread that characterizes Saxon genitives. Consider indeed the difference between (63e, f) and (63g), repeated under (67a, b, c). (67)
a. b. c.
Mary's sister => y = f(x), where f= the sister of A x = Mary Marys youngest sister => y = f(x), where f = the youngest sister of A x = Mary a neighbor's house => y = f(x), where f = the house of A x=// a neighbor//
In (67a, b), the sister-of/youngest-sister-ojfunctions apply to a constant individual (denoted by Mary), and therefore they yield constant individuals, hence the definite interpretation. Compare (67c), where the /zowse-q/'function applies to the variable x contributed by the indefinite genitive DP, a neighbor. The overall DP has an indefinite, that is, variable, meaning, because the values of the house-of function vary depending on the values of the variable to which it applies. Under this analysis, no determiner on the head N is needed in order to construct the denotation of the overall DP, hence no mechanism of feature percolation need be assumed. In sum, the phenomenon known as (in)definiteness spread does not rely on a syntactic mechanism that provides the main N with determiner features, but instead is the consequence of the functional analysis of the head N. This analysis straightforwardly extends to Romanian genitives: the definite article on the head noun does not indicate that the overall possessive DP is definite, but rather that the head noun denotes an (e,e) function; the overall DP itself counts as (in)definite depending on whether the genitive itself is (in)definite. As already observed hi previous sections, the definiteness spread that characterizes CSNs can be identified with the (in)definiteness spread found with Saxon genitives: in beyt ha-iS, the house-of function applies to the constant individual denoted by ha iS, hence the definite reading of the
220 Comparative studies in Romanian syntax overall DP itself However, as already observed by Borer (1996), CSNs and Saxon genitives do not show the same type of the indefiniteness spread: (68)
a.
b.
(69)
a. b.
beyt iS house man 'a house of a man'/ 'a man's house' bney melex sons king 'some sons of a king'/ 'a king's sons' a man's daughter ='the daughter of a man' - 'a daughter of a man1 a man's daughters = 'the daughters of a man' - 'some daughters of a man'
The functional analysis of the head noun accounts for the readings found with the Saxon genitives illustrated in (69), which are characterized by a uniqueness presupposition: a man's daughter presupposes that the man has only one daughter, and a man's daughters refers to the maximal set of the man's daughters. The readings indicated for the Hebrew examples in (68) clearly do not carry uniqueness presuppositions. This interpretive contrast between Saxon genitives and Hebrew CSNs goes against Fassi-Fehri (1989, 1993) and Longobardi (1996), who believe that Saxon genitives and the Hebrew CS share the same kind of "(in)definiteness spread". The difference in interpretation between indefinite Saxon genitives and indefinite CSNs correlates with the fact, already pointed out in section 8 above, that indefinite CSN's are compatible with numerals. In (70) and (71), compare again the unacceptability of numerals with Romanian bare, that is, a/-less, genitives, which contrasts with their grammaticality in Hebrew CSNs. Note that (70) becomes grammatical with the genitive al. (70) (71)
*patrufii unui rege four sons a-GEN.MASC.SG king 'arba'a bney melex four sons king 'four sons of a king'/'the four sons of a king'
The ungrammaticality shown in (70) is expected under the hypothesis that Romanian genitives occupy the SpecDP position, which requires the main N to be interpreted as denoting an (e,e) function: the functional interpretation of the head N is incompatible with numerals, because numerals must compose with property-denoting NPs. Crucially, the Hebrew example in (71) has
(In)definitness spread
221
a semantic interpretation that is clearly different from that of the corresponding Saxon genitive: (71) means 'four sons of a king", whereas a king's four sons means 'the four sons of a king1. The interpretation of the Hebrew example in (71) cannot be obtained via the functional analysis, which correctly accounts for Saxon genitives. Conversely, whatever mechanism could explain the Hebrew data cannot extend to Saxon genitives, since the relevant readings are missing. The interpretive data reviewed here corroborate the syntactic analysis proposed in section 8.: indefinite CSNs cannot be subsumed under the analysis of definite CSNs: whereas definite associates are DPs that occupy SpecDP, indefinite associates are bare NPs embedded as complements of N. Correlatively, definite CSNs rely on the functional denotation of the head N, whereas indefinite CSNs are just BNPs that contain BNP possessors, a configuration that constitutes a particular instantiation of "compound CSNs".
12. EXTENSION TO DEFINITE POSSESSIVES IN FRENCH AND ENGLISH According to Barker (1993), Jackendoff(1974), Milner (1982) and, the generalization stated in (72) holds in French and English. (72)
A definite DP containing a possessor (possessive DP, henceforth) is (in)definite just in case the possessor is (in)definite. [Barker's generalization is in terms of (non)familiarity]
The notion of possessor covers q/^phrases, Saxon genitives as well as possessive adjectives (his, her, etc.). The generalization in (72) is needed in order to account for an observation that goes back to Prince (1979): possessive DPs violate the familiarity constraint. Thus, although possessive DPs are definite expressions, they can be used as first mentions see sequences such as A man came in. His daughter... The generalization in (72) also covers the behaviour of "weak definites" (Poesio's 1994 label): definite descriptions of the form le N de NP/ the NofNP, for example, \sfils d'unpaysanl 'the son of a farmer' do not give rise to the definiteness constraint: (73)
a. b.
? There is the student of a linguist in the garden, n est arrive la fille d'un fermier/ it is arrived the daughter of a farmer/ *la fille du fermier. the daughter of the farmer
Milner's (1982) description of the French data brings in further tests which, although questioned by some French linguists (Corblin 1987; Flaux 1992, 1993), appear to strongly support the hypothesis that definite expressions containing an indefinite genitive indeed take weak readings,
222 Comparative studies in Romanian syntax as shown by the impossibility of left or right dislocation, andthe occurrence in predicate position (after the copula), or in the object position of have: (left or right) dislocation: (74) a. *Un linguiste, je le connais depuis lontemps. a linguist I him know from long-ago b. *Lafille d'un linguiste, je la connais depuis longtemps. the daughter of a linguist I herkown from long-ago c. Jean/le voisin/ce boulanger, je le connais depuis longtemps. Jean/the neighbor/this baker I him know from long-ago subject position of BE+ possessor: (75) a. ??Un appartement/Tappartement d'un voisin est a moi. an appartment/the appartment of a neighbor is to me b. Get appartement/l'appartement de Jean est a moi. this appartment/the appartment of Jean is to me 'This appartment/Jean's appartment is mine.' All existing analyses of possessive DPs gree on the basic empirical generalization stated in (72) and differ only in the way hi which (72) is relevant for the particular implementation that each author chooses: Prince's (1979) "provisionally salient objects", Barker's (1991,1993) distinction between familiarity and saliency and Poesio's (1994) cancellation of familiarity all depend on the "anchoring", stated in (72), of the possessive DP to the possessor. The functional analysis proposed here provides a deeper insight into the phenomenon, by allowing us to understand why the generalization in (72) should hold at all. We have already established above that the functional analysis of the head N is compatible with the definite article: Romanian genitives occupy SpecDP and correlatively the head N denotes an (e,e) function. There is little, maybe no evidence in favor of the idea that French fife-phrases occupy the SpecDP position. Some evidence exists against the idea that English q^-phrases occupy SpecDP: (76)
a. b.
the son of a farmer; the student of a well-known linguist *the fire truck of John
The contrast shown in (76a, b) indicates that o/^phrases cannot be freely generated in a structural, non-thematic position such as SpecDP. They can only be merged as complements of N, an operation that is constrained by the selectional properties of the noun: relational nouns select for a complement (hence the grammaticality of (76a), non-relational nouns do not (hence the
(In)definitness spread
223
ungrammaticality of (76b). We are thus led to conclude that the functional denotation of the head N arises not only hi the context of SpecDP, but also with definite possessive DPs. hi this case then, the functional denotation is triggered by the definite article itself. Recall that Lsbner (1985) argues that all the uses of the definite article correlate with a functional use of the head N. For the purposes of the present paper we need not commit ourselves to this extension. Our main conclusions can be summarized as follows: (77)
The head N of a possessive DP denotes a function from individuals to individuals if: (i) the possessor DP is in SpecDP or (ii) the head N carries the definite article.
This disjunctive generalization indicates that the mapping between syntactic Merge and semantic composition is not homo-morphous: structurally distinct configurations can be interpreted via the same rule of semantic composition. Note, on the other hand, that morphological marking and syntactic structure are correlated in a much stricter way: in both Romanian and English, those DPs that occupy SpecDP carry a special morphology: morphological genitive Case and 's, respectively. Compare this case with French: de is inserted hi front of all adnominal DPs, and there is no clear evidence that the SpecDP position can be structurally defined. Similarly, the morpho-phonological peculiarities of Hebrew CSNs solely indicate that several nouns belong to the same maximal projection, the internal syntax of which may vary.
REFERENCES Abney, S. (1986). Functional elements and licensing. Paper presented at GLOW 1986, Girona. Abney, S. (1987). The English noun phrase in its sentential aspect. Doctoral dissertation, Massachusetts Institute of Technology. Ayoub, G. (1991). La nominalite du nom ou la question du tanwin. Arabica 38, 151-213. Barker, C. (1991). Possessive descriptions. Doctoral dissertation, University of California at Santa Cruz. Barker, C. (1993). Definite possessives and discourse novelty. Proceedings ofCLS 28, volume 1, 26-41. CLSI publications. Barker, C. (1995). Possessive descriptions. CSLI Publications. Borer, H. (1988). On the morphological parallelism between compounds and constructs. In:
224 Comparative studies in Romanian syntax Morphological Yearbook (Booij and J. van Marie, eds.), pp. 23-74. Foris, Dordrecht. Borer, H. (1996). The construct in review. In: Studies in Afroasiatic grammar (J. Lecarme, J. Lswenstamm and U. Schlonsky, eds.), pp. 34-56. Holland Academic Graphics, The Hague. Chomsky, N. (1981). Lectures on Government and Binding. Foris, Dordrecht. Chomsky, N. (1986). Barriers. MIT Press, Cambridge, Mass. Chomsky, N. (1994). Bare Phrase Structure. Occasional Papers in Linguistics 5. Cambridge, Mass: Department of Linguistics-MIT. Chomsky, N. (1995). The Minimalist Program. MIT Press, Cambridge, Mass. Chomsky, N. (1998). Minimalist inquiries: the framework. Manuscript, MIT. Corblin, F. (1987). Indefmi, defini et demonstratif. Constructions linguistiques de la reference. Droz, Geneve. Cornilescu, A. (1993). Notes on the structure of Romanian DP and the Assignment of the genitive Case. University of Venice Working Papers in Linguistics 3(2), 107-133. Cornilescu, A. (1994). Remarks on the Romanian ordinal numeral. Revue roumaine de linguistique 39, 303-337. Crisma, P. (1995). Notes on the structure of Old English DP. Manuscript, University of Venice/ UCLA. Danon (1996). The syntax of determiners in Hebrew, http://www.tau.ac.il/danon/Thesis.html D'Hulst, Y., M. Coene and L. Tasmowski (1997). Last resort strategies in DP: Article reduplication in Romanian and French. Manuscript, UIA, Anvers. Dobrovie-Sorin, C. (1987). A propos de la structure du groupe nominal en roumain. Rivista di grammatica generativa 12, 123-152. Dobrovie-Sorin, C. (1997). Types of predicates and the representation of existential readings. Proceedings of SALT, 33-42. Cornell University Press, Ithaca. Dobrovie-Sorin, C. (forthcoming). De la syntaxe a 1'interpretation, de Milner (1982) a Milner (1995): le genitif In: Cahier Jean Claude Milner (G. Lardreau, ed.). Verdier, Paris. Dobrovie-Sorin, C. and B. Laca. (1998). La genericite entre la reference a 1'espece et la quantification generique. Actes du Colloque Langues et grammaire. (Forthcoming) Fassi-Fehri, A. (1989). Generalized IP Structure, Case and VS word order. In: MIT Working Papers in Linguistics 10 (I. Laka and A. Mahajan, eds.), pp. 75-113. Department of Linguistics-MIT, Cambridge, Mass. Fassi-Fehri, A. (1993). Issues in the structure of Arabic classes and words. Kluwer, Dordrecht. Fassi-Fehri, A. (1997). Layers in the distribution of Arabic dverbs and adjectives and their licensing. In: Perspectives on Arabic Linguistics 11 (M. Eid, ed.). John Benjamins, Amsterdam. Flaux, N. (1992). Les syntagmes nominaux du type le fils d'un paysan: reference definie ou indefinie?. Lefrancais moderne, 113-140. Flaux, N. (1993) Les syntagmes nominaux du type le fils d'un paysan: reference definie ou
(In)defmitness spread
225
indefinie?. Lefranqais moderne, 23-45. Frege, G. (1891). Funktion undBegriff, reeditedin 1969, nsFunktion, Begriff, Bedeutung, 17-39. Vandenhoeck and Rupprecht, Gottingen. Geehoven, V. van. (1996). Semantic incorporation and indefinite expressions. Doctoral dissertation, University of Tubingen. Ghomeshi, J. (1997). Non-projecting nouns and the EZAFE construction in Persian. Natural Language and Linguistic Theory 15, 729-788. Giorgi, A. and G. Longobardi (1991) The syntax of Noun Phrases. Configuration, parameters and empty categories. Cambridge University Press, Cambridge. Grosu, A. (1988). On the distribution of genitive phrases in Rumanian. Lingustics 26, 931-949. Grosu, A. (1994). Three studies in locality and Case. Routledge, London. Heim, I. (1982). The Semantics of definite and indefinite Nps. Doctoral dissertation, University of Massachusetts at Amherst. Jackendoff, R. (1974). An introduction to the X-bar convention. Manuscript, Indiana University Linguistics Club, Bloomington. Kamp, H. (1981). A theory of truth and discourse representation. In: Formal methods in the study of language (J. Groenendijk, T. Janssen, and M. Stokhof, eds.), pp. 52-71. Mathematical Centre Tracts 135, Amsterdam. Kayne, R. (1994). The antisymmetry of syntax. MIT Press, Cambridge, Mass. Kihm, A. (forthcoming). Wolof genitive constructions and the construct state. Proceedings of the 3rd Conference on Afroasiatic languages. Lsbner, S. (1985). Definites. Journal of Semantics 4, 279-326. Longobardi, G. (1994). Reference and proper names: A theory of N-movement in syntax and Logical Form. Linguistic Inquiry 25, 609-669. Longobardi, G. (1996). The syntax ofN-raising: a Minimalist theory. Manuscript, University of Trieste. ManzinijR. and A. Roussou. (1997). A Minimalist theory of 'A-movement and control. Manuscript, London University College. McNally, L. (1995). Bare plurals in Spanish are interpreted as properties. In: Proceedings of the 1995 ESSLLI Conference on Formal Grammar (G. MorrillandR. Oehrle, eds.), pp. 28-40. Miller, Ph. (1992). Clitics and constituents in Phrase Structure grammar. Garland Publishers, New York. Milner, J.C. (1982). Ordres et raisons de langue. Le Seuil, Paris. Milner, J.C. (1995). L'interpretation des genitifs. In: Langues et langage. Problemes et raisonnement en linguistique (J. Bouscaren, J.J. Franckel and S. Robert, eds.). PUF, Paris. Partee, B.H. andM. Rooth. (1983). Generalized conjunction and type ambiguity. In: Meaning, use, and interpretation of language (R. Bauerle, C. Schwarze, and A. von Stechow, eds.), pp. 52-71. Mouton de Gruyter, Berlin.
226 Comparative studies in Romanian syntax Poesio, M. (1994). Weak definites. Proceedings of the 4th conference on semantics and linguistic theory, 28-40. Prince, A. (1979). On the given/new distinction. Chicago Linguistic Society Papers 15. CSLI Publications. Ritter, E. (1987). NSO orders in Modern Hebrew. In: Proceedings ofNELS 17 (J.Mc.Donough and B. Plunkett, eds.), pp. 521-537. Department of Linguistics-MIT, Cambridge, Mass. Ritter, E. (1988). A head movement approach to construct state noun phrases. Linguistics 26, 909929. Ritter, E. (1991). Two functional categories in noun phrases: Evidence from Modem Hebrew. Syntax and Semantics 25, 37-62. Rizzi, L. (1991). Residual V-second and the \vh-criterion. Technical Reports in Formal and Computational Linguistics 2. University of Geneva. Rouveret, A. (1994). Syntaxe du gallois. Principes generaux et typologie. CNRS Editions, Paris. Siloni, T. (1994). Noun phrases andnominalization. Doctoral dissertation, University of Geneva. Siloni, T. (1997). Noun phrases andnominalization. Kluwer, Dordrecht. Siloni, T. (1998). Adjectival constructs and inalienable constructions. www.usc.edu/dept/LAS/hiiguistics/semitic/index.html Vergnaud, J.R. & M.L. Zubizarreta. The definite determiner and the inalienable constructions in French and English. Linguistic Inquiry 23, 595-652. Woisetschlaeger, E. (1983). On the question of definiteness in "an old man's book". Linguistic Inquiry 14, 137-154.
Optimal Romanian clitics
227
OPTIMAL ROMANIAN CLITICS: A CROSS-LINGUISTIC PERSPECTIVE* Geraldine Legendre, Johns Hopkins University
i. INTRODUCTION Romanian shares with other Balkan languages a very rich clitic inventory. This inventory includes nominal clitics, such as the definite article and possessive adjectives which encliticize to the first word of a noun phrase, and clausal clitics which typically procliticize to the verb. Romanian clausal clitics ~ the focus of this paper — include not only familiar pronominal elements, but also tense/aspect auxiliaries, modal particles, and even intensity adverbs (Dobrovie-Sorin 1994; Mallinson 1986; Rivero 1994). These elements are commonly referred to as "clitics" because they are phonologically dependent on a host and they display word order properties that distinguish them from their non-clitic counterparts, as discussed further below. One enduring feature of generative grammar is syntactic movement. For example, elements which by virtue of their thematic properties occupy a right periphery position in a clause may instead surface at the left periphery because they have undergone movement to a higher (leftward) position. This is the case for w/z-phrases in many languages. This has also been claimed for object clitics in Romance following Kayne (1975). While some scholars have since argued against a movement analysis of object clitics, one important assumption behind Kayne's original analysis has gone 'Earlier versions of this paper have been presented at the 1996 Workshop on Comparative Issues in Romanian Syntax held at the University of New Brunswick, Saint John, Canada; at the 1996 Going Romance conference held in Utrecht, the Netherlands; at the 1997 Linguistic Symposium on Romance Languages held at UC Irvine, and at the 1997 Hopkins Optimality Theory Workshop & University of Maryland Mayfest in Baltimore. I would like to thank audiences at these meetings for their comments, criticisms, and suggestions. I am very grateful to my consultants who patiently dealt with all my data questions: Lidia Mangu, Ciprian Chelba, Donka Farkas, and Virginia Motapanyane for Romanian; Boris Nikolov and Marina Todorova for Bulgarian and Olga Tomid for Macedonian. Thanks also to Luigi Burzio and Paul Smolensky for their comments and suggestions.
228 Comparative studies in Romanian syntax virtually unchallenged among generative syntacticians. It is the view that clitic elements are generated in the syntax and as such obey syntactic constraints. The present paper challenges this assumption, arguing instead that clitics instantiate functional features which are realized morphologically as phrasal affixes. On a par with lexical affixes — an alternative way in which functional features may be instantiated -- phrasal affixes are subject to alignment constraints which favors their realization at the edge of some domain. The empirical starting point of the analysis is the existence of a Romanian puzzle from the point of view of current standard assumptions about clitics, functional categories, and verb movement. This puzzle has four parts. First, Romanian questions do not undergo subject-auxiliary inversion (henceforth SAI): (1)
a.
b.
Ce a spus Ion? what aux-3SG said Ion 'What has Ion said?' Cu cine te -ai dus la literal? with whom refl aux-2so go to coast 'With whom did you go to the coast?'
Rather, the subject follows the auxiliary-verb sequence. SA in English (What has John said?) is standardly analyzed as I to C movement of the auxiliary has across the subject NP in SpecIP. What blocks such an analysis in Romanian (*ce a Ion spus) is unclear, apart from the fact that a clitic auxiliary is involved. It is all the more puzzling because Romanian has relatively free word order; in particular, subjects may freely occur in pre- or post-verbal position (Motapanyane 1989,1991). Secondly, examples like (la, b) also raise the issue of the number and type of landing sites for verbal elements. If the lexical verb is in C, where is the auxiliary? One could entertain the view that the subject is in SpecVP and the auxiliary in C. The question then becomes: Where is the past participle? Note that it couldn't be under Tense or Agr, given its non-finite status. Third, positive imperatives show encliticization while questions and declarative statements show proclitization. Compare (2) with (1) and (3). (2)
Lasa- ma \ leave me-IMP 'Leave me!'
Optimal Romanian clitics (3)
229
(Nu) /- am vazut. neg himaux-lSG seen 'I have (not) seen him.'
Why do positive imperatives differ from w/i-questions with respect to clitic placement if they also involve V movement to C, as proposed in Dobrovie-Sorin (1994, 1995), Rivero (1994), Rivero and Terzi (1995)? If, on the other hand, imperatives do not involve V movement, why do they differ from declarative structures like (?), where clitics precede V? Fourth and last, negative imperatives, in contrast to positive imperatives, show procliticization: (4)
Nu ma lasa! neg me leave-IMF 'Don't leave me!'
What, if anything, is common to negative imperatives, questions, and declarative statements (including negative ones)? Romanian negative imperatives do exhibit a change in morphology, from imperative hi (2) to infinitive in (4). What does this have to do with the change in clitic position? In other words, are the two properties — change of clitic position and change in morphology — related or independent of each other? I willpropose the following answers framed in Optimality Theory (OT), as in Prince and Smolensky (1993 ). One, Romanian clitic auxiliaries do not allow SA inversion in questions because these clitics do not have the status of head which is required for SA inversion; rather they are phrasal affixes instantiating functional features which might in some other language be realized as lexical affixes. Using the usual terminology, V itself moves to C with the result that overt subjects appear postverbally. Two, a consequence of removing clausal clitics from the syntax is that the question of landing sites for the auxiliary and the past participle does not even arise. Three, clitic placement differs in questions and positive imperatives for two reasons: (a) questions but not imperatives involve verb movement; (b) the position of clitic pronouns is regulated by the ranking of a set of alignment constraints which favor realizing all features at the left edge of a domain which provisionally can be assumed to be the clause. A competition for this very spot ensues among the various features, which is resolved by ranking the constraints hi a language-particular order. Hence, alignment constraints are violable. Four, Romanian negative imperatives exhibit procliticization rather than encliticization because of the constraint ranking. The difference between positive and negative imperatives lies in which constraints are fatal. Briefly, the presence of [neg], which more than any other feature needs to be at the left edge of the clause, changes the character of the competition. Finally, the morphology of Romanian negative imperatives is non-finite rather than
230 Comparative studies in Romanian syntax finite because, among other things, the top-ranking of the constraint on [neg] forces the feature [finite] of the input to be unparsed. The view defended here shares the late- and post-GB claim that word order phenomena are to a large extent grounded in morphology. One important difference lies in the present claim that the influence of morphology is not mediated by syntax but rather is direct. As we shall see, some crucial aspects of word order in individual languages result from the interaction of syntactic constraints with morphological constraints rather than from a syntacticization of inflectional morphology. The second difference is that constraint satisfaction crucially involves optimization. Its present implementation follows the basic tenets of OT. In OT, cross-linguistically, variation is predicted to be the norm since the constraints are re-rankable. In the limited scope of this paper, I will demonstrate how reranking of the constraints proposed in the Romanian analysis yields the different patterns observed in two Balkan languages, Macedonian and Bulgarian, and one Romance language, Italian. Overall, the view of cliticization defended here bears strong affinity with the view that clitics are phrasal affixes inserted post-syntactically in the morphological component of the grammar (Anderson 1992, 1993; Klavans 1985). The OT framework adopted here makes it possible however, to dispense with the standard serial view and instead focus on the interaction between the morphological properties of clitics and the syntactic properties of other elements they interact with. The rest of the paper is organized as follows. Section 2. is devoted to a comparative analysis of Romanian clitics. Section 2.1. argues for the non-syntactic status of clitic auxiliaries and informally introduces the basic idea of optimization. Section 2.2. summarizes the basic claims of OT. Section 2.3. has two foci. One is clitic clustering, a defining property of clitics which forms the core of the present proposal. The other is finiteness, which interacts with clitic placement in Balkan. Finiteness is argued to behave like a clitic; hence it naturally falls under the same analysis. Language-internal variation is discussed in section 2.4. and cross-linguistic variation in section 2.5. Section 3. focuses on verb movement in Romanian questions and the interaction between structural and morphological constraints. Section 4. focuses on Romanian imperatives; it includes a comparative study of Italian imperatives. Section 5. closes the paper with a detailed summary of the proposal.
2. A COMPARATIVE STUDY OF ROMANIAN CLITICS 2.1. Clitic vs. Non-clitic Auxiliaries In Romanian, possibility is expressed by means of the modal auxiliary aputea 'can, may' followed by an infinitival lexical verb. In declarative sentences, the subject typically precedes an inflected
Optimal Romanian clitics
231
form ofaputea: (5)
a.
b.
Ion putuse veni. Ion could-3so come 'Ion had been able to come.1 Poate Ion veni maine? can-3SG Ion come tomorrow 'Can Ion come tomorrow?'
Romanian questions follow various strategies, including a purely intonational means. One strategy of particular interest here involves SA inversion, as shown in (5b). Under standard assumptions, SA inversion is analyzed as movement of the auxiliary to C (the issue of where the auxiliary is generated need not concern us here). In this respect, aputea behaves like main verbs which also move to C in questions: (6)
a.
b.
Vine Ion? come-3so Ion 'Is John coming?' De ce atarna slanina? from what hang3SG bacon-the 'What is the bacon hanging from?'
(Mallinson 1986,10)
There are (at least) two additional distributional arguments hi favor of assigning the same syntactic status to main verbs and aputea. First, the pronominal feminine singular clitic o must precede the lexical verb in simple tenses (i. e., hi the absence of a clitic auxiliary), as shown hi (7a). O must also precede putea (7b). (7)
a.
b.
Ion o apreciaza.// *Ion apreciaza o. John her appreciates Ion appreciates her 'Ion appreciates her.' O pot vedea.//*Pot vedea-o. her can-ISG see can see her 'I can see her.'
The remaining evidence lies in the distribution of the adverb mai 'again' (or 'no longer1 when used with the negative particle nu). According to Dobrovie-Sorin (1994), Mallinson (1986), andRivero (1994), mai 'again' is one of a small class of monosyllabic VP adverbs which have the status of clitic. Evidence for its clitic status comes from the fact that the distribution of mai is different from
232 Comparative studies in Romanian syntax that of regular VP adverbs. The standard position in a simple tense for a VP adverb like des 'often' is post-verbal, as shown in (8a). Mai always precedes the main verb, as shown in (8b). As expected, mai must also precede putea, as shown in (8c). (8)
a.
b.
c.
Elevii mei vad des filme bune. students my see-3PL often movies good 'My students often see good films.' Ion il mai viziteaza. Ion him still visit-3SG 'Ion is still visiting him,' Nu mai poate scrie. neg still can-3so write 'He cannot write again.'
As independently noted in Dobrovie-Sorin (1994, 26), the distribution of mai is especially problematic for Rivero's (1994) approach. Based on the fact that mai immediately precedes the main verb in simple (and compound) tenses, Rivero proposes to base-generate mai on the lexical verb and have the complex clitic+V move to T/Agr. This account, however, cannot be extended to (8c) in which mai precedes the modal auxiliary rather than the lexical verb.1 Romanian makes use of perfect, future, and conditional auxiliaries in compound tenses. They systematically differ from a putea and main verbs with respect to the distribution of o, mai, and SAL In (9a), o follows the main verb. Many speakers consider the alternative order (o a vazui) with o preceding the perfect auxiliary ungrammatical while some scholars characterize it as archaic (de Kok 1989). Mai follows the perfect auxiliary instead of preceding it, as shown in (9b). (9b) suggests that mai is the last clitic in the Romanian cluster which includes the negative particle, clitic pronouns, auxiliaries, and mai, in that order. This is the position taken in Dobrovie-Sorin (1994):
1 In Rivero (1994) poate is generated under AuxP and moves to T/Agr. An account consistent with her assumptions might alternatively base-generate mai on Auxo in (8c), but it would not explain why mai fails to be base-generated on Auxo in (9b). Alternatively, one might incorporate PF movement of mai into the analysis, perhaps along the lines of Halpern (1995). Under Halpern's Prosodic Inversion analysis, clitics in Balkan languages such as Serbo-Croatian, Bulgarian are generated in leftmost clausal position in the syntax and moved to the right at PF the minimum distance necessary to allow them to satisfy their phonological dependency. Extending this analysis to Romanian, however, is problematic in several ways. First, there is no evidence in Romanian verbal clitics for the type of phonological dependency (Wackernagel's second position effects) that underlies the Prosodic Inversion analysis. Hence, a clitic adverb like mai would have to be moved to various degrees of distance away from its syntactic position, depending on which auxiliaries are present in the structure. Second, there is no syntactic evidence independent of the pattern to be explained for generating clitic adverbs like mai at the left edge of the clause, the left edge of AuxP, or the left edge of VP. Third, PF movement would have to be idiosyncratically restricted to a subclass of clitics. I conclude that PF movement is not a viable solution to the problem posed by the distribution of mai in the context of clitic and nonclitic auxiliaries.
Optimal Romanian clitics (9)
a.
b.
233
A vazut-o. aux-3so seen her 'He has seen her.' Nu /am maivaznt. neg him aux-ISO still seen 'I have not seen him anymore.'
Finally, these auxiliaries may not undergo SAI, as shown hi (10) for the perfect auxiliary. (10)
a.
b.
A venit copilul de la magazdn? aux-3so come child-the from the shop 'Has the child come from the shop?1 ( Mallinson 1986, 5) Ce a spus Ion? what aux-3so said Ion 'What has Ion said?'
Placing the subject between the auxiliary and the main verb leads to ungrammaticality. Note, however, that there is SAI in (10), albeit between the non-finite main verb (appearing as a past participle) and the subject. This shows that inversion is not conditioned by the mere presence of a finite verbal element. Another way of characterizing the phenomenon is to say that the clitic auxiliary is completely inactive in (10): whether it is present or absent in the structure does not affect other elements. This amounts to claiming that the clitic auxiliary is not present in the syntax. Put another way, it is not a head ~ not a node in the tree ~ but merely a phrasal affix. One approach is to stipulate the status (clitic or head) of each functional feature in the lexicon of a given language. An alternative solution more in the spirit of OT goes roughly as follows. Functional features like [perfect], [future], [conditional], [potential], etc. are listed in the lexicon2. Their status is derived from a competition among constraints on realizing features as separate syntactic heads, as lexical affixes (inflection) on existing syntactic heads, or as phrasal affixes (clitics) on existing phrases. To take an example, hi Romanian the constraint on realizing the feature [potential]as a separate head is higher ranked that constraints realizing it as affixes (lexical and phrasal). Heucsputea is realized as a node in the tree. For the feature [perfect], however, the constraint realizing it as a phrasal affix outranks the other two constraints. Hence, [perfect] is realized as a clitic. Nothing in the present analysis hinges on choosing among these alternatives, hence I take up the analysis at the point where the status of :
See Everett (1996) for an independent claim that the lexicon contains ((^features.
234 Comparative studies in Romanian syntax particular features has been determined.3 There is considerable pre-theoretical evidence for the view that clitics are phrasal affixes and need to be differentiated both from words and lexical affixes (though some cases of clitics may in fact be lexical affixes; this is the case for Romanian negative ne- 'un-' and some adverbs). To mention only a few taken from the work of Zwicky and Pullum (1983) and Zwicky (1985), lexical affixes show a high degree of selection with respect to their hosts; clitics, in particular second-position varieties, don't. Clitics can attach to material already containing clitics; lexical affixes can't. Syntactic rules can affect affixed words but not clitic groups nor clitics themselves (see also Anderson 1992, 1993; Klavans 1985; Miller 1992). Moreover, clitic pronouns do not display any of the standard structural properties of nominal categories: they do not take a Spec or any complements; they cannot be modified nor conjoined.4 In addition, clitics obey word-order restrictions not constraining their non-clitic counterparts. Aparticularly challenging class is secondposition clitics, where second-position must be prosodically defined (e.g. Serbo-Croatian). Anderson (1993) and Halpern (1995), among others, have explicitly argued that Serbo-Croatian clitics cannot be handled in the syntax. To return to Romanian, the basic claim here is that the auxiliary/wtea heads a syntactic projection while perfect, future, and conditional auxiliaries do not. SAI requires a verbal head (in C). In Romanian, auxiliaries other thanputea are not syntactic heads. Hence, they do not trigger SAI. The specific morphological proposal made here is that phrasal affixes, on a par with lexical affixes, are subject to alignment constraints, a claim previously made in Anderson (1996) and Legendre (1996). Applied to features, alignment constraints like EDGEMOST (Prince and Smolensky 1993) favor aligning their phonological realization at the (left) edge of a particular domain. (11) states that the domain is the extended V projection of V. This will be motivated in the course of the analysis. (11)
3
EDGEMosi(F, LEFT)=E(F): The PF realization of a feature [F] is left-aligned with the edge
Under the traditional view of morphology and phonology as post-syntactic components, clitics would be morpho phonological Spell Outs of the functional features contained in the lexicon or phrasal affixes (and hence differ from word-level affixes merely in terms of the domain of affixation). Such a proposal is in fact worked out in Anderson (1992) and Klavans (1985). From the perspective of OT, optimization is global (i.e., across components of the grammar) rather than serial. Thus, the present account will focus on the interaction between morphological constraints on clitics and syntactic constraints on syntactic elements they interact with. 4 The view that clitics occupy syntactic positions historically derives from the fact that in some Romance languages, clitic pronouns satisfy Case properties (Kayne, 1975). As is well known, this view is not without problems. Several Balkan languages, including Romanian and Macedonian, require clitic doubling in some contexts, casting doubts on a simple Case connection. In the context of an analysis of French Complex Inversion, Legendre (1998a) provides a preliminary account of Case that incorporates Case visibility, Case agreement, and Case economy.
Optimal Romanian clitics
235
of the extended V projection of the head [F] is associated with. Such edge-alignment constraints are entirely familiar from the OT morpho-phonology literature. They are created by a generalized constraint schema called ALlGN(Categoryj, Edge,; Category^ Edge2) (McCarthy and Prince 1993a,b). Alignment constraints have recently been extended to Focus constructions (Costa, forthcoming, Legendre 1998a, Samek-Lodovici 1996, 1998). Their relevance to clitics has been amply demonstrated in Klavans (1985), albeit in a completely different framework. As is well known, one trait shared by Balkan languages is the existence of clitic auxiliaries. In this respect, it is interesting to note that their position in the clause is subject to variation. The data in (12) show a minimal contrast between Romanian and Bulgarian, two null-subject languages which make use of clitic auxiliaries in compound tenses. (12)
a. R
Am
plecat.
aux-lsoleft 'I have left.' b. B
Procelsinn read
knigata.
aux-1 so book-the
'I have read the book.' In the absence of any other clitics, the perfect auxiliary is clause-initial in Romanian but in second position in Bulgarian. At first glance, EDGEMOST(PERF) seems simply to be satisfied in Romanian but violated in Bulgarian. It is not that simple, however, because some clitic auxiliaries do appear clause-initially in Bulgarian. This is the case for the future auxiliary ste: (13)
a. R
Va
mai vedea-o.
will-1SG again see
her
'He will see her again.' b. B
Ste mu
go dadete.
will him
it
gave-2so
'You will give it to him,' On the one hand, (13b) shows that Bulgarian is not a strict Wackernagel language. On the other, it shows the effect of EDGEMOST(FUT), completely parallel to its role in Romanian. We can make sense of the contrast between Romanian and Bulgarian in (12) and the contrast within Bulgarian between (12b) and (13b) hi exactly the same terms. First, the position of auxiliary clitics derives from the interaction of two constraints rather than the effect of a single constraint like
236 Comparative studies in Romanian syntax EDGEMOST(F). The second constraint, NoNlNiTlAL(F), has a long history in Romance linguistics where it is known as the Tobler-Mussafia Law. In our terms, it is simply a constraint which disfavors clitics in intonational phrase-initial position. Cross-linguistic evidence that the domain is prosodic rather clausal or phrasal is discussed in detail in Legendre (in press a): (14)
NoNlNlTlAL(F): [F] is not realized in intonational phrase-initial position.
Second, consider how the interaction of constraints (11) and (14) results in the pattern in (12). Note that the two constraints can never be satisfied at once. If a structure violates NONINITIAL(F), EDGEMOST(PERF) will be satisfied; this is the case in Romanian. Conversely, if a structure violates EDGEMOST(PERF), NONINITIAL(F) will be satisfied; this is the case in Bulgarian. In other words, (12a) and (12b) can be simply derived from alternative rankings of two constraints on phrasal affixes: EDGEMOST(PERF) » NON!NITIAL(F) in Romanian versus NON!NITIAL(F) » EDGEMOST(PERF) in Bulgarian. In other words, constraints are violable and hierarchically ranked — the most fundamental claim made by OT. The data in (13) involve a different functional feature: [future]. In both languages, the feature is realized as a phrasal affix in domain-initial position. Hence, EDGEMOST(FUT) must outrank NONlNITIAL(F):
(15) Preliminary Rankings: a. Romanian: EDGEMosr(Fur),EDGEMOCT(pERF)>>NONiNrriAL(F) b. Bulgarian:
EDGEMOST(FUT)»NONINrnAL(F)»EDGEMOST(PERF)
The symbol» signals strict ranking. The use of a comma, as in the Romanian ranking, signals that the two EDGEMOST constraints are unranked with respect to each other. The future and the perfect auxiliary do not co-occur, hence there is no evidence for their relative ranking. In summary, I have argued that Romanian clitic auxiliaries differ from non-clitic auxiliaries in that the former do not head syntactic projections. As phrasal affixes, their insertion into a clause is regulated by a hierarchy of alignment constraints which compete for the left clausal edge. This renders them syntactically inactive; hence they do not participate in SA inversion. Non-clitic auxiliaries head syntactic projections of then- own, and as such, do participate in SA inversion. Romanian clitics appear in clause-initial position while many of their Bulgarian counterparts show Wackernagel effects. This follows from alternative rankings of two types of PF alignment constraints, EDGEMOST(F) and NONlNiTlAL(F).
Optimal Romanian clitics
237
2.2. Basic Claims of Optimality Theory The analysis of two clitic auxiliary patterns in Romanian and Bulgarian in section 2.1. illustrates some basic claims of OT: (i) Constraints are universal and violable in well-formed structures. This is possible because constraints are ranked with respect to one another. A lower-ranked constraint may be violated so that a higher-ranked one may be satisfied. One further aspect will be demonstrated below: a given constraint violated by a grammatical sentence in one context may be fatal to an ungrammatical one in another because this constraint may interact with constraints that are relevant in one context but not in another. (ii) The optimal candidate (and only the optimal candidate) is grammatical. (iii) The relative ranking of constraints is determined on a language-particular basis. Thus a grammar is a particular ranking of universal constraints. Constraint reranking and violability account for cross-linguistic variation: a universal constraint may be violated in one language by virtue of being low-ranked and not violated in another by virtue of being high-ranked (exactly the case of Romanian versus Bulgarian above). (iv) Markedness is inherent to the model: all constraints other than faithfulness constraints (which require that the output be maximally faithful to the input) are markedness statements.5 For example, the markedness constraints that play a central role in the distribution of clitics include alignment constraints which form a family of constraints, EDGEMOST(F), where [F] stands for any functional feature. Specific aspects of markedness result from constraint ranking. Other markedness constraints will be introduced in the course of the analysis. (v) The candidate set is universal. In practice, however, it is convenient to limit the evaluation to a subset of the universal set or the "best of the lot".6 The candidate set consists of alternative structural descriptions of the input. For our purposes, we may assume that the input consists of lexical items, argument structure, and functional features.
3
Faithfulness constraints are unique to OT. They include PARSE (all input material should be present in the output) and FILL (all output material should be present in the input). Legendre et al. (1995,1998) document cross-linguistic effects of these two constraints in the context of w/j-movement. PARSE also plays a role in the present analysis of imperatives (see section 4). 6 The present candidate sets are clearly a subset of the universal set because they satisfy some basic principles of X'theory such as the presence of a head position in each projection and XPs restricted to complement or Spec position. The present analysis assumes that X'-theory principles are in Gen; hence, they are satisfied by all candidates. Ultimately, an OT theory of X' principles must be added.
238 Comparative studies in Romanian syntax
2.3. Clustering and Finiteness Effects Universally, clitics cluster. That is, they typically appear together and the cluster-internal order is fixed. This clustering behavior is one important property clitics share with lexical affixes. One way in which languages differ is with respect to the cluster-internal order of clitics. In Romanian, clitic pronouns precede tense/mood/aspect auxiliaries, as shown in (16a). In Bulgarian, it is, roughly speaking, just the reverse: tense/mood/aspect auxiliaries precede clitic pronouns. (16)
a. R
b. B
L- am vazut. him aux-lso seen 'I have seen him1, Dal sum mu go. given aux-lSG him it 'I have given it to him.'
Thus, languages differ not only in terms of the respective ranking of a particular EDGEMOST(F) constraint with NonInitial(F), as illustrated in section 2.1., but also with respect to ranking within the EDGEMOST(F) family of constraints: EDGEMOST(ACC) outranks EDGEMOST(PERF) in Romanian; Bulgarian has the reverse order. Note however, that the relative order of dative and accusative clitics is shared by Romanian and Bulgarian, as shown in (17). One important way in which this simple picture of Balkan languages hides greater complexity is with respect to finiteness. To fully appreciate the effect of finiteness, it is necessary to bring Macedonian into the picture:
(17)
Mi- o da. me it-FEM give-3so 'He gives it to me.' b. M Ti go dade. you it gave-3so 'She gave it to you.' c. B Pokazax mu go pismoto. (Tomi<51996) showed-1 so him it letter-the 'I showed him the letter.' a. R
All the examples in (17) involve finite verbs. Despite the fact that Macedonian is genetically a South- Slavic language, its basic clitic distribution patterns like that ofRomanian: EDGEMOST(DAT)
Optimal Romanian clitics
239
and EDGEMOST(ACC) outrank NONlNiTiAL(F) with the result that clitic pronouns appear clauseinitially. When non-finite verbs like gerunds are taken into consideration, the pattern changes for Romanian and Macedonian: clitic pronouns followthe gerund in clause-initialposition. The pattern remains the same in Bulgarian: (18)
a. R b. M
c. B
Vazindu-/... 'Seeing him...' Davajki/MM go... giving him it 'Giving it to him...' Davaikimw go... giving him it 'Giving it to him...'
Finite verbs may in fact appear clause-initially in all three languages, but only in the absence of clitic pronouns in Romanian and Macedonian (see (17c) for Bulgarian): (19)
a. R
b. M
Poate scrie. can-3so write 'He can write.' Barav edna marka. looked-ISO one stamp 'I was looking for a stamp.'
An important generalization emerges from(17-19): bare verbsmay appear in clause-initialposition, regardless of finiteness. However, the presence of clitic pronouns affects the placement of finite verbs, at least in Romanian and Macedonian. This suggests that [Tense/finiteness] is an active feature, on a par with features like [accusative] and [dative], and that alignment constraints should be extended to [Tense/finiteness] or [T] (to avoid confusion with [F], which stands for any other functional features). Note that I am abstracting away from the actual complexity of finiteness (subsuming [Tense], [person], and [number]) which does not affect the point to be made here. An important source of independent evidence for the existence of constraint NoNlNlTlAL(T) is provided by Germanic V2 languages. The striking property of these languages is well known: finite verbs appear in second position in declarative main clauses. Without going into the details necessary to determine the proper ranking ofNONlNiTlAL(T) in these languages, it can be observed that the Germanic V2 phenomenon can be assimilated to the Balkan finiteness distinction with respect to clitic placement despite the fact that NONlNlTlAL(T) is independent of clitics in Germanic
240 Comparative studies in Romanian syntax (see Anderson 1993 for farther discussion). This is possible because clitics, like finiteness, are analyzed as morphological affixes. In some V2 languages, such as Walpiri, finiteness features are in fact realized as a separate clitic auxiliary verb in second position, hi other languages, including Balkan, finiteness features are realized on verbs themselves (unless a clitic realizing a separate feature like [perfect] is also present).
2.4. Language-internal Variation Returning to Romanian finite structures for the moment, we now face the task of determining the Romanian constraint ranking responsible for procliticization to finite verbs. The OT tableau format is used to display the formal competition among alignment constraints. Table 1. Romanian procliticization to finite verbs I: [dat] [ace] [T]
E(dat)
«• a. [v. mi-o da] b. [v, mi da o]
Nln(T)
E(acc)
E(T)
®
®®
®
**i
*
*
Nln(F)
c. [v. da mi-o ]
*!
**
*
*
d. [v, da o mi]
**j
*
*
*
e. [v, o mi da]
*!
**
*
f [v, o da mi]
**!
*
0
Table 1 (Tl) illustrates the graphic conventions of OT. The grammatical output is marked optimal («•). Constraint ranking is indicated by leftmost constraints outranking rightmost ones. Violations of constraints are recorded as * in individual cells; *! are fatal violations for sub-optimal candidates while ® are violations incurred by optimal candidates. To avoid cluttering, it is convenient to omit constraints which are satisfied by ah* candidates and to limit the input specification to relevant functional features.. Because OT defines a grammar as an optimization system, any aspect of grammar may, in principle, be determined by optimization. An important consequence exploited to the fullest in the present analysis is that phrase structure itself is optimized: it is built from the properties of the input (Prince and Smolensky 1993) and any structure not required by the lexical items of the input can be interpreted as violating * STRUCTURE:
Optimal Romanian clitics (20)
241
*STRUCTURE (*STRUC): Avoid structure (Prince and Smolensky 1993; Legendre 1996)
Assuming the basic X'-Theory schema, the smallest phrasal projection for V is V, the constituent which includes V and its complement(s). The next phrasal projection is VP, the constituent which includes V and its Spec. Under this Minimalist view of structure, V is the minimal clausal unit (as in Romanian ploua '(it) ranis'). Following Samek-Lodovici (1996) and Grimshaw and SamekLodovici (1998), I take referential (and expletive) null subjects to be structurally unrealized. Thus, simple null subject clauses correspond to V (if an object is present) while their counterparts in nonnull subject languages are VPs. This is not independently stipulated for each language but rather is a consequence of optimization. For a given input, candidate structures of varying structural complexity are evaluated. To anticipate the discussion of verb movement in section 3., * STRUCTURE outranks alignment constraints. Hence, any candidate containing additional projections fares worse. These sub-optimal candidates are omitted in all tableaux, except where the competition directly pertains to * STRUCTURE. In the first phase of the analysis, we consider only candidates which include a single projection. Tl illustrates two ways in which candidate a wins the competition. One, the highest ranked constraint violated by the optimal candidate a. is outranked by a constraint violated by its competitors; this is the case for candidates c.-f, from which we derive that EDGEMOST(DAT) » EDGEMOST(ACC). Two, given that closeness to an edge is a matter of degree, EDGEMOST constraints are gradient: multiple violations of a single EDGEMOST constraint occur as a given feature [F] is realized further away from the relevant edge; degree of violation is measured in terms of the free morphemes which separate a given clitic from the relevant edge, though nothing depends on this particular way of evaluating gradiency. Note that gradiency is fatal to candidate b.: unlike a., b. violates EDGEMOST(ACC) twice. In fact, the competition between candidates a. and b. yields a second partial constraint ranking: EDGEMOST(ACC) » EDGEMOST(T). This is because candidates a. and b. violate these two constraints but to varying degree: a. violates EDGEMOST(ACC) once and EDGEMOST(T) twice; b. violates EDGEMOST(T) once and EDGEMOST(ACC) twice. For a. to be optimal, its worse violation has to be lower-ranked than b.'s worse violation. This is the case if EDGEMOST(ACC) outranks EDGEMOST(T). The three lexical items in the candidate structures each instantiate one functional feature; as a result, they all violate non-gradient NONINITIAL(F); these violations cancel out, providing no information about the relative ranking of NONlMTiAL(F). Note that NONlNiTlAL(T)'s rank is indeterminate as well in Tl. To recover its relative ranking, it is necessary to consider a structure in which it is surface-violated. This is the case in present tense structures such as (21). (21)
Citesc carji. read-lSG books
242 Comparative studies in Romanian syntax 'I read books.' The finite verb is clause-initial, hi violation ofNONlNlTTAL(T). However, it satisfies EDGEMOST(T). This indicates that EDGEMOST(T) » NONINITIAL(T). This partial ranking is incorporated into Table 2 (T2), which underlies (22a). (22)
a.
b.
Am plecat. have-lSG left 'I have left.1 Nu I- am vazut. neg him-have-lSG seen 'I have not seen him.' Table 2. Romanian clitic auxiliaries
I: [pert] [T]
E(perf)
E(T)
*!
*
a. w [v,a/M plecat] b.
[y- plecat am]
NIn(T)
Nln(F)
®
®
In T2, a single clitic combines two features, [finite] and [perfect] — the phrasal counterpart of lexical portmanteaux. Corresponding alignment constraints are either satisfied or violated at the same tune; hence the two EDGEMOST constraints are unranked with respect to each other, and so are the two NoNlNiTlAL constraints. One possible ranking, based on the fact that candidate b. loses to a., is that both EDGEMOST constraints (violated by candidate b.) outrank both NoNlNiTlAL constraints (violated by candidate a.). In addition, (22b) shows that accusative clitics precede finite perfect clitics and finite verbs. Consequently, EDGEMOST(ACC) » EDGEMOST(PERF) and EDGEMOST(T). From the present perspective on inflectional morphology, the existence of gerund morphology (Romanian ind/ind) is taken to reflect the existence of a feature which we may refer to as [gerund]. Because gerunds are non-finite, constraints on [T] are irrelevant. The optimal structure is one in which the non-finite verbal form precedes the clitic pronouns; hence EDGEMOST(GER) » EDGEMOST(ACC), as shown in T3.
Optimal Romanian clitics
243
Table 3. Romanian gerunds I: [ger] [ace]
E(ger)
•* a. [v, vaziudu-/] b. [v, il vazind]
*!
E(acc)
Nln(F)
®
® *
The basic R ranking is summarized in (23). (23)
Romanian ranking: E(GER) » E(DAT) » E(ACC) » E(PERF) »(E(T) »NlN(T)» NIN(F)
Returning to non-clitic auxiliary structures, one aspect of the distribution of clitics remains to be addressed. It is the fact that clitic pronouns procliticize to the auxiliary rather than the lexical verb. The pattern is given in (24). (24)
// pot vedea. himcan-lSG see 'I can see him.'
This pattern ought to shed light on the domain of EDGEMOST constraints. Each verbal element determines its own V. If the domain of EDGEMOST(ACC) is verbal (V), then EDGEMOST(ACC) is satisfied by each candidate: the clitic is located at the left edge of a (different) V in each candidate. This leads to the wrong outcome because candidate a. (which corresponds to (24)) fares worse than b.: it violates EDGEMOST(T) which is satisfied by b. On the other hand, if the domain of EDGEMOST(ACC) is clausal, then EDGEMOST(ACC) is satisfied by optimal candidate a. and violated by sub-optimal candidate b. The problem with this characterization is that a clausal domain predicts that a clitic will precede overt subjects in Romanian, which is incorrect (see, for example, 8b). What we need is a domain which overlaps with the clausal domain in/vo-drop contexts, is strictly verbal in the presence of an overt subject, and includes what Grimshaw (1991) calls extended projections of V. This unique domain can be characterized as the extended V projection of the verbal head:
244 Comparative studies in Romanian syntax Table 4. Romanian non-clitic auxiliaries I: [ace] [T]
E(acc)
Nln(T)
®
•s- a. [v, il pot [v, vedea]] b. [vpot [v, il vedea]]
E(T)
*!
Nln(F) ®
*
0
A fundamental claim of OT is that constraints are violable. We may verify the validity of the claim, based on the few Romanian patterns discussed so far. Out of the eight constraints listed in (23), six may be violated by optimal candidates (®)7. The only constraints not violated are EDGEMOST(GER) and EDGEMOST(NEG) . Four constraints are active in the sense that their violation is fatal to some candidate (*!): EDGEMOST(GER), EDGEMOST(DAT), EDGEMOST(ACC), and EDGEMOST(PERF)). A given constraint may be fatal to one candidate, for example, EDGEMOST(ACC) in T4. The same constraint may also be violated by an optimal candidate, in Tl, because the two candidates in question belong to different candidate sets. Thus, competition is local in the sense that it is determined by a particular input. The adverb mai 'again' displays idiosyncratic behavior.8 Recall that mat is generally considered to be a clitic for two reasons. One, its pre-verbal position differs from that of standard adverbs, which are post-verbal. Two, like pronominal clitics, it precedes lexical verbs in simple and compound tenses (hence it follows clitic auxiliaries) while it precedes non-clitic auxiliaries. The relevant data is given in (8) and (9). Its distribution hi imperatives differs, however, from that of pronominal clitics. Mai precedes positive imperative verbs while pronominal clitics follow it (Dobrovie-Sorin 1994):
(25)
7
Mai spune- mi\ again tell-lMP me
Only five are actually violated in the Romanian tableaux. The sixth one, EDGEMOST(PERF) is violated by optimal candidates whenever a pronominal clitic co-occurs with a perfect auxiliary, as in (22b). Note that (22b) also shows that EDGEMOST(NEG) is unviolated. 8 A final complication in the distribution of Romanian clitics concerns the feminine singular accusative clitic pronoun o. In some contexts, o has the same distribution as its non-feminine counterparts. This is the case in simple tenses and non-clitic auxiliary structures (where it shows up as a proclitic) and in gerunds (where it shows up as an enclitic): (i) O vede '(I) see her'. With conditional and perfect auxiliaries, o, unlike its non-feminine counterparts, encliticizes to the non-finite verb: (ii) Afvazut-o'tyv/ouldseeher1. As noted in deKok( 1989), Romanian dative and accusative clitics typically undergo desyllabification processes including vowel elision (imi > mi-; ma> m-; ii > i-; il > /-) and diphthongization (i-am dat [Jam]) before a following vowel. This reduction process appears to be general in the context of proclitics. It looks like Romanian imposes severe restrictions on the number of non-stressed syllables which can precede a prosodic (verbal) head. Regardless of what the exact nature of the phonologiccal constraints may turn out to be, it appears that Romanian economizes on phonological material as long as morphemes conveying important featural information are recoverable. O is the only Romanian clitic pronoun which consists of a single vowel. This means that it cannot undergo phonological reduction without loss of corresponding featural information.
Optimal Romanian clitics
245
'Tell me again!' The generalization is clear: mai always immediately precedes the verb. This strict distribution suggest a prefix rather than a clitic status. In other words, mai behaves like French re- 'again' as in relire 'read again'.9 Together, Italian and Romanian point to the unreliability of the orthographic conventions: Italian lo and Romanian enclitics are clitics despite the fact that they are graphically attached to the verb while Romanian mai is a prefix despite the fact that it is not graphically attached to the verb. As a lexical affix, mai is subject to an EDGEMOST constraint whose domain is lexical rather than phrasal: mai must be aligned with the left edge of the head it is associated with.
2.5. Cross-linguistic Variation In OT, all cross-linguistic variation results from reranking of universal constraints. The theory predicts both minimal and wide-ranging rerankings. To begin with, consider a minimal change. Romanian finite clitic auxiliary clitics are allowed in clause-initialposition (T2) because EDGEMOST constraints, in particular EDGEMOST(PEKF) and EDGEMOST(T), outrank NONlNlTlAL(T). If, however, the ranking is reversed to NoNlNlTlAL(T) » EDGEMOST constraints, thenfiniteauxiliary clitics may not appear in clause-initial position. The corresponding competition is sketched out in T5. Table 5. Reranking prediction I: [perfj [T]
NIn(T)
•s a. [v,past participle aux[T, perfj] b. [v, auxjT, perfj past participle ]
*!
E(perf)
E(T)
®
®
NIn(F)
0
The resulting language differs minimally from Romanian. This is because the change will not affect gerunds for which [T] is irrelevant. Crucially, it will not affect finite verbs either. NoNlNiTiAL(T) is not violated by any optimal structure in which clitic pronouns precede the finite verb. As can be
9
A reviewer points out that mai doesn't show a high degree of selection with respect to its host. In particular, mai may modify adjectives or adverbs, in addition to verbs: (i) copilul mai mare 'the older child' (ii) vreau savinamai cunnd 'I want him to come sooner1. A similar observation can be made in French. The lexical affix re- can modify verbs, nouns, and adjectives: (i) reelire 'reeled' (ii) reelection 'reelection' (iii) reeligible 'reelectable'. This suggests that a high degree of host selection is not a reliable criterion for lexical affixes, contrary to the claim made in Zwicky and Pullum (1983) and Zwicky (1985).
246 Comparative studies in Romanian syntax verified in tableau Tl, placing NoNlNlTlAL(T) at the top of the hierarchy will not affect the outcome of optimization. Such a language exists within the Balkan group: Macedonian. As discussed in Tomic (1996), Macedonian clitics follow non-finite verbal forms. A Macedonian sentence corresponding to the optimal structure in T5 is given in (26a). See Legendre (in press b, c) for a fuller analysis of Macedonian: (26)
a. M
b. M
Dojden e. come have-3SG 'He has come.' Barav edna marka. looked-ISG one stamp 'I was looking for a stamp'
In the absence of any clitics, a violation of NONlNiTlAL(T) will not prevent a finite verb in Macedonian from appearing in clause-initial position, as (26b) reveals. This is because the alternative order ~ edna marka barav — involves adding structure to provide a landing site for the moved NP. This candidate violates *STRUCTURE and *t, as shown in T6. Independent evidence for economy constraints outranking alignment constraints comes from questions and imperative structures to be discussed in sections 3 and 4, respectively: Table 6. Macedonian finite verbs
I:[T]
*STRUC
*t
E(T)
®
•*• a. [v. barav edna marka] b. [xp edna marka [v, barav t ]]
NIN(T)
*!
*
The reverse ranking with the more general constraint outranking the more specific one — NoNlNiTiAL(F) » NoNlNiTiAL(T) --, on the other hand, predicts effects of NoNlNlTlAL(T) to be invisible. This is the case in Bulgarian, as repeated in (27): verbs appear in clause-initial position, whether they are finite or not. The consequence is the well-known Wackernagel effect: clitics cluster in second-position. See Legendre (in press a,b) for details:
(27)
a. B
Dal sum mu go. given aux-1 SO him it 'I have given it to him.'
Optimal Romanian clitics b. B
247
Pokazax mu go pismoto. showed-ISG him it letter-the 'I showed him the letter.'
The corresponding competitions are displayed in T7 and T8, respectively. Table 7. Bulgarian non-finite verbs I: [perf] [dat] [ace] [T]
NIN(F)
a. •*• [v- dal sum mu go] b. [v, sum mu go dal ]
®
*! ***!
c. [v.dal mu go sum]
d. [v . mu go sum dal]
*!
E(T)
E(acc)
E(PERF) E(dat)
®®
®®®
*
##
*
**
NIN(T)
® * ***
*
**
Table 8. Bulgarian finite verbs
I: [T] [dat] [ace]
NIN(F)
®
a. •*" [v. pokazax mu go pismoto] b.
[v, mu go pokazax pismoto]
E(dat)
*!
E(acc)
E(T)
®® *
NIN(T)
® **
Under the present analysis, languages with Wackernagel effects (Bulgarian) and languages without (Romanian) are subject to the same constraints but to different constraint rankings. Thus, Wackernagel effects are simply the consequence of the way the competition plays out. There is no need for any special mechanism such as Long Head Movement (Rivero 1994). Partial rankings for Macedonian and Bulgarian are stated in (28a,b). See Legendre (1998b) for details. For comparative purposes, the Romanian ranking is repeated in (28c). (28)
a. Bulgarian ranking: NONlN(F) » E(PERF) » E(DAT) » E(ACC) » E(T) »NONlN(T) b. Macedonian ranking: NONlN(T), E(PERF) » NONlN(F) » E(DAT) » E(ACC) » E(T) c. Romanian ranking: E(DAT) » E(ACC) » E(PERF) »E(T)» NoNlN(T) »NoNlN(F)
Under these rankings, Bulgarian and Romanian are maximally different, due to reverse rankings of NONINITIAL(F) and EDGEMOST(F) constraints. NONlNlTIAL(F) » EDGEMOST(F) induces Wackernagel effects while EDGEMOST(F) » NONlMTlAL(F) induces clause-initial clitics.
248 Comparative studies in Romanian syntax Macedonian lies in between: it shares the subranking EDGEMOST(PERF) » NONlNiTiAL(F) with Romanian and EDGEMOST(F) outranked by a NONlNlTlAL constraint with Bulgarian. Preliminary observations indicate that further re-rankings can be found hi languages of the world. In Balkan, as we have seen, dative precede accusative clitics. In French, it is the reverse order hi the third person. Lummi (Salish) places accusative before nominative clitics while Walpiri exhibits the reverse order. In Bulgarian, the future clitic precedes the perfect which, hi turn, precedes clitic pronouns. In Serbo-Croatian, dative and accusative clitics precede reflexives; in Slovene and (West Slavic) Czech, reflexives precede other clitic pronouns. The variation within the South-Slavic family alone is remarkable. As demonstrated above, the relative ranking of EDGEMOST and NoNlNiTiAL(F) differentiates second position clitic languages from other languages. Second position clitics are found hi many unrelated languages (including Walpiri, Luiseno, Pashto, etc.) and they are not correlated with any particular internal cluster order of clitics. To summarize section 2., I have argued that an OT analysis of the clustering properties of clitics hi terms of universal but violable alignment and minimal structure constraints is all that is needed to handle the cross-linguistically variable cluster-internal order and, to a large extent, their variable position hi the clause. I have also argued that finiteness, its effects on clitic placement in Balkan as well as the Germanic V2 phenomenon, are best understood by assimilating finiteness or [T] to other affixes subject to alignment constraints. This proposal will receive important confirmation from the analysis of imperatives (section 4).
3. VERB MOVEMENT IN ROMANIAN QUESTIONS The focus of section 2. has been the distribution of clitics in finite and non-finite declarative structures. Additional evidence for the validity of the alignment-based analysis rests on investigating its consequences with respect to more complex structures which may involve verb movement. This entails examining the other parts of the Romanian puzzle. We consider questions first (repeated from (10)) and then turn to imperatives: (29)
a.
b.
Ce a spuslon? what has said Ion 'What did Ion say?' A venit copilul de la magazin? has come child-the from at shop
Optimal Romanian clitics
249
'Did the child come back from the shop?' Following standard assumptions, the w/z-phrase ce present in the input to (32a) is located in the Spec position of the highest projection, forming a chain with a coindexed trace in object position. A distinct projection is needed for ce because the VP internal subject hypothesis is adopted here: the subject Ion fills the Spec of the lower VP. Following Grimshaw (1991), all extended projections of V are considered VPs: hence the w/z-phrase is shown to be in the highest VP in subsequent tableaux. Romanian shows overt w/z-movement as well as evidence that the verb is in the head position of the projection which includes the w/z-phrase in Spec position. This is because the overt subject follows the lexical verb. Legendre et al. (1995,1998) argue that languages with overt w/z-movement are languages hi which it is more important for w/z-phrases to be hi a scopal position than to satisfy economy of movement. These languages may be accounted for by ranking * STRUCTURE and *t (Avoid movement traces, Legendre et al, 1995, 1998) below the constraint that has the effect of placing Wz-phrases in the highest Spec position (languages with Wz-phrases in situ have the reverse ranking). Grimshaw (1997a) argues that SAI hi English results from the interaction of OPSPEC and OeHD: (30)
OPERATOR IN SPECIFIER = OPSPEC: Ops are in Spec position. (Grimshaw, 1997a) OBLIGATORY HEADS = OsHD: A projection has a head. (Grimshaw, 1997a)
OPSPEC requires the presence of a projection with an empty Spec into which object w/z-phrases move. The Spec position may, alternatively, be occupied by a yes/no Op. This projection has an empty head position which in turn must be filled to satisfy OeHD. The resulting competition can be first observed in yes-no questions with a non-clitic auxiliary: (31)
a. b.
Poate Ion veni mune? Can John come tomorrow?
In (3 la), the subject NP follows the auxiliary just as its English counterpart hi (31b). Non-clitic auxiliaries head their own VP (and select for a VP complement). The presence of a lexical verb and a non-clitic auxiliary hi the input leads to candidate structures minimally consisting of two VPs (candidate a). The higher VP provides a Spec slot for the null illocutionary Op (Op hi T9). Hence, the modal auxiliary may appear hi the highest empty head position without requiring additional structure. All candidates hi T9 satisfy OPSPEC; hence it is omitted.
250 Comparative studies in Romanian syntax Table 9. Romanian yes-no questions with a non-clitic auxiliary
I:[T]
ObHd *Struc
*t
a.«s" [vp Op poate [w Ion veni]] b. [w Op poate [yp [v. veni [w Ion t]]] c. [w Op [w Ion [v, poate [w t veni]]]
*!
*!
*
*
*
E(T)
Nln(T) Nln(F) ®
®
*
*
The minimal structure candidate a competes with candidates b. and c. which exhibit alternative word orders necessitating additional structure. The latter, of course, violate * STRUCTURE. Note that, as with all constraints in OT, computation with *STRUCTURE violations is merely relative (across candidates). Candidates a. and c. violate disjoint sets of constraints; thus, one of the structural constraints violated by c. must outrank all the alignment constraints violated by a. for a. to be optimal. We already know that EDGEMOST(T) » NONlNITlAL(T) in Romanian. Based on T9, the ranking of the structural constraints is thus indeterminate. The one given in T9 is only one possible ranking, but it is partly confirmed below in T10 and partly later in the section devoted to imperatives. Note that the analysis of (3 la) makes a prediction: the presence of a clitic auxiliary will not change the inversion pattern (V-subject) because (a) the inversion pattern is determined by the ranking of a syntactic constraint (OBHD) and (b) clitics are not in the syntax. As proposed earlier, the placement of clitic auxiliaries is partly determined by EDGEMOST(F), which favors them at the left edge of the clause. The competition resulting in the grammaticality of (29a) is displayed in T10. A similar analysis applies to (29b):
251
Optimal Romanian clitics Table 10. Romanian \vh-questions with a clitic auxiliary I: [perf] [T]
OPSPEC
OBHD *STRUC *t
E(PERF)
E(T)
**|
**
®
®®
*
*
d. [w ce [v, spus [w Ion [v a tt]]]]
*
**
e. [w ce [v spus a [vp Ion [v, 1 1]]]]
*
** *! *
«"a. [w ce [v a spus [vp Ion [v-t t]]]] b.
[w Ion [v a spus ce]]
c. [w ce [vp Ion [v a spus t ]]]
f . [VP ce [v. a [vp Ion [v, spus t ]]]
*!
*!
*!
*
*
Candidates a. and b. illustrate a competition between movement and in-situ w/z-strategies. Candidate b. loses because of the high-ranking of OPSPEC in Romanian. The high-ranking of OPSPEC, in turn, requires an additional projection whose Spec is the landing site for moved whphrases like ce. This leads to violations of * STRUCTURE for all candidates except the in-situ candidate structure b. The additional projection needed to house ce, in turn, forces verb movement to satisfy OBHD. If such movement fails to occur, the result is sub-optimal (candidate c. and f). XP and X movement incur *t violations which cancel out for candidates a, d, and e. TIO illustrates the fact that lower-ranked constraints determine the outcome of the competition whenever violations of higher-ranked constraints cancel out. This is the case for candidates d. and e. which are eliminated because of EDGEMOST(PERF) violations. For example, in candidate d., [perf] is not realized at the left edge of the extended V projection of the verbal head, that is, the highest V. TIO yields the following partial ranking: OPSPEC, OBHD » * STRUCTURE. One question TIO raises is why, given the high ranking of E(PERF), the clitic auxiliary does not surface preceding the w/z-phrase ce rather than the past participle spus, thus satisfying EDGEMOST(PERF) and EDGEMOST(T) perfectly. I believe that the answer lies hi prosodic constraints which ensure that clitics are prosodically parsed with the verbal head. Such prosodic constraints are clearly at work hi Bulgarian and Macedonian, as demonstrated hi Legendre (hi press a,b). A better understanding of the prosodic properties of Romanian is needed however before a fuller account can be provided. To summarize, Romanian does have verb movement, but only as a consequence of w/z-movement or its analog myes-no questions. The pattern of inversion in Romanian questions is the same hi the presence or absence of a ch'tic auxiliary while it differs in the presence of a non-clitic auxiliary. Under the present analysis, this follows from the claim that non-clitic auxiliaries head syntactic projections of their own but clitic auxiliaries do not. The pattern of Diversion is to a large extent
252 Comparative studies in Romanian syntax determined by a set of three conflicting structural constraints: OeHo vs. *STRUCTURE and *t. OsHD outranks both *STRUCTURE and *t; hence Romanian favors verb movement with the result that the verb precedes the subject.10 The relative ranking of morphological alignment constraints previously established on the basis of declarative sentences ensures that phrasal affixes appear within the extended V projection of V in questions as well. Thus, there are two distinct reasons to front elements: OBHD in syntax, EDGEMOST(F) in PF. Note that this proposal is different from Prosodic Inversion ( Halpern 1995) in that it does not rearrange at PF elements that are positioned in the syntax first.
4. A COMPARATIVE STUDY OF ROMANIAN IMPERATIVES Languages with pre-verbal object clitics in declarative sentences frequently show post-verbal object clitics hi positive imperatives. This is, for example, the case in Romanian: (32)
a. b.
L-atn vazut. 'I have seen him.' Lasa ma! 'Leave-IMP me!'
This kind of alternation in object clitic position has been repeatedly taken as evidence for verb movement to a position higher than that containing the clitic pronouns (e.g. Rivero 1994; Rivero and Terzi 1995; Rooryck 1992; Zanuttini 1997). For example, Romanian is one of the languages for which Rivero and Terzi posit a strong imperative feature in C inducing verb movement prior to Spell Out. Under their analysis, languages which have a strong imperative feature in C are languages which show true imperative morphology only in positive imperatives; languages in which both positive and negative imperatives show true imperative morphology (e.g. Bulgarian) have an imperative feature in I.11 Romanian is of the first kind, since negative imperatives show surrogate
10 As in French, there is an alternative and more colloquial way in Romanian of formulatingj'es-Ho questions which consists of assigning special intonation to a statement in order to express interrogative illocutionary force. These necessitate a separate account which builds on well-attested tonal means of expressing functional features crosslinguistically. 11 Rivero and Terzi (1995) claim that "imperative Vs with distinctive morphology either have a distinctive syntax (Modern Greek, Spanish), or distribute like other Vs (Serbo-Croatian, Ancient Greek)... the second type is the Wackernagel language". The distinctive syntax which characterizes the first type consists of two properties: (i) Modern Greek and Serbo-Croatian do not allow true imperatives to be negated, they resort to surrogate morphology; (ii) imperatives show encliticization while declaratives show procliticization. Macedonian is a real challenge to their analysis: the morphology is specific to imperatives and clitics follow the verb in both positive and negative imperatives. Thus, Macedonian does not fit the first type since it allows true imperatives to be negated. Nor does it fit the second type: finite declaratives show a different syntax, that is, procliticization. Moreover, Macedonian is not
Optimal Romanian clitics
253
infinitive morphology: (33)
Nu mS lasa! 'Don't leave-lNF me!'
Interestingly enough, other Romance languages (e.g. Italian and Spanish) display the same change in morphology as Romanian but without the shift in clitic position. This is shown for Italian in (34).12 (34)
a.
h.
Leggilo! Read-IMP-it 'Read it!' Non leggerlo! not read-INF-it 'Don't read it!'
All analyses referred to in the previous parAgraph, despite differences in implementation, share the view that verb movement in imperatives is morphologically conditioned. The alternative view that verb movement is structurally but never morphologically conditioned predicts that imperatives do not involve verb movement. The issue of whether imperatives are finite or non-finite must also be addressed. Contra Rivero and Terzi (1995) and Zanuttini (1997), I propose that we derive the answer from actual imperative morphology but leave aside its surrogate or non-surrogate status." On an analogy with gerund and perfect forms, imperative morphology realizes an [imperative] feature. Unlike [ger], which is realized as non-finite morphology, and [perfj, realized as finite morphology, [imp] maybe finite or non-finite hi Romanian. Imperative morphology is finite if it carries inflection reflecting distinctions of person and number. The inflection may be specific to imperatives or make use of surrogate
a Wackernagel language under Rivero and Terzi's definition. See Legendre (in press b, c) for a detailed analysis of Macedonian. 12 Italian optionally allows the Romanian order neg-clitic-verb non lo leggere!, at least in some dialects and/or registers (Luigi Burzio, personal communication). This means that certain varieties of Italian have the "Romanian" ranking EDGEMOST(ACC) » EDGEMOST(IMP). 13 Both Rivero and Zanuttini consider morphology surrogate if imperative inflection is identical to a second person form in another tense/mood. For them, only a subclass of Romanian positive imperatives can be said to exhibit surrogate morphology (classes 1 and 5, as shown in (35)). Note that the alternative view I develop in the main text is embedded in a larger OT-based perspective on morphological forms whereby a language, whenever possible, exploits existing resources to mark new categories. In forthcoming work, I analyze the actual morphology of imperatives in Romance and Balkan as reflecting a competition between their syntax (no subject, hnce no Agreement) and their semantics (second person).
254 Comparative studies in Romanian syntax morphology considered finite independently of imperatives, that is, indicative or subjunctive. It is non-finite if it makes use of a surrogate infinitive. Romanian singular imperative morphology is finite (and completely surrogate) except in the presence of the negative marker mi, in which case it is non-finite (identical to infinitive): *. (35)
Conjugation class
Infinitive
Pos. Imper. (so)
Neg. Imper. (so)
1 5 2 3
'sing1 canta 'kill' omori 'see' vedea 'go' merge
canta (= 3s indie, pres.) omoara (= 3s indie, pres.) vezi (=2s indie, pres) mergi (=2s indie, pres.)
nu canta nu omori nu vedea nu merge
4
'die'
mori
nu muri
muri
(=2s indie, pres.)
In OT terms, the variation in finiteness means that, universally, both finite and non-finite candidates compete. In Romanian (as well as Italian) then, a finite candidate wins the positive imperative competition; a non-finite candidate wins the negative imperative competition; in Bulgarian, a finite candidate wins both competitions. Sadock and Zwicky (1985) report the existence of a rare case of imperatives: in Yokuts (a native language of California), the imperative form is in most dialects a bare verb stem. This suggests that the optimal candidate in Yokuts is non-finite, as predicted possible by the theory (no information is available on negative imperatives). This rare case also reveals that, universally, finite imperative morphology is less marked than non-finite morphology (regardless of whether finite morphology is specific to imperatives or not). This suggests two OT avenues. One, the input to imperatives contains both features [imp] and [T]. Two, non-finite imperative morphology follows whenever the feature [T] is not parsed. That is to say, non-finite candidates violate an Input-Output faithfulness constraint called PARSE(T): (36)
PARSE(T): The input feature [T] must be parsed in the output.
One final aspect of the morphology needs consideration. Finite imperative morphology, at least in most Romance and Balkan languages, is realized as verbal inflection rather than some sort of particle (as is the case in some other languages). This means that a single form instantiates two features, [imp] and [T]. As expected, such forms are constrained both by EDGEMOST(IMP) and EDGEMOST(T). Following a suggestion made by Paul Smolensky (personal communication), I propose that such forms are also constrained by the conjunction of EDGEMOST(IMP)& EDGEMOST(T) which outranks the simpler constraints. Such conjunctions have been argued to operate both in phonology (Smolensky 1993, 1995, 1997) and syntax (Legendre et al, 1998; Legendre, in press a). They are known as "local conjunctions" in the literature because the conjoined constraint is violated only when the conjuncts are both violated within a common local
Optimal Romanian clitics
255
domain. Further discussion of local conjunctions follows the presentation of the candidate set in Til which concerns positive imperatives in Romanian. Recall that that the forms lasa and ISsa are finite and non-finite, respectively. This has important consequences for the applicability of certain constraints to a given candidate. Non-finite candidates are annotated with -T in the next few Tables: Table 11. Romanian positive imperatives I: [imp] [T] [ace]
*Struc
*t
E(imp)
Parse (T)
E (ace) E(IMP) E(T) Nin(T)
*!
® *
&E(T)
•»• a. [v lasa ma] a'. [v, lasa ma] -T
*!
b. [v, ma lasa]
*!
b'. [v, ma lasa ] -T c. [v lasa [v, ma t ]]
*!
*
c'.[v,lasa [v, mat]] -T
*!
*
* *
® *
*
*
* *
*
The candidate set includes verb movement candidates c. and c'. Given the relative ranking of the relevant constraints ~ * STRUCTURE, *t » EDGEMOST constraints as shown in section 3. — candidates c. and c'. are immediately disqualified. We know that the optimal candidate a violates EDGEMOST(ACC) and NONINITIAL(T). The question is: why aren't a'., b., or c'. optimal instead? They violate constraints that are lower ranked than EDGEMOST(ACC). The answer must be that additional constraints are at work. One is PARSE(T). Ranking PARSE(T) higher than the highestranked constraint violated by optimal a. says that it is preferable in Romanian to have encliticization in imperatives (and violate EDGEMOST(ACC)) than switch to non-finite morphology (and violate PARSE(T)). This partial ranking eliminates two non-finite candidates a', and b'. This leaves candidate b. which does not violate PARSE(T) and yet is non-optimal. This is where local conjunction comes in. Note that candidate b. violates both EDGEMOST(IMP) andEDGEMOST(T). It is conceivable that there is a stronger requirement on a morpheme which instantiates more than one feature, formally characterizable as the conjunction of EDGEMOST(IMP) and EDGEMOST(T). From the competition between candidates a. and b. we can derive one relative ordering of the conjoined constraint: EDGEMOST(iMP)&EDGEMOST(T) » EDGEMOST(ACC). One important restriction on conjoined constraints is their local domain. It is not the case than any two constraints can be
256 Comparative studies in Romanian syntax conjoined. Only constraints relevant to a single local domain can be.14 My claim, then, is that portmanteau morphemes are the locus of such constraint interaction. It is a welcome possibility, I believe, for synthetic morphology has, to the best of my knowledge, never received any formal consideration by syntacticians incorporating inflectional morphology to syntax. Of course, the theory predicts several types of rerankings. Two important ones are considered here. The reverse order, clitic-verb, is predicted if the conjoined constraint is demoted below EDGEMOST(DAT), EDGEMOST(ACC). Under this reranking, candidate b. is the optimal one in Tl 1. Several cases of procliticization hi positive imperatives are hi fact acknowledged hi Rooryck (1992). Note that two of the languages belong to the Balkan group (Tsakonian is a Modern Greek dialect): (37)
a. Brazilian Portuguese
Me da! 'Give me'
b. Tsakonian
Mou pe! 'Tell me' A/ethuaj! 'Tell me'
c.Albanian
Another reranking prediction is made by reranking PARSE(T) below EDGEMOST(ACC) or EDGEMOST(IMP). It yields procliticization with non-finite verbal morphology; that is, candidate b'. becomes optimal. It remains to be seen whether Yokuts instantiates this ranking. Under the present analysis, positive imperatives hi Romanian are finite because, among other things, they satisfy PARSE(T). However, PARSE(T) is violated in Romanian negative imperatives. Can we also derive the pre-verbal position of clitics in negative imperatives from the ranking in Tl 1? The answer is yes, as shown hi T12.
14 Note that the Romanian plural imperative is finite in both positive and negative contexts; hence reference to number features must ultimately be incorporated to the present account.
257
Optimal Romanian clitics Table 12. Romanian negative imperatives I: [neg] [imp] [T] [ace]
E (neg) E(imp)& E(T)
PARSE (T) E (ace) E (imp) E(T) ®
•S'a. [v nu ma lasa] -T **i
a', [y nu ma lasa ]
*
b. [v/ «M lasa m£\ -T b'. [V' HM lasa ma ]
*!
c. [vlasa numa\ -T
*!
c'. [v lasa AIM /waj
*!
*
®
®®
®®
*
**
**
**!
*
**
*
*
** **
Consider how this result is achieved. Assume that the alignment constraint on the feature [neg] outranks all other EDGEMOST constraints (nu precedes all other clitics discussed here). The fact that candidate c. is sub-optimal shows that EDGEMOST(NEG) outranks PARSE(T), violated by optimal candidate a. Whenever EDGEMOST(NEG) is satisfied, all other competing EDGEMOST constraints are violated. In particular, finite candidates violate EDGEMOST(IMP)&EDGEMOST(T) while non-finite candidates vacuously satisfy the conjoined constraint. Under the ranking EDGEMOST(IMP)&EDGEMOST(T) » PARSE(T), all finite candidates are eliminated from the competition. This leaves the non-finite ones, a. and b., which differ with respect to the next lowerranked constraint, EDGEMOST(ACC). Because of gradiency, only candidate a survives this step in the optimization process. Hence, candidate a. is optimal. The remainder of this section is devoted to imperatives in languages like Italian which share the morphological properties of Romanian but differ in the placement of clitics in negative imperatives. I am assuming with Zanuttini (1997) that non is a head, and that PARSE(NEGScoPE) is undominated. T13 shows that minimal reranking predicts the Italian pattern. The forms leggi and legger(e) are finite and non-finite, respectively:
258 Comparative studies in Romanian syntax Table 13. Italian negative imperatives I: [imp] [T] [ace]
E(imp)&E(T)
a. [v non [lo leggere]] -T a'.[ v . non [lo leggi]]
E (ace)
**i
*
**
*
**i
*
**
*
©
®®
*
**
* **i
•*• c. [v< non [legger/o]] -T c'. [v, non [leggi/o]]
E (imp)
* **i
b. [v non lo [leggere]] -T b'.[v, non lo [leggi]]
Parse(T)
® *!
E(T)
**
**
0
Recall that hi Romanian, candidate a. wins hi T12: it violates EDGEMOST(IMP) twice but higherranked EDGEMOST(ACC) only once; its closest competitor b. violates EDGEMOST(ACC) twice. In Italian (T13), the optimal candidate is c. Unlike its closest competitors, a. and b., c. violates EDGEMOST(IMP) only once versus twice for a. and b. On the other hand, c. violates EDGEMOST(ACC) twice which is violated only once by a. and b. Compared to Romanian, the Italian pattern follows simply from the reranking ofEDGEMOST(lMP) and EDGEMOST(ACC). In Romanian, EDGEMOST(ACC) » EDGEMOST(IMP). In Italian, it is the reverse: EDGEMOST(IMP) » EDGEMOST(ACC). Consider the prediction that the ranking in T13 makes for Italian positive imperatives. In the absence of non, EDGEMOST(IMP)&EDGEMOST(T) can be satisfied. PARSE(T) can also be satisfied by candidates which satisfy the conjoined EDGEMOST constraint. The competition trickles down to the very ranking established on the basis of T13; that is, EDGEMOST(IMP) outranking EDGEMOST(ACC). The optimal candidate must be the one which satisfies EDGEMOST(IMP), that is a structure hi which the accusative clitic follows the verb. This is indeed the correct prediction, as shown hi (37a). The corresponding Table is given in T14.
259
Optimal Romanian clitics Table 14. Italian positive imperatives
I: [imp] [T] [ace]
*Struc
*t E(imp)
Parse (T) E(IMP)
E (ace)
*!
® *
E(T) *In(T)
&E(T) «-a. [y-leggi/o] a'. [v,legger/o] -T
*!
b. [v/oleggi]
*!
b'. [v, lo leggere] -T c. [v leggi [v, lo t]] *!
*
®
*
*
*
* 0
To summarize the main points of section 4., I have proposed that the Romanian order verb-clitic in positive imperatives is regulated by a conjoined constraint, EDGEMOST(iMP)&EDGEMOST(T), which outranks the constraints it is a conjunction of, as well as EDGEMOST(ACC). The only way the conjoined constraint can be satisfied is by having clitic pronouns follow the verb. Wherever the conjoined constraint is inapplicable, the ranking EDGEMOST(ACC) » EDGEMOST(T) forces the decision and yields procliticization (i.e., in declaratives and questions). In negative imperatives, EDGEMOST(IMP)&EDGEMOST(T) can never be satisfied due to the presence of the negative particle. The solution is to be unfaithful to the input and violate PARSE(T). The best candidate becomes the one which best satisfies EDGEMOST(ACC) » EDGEMOST(IMP), namely procliticization. Thus, finite and non-finite candidate structures compete for optimization in all imperatives. Finally, I have illustrated minimal reranking effects hi Italian. Dobrovie-Sorin (1994, 1995) offers an alternative analysis in which comparatively few functional projections are also posited. Her analysis is specifically designed to address the topic of the present paper. Yet, it differs from the present analysis in some significant ways. For one, she analyzes both questions and imperatives in terms of verb movement which has the consequence of generating traces which are subject to the inviolable ECP. In her analysis, clitics are Xs, hence they block antecedent-government of the t(race) in I by its coindexed antecedent V in C under Relatived Minimality (Rizzi 1990). Her solution is to apply "merging", an operation by which adjacent functional X categories into one X. Clitics are generated adjoined to IP, which can meet the conditions for merging only if IPs are assumed to be Spec-less. Several of these moves are theoretically costly: (a) Adjunction of X to IP violates the Structure Preservation Hypothesis, (b) The Spec-less IP structure violates standard X' Theory, (c) Merging in questions is a case of lowering, (d) The clitic trace left behind by merging k imperative structures must be assumed to be invisible for antecedent government. To derive the fact that clitic pronouns cannot intervene between the clitic auxiliary and the lexical verb, Dobrovie-Sorin proposes that Romanian auxiliary
260 Comparative studies in Romanian syntax structures are structurally distinct from their counterparts in other Romance languages. Instead of heading an AuxP projection dominating VP, they are assumed to be adjoined to a CP whose head is occupied by the proposed V-I. Following V-I to C movement, the clitic pronoun ends up in postverbal position, which can be observed for o 'her' in the presence of an auxiliary. She derives the pre-verbal position via a local rule of clitic climbing characterized as a morpho-phonological process. It is rather clear that the theoretically undesirable moves she makes are tied to a framework of inviolable constraints and to the cumbersome syntacticization of ch'tics and morphology. As noted above, she, in fact, violates a number of principles of the theory she assumes. With respect to the important issue of cross-linguistic variation, she explicitly rejects Rivero's view that languages may differ in terms of then" functional projections. What she ends up with — a constituent structure truly idiosyncratic to Romanian ~ does not seem to lead to a more constrained typology: if Romanian can be idiosyncratic in a particular way, why can't another language be idiosyncratic in some other way?
5. CONCLUSION In conclusion, it is useful to step back from particular details of the optimization process and highlight the main aspects of the proposed theory of clitics. It is grounded in the characteristic clustering property of clitics. It builds on earlier claims that clitics belong to the morphology rather than to the syntax (Anderson 1992;Klavans 1985). Like lexical affixes, clitics instantiate functional features and they are subject to alignment constraints. The only significant difference between lexical affixes and clitics is the domain at the edge of which a particular feature must be realized. If the domain is the word, the feature is realized as a lexical affix; if the domain is clausal, the feature is realized as a phrasal affix or clitic. The main departure from the earlier work lies in the claim that alignment constraints are violable, a claim independently made in Anderson (1996). Different alignment constraints regulating the realization of distinct features compete for the left edge of the clause; hence they must be violable. Which one prevails depends on two factors: the input to optimization which contains the features themselves and thus determines which constraints are applicable in a particular context, as well as the ranking of the alignment constraints themselves. The latter constitutes a partial grammar of a given language. Partial rankings for Romanian, Macedonian, Bulgarian, and Italian are proposed. They exemplify the OT theory of cross-linguistic variation, typology by reranking. From the present perspective, the Romanian puzzle pertaining to its clitic elements, as well as some of its cross-linguistic extensions, can be solved in a comparatively simple fashion. Consider each part of the puzzle in turn.
Optimal Romanian clitics
261
(1) Why doesn't a relatively large class of Romanian auxiliaries allow SAI in questions? As is well known, SAI is a syntactic operation whereby a verbal head moves to an empty head (C) position. The answer is that clitic auxiliaries are not heads, hence they do not undergo SAI. Instead, the lexical verb itself moves to C with the result that the subject appears after the lexical verb rather than the auxiliary. (2) The standard view of syntax requires a landing site for all verbal elements, including the clitic auxiliary. If the lexical verb is in C, as revealed by the position of overt subjects, where is the clitic auxiliary? To solve the problem, Dobrovie- Sorin (1994) posits a complicated constituent structure. In contrast, the issue does not even arise under the present theory. (3) Why does clitic placement vary in questions and positive imperatives? The answer is two-fold. One part involves the issue of verb movement: questions involve movement of V to C to satisfy a high-ranked structural constraint, OBHD; imperatives do not involve movement. The other part of the answer lies with constraint ranking. All morphological features are subject to EDGEMOST(F), including features like [accusative] and [T]. Under the ranking EDGEMOST(ACC)»EDGEMOST(T), clitic pronouns precede the finite verbal element. Encliticization in positive imperatives result from a conjoined constraint EDGEMOST(IMP)&EDGEMOST(T) outranking EDGEMOST(ACC). (4) Why do negative imperatives exhibit procliticization rather than encliticization on the one hand and infinitive rather than finite morphology on the other? Clitic pronouns are pre-verbal in negative imperatives because of the ranking of alignment constraints: EDGEMOST(ACC)»EDGEMOST(IMP). Infinitive morphology is interpreted as a failure to parse [T]. In negative imperatives [T] is unparsed because EDGEMOST(IMP)&EDGEMOST(T) is undominated: EDGEMOST(IMP)&E(DGEMOSTT) » PARSE(T). Violating PARSE(T) renders the conjoined constraint vacuous. Thus the morphology is tied to the relative ranking of PARSE(T) while the placement of clitics is tied to the relative ranking of EDGEMOST(F). Contra Zanuttini (1997), the two features are independent from one another. This is a desirable result because, cross-linguistically, four types of negative imperatives are attested: infinitive morphology with procliticization (Romanian), infinitive with encliticitization (Italian, Spanish), finite morphology with procliticization (Bulgarian*, Albanian), and finite morphology with encliticization (Slovak*, Czech*, Serbo-Croatian*, and Macedonian). Starred are so-called Wackernagel languages (analyzed here as cases where NONlNlTlAL(F) » EDGEMOST(F)). The sample is quite small and I suspect that second-position clitic languages with infinitive imperative morphology also exist. If this turns out to be true, then there is no correlation between finite morphology, clitic placement in negative imperatives and Wackernagel effects, as predicted by the current analysis, in which they are respectively governed by the relative ranking of three independent constraint families: PARSE(F), EDGEMOST(F), AND NONlNlTlAL(F).
262 Comparative studies in Romanian syntax
REFERENCES Anderson, S. (1992). A-Morphous morphology. Cambridge University Press, Cambridge. Anderson, S. (1993). Wackernagel's revenge: Clitics, morphology and the syntax of second position. Language 69, 68-98. Anderson, S. (1996). How to put your clitics in their place or why the best account of secondposition phenomena may be something like the optimal one. The linguistic review 13,165191. Costa, J. (to appear). Word order typology in Optimality Theory. In: Optimality-theoretic syntax (G. Legendre, J. Grimshaw and S. Vikner, eds.). MIT Press, Cambridge, Mass. Dobrovie-Sorin, C. (1994). The syntax of Romanian: Comparative studies in Romance. Mouton de Gruyter, Berlin. Dobrovie-Sorin, C. (1995). Clitic clusters in Rumanian: Deriving linear order from hierarchical structure. In: Advances in Roumanian linguistics (G. Cinque and G. Giusti, eds.), pp. 5581. John Benjamins, Amsterdam. Everett, D. L. (1996). Why there are no clitics: An alternative perspective on pronominal allomorphy. The Summer Institute of Linguistics and the University of Texas at Arlington Publications in Linguistics. Grimshaw, J. (1991). Extended projections. Manuscript. Brandeis University. Grimshaw, J. (1997). Projection, heads, and Optimality. Linguistic inquiry 28, 373-422. Grimshaw, J. & V. Samek-Lodoyici. (1998). Optimal subjects and subject universals. In: Is the best good enough?Optimality and competition in syntax (P. Barbosa, D. Fox, P. Hagstrom, M. McGinnis, and D. Pesetsky, eds.), pp. 193-219. MIT Press, Cambridge, Mass. Halpern, A. (1995). On the placement and morphology of clitics. CSLI Publications, Stanford. Kayne, R. (1975). French syntax. MIT Press, Cambridge, MA. Kok, A. de. (1989). On the conjunct object pronoun systems of Rumanian and French: A comparative analysis. International journal of Rumanian studies 7, 29-29. Klavans, J. (1985). The independence of syntax and phonology in cliticization. Language 61, 95120. Legendre, G. (1996). Clitics, verb (non)-movement, and optunahty in Bulgarian. Technical Report JHU-CogSci-96-5, Baltimore, MD: Johns Hopkins University. Legendre, G. (1998a). Optimization hi French stylistic and complex Inversion. Paper presented at the Rutgers University OT Workshop, May 1998. Legendre, G. (1998b). Generalized optimality-theoretic alignment: The case of South Slavic clitics. Paper presented at the University of Essex. November 1998. Legendre, G. (hi press a). Morphological and prosodic alignment of Bulgarian clitics. In: Optimality Theory: Syntax, phonology, and acquisition (J. Dekkers, F. van der Leeuw and
Optimal Romanian clitics
263
J. van de Weijer, eds.). Oxford University Press, New York. Legendre, G. (in press b). Morphological and prosodic alignment at work: The case of South Slavic clitics. In: Proceedings ofWCCFL XVII (S.J. Blake, E.-S. Kim and K.N. Shahin, eds.). CSLI Publications, Stanford. Legendre, G. (in press c). Second-position clitics in a verb-second language: Conflict resolution in Macedonian. In: Proceedings ofESCOL 1997 (S. Avrutin and D. Jonas, eds.). Cornell University: GLC Publications. Legendre, G., C. Wilson, P. Smolensky, K. Homer and W. Raymond. (1995). Optimality and \vhextraction. In: Papers in Optimality Theory (J. Beckman, S. Urbanczyck and L. Walsh, eds.), pp. 607-636. UMOP 18, University of Massachusetts. Legendre, G., P. Smolensky and C. Wilson. (1998). When is less more? Faithfulness and minimal links in w/z-chains. In: Is the best good enough? Optimality and competition in syntax (P. Barbosa, D. Fox, P. Hagstrom, M. McGinnis, and D. Pesetsky, eds.), pp. 249-289. MIT Press, Cambridge, Mass. Mallinson, G. (1986). Rumanian. CroomHelm, London. McCarthy, J. and A. Prince. (1993 a). Prosodic morphology I: Constraint interaction and satisfaction. MIT Press, Cambridge, Mass. McCarthy, J. and A. Prince. (1993b). Generalized alignment. Yearbook of Morphology 79-153. Miller, P. H. (1992). Clitics and constituents in Phrase Structure Grammar. Garland Press, New York. Motapanyane, V. (1989). La position du suject dans une langue a 1'ordre SVO/VSO. Rivista di Grammatica Generativa 14, 75-103. Motapanyane, V. (1991). Theoretical implications of complementation in Romanian. Doctoral dissertation, University of Geneva. Prince, A. and P. Smolensky. (1993). Optimality Theory: Constraint interaction in generative grammar. Manuscript, Rutgers University and University of Colorado. Rivero, M-L. (1994). Clause structure and V-movement hi the languages of the Balkans. Natural Language and Linguistic Theory 12, 63-120. Rivero, M-L and A. Terzi. (1995). Imperatives, V-movement and logical mood. Journal of Linguistics 31, 301-332. Rizzi, L. (1990). Relativized minimality. MIT Press, Cambridge, Mass. Rooryck, J. (1992). Romance enclitic ordering and universal grammar. The linguistic review 9, 219-250. Sadock, J.M. and A.M. Zwicky. (1985). Speech act distinctions in syntax. In: Language typology and syntactic description: Clause structure (T. Shopen, ed.), pp. 155-196. Cambridge University Press, Cambridge. Samek-Lodovici, V. (1996). Constraints on subjects: An optimality-theoretic analysis. Doctoral dissertation, Rutgers University, New Brunswick.
264 Comparative studies in Romanian syntax Samek-Lodovici, V. (1998). Opposite constraints: Left and right Focus-alignment in Kanakuru. Lingua 104, 111-130. Smolensky, P. (1993). Harmony, markedness, and phonological activity. Paper presented at the Rutgers Optimality Workshop 1, New Brunswick, N.J. Smolensky, P. (1995). On the internal structure of the constraint component Con of UG. Paper presented at UCLA. Smolensky, P. (1997). Constraint interaction hi generative grammar II: Local conjunction, or random rules hi UG. Paper presented at the 1997 Hopkins Optimality Theory Workshop & University of Maryland Mayfest, Baltimore, MD. Tomic, O. M. (1996). The Balkan Slavic clausal clitics. Natural Language and Linguistic Theory 14, 1-58. Zanuttini, R. (1997). Negation and clausal structure: A comparative study of Romance languages. Oxford University Press, New York. Zwicky, A. M. (1985). Clitics and particles. Language 61, 283-305. Zwicky, A.M. and G. Pullum (1983). Cliticization vs. inflection: English n 't. Language 59, 502513.
Parameters for Focus in English and Romanian
265
PARAMETERS FORFocus IN ENGLISH AND ROMANIAN Virginia Motapanyane, University of New Brunswick-Saint John
i. INTRODUCTION Focus interpretation involves prosodic stress on a constituent which surfaces in situ, as in (la), or in a preposed position, as in (Ib). (1)
a. b.
I could give away this book (not that one), This book I could give away (not that one).
Standard definitions of focus (e.g. Rochemont and Culicover, 1990) associate it with "new information", yielding different types offoci(i.e. contrastive, non-contrastive, presentational; see King, 1995) which have an impact on the mapping of the sentence. In particular, constituents specified for [Focus] must obtain scope over VP and/or IP. Syntactic scope entails movement, which implies overt dislocation as in (Ib), or formal feature (FF) raising, as hi (la). Overt movement, as in (Ib), gives rise to important variation, even within the same grammar. For example, English fronting to Focus consists of topicalization, as in (Ib), or clefting, as in (2).1
*I would like to thank Gabriela Alboiu and Catherine Rudin for useful comments; for patient proof-reading I am grateful to Julie Brittain and Rod Hill. This research was supported by the SSHRC grant 410951417. 'A terminological explanation is in order: "topicalization" in this paper, as in (i), is equivalent to "focus topicalization" in Gundel (1974), who distinguishes it from "topic topicalization", such as illustrated in (ii). The latter type is also identified as English left dislocation (ELD) in Cinque (1990), who treats it as distinct from the Romance configurations with Clitic left dislocation (CLLD). In CLLD structures we obtain a Topic (versus focus) reading for the constituents located in the left periphery of the sentence, (i) That car she wanted to buy (not the other one), (ii) That car, she wouldn't want it as a present!
266 Comparative studies in Romanian syntax (2)
a. b.
It is a book (that) I could give away, It is John who wants it.
In this paper I discuss operations which, on a par with the topicalization and clefting in (2), induce pre-verbal Focus through overt fronting of focused constituents. Empirical data for this discussion come from English and Romanian. The analysis of pre-verbal Focus constructions in these two languages aims to account for cross-linguistic variation in terms of parametric settings, and to provide arguments for the visibility of [Focus] features during the computation. English and Romanian offer relevant samples for a comparative analysis due to their systematic contrasts in mapping out pre-verbal Focus. The operations deriving pre-verbal Focus in English induce ungrammaticality in Romanian, as shown by the contrast between (lb)/(3a) and (2a, b)/(3b, c), respectively. (3)
a. b. c.
*7Aceasta carte Ion ar putea-(o) dona. this book-FOC Ion would could it give-away *E o carte ca Ion ar putea-(o) dona. is a book-FOC that Ion would could it give-away *E Ion cine vrea asta. is Ion who wants this
Pre-verbal Focus in Romanian seems to target a lower scope position compared to English. Let us assume, for the tune being, that pre-verbal subjects occupy the same position in both languages. Then, (4) shows that Focus is hierarchically lower in Romanian, following the subject, whereas it precedes the subject in English, as shown in (lb)2. (4)
Ion o carte ar putea dona (nu bani). Ion a book-FOC would could give-away not money 'It is a book Ion could give away, not money.'
The location of pre-verbal Focus in Romanian remains constant in various contexts, such asyes-no interrogatives and subordinate clauses, as in (5a, c). This property further deepens the gap between 2
Other constituents, namely, those which undergo CLLD, may precede Focus in (4), as shown in (i). However, these configurations do not serve a comparative purpose, since they are absent from English, (i) Cartea mdine vreau s- o citesc. book-the tomorrow-Foe want-1 so SA it read-SUBJ 'It is tomorrow I want to read the book.' The object in (i) receives a Topic reading, which could also extend to the subject in (4), as explained in the next section.
Parameters for Focus in English and Romanian
267
English and Romanian: only the former requires obligatory Subject Auxiliary Inversion (SAI) hi yes-no interrogatives (5b), and tends to limit pre-verbal Focus to root clauses. When embedding occurs hi English, it is more easily accepted with clefting, as hi the translation for (5c), than with topicalization (5d). (5)
a.
b. c.
d.
Oare Ion mapna voia, saucasa? Q Ion car-FOC wanted or house 'Was it a car John wanted or a house?' *It was a car Ion wanted or a house? Ma nitreb daca Ion ma§ina voia. me ask if Ion car-FOC wanted 'I wonder if it was a car Ion wanted.' *I wonder if a car Ion wanted.
The contrastive behavior of pre-verbal Focus constructions in English and Romanian is discussed in the framework of the Minimalist Program (Chomsky 1995). Cross-linguistic variation hi this theoretical framework is morphologically driven: it arises at the level of formal features which enter the numeration, and undergo Select and Merge. The classes of syntactically relevant features are limited to formal features, whereas semantic features are invisible for computations. Formal features in this framework include categorial features and functional features which merge hi the extended domain of lexical categories (e.g. V or N) for checking purposes. [Focus] features do not belong to the class of formal/categorial features, since they do not define a lexical category (e.g., in the way [wh] features define the w/j-phrases) and are not obligatory as checking devices on the verbal or nominal morphology (e.g., as [tense] features are hi the functional domain of V). In fact, Chomsky (1995, 220) considers that Focus belongs to "surface effects", external to the derivation, and accessed at the interface. This paper supports the theoretical assumptions of the Minimalist Program by pointing out that [Focus] features must combine with semantically related features as [wh] and [tense]. These two features extend then- formal status to [Focus] and thus ensure its transition into the grammar as a two-feature set: [Focus/wh] or [Focus/tense]. Supporting evidence for this hypothesis comes from two sources: (a) current analyses ofFocus constructions hi English, summed up in section 2., which point to the correlation between Focus and w/i-movement; and (b) empirical data from Romanian, discussed in section 3., which indicate a strict dependency between fronting to Focus and the checking configuration for [tense] features. This hypothesis leads to a parametric approach, defined within the concept of local economy, which allows for marked and unmarked parametric settings. The unmarked parametric setting for Focus in English is [Focus/wh], inducing clefts, as hi (2); the
268 Comparative studies in Romanian syntax unmarked parametric setting for Focus in Romanian is [Focus/tense], inducing fronting to SpecTP, as in (4). The reverse setting is marked, and it leads, in English, to topicalization, as in (Ib), or to marked CP constructions in Romanian. Section 4. argues that the structural configuration for topicalization in English is similar to Romanian Focus constructions, with fronting to SpecTP. Section 5 reappraises clefts and negative constituent proposing, showing that both operations involve derivations distinct from topicalization. Aside from accounting for variation in Focus constructions, the parametric approach to Focus sheds new light on cross-linguistic variation in interrogative clauses. In section 6., the systematic contrast between Wz-movement in English and Romanian is reduced to the checking configuration for the [wh] and the [Focus] features specified on w/z-phrases. In conclusion, the analysis of [Focus] as syntactically dependent on the formal features [wh] or [tense] leads to a uniform treatment of variation in Focus and interrogative clauses, while also complying with the Minimalist restriction of syntactically relevant features to formal features.
2. THE CORRELATION BETWEEN [Focus] AND [WH] 2.1. Current Analyses of Clefting and Topicalization Current studies of English Focus constructions agree in defining Focus movement as a form ofwhmovement. They emphasize the correlation between the syntax of w/z-phrases and the syntax of Focus(Chomsky 1977,Rochemont 1978,1986; cf. Mueller and Sternefeld 1993); in particular, the shared requirement for movement to a scope position is emphasized. Chomsky (1977) considers that the ie-phrase in clefts such as (2a), repeated for convenience, adjoins to CP and is co-indexed with a Null Operator in SpecCP.3 The Null Operator binds the variable of the focused constituent, as in (6a). In (2b), the relative who, co-indexed with the focused constituent, behaves as a proper operator, binding the trace of the focused constituent, as in (6b). In a different theoretical framework, Kayne (1994) argues for an alternative analysis: in a construction like (2a), be selects the cleft, in which the clefted phrase moves to the SpecCP, whose head C is the relative that; the relative has quantificational features and licenses its Spec as a position for structural operators. So, in (2a) the phrase attracted to SpecCP for feature checking binds its variable as in (6c). (2)
a.
It is a book (that) I could give away.
*The original S/S' notation in Chomsky (1977) has been converted into current terminology, that is IP/CP.
Parameters for Focus in English and Romanian
(6)
b. a. b. c.
269
It is John who wants it. [CPIt is a book, [CPOPj (that) [n> I could give away t; ]]] [CPIt is Johnt [CP whOj [,p t; wants it ]]] It is [CP a book, [c (that) [w I could give away ts ]]]
Topicalization, as in (Ib), raised more controversial discussions in the literature. For Chomsky (1986) the dislocated constituent adjoins to CP and is coindexed with a Null Operator in SpecCP. The structure must be interpreted as a CP small clause in which the constituent in Topic is a notional subject-of-predication, and the CP is a sentential predicate derived by the empty operator. On the other hand, Rochemont (1986) argues, on the basis of root versus embedded asymmetries, that topicalization involves adjunction to IP. Drubig (1992) restates the CP analysis of topicalization by arguing that Focus topicalization dislocates constituents to SpecCP in structures with a recursive CP. A more recent analysis is proposed in Mueller and Sternefeld (1993), where a CP/TP unit, in which T is a functional head for Topic, parallels the recursive CP in Drubig (1992). From this short summary of current research, it appears that both clefting and topicalization in English target a left-peripheral A'-position compatible with operators, and generally identified as SpecCP. Thus, movement to SpecCP motivates a unified treatment of pre-verbal Focus constructions and interrogative clauses in English. On the basis of semantic considerations, several studies in the GB framework have proposed a Focus Phrase (Horvath 1995) and a Focus Criterion (Brody 1990) to account for the array of syntactic operations in Focus constructions. However, the separate manifestation ofFocus in syntax has been questioned in other studies. For example, Haegeman (1995) shows that in several languages (Classical Arabic, Gun) the morphology for Focus overlaps with the morphology for question formation (see also Motapanyane 1997 for Bulgarian //'). In these languages, the Focus Operator belongs to the domain of "affective" operators, as do negative and interrogative operators. Derivations induced by these operators comply with a similar well-formedness condition (defined in Rizzi 1990), requiring a Spec-head configuration, in which the Spec is also a position for structural operators. Along these lines, Haegeman points out that Focus and interrogative operators are syntactically non-distinct, since they obtain the Spec-head relation in the same position. This analysis agrees, in spirit, with the analyses on Focus and w/z-movement in English summed up at the beginning of this section.
2.2. Focus and WH-movement in Romanian Since the correlation between [wh] and [Focus] is considered to be universal, it is expected to extend to Romanian as well. It appears, however, that the overlap between wh-movement and
270 Comparative studies in Romanian syntax Focus movement in this language is highly restricted. The word order in (4) and (5) may suggest that fronting to Focus follows the pattern of Hungarian, which has a structural pre-verbal position, distinct from SpecCP and targeted by w/z-phrases and Focus phrases alike.4 This is a tempting approach, since the Hungarian word order is also possible in Romanian, as shown in (7). In (7), the optional question morpheme oare occupies the head C (see the Appendix to this paper), and both the focused constituent in (7a) and the w/i-phrase in (7b) move to the same clause-internal position. (7)
a.
b.
(Oare) Ion pe Maria o invitase? Q Ion/?e-Mary-FOC her invited 'Was it Maria whom Ion had invited?' (Oare) lonpe cine invitase? Q Ion pe-v/hom invited 'Who had Ion invited?'
Let us assume that clauses are Tense projections (TP), as proposed in the Minimalist Program (Chomsky 1995, 355-67). Following (7), both the Focus and the w/i-movement seem to target a position within TP, presumably SpecTP, as in Hungarian. Nevertheless, the examples in (8) and (9) show that the unmarked target for w/z-movement in Romanian is SpecCP. First, Romanian has multiple w/i-movement, which yields ungrammatical results within TP (8a), and targets only SpecCP, as in (8b). (8)
a. b.
c.
*(Oare) loupe cine unde invitase? Q Ion /?e-whom where invited Pe cine unde, (oare), invitase Ion? pe-whom where Q invited Ion 'Whom did Ion invite where?' Pe cine oare (*Ion) invitase (Ion)? pe-wbiomQ Ion invited Ion 'Whom did Ion invite?'
In (8c), the unmarked word order w/i-phrase - oare further indicates that the w^-phrase lands in
4
An account along these lines is implied in Kiss (1997) where Romanian appears on the list of Topic-prominent languages and on the list of languages with structural Focus, predicting that Focus movement in Romanian follows the Hungarian pattern. It is true that typologically Romanian belongs to the two language groups, on a par with Hungarian. However, Romanian wA-movement rules differ from Hungarian, in that the landing site for w//-phrases and focused phrases is distinct. From this point of view, Romanian behaves like Bulgarian and Russian.
Parameters for Focus in English and Romanian
271
SpecCP, higher than oare in C. Moreover, (8c) shows that subject verb inversion (SVI) is obligatory in w/i-interrogatives, a phenomenon that occurs systematically in Romanian when a quantified phrase (wh or other) moves to SpecCP, but not when it moves to SpecTP. Another series of data show that bare quantifiers in pre-verbal subject position cooccur with focused constituents (9a) but not with w/i-phrases (9b), unless SVI is observed, as in (9c). As mentioned for (8c), SVI follows from wft-movement to SpecCP, resembling SAI in English: (9) a. Nimeni nici o atenpe nu-i dadea. nobody not an attention-FOC not him/her gave 'Nobody was paying the least attention to her/him.' b. *Nimeni ce nu- Ji va aduce? nobody what not you will bring c. Ce nu- £ va aduce nimeni? what not you will bring nobody 'What will no one bring you?' The examples in (8) and (9) show that w/z-phrases undergo overt movement to SpecCP. Only a reduced set of constructions may use an alternative landing site clause internally, as in (7). So (7a, b) represent the extent to which the overlapping between w/z-movement and fronting to Focus occurs in Romanian. In most cases, w/z-phrases appear in SpecCP, whereas focused constituents surface in the clause-internal position illustrated in (7a).
2.3. Exclusion of Pre-verbal Focus in SpecCP The examples in (7) to (9) suggest that Romanian is a structural hybrid between the English and the Hungarian patterns which display one common target position for both wh- and Focus movement (i.e., either SpecCP in English or SpecTP in Hungarian). That is, in Romanian \vhphrases target SpecCP, whereas focused constituents presumably target SpecTP. This section will identify the factors which prevent fronting to Focus within CP, whereas section 6 will show why w/z-phrases may not usually land hi SpecTP in Romanian. A natural account for the dissociation between Focus and the head C in Romanian follows from the properties of w/z-movement in this language, which prevent the derivation of clefts, as in English. Notably, SpecCP in Romanian is not a position for structural operators (although it is compatible with intrinsic operators), and it does not license empty operators, such as are needed for clefting.
272 Comparative studies in Romanian syntax Dobrovie-Sorin (1990, 1994) shows that w/z-phrases, bare quantifiers and quantified NPs in Romanian differ with respect to their intrinsic properties and that this is reflected in the syntax: when moving to a scope position (presumably SpecCP), some of them form operator-variable chains, while others form chains with clitic pronouns. The latter lack quantificational properties, disabling SpecCP as a position for structural operators. This contrast is illustrated in (lOa, b). (10)
a.
b.
Nugtiam pecine, ai invitattj. not knew-1 SO pe-whom have-2SG invited 'I did not know whom you invited.' Nu §tiam pe caret 1,- ai invitat t;. not knew- ISG pe-which him have-2SG invited 'I did not know which one you invited.'
The generalization is that the compatibility with SpecCP of both cine 'who', a [+qu] element, and care 'which', a [-qu] element, indicates the negative setting for the structural quantifier parameter in Romanian. This setting is possible because the language has object ch'tic doubling, and movement of (-qu) and (+qu) w/z-phrases creates two types of chains (i.e. with or without clitic doubling). Thus, the negative setting for the quantifier parameter in Romanian, where SpecCP hosts [+/- qu] elements, contrasts with the positive setting hi English, where SpecCP hosts only [+qu] elements5. An important consequence of this parametric choice is the absence of Null Operators in SpecCP. If clefts necessarily involve a Null Operator (Chomsky 1977), as in (6a), then fronting to Focus to SpecCP would lead to illicit derivations in Romanian, as in (11). (11)
a. b. c.
*Ema$inaca Paul a cumparat. is car-FOC that Paul has bought *E Paul cine a cumparat ma§ina. is Paul-FOC who has bought car-the Paul, e celj care; a cumparat ma§ina. Paul-FOC is that which has bought car-the 'Paul is the one who bought the car.'
When SpecCP licenses an Null Operator, a complementizer like that may also associate with the respective [+qu] head C, as in clefts, and acquire a relative reading. Co, the Romanian counterpart
3
Dobrovie-Sorin (1994) argues that the relative which is intrinsically [-qu], on a par with Romanian care. However, which lands in a [+qu] Spec,CP, and acquires the quantificational features of the landing site. So which ends up behaving as who, instead of displaying the syntax of the Romanian care.
Parameters for Focus in English and Romanian
273
of that, cannot undergo the same "relativization" process because SpecCP in Romanian does not contain a Null Operator to license it. Thus, the properties of SpecCP rule out that-clQ&s in Romanian, as in (1 la). Along the same line, the w/z-phrase in (lib) is an intrinsic operator in SpecCP and it should be able to license clefts, as seen in English (2b). However, cine, unlike its English counterpart who, cannot function as a relative. The only relative available is care 'which', intrinsically a [-qu] element creating doubling clitic chains as in (lOb) and, therefore, unsuitable for clefts. The grammatical turn for (1 la, b) in Romanian presents the relative care in a pseudo-cleft, as in (1 Ic), where the 6e-phrase is de-quantified through the insertion of a demonstrative pronoun which selects the relative clause. Summary of section 2.: Focus and w/z-structures present a strong correlation in English, consisting of a similar target for movement; that is, SpecCP. Romanian presents such a correlation only in restricted environments which have a clause internal position as common target, as illustrated in (7). Elsewhere, only Focus lands in this position, whereas w/z-phrases land in SpecCP, as shown in (8) and (9). The difference between Focus hi SpecCP in English and Focus in TP in Romanian coincides with the possibility of clefting only in English, but not in Romanian (1 la, b). The source of this cross-linguistic variation has been attributed to the properties of SpecCP, which qualifies as a position for structural operators in English (then, it licenses clefts), but not in Romanian, as shown in (10).
3. THE CORRELATION BETWEEN [Focus] AND [TENSE] The negative setting for the structural quantifier parameter seems to prevent the association of [Focus] with [wh], as in clefts, so focused constituents andw/z-phrases do not both target SpecCP hi Romanian. Moreover, since SpecCP is not an obligatory landing site for [+qu] elements, another (lower) Spec in the hierarchical structure may attract wh- or Focus movement. The hypothesis developed in this section is that a non-argumental SpecTP serves as a landing site for Focus movement, and that this SpecTP is projected when the [Focus] feature merges with [tense]. Aside from the landing site (i.e. SpecCP in English versus SpecTP in Romanian), fronting to Focus triggers similar derivations in both languages, in which the head carrying the [focus] feature (i.e. C or T) projects two Spec positions for multiple feature checking. Along these lines, the strategies for fronting to Focus in English and Romanian contrast only in the parametric choice of the syntactic carrier for [Focus] features; that is, [wh] in C in English, as opposed to [tense] in T in Romanian,
274 Comparative studies in Romanian syntax 3.1. The Checking Configuration for the [Focus/tense] Feature This section will propose an analysis of fronting to Focus in Romanian, which displays the [Focus/tense] setting. The word order in pre-verbal Focus constructions, as illustrated in (4), is subject Focus-verb. Several tests indicate that the pre-verbal Focusposition intervenes between the subject in SpecTP and the inflected verb in T. A first set of tests, in (12) and (13), confirm that preverbal subjects are in SpecTP, whereas (14) shows that the Focus position is also a SpecTP. Pre-verbal subjects in Balkan languages have been analyzed as Topics adjoined to a maximal projection (Rudin 1993, Izvorski 1994). Romanian subjects may receive a Topic reading, in which case they adjoin to CP, as in (12a), or to TP. However, the same subject with a neutral reading surfaces in a pre-verbal position within TP. The distinction between Topic subjects and neutral subjects becomes clear when the subject is a bare quantifier: bare quantifiers cannot occur in Topic, as in (12b), above oare; also, they disallow concurrent operator-variable chains, such as formed by mereu 'always' and ceva 'something' in (12c). (12)
a.
b. c.
Ion oare §i-a cumparat ma§ina? John Q himself-DAT has bought car "Did John buy himself a car?" *Cineva, oare §i- a cumparat ma§dna? somebody Q himself/herself has bought car *Mereu CEVA ne aduce Ion. always something us-DAT brings John
The ungrammaticality of (12b) and (12c) contrasts with the examples in (13a, b) and indicates that the latter have the subject bare quantifier in a SpecTP for subjects, not in Topic. Motapanyane (1994) shows that these bare quantifiers occupy an argumental position (SpecIP or SpecTP according to current notation), as opposed to a non-argumental Topic position. Crucially, an argumental SpecTP for subjects does not interfere with movement to a non-argumental scope position in the same domain; so, bare quantifiers or quantified NPs may co-occur as subject and Focus, as illustrated in (13a, b), because the chains they create observe different conditions (i.e., A versus A' types).6 'Cornilescu (present volume) offers an analysis of pre-veibal subjects which predicts that all the bare quantifiers in pre-verbal subject position must be in Focus. Examples such as (13) bring empirical counter-arguments: two bare quantifiers (or quantified NPs) concur in subject and Focus positions, and a Focus reading on the former is excluded. Further examples attesting the same word order follow: (i) Un copil altceva mi-ar cere. a child something-else-FOC me would ask ' A child would ask me for a different thing.'
Parameters for Focus in English and Romanian (13)
a.
Cineva
trei pachete
ii
275
trimisese!
sombody three parcels-FOC him sent 'Someone had sent him three parcels!' b.
Nimeni nimic
nu Ji-
ar
face fara
pile.
nobody nothing-Foe not you would do without connections 'There's not one thing anyone would do for you if you don't have connections.' Free alternation of bare quantifiers and other nominal classes in the pre-verbal subject position indicates that all the pre-verbal subjects with neutral reading occupy an argumental SpecTP, and they can equally precede the constituent in Focus. Verbs move overtly to T in Romanian, leading to optional S VO/VSO. Thus, the pre-verbal subject in SpecTP in (13a, b) enters in a Spec-head relation with the verb in T. The word order shows that the pre-verbal Focus, an A'-position targeted by overt movement of quantified elements, intervenes between these two constituents. Restriction on multiple Focus, as in (14), further indicates that this is a Spec position, as opposed to adjunction to a maximal projection. (14)
a.
*Paul mafina, loaneij
if
ar
fi
dat- oi5 nu bicicleta
Mariei.
Paul car-the loana-DAT her would be given it not bicycle-the Maria-DAT b.
Ma§ina, loanei, Paul iar fi dat- o cu placere. car-the loana-DAT Paul her-DAT would be given it with pleasure 'As for the car, for Joan, Paul would have given it to her with pleasure.'
In (14a) two constituents in Focus lead to ungrammaticality. However, then- concurrent clitic doubling chains should allow for multiple fronting on a par with multiple Topic in (14b). Then the restriction in (14a) must follow from the scope properties of the position targeted by movement to Focus, and not from the chain type allowed. Scope positions correspond to Specs, since they have to enter into a checking relation with the head which carries the scope-bearing feature. Therefore, Focus must be a Spec,TP.
(ii)
(iii)
Altcineva penimeni n- ar invita. someone-else pe nobody-FOC not would invite 'Someone else wouldn't invite anybody.' Multefete nimic nu voiau si minance. many girls nothing-Foe not wanted SA eat-susj.3PL 'Many girls did not want to eat anything.'
276 Comparative studies in Romanian syntax 3.2. TP with Multiple Specs On the basis of the tests in (12) to (14), fronting to Focus in Romanian undergoes the derivation in (15).
[V+T] The multiple Spec analysis in (15 ) draws attention to tw o facts with theoretical consequences. First, the subject and the Focus positions equidistant to the head should be interchangeable in linear order. The ungrammaticality of (16) shows that this prediction does not hold for Romanian. (16)
a. b. c.
d.
*Azi Paulsa dus, sauieri? today-FOC Paul REFL has gone or yesterday *Oare azi Paul s- a dus, sau ieri? Q today-FOC Paul REFL has gone or yesterday ?Oare-o waging Paul a achitat, sau o casa? Q a car-FOC Paul has paid or a house 'Is it a car Paul has paid for or a house?' ??O mapna Paul a achitat sau o casa? a car-FOC Paul has paid or a house
The restriction in (16) will be approached in terms of checking theory, looking, in particular, at the checking mechanism of the features grouped under the head T. First, let us identify the features of T in (15), following the inventory proposed in Chomsky (1995), then we consider how the conditions on checking are met in this configuration. T has a series of verbal and nominal features, the former being checked by verb movement to the head T, while the latter trigger movement of a nominal constituent to the related Spec. The nominal features include [Case] and a predicative feature which replaces the Extended Projection Principle (EPP) of the previous framework. [EPP], being nominal in nature, is checked against the nominal
Parameters for Focus in English and Romanian
111
[D] feature of a constituent in SpecTP7. To these, the configuration in (15) adds [Focus], which is neutral with respect to verbal or nominal properties. The crucial factor is that the [EPP] feature must be checked from the highest SpecTP, which corresponds to the initial position in the predication. Then, feature checking in (15) may be implemented in two ways, as proposed in (17).
(17) a.
[Case/EPP], [Focus]
[Focus/EPP],[Case]
In (17) the NP subject does not carry [Focus] features. If the subject moves to the highest SpecTP, as in (17a), it checks the [Case] and the [EPP] feature of T, and another constituent moved to the lower SpecTP checks the [Focus] feature of T. The focused constituent is unrestricted in nature. The alternative, in (17b), is that the subject moves to the lower SpecTP. From this position, the subject NP can check only the [Case] feature of T. In this configuration, the focused constituent that moves to the highest SpecTP must check both the [Focus] and the [EPP] feature of T. Since [EPP] implies [D] features, only constituents with nominal properties can perform this double checking task. Thus, (17b) restricts the class of focused constituents to those with nominal properties, whereas (17a) allows for fronting to Focus of any type of constituents. The equivalent to (17a) is (13)7(15), which is the preferred version, whereas (17b) corresponds to (16). In (16a, b) the adverb fails to check the [EPP] feature of T, whereas a DP, as in (16c, d) has more success in derivation. The aggravation in (16d) may stem from parsing, which, in the absence of the interrogative morpheme oare, would situate the focused constituent in SpecCP, yielding a configuration which requires SVI, a condition which is not met in our structure. The second fact revealed by the structure in (15) is that the Focus position may head A'-chains of two types. The examples in (18) illustrate this double compatibility.
7
One may argue that the checking condition on [EPP] restricting the class of checking constituents to nominals is established on the basis of English structures, whereas Romanian is more flexible with respect to pre-verbal positions. However, the word order does not provide counter-evidence. A proof in this respect would come from a configuration in which, for example, a PP in clause-initial position would exclude the insertion of a (non-Focus) DP subject between it and the verb, on the grounds that the PP and the DP are competing for the same [EPP] related position. Such a configuration cannot be obtained in Romanian, so until further evidence becomes available, the condition on [EPP] checking in English holds in Romanian as well.
278 Comparative studies in Romanian syntax (18)
a.
b.
Paul scrisorile, tej voiatj, nu banii. Paul letters-the-FOC them wanted not money-the 'It was the letters Paul wanted, not the money.' Paul scrisori^ (*le;) voia ts de la tine, nu bani. Paul letters-FOC them wanted from you not money 'It was letters that Paul wanted from you, not money.'
Looking at (18), one may think that definite nouns are base-generated in the Focus position and head ch'tic doubling chains, whereas indefinite nouns undergo movement to the same position.8 If this were the case, base-generation of constituents in Focus would raise important questions for the analysis in (15), since Merge in SpecTP would disable subsequent movement of subjects to SpecTP. However, extraction tests show that, despite the presence of clitic pronouns, the chain in (18a) is formed through movement. Consider the contrast hi (19). (19)
a.
b.
c.
lonmapnOi spunea c- ar fi vrut- Ojtj Maria. Ion car-the-FOC said that would be wanted it Maria 'It was the car Ion said that Maria would have wanted.' loupe oricine, spunea c- ar invita t; Maria. Ion /7e-anybody-FOC said that would invite Maria 'John said that Maria would invite anybody.' *Ion ma$ina spunea ca Maria ar fi vrut-o. Ion car-the-FOC said that Maria would have wanted it
Extraction across the complementizer ca 'that' is allowed in Romanian, and it includes fronting to Focus to the matrix, as in (19a, b), where both definite nouns and bare quantifiers undergo the movement. Note that the condition for grammaticality in (19a, b) is SVI hi the embedded clause, failing which the sentence is ruled out, as in (19c). SVI arises when there is w/z-movement to SpecCP (see also the discussion around the examples hi (9)). As further mentioned for the structural quantifier parameter (the examples hi (10)), w/i-movement to SpecCP may form chains with or without quantificational properties. Along these lines, (19a, b) indicate that both definite and indefinite DPs move cyclically from the embedded clause, via the embedded SpecCP, and land in the SpecTP for Focus hi the matrix clause. This qualifies as Wi-movement and it triggers SVI in the embedded clause even when the resulting chain displays clitic doubling (19a).
8 A reviewer suggests that both definite and indefinite DPs create the same type of chain, with the only difference that lexical clitics mark the definiteness feature of the former class. However, Gierling (1996, 1997) demonstrates that defmiteness and clitic doubling do not overlap, since [+defmite]/[+human] DP objects may occur without clitic doubling in Romanian. It is rather the case that clitic doubling stems from certain feature distribution within VP, in particular, the realization of Focus with scope over VP.
Parameters for Focus in English and Romanian
279
To sum up this section, the option [Focus/tense] implies derivations with multiple SpecTP, as hi (15). The hierarchical order of Specs depends on the checking conditions of the [D]/ [EPP] features of T and the [Focus] feature of T, as shown in (17). Focus movement to SpecTP is a form of \vhmovement hi Romanian, although the target position is different; that is, Focus movement, on a par with ^-movement, creates either operator-variable chains, as in (18b), or clitic doubling chains, as in (18a).
4. MARKED PARAMETRIC SETTINGS Empirical data discussed for English and Romanian indicate that the two languages differ in then' parametric setting for [Focus] insofar as [Focus] associates with [wh] in English but with [tense] in Romanian. From the perspective of local economy (Collins 1997), both [Focus/wh] and [Focus/tense] options are equally present hi the numeration, and equally undergo Select and Merge. However, one option systematically leads to convergent derivations, whereas the opposite option causes derivations to crash. Successful derivations depend on the interaction between the selected feature and other grammar-internal constraints. In the case of pre-verbal Focus, the relevant constraint concerns the properties of scope positions hi the language. The contrastive properties of SpecCP hi English and Romanian entail contrastive results when Merge applies to [Focus/wh] and [Focus/tense], so that the former succeeds hi English rather than hi Romanian, and vice-versa. However, the parametric settings are not exclusive at the internal level of a language. That is, specific configurations may allow [Focus/tense] to succeed hi English, or [Focus/wh] in Romanian.
4.1. Marked [Focus/qu] in Romanian The examples hi (20) illustrate marked occurrences of fronting to Focus hi Romanian. (20)
a.
b. c. d.
Mapna oare a cumparat Paul, sau casa? car-FOC Q has bought Paul or house 'Is it a car Paul bought or a house? *Nu §tiam mapna daca cumparase Paul, sau casa. not knew-lso car-FOC whether bought Paul or house Ma§in£ oare (*Paul) a cumparat (Paul), sau casa? car-FOC Q Paul has bought Paul or house ??Mz£f/iaj oare (*Paul) a cumparat-Oj (Paul) tf, sau casa? car-the-FOC Q Paul has bought it Paul or house-the
280 Comparative studies in Romanian syntax If the yes-no question marker oare occupies the head C (see Appendix), then the focused constituent in (20a) moves to a position above this, presumably in SpecCP. Focus movement to SpecCP is possible only in root yes-no questions, and induces ungrammaticality in other contexts, as in (20b).When focused constituents target SpecCP, SVI is obligatory, as shown in (20c). In (20d), a definite DP undergoing the same fronting process as the indefinite in (20a) yields an almost ungrammatical sentence. The examples in (20) point to the following analysis: only oare may license fronting to Focus to SpecCP, whereas its absence induces ungrammaticality, as shown in (16d). It seems that C containing oare may have active [+qu] features ( see Appendix), which attract [focus], and allow for a complex [Focus/qu] feature as in (20). The feature configuration of C in this context implies a quantificational component, which would explain the restriction to indefinites observed in (20a) versus (20d). A similar marked setting occurs in English, where [Focus/tense] succeeds under certain conditions. The next section will argue that specific contexts in which this parametric setting is successful involve topicalization, as opposed to clefting.
4. 2. Marked [Focus/tense] in English as Topicalization Topicalization involves Left Dislocation of a [Focus] marked nominal constituent to a scope position outside the clause. The restricted class of constituents that can be dislocated in this way contrasts topicalization with clefting, which applies to any type of constituents. If topicalization is movement to SpecCP, as in Drubig (1992), or adjunction to IP, as in Rochemont (1986), the restriction to nominals is puzzling.9 On the other hand, if topicalization involves Left Dislocation to SpecTP, the restriction to nominals is motivated: only constituents with [D] features can check the [EPP] feature of T, by moving to the highest SpecTP.10 Along these lines, the analysis in (15) extends to topicalization structures, so that (Ib) has the representation in (21). The analysis of SpecTP as the target for topicalization (rather than an adjoined position to CP/IP as in the analyses summed up in section 2.) can capture the restriction
"The term "nominal constituents" includes any constituent in which the [+D] feature can check the [D] feature of T. Some PPs seem to have that ability, as shown in (i), while others do not, as in (ii). Presumably, the [D] feature percolates to the P head only in (i) and becomes accessible for checking against T. (i) To John I want to give it (not to Mary). (ii) *On vaccation he wants to go there (not on an assignment). "Collins (1997) argues that constituents other than DP/NP may check the [EPP] feature of T. However, this has to be licensed through overt head adjunction of V to T, as is the case in locative and quotative inversion. There is no overt V-movement to T in topicalization structures; therefore, the restriction on the class of constituents that may check [EPP] must hold.
Parameters for Focus in English and Romanian
281
oftopicalization to nominate, through compliance with the feature checking process of the features ofT: (1) b. This book I could give away.
(21)
TP
^^
Focus
T
Subject ^— T [focus/tense]
The hierarchical order of Specs contrasts (21) and (15). This is expected, since pre-verbal Focus (be it fronting as in (15) or Left Dislocation as in (21)) involves a form of w/z-movement. As pointed out in section 3., w/j-movement observes different conditions in English and in Romanian. In particular, the highest non-argumental Spec in English (usually SpecCP) behaves as an obligatory position for structural operators; that is, the site for elements which take scope over the sentence. In the absence of SpecCP, the role of scope position is implemented by the highest nonargumental Spec in the structure, which corresponds to SpecTP in (21). Since SpecTP for Focus is the highest position within TP, it must also check the [EPP] feature of T. Thus, (21) corresponds to the configuration in (17b), in which the class of constituents that can be fronted to SpecTP for Focus is restricted to nominals, the [D] feature of which checks [EPP]. As mentioned in the previous sections, non-argumental SpecCP/SpecTP have a different status in Romanian (i.e., they are compatible with operators and non-operators alike), and this is reflected hi the derivational pattern, which chooses (17a) as a non-marked configuration. Thus, (16c) is the Romanian counterpart to English topicalization as in (Ib). Further empirical support for the analysis of topicalization in (21) comes from the behavior of negative quantified DPs, as in (22). (22)
a. b.
*No one John invites without a purpose. No one would John invite without a purpose.
It has been noticed in previous examples, such as (Ib), that topicalization yields a subject-verb word order. The ungrammaticality of (22a) shows that when the "topicalized" constituent is a negative quantified DP, the subject-verb order rules out the sentence. Indeed, the grammatical
282 Comparative studies in Romanian syntax version in (22b) displays SAL Current analyses of negative constructions (see Haegeman 1995 for an overview) claim that abstract [neg] features associate only with C, versus T. From this perspective, the negative construction in (22) involves a CP, and fronting of the negative quantified DP targets SpecCP. This characterization of the negative construction is compatible with the environments which require SAL Since these environments contrast with those in which other classes of DPs are topicalized through the obligatory presence/absence of SAI, we conclude that only the former involve a CP, whereas the latter must lack this level, which is responsible for SAI. Thus, proposing of negative quantified DPs must also be kept distinct from topicalization. Summary of section 4.: Marked parametric settings are possible in both English and Romanian in certain contexts. In Romanian, [Focus] may merge with the question marker oare in C, licensing SpecCP as a Focus position. In English, [Focus] merged with [tense] induces successful derivations by projecting a SpecTP as a Focus site, compatible with nominal constituents only. The latter case yields the topicalization structures, which have the configuration hi (21), similar to the configurations for fronting to Focus hi Romanian discussed for (17).
5. CLEFTS AND NEGATIVE CONSTITUENT PREPOSING Sections 3. and 4. proposed a representation of Focus constructions hi which the [Focus] feature associates with [tense] hi T. The ensuing derivation presents a TP with multiple Specs, each Spec checking one of the two-feature set hi T. Checking theory applies evenly to all the heads with double specification. Hence, it is expected that double specification on C, that is [Focus/wh] in English Focus constructions, must give rise to configurations with multiple Specs as well. This prediction prompts a reappraising of clefts, as proposed hi 5.1. Also, the analysis of topicalization as involving a TP with multiple Specs implies that the configuration for topicalization is distinct from configurations with negative constituent preposing. This structural distinction will be explored hi 5.2.
5.1. Revisiting Clefts Fronting to Focus hi Romanian and topicalization hi English entail structures with multiple Specs due to the complexity of feature specification on T: a Spec for [tense] features, and a Spec for the [Focus] feature. Thus, (15) and (21) reflect compliance with Checking theory. Similar conditions for checking must be operative in structures with the option for [Focus/wh] features specified on C. Since the focused phrase is specified only for [Focus], another constituent must check the [wh]
Parameters for Focus in English and Romanian
283
feature of C leading to a structure with multiple SpecCP. Clefting is an example of a multiple SpecCP structure (as predictable from the analyses in (6)) in which the focused constituent is merged hi the highest SpecCP, and does not interfere with movement of w/z-phrases to SpecCP, as suggested in (23a). TTwtf-relatives have the [wh] feature of C checked by a Null Operator hi SpecCP, as in (23b). Absence of verb movement to C in (23) must be related to the absence of interrogative features in C, which are always associated with [V] features.
(23) a. CP
b.
[focus/wh]
[focus/wh]
The structure hi (23) mirrors the structure of TP with [Focus/tense] features on T, as proposed hi (15 ), (21). This is hi line with the hypothesis defended hi this paper, that pre-verbal Focus entails a similar syntactic pattern hi English and Romanian, and the systematic contrasts follow from parameter settings, which constrain fronting to Focus to being either within CP or within TP.
5.2. Negative Constituent Proposing Movement of negative constituents hi Focus structures is consistent with the cross-linguistic variation discussed so far as they land in SpecCP in English and hi SpecTP hi Romanian. English constructions with negative polarity items and proposing of negative constituents led to analyses hi which [neg] features with sentential scope are specified on two functional heads: on C when they are abstract, and on T (or Neg) when they are spelled out as the morpheme not (see Drubig 1992 vs. Rizzi 1990). Abstract [negation] features are checked either through verb movement to C, as in (24a), or through movement of a negative constituent to SpecCP, as hi (24b). (24)
a. b.
Do you like anyone? With no job would John be happy. (Drubig 1992, 8)
The fronted constituent in (24b) carries both [neg] and [Focus] features and undergoes wh-
284 Comparative studies in Romanian syntax movement, which triggers SAL Consistency with (23) entails an approach to (24b) in terms of Checking theory. The head C has quantificational [wh] features, which attract abstract [neg] and [Focus] features. [Neg] features also entail [V] features in C. The focused constituent checks the [neg], [Focus] and [wh] features through movement to SpecCP, whereasthe [V] feature is checked through verb movement to C, as in (25).
(25) Spec [with no job]j C TP [wh/Focus], [neg], [V] would John be happy tj
As is predictable from the parametric approach proposed in this paper, Romanian Focus constructions with negative constituent proposing have the pattern of positive Focus constructions with movement to SpecTP. [Neg] features ic this language always associate with [tense]. So subjects may always precede the focused negative constituent, as in (26), which has the structure of(15). (26)
Nimeni/Ion nimic nu £- ar face fira pile. nobody/Ion nothing-FOC not you would do without connections 'There isn't a thing anyone/Ion would do for you if don't have connections.'
To conclude this section, negative constituent proposing is distinct from topicalization insofar as it is triggered by [Focus] association with [wh/neg] in C, as opposed to [Focus/tense] in T. The former configuration entails SAI, whereas the latter does not. In Romanian, [neg] features invariably associate with [tense], so focused negative constituents induce the same derivations as fronting to Focus in positive sentences, and as illustrated in (15).
6. INTERROGATIVE CLAUSES FPTz-phrases (in interrogatives and free relatives) have been characterized in the literature as intrinsically marked for Focus, so that vc/z-movement is movement to a Focus position (see Rochemont and Culicover 1990). Double specification of w/z-phrases for [wh] and [Focus] corresponds to double checking of the respective features specified on functional heads. While this requirement is universal, how it is implemented depends on language internal properties. Let us
Parameters for Focus in English and Romanian
285
follow the above mentioned studies by considering that only C heads carry [wh] features, and this property attracts w/z-phrases to the CP level. Then the parametric approach to Focus developed in this paper predicts that the movement of w/z-phrases must vary cross-linguistically according to the type of feature checking applying to the [wh] features (i.e., through overt movement or FF raising) and to the functional host for [Focus]. In particular, w/z-phrases check both features in C in English, as in (27a), whereas they implement dissociated checking in Romanian, as in (27b): [Focus] checks [Focus/tense] in T, while [wh] undergoes further raising to check the [wh] of C. Thus, (27) represents the cross-linguistic variation induced by the focus parameter in interrogative clauses.
(27) a.
CP
b.
CP
Spec Wz-phrase C [Focus/wh]
Different locations (either C or T) for feature checking in (27) entail variation in the structure of interrogative clauses. Crucially, the configuration in (27) shows that the features of T affect the \vhmovement only in Romanian, not in English. Distribution of [Focus] and [wh] features in Romanian between different functional heads imposes dissociated checking, which may, theoretically, result in two positions for the H'/z-phrase: SpecTP or SpecCP. The w/z-phrase may land in SpecTP, after checking the [Focus] feature, with further (FF) raising to C, to check the [wh] feature (that is, only the [wh] feature of the constituent moves to C, instead of the whole constituent to SpecCP). This possibility was presented in the marked example in (7b), repeated for convenience. (7)
b.
(Oare)Ion/?ec/ne invitase? Q Ion /7e-whom invited 'Who was the one Ion had invited?'
However, the unmarked interrogatives present obligatory movement of the constituent to SpecCP, via SpecTP, as in (27b), as opposed to FF raising from T to C. Therefore, we witness an
286 Comparative studies in Romanian syntax interaction be I ween the option for constituent or FF raising on one hand, and the option for hosting [Focus] features on the other hand (i. e., C or T)11. For Romanian, the conclusion is that w/z-phrases with sentential scope check the [Focus/tense] feature through obligatory constituent raising to Spec,TP, with further raising to check the [wh] feature of C. The latter may occur through constituent raising to Spec,CP or FF raising to C. The contrast between the checking configurations in structures with w/z-phrases in English and Romanian, as represented in (27), accounts for the following cross-linguistic variation: (i) banning of wh in situ in Romanian, as in (28a), since [Focus/tense] in T is computed only overtly. (28)
a.
*A plecat unde? has left where
(ii) checking of [Focus/tense] through overt movement also accounts for obligatory movement of all the w/i-phrases in multiple questions (28b). Presumably, C hi (28b) projects a multiple Spec structure for the two w/z-phrases12. (28) b.
Cine unde a plecat? who where has left 'Who left where?'
(iii) V2 effects in embedded interrogatives (28c), which preserve the structure in (27b). Locality constraints on chains disallow lexical material between SpecTP and SpecCP when movement is cyclical between these two positions, as hi (27b). However, this locality constraint does not hold when the w/i-phrase moves directly from its base-generated position to SpecCP, as in the English equivalent structures: (28)
c.
Nu §tiam unde (*Ion) a plecat (Ion), not knew where Ion has gone Ion 'I did not know where John had gone.'
"Languages in which the interrogative clauses have the pattern in (27b) may present further cross-linguistic variation, according to the strength of [wh] in C: strong [wh] features attract overt movement of the wA-phrase to SpecCP, whereas weak [wh] features allow the w/j-phrase to land in SpecTP, as in (7b). Hungarian seems to adopt this latter pattern, so that both iv/i-phrases and focused constituents land in the same structural position (compare (7a, b)) before Spell Out. 12 C Rudin (1988) considers that the two w/i-phrases share the same SpecCP.
Parameters for Focus in English and Romanian
287
The set of properties illustrated in (28) have the opposite value in English, where w/i-phrases do not check two functional heads. Thus, English allows for w/i-phrases in situ, disallows multiple whmovement, and does not display V2 in embedded interrogatives.
s. CONCLUSIONS In the framework of the Minimalist Program, this paper argues that [Focus] features surface in the grammar in conjunction with semantically related formal features, in particular, [wh] and [tense]. Cross-linguistic variation arises from the parametric choice of the functional host for [Focus], that is, C with [wh], or T with [tense]. This hypothesis is tested on English and Romanian, with systematic contrasts in pre-verbal Focus constructions which can be reduced to parameter settings, opposing the [Focus/wh] option in English to the [Focus/tense] option in Romanian. Subsequent application of the parametrized pre-verbal Focus analysis leads to a uniform account for crosslinguistic variation in a variety of structures involving w/i-movement. It appears that pre-verbal Focus constructions in English and Romanian rely on similar syntactic structures with multiple Specs, as in (23) and (15), needed for checking on heads with two sets of features. Variation occurs in the distribution of the two-feature set, specified on C in English and on T in Romanian. This follows from language internal properties, in particular, the properties of scope positions and of A'-chains hi the two languages. Since fronting to Focus is considered a form of w/i-movement, the multiple Spec structures in (15) and (23) have consequences for interrogative clauses and constructions with negative constituent preposing. Fronting of negative constituents in Focus constructions confirms the contrastive parametric setting as it targets SpecCP in English versus SpecTP in Romanian. Also, w/z-phrases undergo overt movement to SpecCP in one step in English, as in (27a), but in two steps in Romanian, where they transit through SpecTP to check the [Focus] feature, as in (27b). Important cross-linguistic variation follows from the different checking procedure hi structure with whphrases. For example, wh in situ, V2 effects hi embedded w/z-interrogatrves, and multiple whmovement yield contrastive results in the two languages. While arguing for opposite parameter settings hi pre-verbal Focus constructions, this analysis leaves room for internal variation: both parametric options may occur in a language, according to the structural context. Topicalization illustrates this possibility for English, since Select and Merge of [Focus/tense] leads to successful derivations in restricted contexts, as hi (21), which replicates the Romanian structure in (15). However, interacting constraints prompt the [Focus/wh] option as the most productive hi this language. The empirical data analyzed hi this paper point to the fact that [Focus] features become syntactic elements only in conjunction with formal features, that is, [wh] or [tense], which are semantically related to [Focus]. This approach is in line with the theoretical assumptions of the
288 Comparative studies in Romanian syntax Minimalist Program which restrict syntactic relevance to formal features.
APPENDIX: NOTES ON THE QUESTION MARKER OARE The morpheme oare has optional occurrences and seems to carry reinforcing interrogative features. Thus, at first sight, oare enhances any question, be it a yes-no question, as in (la), or a whquestion, as in (Ib). (1)
a.
b.
(Oare) nimeni n - o va invita pe Maria? Q nobody not her will invite pe-Maria 'Will anybody invite Mary?' Unde (oare) se va duce Maria? where Q REFL will go Maria 'Where will Mary go?'
The placement ofoare in the clause seems quite flexible, as shown in (2), although the intonation on oare is not equally neutral in all the positions. (2)
(oare) Maria (oare) a (*oare) pus (oare) scrisorile (oare) la (*oare) po§ta (oare)? Q Maria Q has Q put Q letters-the Q at Q mail Q 'Did Maria mail the letters (I wonder)?'
As the word order in (2) indicates, the only positions denied to oare are those which are also denied to full fledged adverbs. Compare the word orders in (2) and (3), while keeping in mind that the intonation on the adverb also vary considerably from one position to another. (3)
(ieri)
Maria (ieri)
a(*ieri)
pus(ieri)
yesterday Maria yesterday has yesterday put yesterday scrisorile (ieri) la (*ieri) po§ta (ieri)? letters-the yesterday at yesterday mail yesterday 'Has Maria mailed the letters yesterday?' On the basis of (2) and (3), one might conclude that oare is an adverb. However, the syntactic behavior ofoare is not typical of adverbs. First, oare cannot take modifiers, whereas adverbs do, as shown in (4). (4)
a.
*mai/*prea/*cam/*abia oare
Parameters for Focus in English and Romanian
b.
289
more/ too/ quite/ hardly Q mai/ cam mereu; mai/ prea mult; more/quite frequently; more/quite much; mai/ abia acum; mai/abia ieri more/hardly now; more/hardly yesterday 13
Second, oare cannot cooccur with the lexical complementizers ca, ca 'that', whereas adverbs do, as shown hi (5). (5)
a.
b.
Mi-a spus ca (ieri/ *oare) a vazut-o (ieri/*oare). me has said that yesterday/Q has seen- her yesterday/Q 'He told me that he had seen her yesterday.' A§ fi preferat ca (maine/ *oare) Maria sa vina mai devreme. would be preferred that tomorrow Q Maria SA come more early 'I would have preferred for Mary to come earlier tomorrow'
Finally, oare may occur between a w/i-phrase and the verb, as in (6a), a position from which adverbs and other lexical phrases are excluded, as shown in (6b). (6)
a.
b.
Pecine oare invitase Maria ieri? pe whom Q invited Maria yesterday 'Whom had Maria invited yesterday?' Pe cine (*ieri/ *Maria) invitase (Maria)? pe whom yesterday/Maria invited Maria
The positional contrast in (6) may lead one to think that oare may be a short/clitic adverb, which cliticizes on V, yielding an [adv-V] complex. That is, short adverbs, such as mai 'more', abia 'hardly' in (7) do not adjoin to a maximal projection, but only to heads; hence, they do not affect the locality constraint on the w/z-chain in the way full fledged adverbs do. (7)
a.
Pe cine mai/ abia invitase Maria? /je-whom more/hardly invited Maria 'Whom else did Maria invite?//Whom has Maria just invited?'
However, there are grounds for denying oare a clitic adverb status. First, short clitics do not have the same distribution as oare, as (7b) shows when compared to (2), repeated here for convenience. 13
Cornilescu (present volume) treats abia 'hardly' as a full fledged adverb in the examples (19) to (21).
290 Comparative studies in Romanian syntax (7)
(2)
b.
(*mai) Maria (*mai) a (mai) pus (*mai) more Maria more has more put more scrisorile (*mai) la (*mai) po§ta (*mai)? letters-the more at more mail more 'Did Maria mail the letters at all?' (oare) Maria (oare) a (*oare) pus (oare) scrisorile (oare) la (*oare) po§ta (oare)? Q Maria Q has Q put Q letters-the Q at Q mail Q 'Did Maria mail the letters (I wonder)?'
Second, short/clitic adverbs are obligatorily adjacent to the inflected verb, whereas oare can be separated from the verb by other constituents, as shown in (8). (8)
a.
b.
Oare pe cine a invitat Maria? Q pe whom has invited Maria 'Whom did Maria invite?' Oare mdine vrea sa vina? Q tomorrow-Foe wants SA come 'Is it tomorrow he wants to come?'
In (8) oare is separated from the verb by a w/i-phrase or a constituent in Focus. The word order in (8) recalls the structures discussed in the section 2.2. of this paper, where some w/z-phrase were shown to land in SpecTP. So far, the tests in (4) to (6) have confirmed that oare cannot be a full fledged adverb, whereas the tests in (7) and (8) demonstrated that it cannot be a short/clitic adverb either. By elimination, oare must be located in a functional head. The definition of oare as a head would also be compatible with the definition of its Bulgarian counterparts (dali, -If), as question morphemes base-generated in C (Rivero 1993; Rudin 1990-91). If this analysis is on the right track, then the examples in (4) to (7) are easily accounted for, while the distribution of oare in (2) calls for an explanation. Base-generation of oare in C would allow w/i-phrases to land in SpecCP, as in (6), because oare in C does not interfere with the locality condition on the w/z-chain headed from a Spec (i.e. SpecCP) position. This explanation holds not only for root clauses, as in (6), but also for embedded w/j-clauses, as in (9). (9)
Ma intreb cu cine oare so fi casatorit Ion? REFL ask with whom Q REFL will be married Ion 'I wonder whom Ion has married?'
Parameters for Focus in English and Romanian
291
However, base-generation of oare in C raises two empirical questions: (a) why does oare surface in so many positions in the sentence, as seen in (2); and (b) why does oare appear in indirect interrogatives with daca 'if, as in (12)? The explanation proposed here relies on the feature specifications of oare, in the pre-syntactic component, which will attract various constituents to Speooare for checking purposes, once oare is merged in the syntax. I propose that in (2) oare is merged in C, and carries a complex [qu/X] feature. The phrases attracted to SpecCP to check this feature vary in type and complexity, that is, w/2-phrases, DPs, VPs, whole sentences. This definition is compatible with the semantic value of oare, which feeds more than strictly interrogative features to the interpretation. By itself, as in the dialogue in (10), oare conveys the notion of doubt and/or irony, while it could also question the previous statement. (10)
-Ionsa hotaratsase apuce delucru. Ion REFL-has decided SA REFL start-SUB J. 3 so of work 'Ion has decided to start working.' - Oare?/Oare! 'Is that right?/No kidding!/! wonder.'
In (10), the interpretation of oare could be interrogative (e.g,. 'Is it?') but also purely exclamative (e.g. 'Right!') or dubitative (e.g., 'Iwonder'), depending on the ascending or descending intonation on the morpheme. Then, by itself, oare is not an interrogative element, but rather a carrier of some discourse "affective" features. These features are added to the [+qu] value when oare is merged in an interrogative syntactic environment. Hence, it is reasonable to define oare as a discourse element underspecified for [qu], this underlying feature being substantiated through the association of oare with a syntactic carrier of [+qu] features. The association of oare with a [+qu] head triggers a free riding association of the "affective" features of oare with the same head, so that the discourse features become visible for syntactic computation in the same way in which the grammaticalization of Focus was described in this paper. The discourse features may be [Topic], [Focus] or others, yielding a complex feature [qu/X] with formal status. As C is the head associated with [+qu] features, we expect oare to merge with this head, or "parasite" on elements compatible with this head. Constituents in the SpecCP of oare must check not only the [+qu] feature of C, but also the discourse [X] feature of oare. These constituents may be generally characterized as the notional subject-of-predication, about which the rest of the oare-sentence is predicated. For example, the SpecCP of oare hosts a TopicP in (1 la), a focused PP in (1 Ib) and a TP in (1 Ic).
292 Comparative studies in Romanian syntax (11)
a.
b.
c.
[leri, in pare, copiilor] oare le- a yesterday in park children-the-DAT Q them-has dat tata debaut saunu? given father of drunk-SUP or not 'In the park, yesterday, did father give the children anything to drink or not?' Ion [la munte] oare voia sa piece sau la mare? Ion to mountain-FOC Q wanted SA go or to sea 'As for Ion, did he want to go to the mountains or to the sea?' [Mai vine] oare Ion? more comes Q Ion 'Is Ion still coming?'
Defining oare as a C morpheme with [qu/X] features explains its incompatibility with the lexical complementizers ca, ca 'that' on different grounds: these complementizers, illustrated in (5), have [-qu] features and would not attract oare. However, a lexical complementizer with adequate [+qu] features would allow for concurrence with oare. This is the case of daca 'if in (12). (12)
Ma intreb daca, (oare), n -ar fi (oare) mai bine (oare) sa plecam acasa. REFL ask if Q not would be Q more well Q SA go-suBJ. IPL home 'I wonder if it wouldn't be better to go home.'
In (12) daca 'if checks the [qu] feature of C, and "feeds" the underspecified [qu] feature of oare, yielding a recursive CP projection, as in (13).
In (13), the oare segment is selected and licensed by the complementizer daca 'if. In its in situ C position, oare has its discourse features checked by the constituents moved to SpecCP2, which may be: the NegP n-arfi 'wouldn't be' or the entire clause (TP) n-arfi mai bine 'wouldn't it be better'. Alternatively, SpecCP2 may be empty; in this case, the discourse features of oare must raise to daca for checking in LF. The latter option corresponds to adjacency in word order between these two C elements, accompanied by intonational breaks around oare, graphically represented through commas in (12).
Parameters for Focus in English and Romanian
293
To sum up this analysis, the morpheme oare is not an interrogative but a discourse marker, h'censed in syntax by the [+qu] features of C. The head C containing oare carries a complex [qu/X] feature, where X stands for "affective" discourse features (e.g. Topic, Focus, Doubt etc.). Since oare has the capacity to grammaticalize the discourse features, it yields derivations with SpecCP attracting various constituents which can check the complex [qu/X] feature. These configurations occur in any environment with C [+qu], including root and embedded clauses, yes-no questions and whinterrogatives.
REFERENCES Brody, M. (1990). Some remarks on the Focus field in Hungarian. UCL Working Papers in Linguistics 2, 201-225. Chomsky, N. (1977). On Wh-Movement. In: Formal Syntax (P.W. Culicover, T. Wasow and A. Akmajian, eds.), pp. 71-132. Academic Press, New York. Chomsky, N. (1986). Barriers. MIT Press, Cambridge, Mass. Chomsky, N. (1995). The Minimalist Program. MIT Press, Cambridge, Mass. Collins, C. (1997). Local Economy. MIT Press, Cambridge, Mass. Dobrovie-Sorin, C. (1990). Clitic doubling, wh-movement, and quantification in Romanian". Linguistic Inquiry 21, 351-397. Dobrovie-Sorin, C. (1994). The Syntax of Romanian. Comparative Studies in Romance. Mouton de Gruyter, New York. Drubig, H. B. (1992). On Topicalization and Inversion. Arbeitspapiere des Sonderforschungsbereichs 340, Tuebingen. Gierling, D. (1996). Further parallels between clitic doubling and scrambling. In: Proceedings of the 13th Eastern States Conference on Linguistics '96 (A. D. Green and V. Motapanyane, eds.), pp. 113-123. Cornell University, CLC Publications, Ithaca, NY. Gierling, D.(1997). Clitic doubling, Specificity and Focus in Romanian. In: Clitics, pronouns and movement (J. Black and V. Motapanyane, eds.), pp. 63-85. John Benjamins, Amsterdam. Haegeman, L. (1995). The syntax of Negation. Cambridge University Press, Cambridge. Horvath, J. (1995). Structural Focus, structural Case, and the notion of Feature-Agreement. In: Discourse configurational languages (K. Kiss, ed.), pp. 28-64. Oxford University Press, New York. Izvorski, R. (1994). Yes-no questions in Bulgarian: Implications for phrase structure. Paper presented at the 9th Conference on Balkan and South Slavic Linguistics, Literature and Folklore, Bloomington, Indiana.
294 Comparative studies in Romanian syntax Kayne, R. (1994). The antisymmetry of syntax. MIT Press, Cambridge, Mass. King, T. H. (1995). Configuring Topic and Focus in Russian. CSLI Publications, Stanford. Kiss, K. (1997). Discourse-configurationality in the languages of Europe. In: Eurotyp. Vol.2, Constituent order in the languages of Europe (A. Siewierska, ed.). Mouton de Gruyter, Berlin. Mueller, G. andW. Sternefeld. (1993). Improper movement and unambiguous binding. Linguistic Inquiry 24, 461-507. Ortiz de Urbina, J. (1989). Parameters in the grammar of Basque: A GB approach to Basque syntax. Foris, Dordrecht. Motapanyane, V. (1994). An A-position for Romanian subjects. Linguistic Inquiry 25, 729-734. Motapanyane, V. (1997). Preverbal Focus in Bulgarian. Journal of Slavic Linguistics 5, 265-299. Rivero, M-L. (1993 ). Bulgarian and Serbo-Croatian Yes-No Questions: V-raising to -LI versus -LI hopping. Linguistic Inquiry 24, 567-575. Rizzi, L. (1990). Speculations on Verb Second. In: Grammar in Progress: GLOW essays for Henk van Riemsdijk (J. Mascaro and M. Nespor, eds.), pp. 375-386. Foris, Dordrecht. Rochemont, M. (1978). A theory of stylistic rules in English. Ph.D. Dissertation, University of Massachusetts. Rochemont, M. (1986). Focus in generative grammar. John Benjamins, Amsterdam. Rochemont, M. and P. Culicover. (1990). English Focus constructions and the theory of grammar. Cambridge University Press, Cambridge. Rudin, C. (1988). On multiple questions and multiple wh-fronting. Natural Languages and Linguistic Theory 6, 445-501. Rudin, C. (1993). On Focus position and Focus marking in Bulgarian questions. Paper presented at the Formal Linguistic Society of Mid-America IV, Iowa.
The syntax and morphology of Romanian imperatives
295
THE SYNTAX AND MORPHOLOGY OF ROMANIAN IMPERATIVES * Mihaela Pirvulescu and Yves Roberge, University of Toronto
1. INTRODUCTION Various works in the Government and Binding framework establish a close relationship between surface word order, (inflectional) affix order and the syntactic structure of the corresponding sentence such that the surface order is taken to reflect the presence and position of lexical and functional categories in the structure. In recent years, this approach has been applied to imperative verbal morphology in order to examine the syntax of imperative constructions. The aim of this paper is to offer a detailed description and analysis of the verbal morphology of Romanian imperatives, and of the -e/-i alternation found on a certain class of verbs and which appears to be based in (surface) transitivity alternations. We first describe the facts in the larger context of a discussion of Romanian verbal morphology and then consider their relevance for the syntax of imperative constructions in the Minimalist framework (Chomsky 1995) with particular emphasis on the transitivity issue. We argue that most morphosyntactic properties of the Romanian imperative constructions follow from Tense considerations.
2. ROMANIAN VERBAL MORPHOLOGY 2.1. General description Romanian has five personal moods (indicative, subjunctive, conditional, imperative and
* We would like to thank David Heap and an anonymous reviewer for their comments on earlier drafts of this paper. Funding for this work was provided in part by the Social Sciences and'Humanities Research Council of Canada (Pirvulescu 752-97-2104 and Roberge 410-94-028).
296 Comparative studies in Romanian syntax presumptive) and four non-personal moods (infinitive, past participle, gerund and supine). The inflection can be realized by synthetic or analytic means: ven
-ea come -THEMATIC VOWEL 'I was coming.' am venit
(1)
b.
AUX
-m -1SG
COme-PAST PARTICIPLE
'I came.' The general structure of synthetic formations is the following: (2) Root + Root extension/Thematic vowel + Tense + Number-Person The simplest classification of Romanian verbs is in five classes determined by the infinitive endings. The infinitive is made of a particle a 'to' and five different thematic vowels which attach to the radical:
(3) I -a a minca 'to eat'
II
III
rv
-ea a tacea 'to shut up'
-e a merge
-i a fugi
'to go'
'to run'
V -f a cohort 'to descend'
2.2. Imperative morphology Romanian has two imperative forms, 2nd singular and 2nd plural: Table 1. Romanian imperative forms SING
PL
affirmative: 'sing!' canta! (~3so indicative) cantaji! (~2PL indicative)
negative: 'don't sing!' nu canta! (-infinitive) nu cantati! (~2PL indicative)
In the affirmative, the imperative uses forms from the present of the indicative paradigm: for singular, the 3rd person singular of the present of the indicative and for the plural, the 2nd person
The syntax and morphology of Romanian imperatives 297 plural of the present indictive.1 In the negative, the imperative uses the infinitive forms for the singular and the 2nd person plural of the present indicative for the plural forms.
2.3. The morphological marking of imperative singular Traditionally, imperative 2nd person singular is said to be homonymous with the 3rd person form of the present indicative. The analysis of the 3rd person singular of the present indicative is, in general, the following (see ffiescu 1983, lordan et al. 1967, Irimia 1976, GuJu-Romalo 1985): (4)
Root + -eaz + lucr -eaz 'He works.'
number-person -a
However, this analysis is challenged in Diescu (1991) where she considers the 3rd person singular to be zero marked. The ending is considered a thematic vowel. In this paper we will adopt the first type of analysis: the 3rd person singular of the present indicative is positively marked for person and number. The affixes are the following: (5)
-a for the verbs in -a and some verbs in -i (strong verbs) and in -/'; -e for the other verbs (the verbs in -ea, -e, -i, -i).
Arguments in support of this view can be found. First, that the 3rd person singular ending is not a thematic vowel seems clear when we look at the verbs in -/ or -i since it does not correspond to the thematic vowel -i/-f used hi the infinitive and plural forms. Compare (6a) with (6b). (6)
a. b.
a ven/7 'to come' el vine 'he comes'
noi ven/m 'we come'
a hotarf / 'to decide' el hotara§te 'he decides'
noi hotardm 'we decide'
Second, the affixes for number-person do not bear any stress, while thematic vowels are stressed. The stressed vowel is underlined in the examples below:
1 An anonymous reviewer points out that since indicative and subjunctive morphology are distinct only in the 3rd person, imperative morphology could in fact be based on the subjunctive. While this is true, it does not seem to affect directly the analysis that is proposed here.
298 Comparative studies in Romanian syntax (7)
el vme 'he comes'
noi ven/m 'we come'
Third, thematic vowels are in complementary distribution with -ez/-esc sequences (-ezl-esc mark the singular and the 3rd person plural, while the thematic vowel marks 1st and 2nd person plural): (8)
el lucreaza 'he works'
noi lucram 'we work'
It can thus be assumed that the imperative singular has the same affix as the 3rd person singular of the present indicative.
2.4. The alternation in the singular endings In (9) we present the Romanian imperative endings for the verb classes introduced above. (9) class SING
I (-a) (eaz)a
H ffl IV (-ea) (-e) (-i) -e/-i -e/-i -(esc)-e/-i
V (-1) (-asc)-a/-e
For the 2nd person singular we find two sets of endings: (a) endings which are homonymous with the 3rd person singular of the present indicative, -(eaz)a,-(esc/asc)e; (b) and an ending which is identical to the 2nd person singular of the present of indicative, -/. Each set of endings distribute according to the verb class, but the choice between the two sets depends, for some conjugations, on the verb's transitivity or intransitivity (see GLR 1966, Gu{u-Romalo 1989). The alternation concerns the endings -e and -/ and appears for verbs in classes n, ffl and IV: transitive verbs take -e whereas intransitive verbs take -/. This -e/-i alternation is illustrated in the following examples:
(10)
class n SING
tad 'shut up'
m
rv
scoate/mergi dormi/inghite 'take out/go' 'sleep/swallow'
Moreover, the same verb may take either -e or -/' depending on whether it is transitive or intransitive:
The syntax and morphology of Romanian imperatives (11)
infinitive a arde to burn
transitive arde-l!
299
intransitive arzi!2
burn-it!
bum (you)!
2.4.1. The diachrony of Romanian imperative forms. In Old Romanian (from around the Vth to the EX* century), for the first three classes, the form of 2nd person singular imperative was identical to the 3rd person indicative. For the fourth class, the imperative singular was identical with the 2nd person indicative. (12)
Verb classes in Old Romanian
i
ii
m
rv
-a
-e
-e
-i
The imperative ending -/' of the intransitive verbs of conjugations n and HI in Modern Romanian is said to derive from the original -e by a phonetic process of raising of the final unstressed -e. The transitive verbs in -e were exempt from this change because of their frequent use with a consonantal postverbal clitic. For conjugation IV, transitive verbs which originally had -/ in the second person singular, took, by analogy with classes n and HI, the -e ending. Intransitive verbs in -/ kept their ending (see GLR 1966, Rosetti et al. 1969). 2.4.2. Previous explanations. Synchronically, the -/ ending characterizes, according to Irimia (1976), disyllabic verbs which have a radical ending in an affricate or a sonorant. However, there are other verbs which take the -/ ending (verbs with the radical in d, b, m, s, i). The description of the alternation between -e and -/' relies on the pronominal enclitic: the -e is said to be conditioned by the presence of the pronominal enclitic (Diescu 1991, Lombard and Gadei 1981). There are 13 verbs in this situation (see Appendix) and, according to Lombard and Gadei (1981) a verb which normally would take -/, takes -e when it forms a phonetic word with the following clitic: (13)
arzi\ burn
-*
arde-l\ burn it
The alternation is therefore assumed to have a phonological status, similar to other
2
It is widely accepted that the change from -d- to -z- is not morphological but corresponds rather to one of the many changes affecting the final consonant of verbal roots.
300 Comparative studies in Romanian syntax phonological processes occurring at the end of the word.3 The distribution of the two endings would be the following:
(14)
[ j ]# arzil burn
[e] + C# arde-l! burn-it
However, this account poses the following problem: we can find constructions with a direct object NP instead of a clitic which also take the ending -e as in (15a, b). The example in (15c) shows further that the -/ ending does not simply occur before a phonological pause: (15)
a.
b.
c.
Arzi! burn-AFF 'Burn!' Fierbi! boil-AFF 'Boil!' Na, fierbi so boil-AFF 'Boil now in sorrow!'
Arde toate documentele! burn-AFF all documents-the 'Bum all the documents!' Fierbe macaroanele! boil-AFF pasta-the 'Boil the pasta!' acuma de necaz! now of sorrow
These examples show that it would be problematic to assume that the alternation is exclusively phonological in nature.
3. ANALYSIS 3.1. Previous analysis Rivero (1994) presents an analysis of the Romanian imperative based mainly on the assumption that the illocutionary force of imperatives is to be accounted for through the presence of an imperative affix in C°. She proposes the syntactic structure in (16b) for the Romanian imperative in (16a).
' As in: iubiji!
-»
iubip-1!
[jubitS'] -*
[jubitSil] (short [i] becomes syUabic if it is foUowed by the clitic)
love-2PL
love-2PL + CLITIC
The syntax and morphology of Romanian imperatives 301 (16)
a. b-
Prinde-o! catch-it [CP C [T/AgrP CL [T/AgrP T/Agr [w VO]]]]]
In (16b), V moves to T/Agr to pick up a person feature specification and then to C° to give morphological support to the imperative affix. Following Joseph and Philippaki-Wartburton (1987), Rivero assumes a distinction between true and surrogate imperatives according to which true imperative verbs are usually restricted to 2nd person and are identified by a morphology not shared by the same person in any other tense in the system. Surrogate imperative verb forms, on the other hand, have a morphology identical to the same person of another tense and need not be restricted as to person. This distinction is illustrated in (17). (17) True: Surrogate:
a. b.
Vorbe$te\ (2SG.IMPER = 3SG.IND.PRES) speak-SG Vorbip\ (2PL.IMPER = 2PL.IND.PRES) speak-PL
The distinction appears relevant to the syntax of imperatives and would seem to constitute evidence in favor of the view that morphology and syntax interact in a non-trivial manner. This is supported by the fact that verb movement in true imperatives belongs to the same type of non-finite raising (Long Head Movement) as participles and gerunds. The verb in surrogate imperatives, on the other hand, behaves like a regular tensed verb. This accounts for the following properties: a) true imperatives are reserved for root environments; b) true imperatives cannot be negated, while surrogate imperatives are unrestricted in this respect; c) in true imperative constructions clitics are obligatorily postverbal. Dobrovie-Sorin (1994) considers the affirmative imperative as a construction which shows Vpreposing. Therefore, in the structure hi (18), V-preposing moves the verb to I and then [V-I] moves to C, stranding the clitic.
(18)
302 Comparative studies in Romanian syntax This operation explains the post-verbal position of the clitic in imperatives. While accounting for the position of clitics in non-negated imperatives, this analysis cannot straightforwardly be extended to negated imperatives, in which the clitics are pre-verbal. It could be that in the negative construction, V-preposing cannot apply; this would account for the pre-verbal position of the clitics. But if verb movement to C is motivated by checking requirements and is responsible for the illocutionary force, then the question is how one would account for the illocutionary force of negated imperative constructions? As far as we understand it, this analysis does not provide an answer. While maintaining a distinction between surrogate and true imperatives, Zanuttini (1997) proposes that it is purely morphological and is not to be attributed to different syntactic structures. For Zanuttini, besides the characteristics already mentioned, there is another major difference between true and surrogate imperatives. True imperatives exhibit poor morphological specifications consisting of the verbal root followed by the thematic vowel and, in some cases, by an Agreement morpheme. There is no morphological evidence that they are specified for Tense or mood. The focus of Zanuttini's paper is an investigation of the widespread incompatibility found in Romance languages between true imperatives and the negation marker. When negated, imperative verbs revert to a different morphology, normally infinitive, indicative or subjunctive. In order to account for this incompatibility (and the position of clitics) she proposes the structure in (19) for a negated imperative which uses an infinitive morphology (as is the case hi Romanian).4 Here the negation node licenses an auxiliary, not the infinitive, and the clitic moves to the auxiliary position: (19)
neg
(empty auxiliary)
infinitive
clitic
The auxiliary is the realization of a functional category and the contrast illustrated in (20) suggests that this functional category is a Mood Phrase (MP). (20)
1
* OK
neg + true imperatives (mail verb) neg + true imperatives (aux) + infinitive (in various Italian dialects)
This structure seems to be supported by Friulian which uses the infinitive preceded by an overt auxiliary: (i) No sta (a) crodi! (Friulian) Neg stay a to-believe 'Don't believe that!'
The syntax and morphology of Romanian imperatives 303 The syntactic structure of imperatives is represented in (21), with two functional categories: CP which hosts the illocutionary force regardless of the morphological marking (true vs. surrogate) and MP which accounts for the different behavior of surrogate vs. true imperatives in negative structures.
(21)
What (21) shows is that while all imperatives must check the illocutionary feature in C, main verbs from the true imperative paradigm cannot check mood features because of their poor morphological specification (no marking for Tense, aspect or mood). Since the negative marker requires that the feature of the head of MP be checked, true imperatives are not compatible with negation. To sum up so far, most analyses of the imperative construction rely on the existence of a functional category - the complementizer C - which bears a feature corresponding to the illocutionary force. The properties of imperatives (negation, clitics and the distinction between surrogate and true imperatives) follow from verb movement to C to check the imperative feature. Dobrovie-Serin's analysis raises difficult issues and a critique of some of Rivero's positions can be found in Zanuttini (1997). In our analysis, we adopt Zanuttini's proposal with respect to the poor morphological marking of imperatives but we will extend this impoverished structure to the syntactic level (see also Zanuttini 1994).
304 Comparative studies in Romanian syntax 3.2. Proposal 3.2.1. The syntactic structure of imperative constructions. We adopt the general approach to imperatives outlined in Pirvulescu and Roberge (1998). The analysis is based on the following observations. Generally, imperatives represent a mood and a specific illocutionary force; however, we show that a distinction should be drawn between the two. Since almost any linguistic expression, given an appropriate context, can be interpreted as an "order", the injunctive illocutionary force of the imperative should not be attributed to the presence of a feature hi C or in any other functional category. Rather, we postulate that some property of constructions with injunctive force is responsible for the injunctive reading in an appropriate context. The property or properties in question can vary from one construction to the next.5 Adapted to sentences, this approach simply states that a numeration lacking certain categories could lead to a derivation interpretable as injunctive at LF. Non-negated imperative constructions are very close, at least superficially, to simple VP projections. Pirvulescu and Roberge (1998) thus propose that they correspond to derivations containing no Tense and Agr projections. This gives them their injunctive force. Morphologically, imperatives correspond to a conjugation contrast based on Mood. Following Curat (1991), we take Mood to be defined as a function of Tense such that a Mood is a representation of Tense. If Tense corresponds to a functional category in linguistic expressions we can establish a three way distinction : (a) no T node; (b) T specified for [+finite]; (c) T specified for [-finite]. Ignoring other categories but using C for an irrealis feature, we end up with (22).
(22)
a.
5 Curat (1991, 166) points out that whereas Garfow/'waiter!' can be used to call out to the person, Le garfon! 'the waiter!' cannot. In this case then, the lack of Determiner would lead to the injunctive reading.
The syntax and morphology of Romanian imperatives b.
305
TP
The (a, b and c) structures in (22) correspond to the subjunctive, indicative, and infinitive mood respectively. In (22d), V raises to the abstract light verb to check its V feature. This is the minimal structure assumed for non-negated imperatives. We concentrate next on one of the many consequences of this approach, namely the verbal morphology of imperatives.6 3.2.2. The default morphological marking. The simplest assumption consistent with our approach (impoverished syntactic representation) for the homonymy between the 2nd person singular imperative and the 3rd person singular present indicative would be to consider the imperative realization as a default marking in the Phonological Component7. The question we try to answer is why the imperative takes the 3rd person singular present indicative for its singular form and the 2nd person plural present indicative for its plural form and on what grounds those forms could be defined as default forms. Consider the inflection of the verb a gdndi, 'to think':8
6
The question of how clitics are morphologically attached to the verb in imperative constructions is irrelevant to our discussion. 7 The 2"4 person interpretation will be provided by LF mechanisms. * Certain verbs (a avea 'to have', afi 'to be', a duce 'to bring', a face 'to do' and a zice 'to say') have specific forms for the imperative, which are very short. These short forms can be negated even when they are true vs.
306 Comparative studies in Romanian syntax
imperfect
simple past
past perfect
gandesc9
gand-ea-m
gand-i-i
gand-i-se-m
gande§t-i
gand-ea-i gand-ea
gand-i-§i
gand-i-se-^
gand-ea-m
gand-i gand-i-ra-m
gand-i-se gand-i- se-ra-m
gand-i-Ji
gand-ea-{i
gand-i-ra-ji
gand-i-se-ra-Ji
gandesc
gand-ea-u
gand-i-ra
gand-i- se-ra
person present
1SG 2SG 3SG 1PL 2PL 3PL
gande§t-e gand-i-m
Subjunctive mood
Imperative mood
person present
person
present
1SG
sa gandesc
2so
gande§t-e!
2SG
sa gandest-i
2PL
gand-i-Ji!
3SG 1PL
sa gandeasc-a sa gand-i-m
Past Participle
gand-i-t
2PL
sa gand-i-ti
Gerund
gand-i-nd
3PL
sa gandeasc-a
Infinitive
a gand-i
These paradigms can be used to classify the singular affixes in the following way: -0
1 SG [present indicative and present subjunctive] 3SG [imperfect, simple past, past perfect]
-i
3PL [present indicative] 1SG [simple past]
-e
2SG [present indicative, present subjunctive, imperfect] 3SG [present indicative]
-a
3SG [present subjunctive]
-§i
2SG [simple past, past perfect]
From these affixes, -p can easily be eliminated as an imperative morpheme because it carries Tense features, and the imperative is tenseless form. The same could be said of-a, which carries mood features. Of the remaining three affixes, the first two (-0 and -/) are candidates for default
surrogate imperatives. In the negated form, either the positive imperative or the infinitive can be used in free variation, as shown in (i) and (ii), taken from Pirvulescu (1998). (i) afi fii! nufii/nufi to be be! do not be! (ii) a face fa nu fa! / nu face! to do do! don't do! 9 To simplify the segmentation of the verbal forms we keep the sequence -esc-/-est- together with the root.
The syntax and morphology of Romanian imperatives 307 status (following Halle's 1997 approach to default specification). However, none is the correct choice for the imperative singular. To exclude them, we make use of the feature geometry system developed by Ritter (1997), which encodes the notion of markedness and the hierarchy of features. Also according to Ritter (1997), "some nodes in the morphological feature tree correspond to conceptual semantic notions". The Discourse Participant node determines person features while the Individuation node is responsible for number and gender features:10 (24)
Referring expression Discourse participant Speaker [present]
ISO: -0
(25)
Referring expression Individuation Minimal/ Group [past] [present]
3SG/PL: -0
(26) Referring expression Discourse Participant Speaker / Addressee [simple past] (27) Referring expression Individuation Minimal
2SG/1SG: -i Individuation Minimal 3so: -e
We follow Bonet (1991, 16) in establishing the markedness hierarchy: "The more complex in number of nodes the structure [...] is, the more marked it is." This definition suffices to ensure the selection of (27) over (26). In addition, notice that the affixes -/ and -0 depend on the Tense marking for establishing the specific value for the Discourse Participant node and for the Individuation node. For the plural, we have to explain why imperatives take the form of the 2nd person plural present indicative. We consider the plural form to be assigned by default as well. The plural affixes for the present of the indicative are the following:
10 We consider that there are two representations for the 0 morpheme, one for the 1" person and another one for the 3rd person.
308 Comparative studies in Romanian syntax (28)
-m -Ji -0
1PL (everywhere) 1SG (imperfect, past perfect) 2PL (everywhere) 3PL (present indicative)
Other plural morphemes are: -u (3rd person plural imperfect) and -ra (plural simple past and past perfect) but we exclude them because they must both bear Tense specification. The representations for the 1st and 2nd person plural are the following: (29) Referring expression Discourse Participant Speaker (30) Referring expression Discourse Participant Addressee
iPL/lSG: -m Individuation Group / Minimal [imperfect, past perfect] 2PL: -£ Individuation Group
Even though both representations seem equally marked, (29) needs to make reference to Tense in order to establish a specific value for the Individuation node. Choosing (30) over (29) follows from our analysis where no Tense node is present in the syntax. 3.2.3. The object of imperatives. We consider the issue of the internal argument of imperative verbs to be closely linked to the external argument question. Pirvulescu and Roberge (1998) argue that the subject in imperative constructions is the subject of the light verb and not that of the lexical verb. This is due to the lack of a Tense projection which, in turn, prevents assignment of a th-role to the external argument by the lexical verb. In imperative constructions then, the subject is provided by the structure independently of the lexical verb as though it were a small clause. It is further argued, based on extensive empirical evidence, that such a configuration provides an impoverished predication context for the lexical V which consequently triggers a shift towards the complement position. Assuming that predication obligatorily involves an argument and a predicate, imperative constructions need to project an internal argument.11 Therefore, the imperative structure will be as in (31).
11 In the case of unaccusative verbs, the internal argument will also act as the subject. More generally, it seems that what is needed is a complement of some sort. The complement sometimes needs a secondary predicate in order to be interpreted properly: (i) *Arrive e. (ii) Arrive [e on time].
The syntax and morphology of Romanian imperatives
309
Complement Given this, it must therefore be the case that the same structure applies whether the lexical V is transitive or intransitive. In other words, the sentence in (32a) does not correspond to the one in (32c) in terms of the interpretation assigned to the missing object. (32)
a. b. c. d.
I already ate. I already ate it. Eat! Eat it!
In (32a), the object is simply implicit and not projected; it does not have the same reading as (32b). In (32c), on the other hand, the complement is a projected empty object which corresponds to a particular something to be eaten; it has a reading similar to that of (32d). The discourse context ensures recoverability of this object. In a language like English, there is no distinction between the verb forms used in (32a) and (32b); this is also the case in Romanian as shown below with the verbs afierbe 'to boil' and a arde 'to burn'.
(33)
a.
b.
c. d.
Tu fierbi. you boil Tu fierbi macaroanele. you boil pasta-the 'You boil the pasta.' Tu arzi. you burn Tu arzi toate documentale. you burn all documents-the 'You burn all the documents.'
Similarly in English, the verbs in (32c) and (32d) are morphologically identical. In contrast, as
310 Comparative studies in Romanian syntax shown in (34), these are the contexts in which one finds the -el-i alternation for a significant number of Romanian verbs (see also the Appendix to this paper). (34)
a. b.
c. d.
Fierbi! boil Fierbe macaroanele! boil pasta-the 'Boil the pasta!' Arzi! burn Arde toate documentele! burn all documents-the 'Bum all the documents!'
We have already shown that this alternation is not phonological; it must therefore be morphosyntactic. Given the fact that imperative verbal endings do not correspond to subject/verb Agreement and Tense but are rather default realizations, then it is plausible to assume that the endings have become sensitive to the contents of the complement position.12 In the examples in (34), the endings do in fact approximate object-verb Agreement although we do not want to claim that they are.
4. CONCLUSION In this article, we have proposed an analysis of the morphosyntax of Romanian imperatives with an emphasis on verbal morphology. The inflectional paradigms of imperative verbs are construed as resulting from the assignment of default forms where the default forms are determined through feature geometry; among the competing affixes, only the ones that are not specified for Tense or those that are the least marked enter into the formation of imperative verb forms. The relevance of Tense in the determination of the default forms follows from the syntactic structure we have adopted for imperative constructions. Following Zanuttini (1997) and Pirvulescu and Roberge (1998), we take this structure to lack a Tense projection. As a further consequence of
12 That the endings are not based on subject Agreement in imperatives is supported by the fact that while negation, which is Tense dependent, does not affect the verb forms in the indicative, it does in the imperative which reverts to the infinitive form. (i) TV arzi/ Tu nu arzi. but (ii) Minf!/ Nu min$! 'Youburn.'/'Youdonotburn.' 'Lie!'/ 'Do not lie!' In the (ii) examples, the stress pattern (') marks the difference between the imperative and the infinitive.
The syntax and morphology of Romanian imperatives 311 the lack of Tense, we have shown that imperative constructions force the projection of an internal argument position and that this is reflected hi the -el-i alternation found in certain Romanian imperative verb forms.
APPENDIX Intransitive verbs which take the ending -/ for the 2nd person singular: 33 verbs. Underlined verbs have two forms in the imperative 2nd person singular, namely -e/-i. n (-ea) a tacea 'to keep quiet' a zdcea 'to lie' a cadea 'to fall' a §edea 'to sit' aparea 'to seem' a aparea 'to appear'
HI (-e) a rdde 'to laugh' a merge 'to go' a curge 'to flow' a ramdne 'to remain' apldnge 'to cry' a parcurge 'to cover' asuge 'to suck' a create 'to grow up' a ajunge 'to catch up' a fierbe 'to boil' aarde 'to burn' atrece 'to pass' a geme 'to moan'
IV (-//-/) a dormi 'to sleep' a adormi 'to fall asleep' a iep 'to go out' a muri 'to die' apieri 'to perish' a rasari 'to rise' a minfi 'to lie' asorbi 'to sip' a simp 'to feel' a sari 'to jump' a tresdri 'to start' afugi 'to run'
We found two intransitive verbs from the class I (-a): a sta 'to stay', a zbura 'to fly'. We also found four transitive verbs with the imperative 2nd person singular in -/:
ii
ni
rv
a vedea 'to see'
a crede 'to believe' a incepe 'to begin'
a auzi 'to hear'
SeeD/c/ronar invers, (1957), Graur (1968), lordan et al. (1967), Irimia (1976).
REFERENCES Beukema, F. and P. Coopmans (1989). A Government and Binding perspective on the
312 Comparative studies in Romanian syntax imperative in English. Journal of Linguistics 25, 417-436. Bonet, E. (1991). Morphology after syntax: pronominal clitics in Romance. Doctoral dissertation, Cambridge: Massachusetts Institute of Technology. Chomsky, N. (1995). The Mnimalist Program. MIT Press, Cambridge, Mass. Coteanu, I. et al. (1969). Istoria limbii romdne. Editura Academiei, Bucharest. Curat, H. (1991). Morphologic verbale et reference temporelle enfrancais moderne. Librairie Droz, Geneve. Dicponar invers (1957). Editura Academiei, Bucharest. Dobrovie-Sorin, C. (1994). The syntax ofromanian. Mouton de Gruyter, Berlin. (GLR) Gramatica limbii romdne (1966). Editura Academiei, Bucharest. Graur, Al. (1968). Tendinjele actuate ale limbii romdne. Editura §tiintifica, Bucharest. Gu^u-Romalc, V. (1985). Morphologia. In: Limba romdna contemporana (Ciompec, G. et al., eds.), pp. 86-235. Editura Didactica, Bucharest. GuJu-Romalo, V. (1989). La flexion. In: Lexikon der Romanistischen Linguistik, vol. HI (H. Gunter, M. Metzelthi and Ch. Schmitt, eds.), pp. 19-33. MaxNiemeyer Verlag, Tubingen. Halle, M. (1997). Distributed morphology: impoverishment and fission. In: Papers at the interface. MIT Working Papers in Linguistics 30 (B. Bruening, Y. Kang and M. McGinnis, eds.), pp. 425-449. MIT Department of Linguistics, Cambridge, Mass, fliescu, M. (1983). Pour une typologie de la morphologic du verbe rornan. Analyse du verbe roumain. Revue de Linguistique Romane 47, 129-141. Diescu, M. & L. Mourin (1991). Typologie de la morphologie verbale romane. Amoe, Innsbruck. Irimia, D. (1976). Structura gramaticalaa limbii romdne: verbul. Editura Junimea, lassy. lordan I. et al. (1967). Structura morfologicaa limbii romdne. Editura §tiinjifica, Bucharest. Joseph, B. and I. Philippaki-Wartburton. (1987). Modern Greek. CroomHelm, London. Lombard, A. & C. Gadei (1981). Dictionnaire morphologique de la langue roumaine. Editura Academiei, Bucharest. Pirvulescu, M. (1988) Le concept de paradigme dans la morphologie verbale. Manuscript, University of Toronto. Pirvulescu, M. and Y. Roberge (1998). Objects, and the structure of imperatives. Paper presented at LSRL 28, The Pennsylvania State University, April 1998. Ritter, E. (1997). Agreement in the Arabic prefix conjugation: evidence for a non-linear approach to person, number and gender features. In: Proceedings of the 1997 Annual Conference of the Canadian Linguistics Association (L. Blair, C. Bums and L. Rowsell, eds.), pp. 19-33. University of Calgary: Department of Linguistics. Rivero, M.-L. (1994). Clause structure and V-movement in the languages of the Balkans. Natural Language and Linguistic Theory 12, 63-120.
The syntax and morphology of Romanian imperatives Rosetti, Al. et al. (1969). Istoria limbii romdne. Editura Academiei, Bucharest. Zanuttini, R. (1994). Speculations on negative imperatives. Rivista di linguistica 6(1), 67-89. Zanuttini, R. (1997). Negation and clausal structure. Oxford University Press, New York.
313
This page intentionally left blank
Bibliography
315
BIBLIOGRAPHY: GENERATIVE STUDIES ON ROMANIAN SYNTAX (from 1980 to 2000)
Alboiu, Gabriela (1999). Object raising in Romanian. Proceedings from the 1998 CLA Conference (Cahiers Linguistiques d'Ottawa), pp. 1-11. (1999). Contrastive Focus in Romanian. Bucharest Working Papers in Linguistics 1 (forthcoming). (1999). (De)-focusing and object raising in Romanian Canadian Journal of Linguistics 44(1) (forthcoming). (2000). Contrastive Focus and sentence-initial operators in Romanian. Proceedings from the 1999 CLA Conference (Cahiers Linguistiques d'Ottawa), pp. 1-12. Avram, Larisa (1992). A minimalist approach to subject positions in Romanian. Cahiers de linguistique theorique etappliquee 39, 121-128. (1994). A contrastive analysis of modals in English and Romanian. Revue roumaine de linguistique 39(3/4), 267-284. (1994) Auxiliary configurations in English and Romanian Revue roumaine de linguistique 39(5/6), 493-510. (1998). Romanian HAVE: Merge and Reconsider. Revue roumaine de linguistique 1-2, 65-92. (1999). To have or to be? In: Traiani Augusti vestigiapressa sequamur (M. Coene, W. de Mulder, P. Dendale andYves D'Hulst, eds.), pp. 371-392. Unipress, Padova. (1999). Auxiliaries and the structure of language. Editura Universitajii Bucure§ti, Bucharest. (1999). To have or to be?. De bezitsrelatie/Relatia de posesie, 373-391. (1999). Possessive clitics in Romanian. Bucharest Working Papers in Linguistics 1, 11-24. (1999). Masculine and feminine clitics. An acquisition account. Analele Universitapi Bucure§ti. Limbi strSne (forthcoming).
316 Comparative studies in Romanian syntax Avram, Larisa and Martine Coene (2000). Genitive/Dative clitics. FSSBL 3 (forthcoming). Bernstein, Judy (1993). Topics in the syntax of nominal structure across Romance. Doctoral dissertation, CUNY. (1997). Demonstratives and Reinforcers in Romance and Germanic Languages. Lingua 102, 87113. Campos, Hector (1995). Full and reduced clitics in Megleno-Romance. Probus 7, 247-278. Cinque, Guglielmo & Giuliana Giusti (eds.) (1995). Advances in Rumanian linguistics. John Benjamins, Amsterdam. Coene, Martine (1994). Some reflections on the position of the adjective in Romanian. Cahiers de linguistique theorique et appliquee 31, 9-29. (1999). A propos de la nature aspectuelle du groupe nominal a tete zero. In: Actes du colloque de syntaxe et semantique du groupe nominal (N. Flaux, ed.). Arras, (forthcoming). (1999). "Homo ellumbonum", ou existe-t-il un demonstratif provenant du Lathi 'ellum'?. Studii £7 cercetari lingvistice (forthcoming). (1999). Relative clauses with null antecedents and the case matching problem. In: Traiani Augusti vestigiapressa sequamur (M. Coene, W. de Mulder, P. Dendale andYves D'Hulst, eds.), pp. 243-266. Unipress, Padova. (2000). Dative/Genitive Clitics as Last Resort. In: Formal Syntax and Semantics of the Balkan Languages (M. Dimitrova-Vulchanova, ed.). (forthcoming). Coene, Martine; Walter de Mulder; Patrick Dendale and Yves D'Hulst (eds.) (1999). Traiani Augusti vestigia pressa sequamur. Studio linguistica in honorem Lilianae Tasmowski. Unipress, Padova. Coene, Martine; Yves D'Hulst and Liliane Tasmowski (1997). On the syntax of Romanian possessor phrases. Revue roumaine de linguistique 42 (3-4), 149-166. (1999). *Cel al meu versus Cel al studentului. On the intersection between possession, definiteness and differentiation. Bucharest Working Papers in Linguistics 1. (forthcoming).
Bibliography
317
Comorovski, Ileana (1986). Control and obviation in Romanian. In: Proceedings of the Second Eastern State Conference on Linguistics (S. Choi; D. Devitt; W. Janis; T. McCoy; Zhang-Zheng-Sheng, eds.), pp. 47-56. Ohio State University, Columbus. (1986). Multiple Wz-movement in Romanian. Linguistic Inquiry 17, 171-177. (1989). Discourse and the syntax of multiple constituent questions. Doctoral dissertation, Cornell University. Cornilescu, Alexandra (1980). Relative clauses in English and Romanian. Doctoral dissertation, University of Bucharest. (1981). Non-restrictive relative clauses: An essay in semantic description. Revue Roumaine de Linguistique 26 (1), 41-67. (1984). Recent developments in generative theory. The/»ro-drop parameter and the analysis of Romanian and English infinitive constructions. Revue Roumaine de Linguistique 29, 331-345. (1986). On the syntax and semantics of free relative clauses in English and Romanian. In: Linguistics across historical and geographical boundaries (D. Kastovsky and A. Szwedek, eds.), pp. 1165-1182. Mouton de Gruyter, Berlin. (1987). The assignment of the Romanian dative. Revue Roumaine de Linguistique 23 (3), 213224. (1988). The genitive in English and Rumanian. Analele Universitafii Bucuregti, 213-224. (1992). Remarks on the determiner system of Romanian: The demosntratives al and eel. Probus 4, 189-260. (1993). Notes on the structure of Romanian DP and the assignment of the genitive Case. University of Venice Working Papers in Linguistics 3 (2), 107-133. (1994). Remarks on the Romanian ordinal numeral. Towards a unitary description of phrases headed by 'AL'. Revue roumaine de linguistique 39, 303-337. (1995). Rumanian genitive constructions. In: Advances in Roumanian linguistics (G. Cinque and G. Giusti, eds.), pp. 1-54. John Benjamins, Amsterdam. (1995). Concepts of Modern Grammar. Bucharest University Press, Bucharest. (1999). The double subject construction and nominative case checking. Traiani Augusti vestigia pressa sequamur (M. Coene, W. de Mulder, P. Dendale andYves D'Hulst, eds.), pp. 267-284. Unipress, Padova. Cornilescu, Alexandra and Daniela Urdea (1987). Relajii §i elemente anaforice in limba romana: construe^ cu pronumele reflexive 'se' §i 'sine'. Limbaromdn£36, 105-122.
318 Comparative studies in Romanian syntax Dobrovie-Sorin, Carmen (1981). A propos de 1'interpretation des groupes nominaux indefinis dans les structures imperatives. Semantikos 5(1), 55-69. (1981). A propos de 1'interpretation des groupes nominaux indefinis dans les structures imperatives. Revue Roumaine de Linguistique 26(6), 565-577. (1982). Imperatifs, factivite et implication. Lingua 58, 168-189. (1983). Le sujet de 1'imperatif Histoire, epistemologie, langage 5(1), 129-134. (1984). Imperatifs conditionnels. Lefrancais moderne 1/2, 1-21. (1985). Actes de langage et theorie de I'enonciation. Paris: Universite de Paris 7, Collection E.R.A. 642. (1985). Redoublement clitique, relatives et interrogatives en roumain et espagnol. Linguisticae Investigations 9, 269-306. (1986). A propos du contraste entre le passif morphologique et se moyen dans les tours impersonnels. Linguisticae Investigations 10, 289-330. (1987). Syntaxe du roumain. Chames thematiques. Doctoral dissertation, Universite de Paris 7. (1988). A propos de la structure du groupe nominal en roumain. Rivista di grammatica generativa 12, 123-152. (1989). Mouvement \vh et relations de quantification. Actes de la table ronde sur la syntaxe comparee, Paris 1988. CNRS publications. (1989). Emploi du subjonctif et structure de la proposition en roumain. Revue des langues romanes 93(2), 373-400. (1989). Subject anaphors in Rumanian subjunctive clauses. InternationalJournal of Rumanian Studies 7. (1989). Principes generaux et parametrisation: la position typologique des relatives roumaines. Verbum 12(3), 263-280. (1990). Clitic doubling, w/j-movement, and quantification in Romanian. Linguistic Inquiry 21, 351-397. (1990). Auxiliaries and sentence structure in Romanian. In: Grammar in Progress. Glow Essays for Henk van Rif.msdijk (J. Mascaro and M. Nespor, eds.), pp. 101-111. Foris, Dordrecht. (1991). Mouvement w/z et relations de quantifications en roumain. In: Grammaire generative et syntaxe comparee (J. Gueron and J.-Y. Pollock, eds.), pp. 67-82. Editions du CNRS, Paris. (1992). A propos du se/si dh impersonnel en roumain, fran9ais et italien. In: Structure de la phrase et theorie du liage (H.-G. Obenauer and A. Zribi-Hertz, eds.), pp. 551-68. Presses Unrvershaires de Vincennes, Paris. (1994). The syntax of Romanian. Comparative studies in Romance. Mouton de Gruyter, Berlin. (1994). On impersonal se/si in Romanian, French and Italian. In: Paths Towards Universal
Bibliography
319
Grammar. Studies in Honor of Richard S. Kayne (G. Cinque, J. Koster, J.-Y. Pollock, L. Rizzi and R. Zanuttini, eds.), pp. 137-153. Georgetown University Press, Washington, D.C.. (1995). Syntactic configurations and reference: se/si in Romance. In: Grammatical theory and Romance languages (K. Zagona, ed.), pp. 73-86. John Benjamins, Amsterdam. (1995). Clitic clusters in Rumanian: Deriving linear order from hierarchical structure. In: Advances in Rumanian linguistics (G. Cinque and G. Giusti, eds.), pp. 55-82. John Benjamins, Amsterdam. (1995). Entre deux champs theoriques: la variation linguistique. In: Langues et langage. Melanges offerts a Antoine Culioli (J. Bouscaren; J.J. Franckel and S. Robert, eds.), pp. 73-83. PUF, Paris. (1995). On the denotation and scope of indefinites. University of Venice Working Papers in Linguistics 5 (1), 67-114. (1997). Auxiliaires et structure de la phrase en roumain. In: Actes du XIXe Congres International de Linguistique et Philologie Romanes (R. Lorenzo, ed.), pp. 1123-1134. Fundacion Pedro Barri de la Maza, La Coru. (1997). Types of predicates and the representation of existential readings. SALT Proceedings. Cornell University Press, Ithaca. (1997). Les themes entre la syntaxe et la structure de rinformation. Actes de la thematisation. Peter Lang, New York. (1998). Impersonal si constructions in Romance and the passrvization of intransitives. Linguistic Inquiry 29, 399-437. (1998). The typology of pronouns and the distinction between syntax and morphophonology. In: Clitics (Eurotype) (H. van Riemsdijk, ed.). Mouton de Gruyter, Berlin. (1998). Clitics across categories. In: Clitics (Eurotype) (H. van Riemsdijk, ed.). Mouton de Gruyter, Berlin. (1998).Head-to-head merge in Balkan subjunctives and locality. In: Comparative syntax of Balkan languages (B. Joseph; A. Ralfi and M.-L. Rivero, eds.). Oxford University Press, New York. (1998). Du genitif roumain a 1'etat construit. In: Traiani Augusti vestigia pressa sequamur (M. Coene, W. de Mulder, P. Dendale and Yves D'Hulst, eds.), pp. 299-318. Unipress, Padova. (1998). De la syntaxe a 1'interpretation, de Milner (1982) a Mttner (1995): le genitif. In: Cahier Jean Claude Milner (G. Lardreau, ed.). Verdier, Paris. Dobrovie-Sorin, Carmen and Brenda Laca (1998). La genericite entre la reference a 1'espece et la quantification generique. Actes du Colloque langues etgrammaire 3. Presses universitaires de Vincennes.
320 Comparative studies in Romanian syntax Dumitrescu, Domnija (1991). General considerations about echo questions in Spanish and Romanian: Towards defining the concept. Revue Roumaine de Linguistique 36 (Part I: 141-167; Part II: 279315). Eeden, W. van (1988). On preposition and prepositional phrases in Rumanian. InternationalJournal of Rumanian Studies 6(2), 93-105. Farkas, Donka (1981). Word order in Rumanian main clauses. Folia Slavica 4(2-3), 254-262. (1982). Intensionality and Romance subjunctive relatives. Indiana University Linguistics Club, Bloomington, IN. (1984). Subjunctive complements in Rumanian. In: Papers from the Xllth Linguistic Symposium on Romance Languages (P. Baldi, ed.), pp. 354-372. John Benjamins, Amsterdam. (1985). Intensional descriptions and the Romance subjunctive Mood. Garland Publishing Inc., New York. (1985). Obligatorily controlled subjects in Romanian. In: Proceedings of the 21st Annual Meeting of the Chicago Linguistics Society (W.H. Eilfort; P.D. Kroeber and K.L. Peterson, eds.), pp. 91-100. CLS, Chicago. (1988). On obligatory control. Linguistic and Philosophy 11, 27-58. (1988). On the morpho-syntax of subjunctive clauses in Rumanian. InternationalJournal of Rumanian Studies, 6(2), 7-19. (1990). Two cases of underspecification in morphology. Linguistic Inquiry 21, 539-550. (1992). On obviation. In: Lexical matters (A. Szabolcsi and I. Sag, eds.), pp. 85-109. CLS, Chicago. (1992). On the semantics of the subjunctive complements. In: Romance languages and modern linguistic theory (P. Hirschbhuler and K. Koerner, eds.), pp. 69-105. John Benjamins, Amsterdam. (1996). Word order in Romanian subjunctive clauses. In: Autolexical syntax (E. Schiller; E. Steinberg and B. Need, eds.), pp. 87-103. John Benjamins, Amsterdam. Farkas, Donka and Ahnerindo Ojeda (1983). Agreement and coordinate NPs. Linguistics 21, 659-673. Farkas, Donka and Kostas Kazazis (1980). Clitic pronouns and topicality in Romanian. In: Papers from the 16th Regional meeting of the Chicago Linguistic Society (J. Kreiman and A. Ojeda, eds.), pp. 75-82. Chicago
Bibliography
321
Linguistic Society, Chicago. Farkas, Donka and Draga Zee (1995). Agreement and pronominal reference. In: Advances in Roumanian linguistics (G. Cinque and G. Giusti, eds.), pp. 83-102. John Benjamins, Amsterdam. Gierling, Diana (1996). Clitic doubling: typology, syntax and information structure. University of Tubingen: Papers on typology of information structure series (monograph). (1996). Further parallels between clitic doubling and scrambling. In: Proceedings of the 13th Eastern States Conference on Linguistics '96 (A. D. Green and V. Motapanyane, eds.), pp. 113-123. Cornell University- CLC Publications, Ithaca, NY. (1997). Clitic doubling, specificity and Focus in Romanian. In: Clitics, pronouns and movement (J. Black and V. Motapanyane, eds.), pp. 63-85. John Benjamins, Amsterdam. Giusti, Giuliana (1991). La sintassi dei nominal! quantificati in romeno. Rivista di grammatica generativa 16, 2957. (1992). La sintassi dei sintagmi nominali quantificati. Un studio comparativo. Doctoral dissertation, University of Padova/University of Venice. (1993). Enclitic articles and double definiteness: A comparative analysis of nominal structure in Romance and Germanic. University of Venice Working Papers in Linguistics 3(1), 8394. (1993). Enclitic articles and double definiteness: A comparative analysis of nominal structure in Romance and Germanic. The Linguistic Review 11, 231-255. (1995). Heads and modifiers among determiners: Evidence from Rumanian. In: Advances in Rumanian linguistics (G. Cinque and G. Giusti, eds.), pp. 103-126. John Benjamins, Amsterdam. (1997). The categorial status of determiners. In: The new comparative syntax (L. Haegeman, ed.), pp. 95-123. Longman, London. Grosu, Alexandru (1988). On the distribution of genitive phrases in Rumanian. Linguistics 26, 931-949. (1988). On an asymmetry in the distribution of island constraints. Lingua 74, 167-184. (1992). Une etude de celui et de eel dans le cadre de la Theorie de la Projection Etendue. Travaux de linguistique 24, 5-19. (1994). Three studies in locality and Case. Routledge, New York. (1995). Free relatives with "Missing Prepositions" in Rumanian and Universal Grammar. In:
322 Comparative studies in Romanian syntax Advances in Rumanian linguistics (G. Cinque and G. Giusti, eds.), pp. 127-160. John Benjamins, Amsterdam. (1995). Overt and Null Operators in French and Romanian. In: Tendences recentes en linguistique franqaise et generate (H. Bat-Zeev Shyldkrot and L. Kupferman, eds.), pp. 201-210. John Benjamins, Amsterdam. (1995). Irrealis free relatives in Romanian. In: A Festschrift for Solomon Marcus (A. Paun, ed.). Editura Academiei Romane, Bucharest. (1999). An unusual Romanian construction and its theoretical implications. In: Traiani Augusti vestigia pressa sequamur (M. Coene, W. de Mulder, P. Dendale and Yves D'Hulst, eds.), pp. 97-108. Unipress, Padova. Grosu, Alexandru and Julia Horvath (1984). The GB theory and Raising in Rumanian. Linguistic Inquiry 15, 345-353. (1987). On non-finitness hi extraction constructions. Natural Language and Linguistic Theory 5, 181-196. (1987). On the notion "head": Evidence from interrogatives and free relatives. Theoretical Linguistics 14, 35-64. lonescu, Rodica (1990). Rumanian supine constructions. M.A. thesis, University of Ottawa. Isac, Daniela (1994). Sentence negation in Romanian. Revue Roumaine de Linguistique 29 (3/4), 371-393. (1996). Sentence structure and verb movement in Romanian. Cahiers de linguistique theorique etappliquee 29, 99-107. (1997). NOT in the mood. Revue roumaine de linguistique Forthcoming. (1998). Sentence negation in English and Romanian. Doctoral dissertation, University of Bucharest. (1999). Negation, Mood and Tense. Proceedings of the 1st conference on "Linguistic theory in Eastern European Languages ", (1999). Two Romanian moods. In: Traiani Augusti vestigia pressa sequamur (M. Coene, W. de Mulder, P. Dendale and Yves D'Hulst, eds.), pp. 443-460. Unipress, Padova. (1999). The (non) distinctive syntax of Romanian imperatives. Revue roumaine de linguistique, Forthcoming. Jong, Jelly Julia de (1996). The case of bound pronouns in peripheral Romance. Groningen Dissertations in Linguistics 16. Grafisch Bedrijf Ponsen and Loojinen BV, Wageningen.
Bibliography
323
Joseph, Brian (1982). A new convergence concerning the Balkan loss of the infinitive. Indogermanische Forschungen 85, 176-187. (1983). The synchrony anddiachrony of the Balkan infinitive: A study in areal, general, and historical linguistics. Cambridge University Press, Cambridge. Kempchinsky, Paula (1986). Romance subjunctive clauses and Logical Form. Doctoral dissertation, UCLA. (1987). Directionality of government and nominative case assignment in Romanian. In: Studies in Romance linguistics (C. Kirschner and J. DeCesaris, eds.), pp. 209-224. John Benjamins, Amsterdam. (1998). Mood phrase, Case checking and obviation. In: Romance linguistics: Theoretical perspectives (A. Schwegler et aL, eds.), pp. 143-154. John Benjamins, Amsterdam.
Kok, A. de (1989). On the conjunct object pronoun systems of Rumanian and French: A comparative analysis. International Journal of Rumanian Studies 1. Laca, Brenda and Liliane Tasmowski (1994). Referentialite du pluriel indefini dans les langues romanes. Faits de langues 4, 97-104. (1994). Le pluriel indefini de 1'attribut metaphorique dans les langues romanes. Lingvisticae Investigationes 18, 27-48. Lascu, Corina (1999). Roles semantiques et marquage casuel dans 1'analyse des deverbaux en roumain et en fran9ais. In: Traiani Augusti vestigiapressa sequamur (M. Coene, W. de Mulder, P. Dendale and Yves D'Hulst, eds.), pp. 337-352. Unipress, Padova. Legendre, Geraldine (2000). Positioning Romanian Verbal Clitics at PF: An Optimality Theoretic Analysis. In: Clitics in Phonology, Morphology and Syntax (B. Gerlach and J. Grijzenhout, eds.). John Benjamins, Amsterdam, (forthcoming). (2000). On the status and positioning of verbal clitics. Natural Language and Linguistic Theory (forthcoming). Lema, Jose and Maria-Luisa Rivero (1989). Long Head Movement: ECP vs. HMC. In Proceedings ofNELS 20, University of Massachusetts: GLSA, 333-347.
324 Comparative studies in Romanian syntax (1992). Inverted conjugations and V-second effects in Romance. In: Theoretical analyses in contemporary Romance linguistics (C. Laeufer and T. Morgan, eds.), pp. 311-328. John Benjamins, Amsterdam. Manoliu-Manea, Maria (1983). 'Ca nu-s fata de gasit. . .': Le Supin et la voix en roumain. In: Verbe etphrase dans les langues romanes: Melanges offerts a Louis Mourin (E. Roegiest and L. Tasmowski, eds.), pp. 247-253. University of Ghent, Ghent. Motapanyane, Virginia (1989). La position du sujet dans une langue a 1'ordre SVO/VSO. Rivista di grammatica generativa 14, 75-103. (1990). The switch from infinitive to subjunctive clauses in Romanian. In: PAMAPLA 14 (J. Black, ed.), pp. 115-131. Memorial University of Newfoundland, St. John's NF. (1991). On the representation of the Object-Agreement phrase. In: PAMAPLA 15 (W. J. Davey and B. LeVert, eds.), pp. 73-82. University College of Cape Breton, Sydney, NS. (1992). Romance clitics and the structure of AgrP. Linguistica Atlantica 14, 73-87. (1992). On super raising hi null subject languages. The Penn Review of Linguistics 16, 93-101. (1992). A-chain formation and super raising. In: CLA Annual Conference 1992 (C. Dyck; J. Gomeshi and T. Wilson, eds.), pp. 205-214. Toronto Working Papers in Linguistics, Toronto. (1994). Preverbal positions hi VSO languages. Proceedings of the WECOL 93, vol.6., 303-311. Fresno, Ca: Department of Linguistics, California State University. (1994). On preverbal positions hi Romanian. Canadian Journal of Linguistics 39(1), 15-36. (1994). An A-position for Romanian subjects. Linguistic Inquiry 25, 729-734. (1995). Low Focus position hi Balkan languages. In: CLA Annual Conference 1995 (P. Koskinen, ed.), pp. 373-384. Toronto Working Papers hi Linguistics, Toronto. (1995). NP-movement from finite clauses hi Rumanian. In: Advances in Rumanian linguistics (G. Cinque and G. Giusti, eds.), pp. 161-170. John Benjamins, Amsterdam. (1995). Theoretical implications of complementation in Romanian. Unipress, Padova. (1997). A Focus parameter. In: Papers from the 1997 Mid-America Linguistics Conference (Li Xingzhong, L. Lopez and T. Stroik, eds.), pp. 238-246. Department of Linguistic, University of Missouri-Columbia, Columbia, MO. (1998). C/T Merge Location for Focus. Linguistica Atlantica 20, 109-122. (1998). Focus, checking theory and fronting strategies hi Romanian. Studia Linguistica 52(3): 227-243. (1998). Evidence for Focus features. In: The Minimalist parameter (G. Alexandrova and O. Arnaudova, eds.), pp. 243-256. John Benjamins, Amsterdam.
Bibliography
325
(1998). Object-to-subject raising in toMg/z-constructions. Revue roumaine de linguistique 43 (1/2), 37-52. Motapanyane Virginia and Larisa Avram (1999). A Romanian modal at the cross-roads between Romance and Balkan. Revue roumaine de linguistique 44 (forthcoming). (2000). The syntax ofputea and its mixed typology. Belgian Journal of Linguistics (forthcoming). MyhiU, John (1987). Two sources of VS order in Rumanian. In: Studies in Romance languages (C. Neidle and R. Nunez, eds.), pp. 215-224. Foris, Dordrecht. Niculescu, Alexandru (1983). Le vocatif roumain. In: Verbe etphrase dans les langues romanes: Melanges offerts a Louis Mourin (E. Roegiest and L. Tasmowski, eds.), pp. 255-260. University of Ghent, Ghent. Prrvulescu, Mihaela (2000). Le concept de paradigtne dans la morphologic verbale. Doctoral dissertation, University of Toronto (forthcoming). Pirvulescu, Mihaela and Yves Roberge (1999). Objects and the structure of imperatives. In: Formal Perspectives on Romance Linguistics (J.-M. Authier, B. E. Bullock and L. A. Reed, eds.), pp. 211-226. John Benjamins: Amsterdam. Popescu-Ramirez, Liliane and Liliane Tasmowski-De-Ryck (1988). Thematicite et possessrvite en roumain. Lingvisticae Investigationes 12(2), 303-335. (1988). Le complement du nom et le possessif. International Journal of Rumanian Studies 6(2), 81-93. (1992). "Je peux partir / Je peux que je parte". Hommages a Nicolas Ruwet, 464-474. C&C, Ghent. Reinheimer, S. and Liliane Tasmowski (1997). Flexiunea verbala romaneasca rn perspectiva romanica. Limbap Literatura. Forthcoming. (1999). Verbes reguliers, pseudo-reguliers et irreguliers dans les langues romanes. In: Actes du
326 Comparative studies in Romanian syntax 16e colloque international de lexique et grammaire compares des langues romanes (J. Klein and B.Lamiroy, eds.). CILL (forthcoming). Rivero, Maria-Luisa (1987). La teoria de las barreras y las completivas del rumano. In: Sintaxis de las lenguas romanicas (M. Fernandez-Lagunilla and V. Demonte, eds.), pp. 329-353. El Arquero, Madrid. (1989). Barriers and Rumanian. In: Proceedings of the Linguistic Symposium on Romance Languages XVII (J. de Cesaris and C. Kirschoner, eds.), pp. 289-312. Johb Benjamins, Amsterdam. (1991). Exceptional Case Marking effects in Rumanian subjunctive complements. In: New analyses in Romance languages (D. Wanner and D. A. Kibbee, eds.), pp. 273-298. John Benjamins, Amsterdam. (1994). Clause structure and V-movement in the languages of the Balkans. Natural Language and Linguistic Theory 12, 63-120. Rizzi, Luigi (1991). Residual Verb Second and the Wh-Criterion. Technical Reports in Formal and Computational Linguistics 2. Universite de Geneve, Geneva. Rudin, Catherine (1988). On multiple questions and multiple wh fronting. Natural Languages and Linguistic Theory 6, 445-501. §tefinescu, loana (1982). The transformational rule of raising in Romanian. Revue roumaine de linguistique 27, 365-395. (1992). On diminutive suffixes. Folia Linguistica 3/4, 339-356. (1995). Psychological verbs with dative experiencers and nominative themes: The case of Romanian. CUNY Forum 19, 109-148. (1997). The syntax of agreement in Romanian. Department of Linguistics, MIT Working Papers in Linguistics 14, Cambridge, MA. Steriade, Donca (1980). Clitic doubling in the Romanian \vh constructions and the analysis of topicalization. In: Papers from the 16th Regional Meeting ofCLS (J. Kreiman and A. Ojeda, eds.), pp. 282-297. Chicago Linguistic Society, Chicago, II.
Bibliography
327
Tasmowski, Liliane (1987). La reduplication clitique en roumain. In: Rdtoromanisch und Rumdnisch. Akten der Theodor Gartner-Tagung (G. Plangg and M. Diescu, eds.), pp. 377-399. Innsbruck: Romanica Aenipontana XIX. (1988). Complementation in Rumanian (Part I). InternationalJournal of Rumanian Studies 6(2). (1989). Complementation in Rumanian (Part n). InternationalJournal of Rumanian Studies 7(1). (1989). Cours de grammaire roumaine. Communication and Cognition, Ghent. (1990). Les adjectifs demonstratifs fran9ais et roumains dans la phrase et dans le texte. In: Aux confins de la grammaire (P. Cadiot and A.Zribi-Hertz, eds.). Langages 97, 82-100. (1990). Les pronoms demonstratifs en francais et en roumain. Revue roumaine de linguistique 35, 375-383. (1992). Infinitif et subjonctif apres aputea, a §ti et a vrea dans quelques textes roumains du 19.siecle. In: Beitrage zur rumdnische Sprache im 19 Jahrhundert (E. G., P. Stein and B.Weber, eds.), pp. 261-270. Niemeyer, Tubingen. (1992). (Co)reference et reference disjointe dans la phrase et dans le texte. Studii §i CercetSri Lingvistice 43, 491-499. (1993). Entre anaphore et deixis: 1'emploi des pronoms dans La chronique de Moldavie de Grigore Ureche. Actes du 20e congres international de linguistique etphilologie romanes, vol. 1, section I, 751-763. Francke, Tubingen. Tasmowski, Liliane and Paul Verluyten (1982). Linguistic control of pronouns. Journal of Semantics 1, 323-346. Terzi, Arhonto (1992). PRO infinite clauses. A study of the inflectional heads of the Balkan languages. Doctoral dissertation, CUNY.
This page intentionally left blank
Index
329
NAME INDEX Abney, 172, 178, 180, 223 Alboiu, 2, 3, 1, 12, 27, 33, 45, 265, 315 Alexiadou, 85, 87-91, 97, 113, 131 Anderson, 230, 234, 240, 260, 262 Avram, 17-19, 45, 83, 135, 315, 316, 325 Ayoub, 194, 223 Baker, M, 131, 172 Barker, 217, 221, 223 Bayer, 3, 44, 45, 49, 51, 56, 58, 59, 61-65, 78, 79 Bech, 54, 79 Beghelli, 58, 59, 67, 79 BeUetti, 23, 45, 110, 131 Benveniste, 173 Bernstein, 2, 4, 45, 139, 150, 173, 316 Beukema, 311 Bonet, 307, 312 Borer, 40, 45, 88, 131, 145, 173, 177, 193-195, 197, 206, 210-213, 220, 223, 224 Bouchard, 40, 45, 81 Brody, 59, 79, 269, 293 Burzio, 96, 131,227,253 Byck, 98, 101, 131 Cardinaletti, 25, 45, 47, 49, 83, 84, 87, 88, 94, 98, 104, 114, 121, 125, 129, 131-133 Chomsky, 6, 32, 43, 45, 46, 49, 50, 53, 54, 56, 57, 59, 60, 63, 65-67, 74, 78-80, 85, 131, 132, 136, 137, 158, 173, 181, 184, 195, 224, 267-270, 272, 276, 293, 295, 312 Cinque, 7, 10, 12, 26, 46, 47, 55, 56, 79, 85, 87, 89, 90, 97, 118, 123, 131, 151, 153, 157, 167, 173, 174, 262, 265, 316, 317, 319, 321, 322, 324 Coene, 2, 3, 6, 135, 166, 173, 186, 187, 193, 224, 315-317, 319, 322, 323 Collins, 279, 280, 293 Comorovski, 29, 30, 46, 317 Coopmans, 311
330 Comparative studies in Romanian syntax Corblin, 221, 224 Cornilescu, 3, 3-5, 7, 22-25, 27, 28, 36, 40, 46, 83, 121, 132, 135, 139, 143, 145, 151, 173, 184, 195,224,274,289,317 Corver, 49, 167, 173 Costa, 235, 262 Coteanu, 312 Crisma, 202, 224 Curat, 304, 312 Danon, 193, 208, 224 Davison, 79 Dekydtspotter, 123, 125, 132 Delfitto, 151, 164, 165, 173 Deprez, 132 Diesing, 132 Dobrovie-Sorin, 2, 3, 5, 6, 9-13, 15, 17-26, 28, 31, 33, 35, 36, 40, 43, 45, 46, 49, 83, 84, 88, 90, 92, 96, 106, 107, 114, 121, 123, 131, 132, 135, 139, 145, 173, 177, 182-184, 191-193, 195, 203, 224, 227, 229, 231, 232, 244, 259, 261, 262, 272, 293, 301, 312,318,319 Drubig, 269, 280, 283, 293 D'Hulst, 2, 3, 6, 135, 151, 164, 173, 316 Everett, 233, 262 Farkas, 2, 3, 33, 35-37, 46, 227, 320, 321 Fassi-Fehri, 193, 195, 212, 220, 224 Ferguson, 164, 173 Flaux, 221, 224, 316 Fox, 65, 79, 262, 263 Frege, 217, 225 Gallman, 61 Gawron, 170, 174 Ghomeshi, 225 Gierling, 12, 46, 278, 293, 321 Giorgi, 183, 225 Giusti, 2, 4, 7, 8, 46, 47, 139, 149, 150, 166, 173, 174, 262, 316, 317, 319, 321, 322, 324 Gorell, 63, 79 Graur, 311, 312 Grimshaw, 57, 65, 79, 241, 243, 249, 262 Grosu, 2-5, 7, 35, 37, 42-45, 47, 49, 61, 79, 139, 143, 147, 149, 151, 174, 177, 179, 180, 183, 188, 194, 202, 204, 225, 321, 322
Index Gueron, 47, 66, 79, 132, 318 Gu u-Romalo, 297, 298, 312 Haegeman, 48, 133, 174, 269, 282, 293, 321 Haider, 35, 47, 79 Halle, 312 Halpeni, 232, 234, 252, 262 Harris, 174 Heim, 216, 225 Hoekstra, 20, 47, 132 Homer, 263 Hornstein, 50, 80, 133 Horvath, 2, 35, 37, 42, 43, 47, 49, 269, 293, 322 ffiescu, 297, 299, 312, 327 lordan, 297, 311, 312 Irimia, 297, 299, 311, 312 Isac, 30, 31, 47, 83, 91, 132, 322 Izvorski, 274, 293 Jackendofl^ 218, 221, 225 Jong, 11,47,322 Joseph, 3, 301, 312, 319, 323 Kamp, 216, 225 Kayne, 5, 46, 47, 54, 62, 74, 80, 132, 183, 184, 225, 227, 234, 262, 268, 294, 319 Kempchinsky, 33, 35, 36, 47, 323 Kennedy, 66, 80 Kihm, 189, 225 King, 59, 60, 62, 68, 69, 72-74, 160-166, 196, 197, 199, 201, 208, 210, 212, 220, 265, 294 Kiss, 27, 39, 47, 58, 80, 270, 293, 294 Klavans, 230, 234, 235, 260, 262 Koizumi, 88, 132 Roster, 46, 49, 79, 80, 319 Kraskow, 30, 47 Laca, 203, 224, 319, 323 Laka, 224 Lappin, 49, 80 Larson, 80 Lasnik, 97, 132 Legendre, 3, 45, 227, 234-237, 241, 246, 247, 249, 251, 253, 254, 262, 263, 323 Lema, 9, 47, 323
331
332 Comparative studies in Romanian syntax Lombard, 299, 312 Longobardi, 49, 139, 145, 174, 182, 183, 187, 195, 196, 202, 212, 220, 225 Lsbner, 218, 223, 225 Mahajan, 12, 47, 132, 224 Mallinson, 227, 231, 233, 263 Manzini, 18,47, 182,225 Marantz, 136, 174 May, 1-3, 6, 8, 9, 11-13, 18-20, 24, 29, 30, 32, 34, 35, 39, 41-44, 49, 50, 53, 54, 60, 61, 64-69, 72, 73, 78-80, 83, 84, 86, 90-94, 96-99, 101-106, 108, 111-121, 124-128, 140, 144, 146, 149, 154, 156-158, 160, 164, 165, 167, 170, 171, 180-184, 188, 190, 192, 193, 200, 202-204, 206-209, 213, 215, 217, 218, 223, 227, 228, 233, 234, 237, 239, 240, 242, 244, 245, 248, 249, 253, 260, 262, 266, 270-274, 277-280, 282, 284-287, 289, 291, 292, 298 McCarthy, 235, 263 McNally, 182, 203, 225 Miller, 192, 225, 234, 263 Milner, 191, 218, 221, 224, 225, 319 Moritz, 61, 80 Motapanyane, 1-3, 1, 18, 19, 22-27, 30, 33, 35, 36, 38, 40, 42, 43, 45-47, 49, 83, 96, 114-116, 121, 132, 227, 228, 263, 265, 269, 274, 293, 294, 321, 324, 325 Mueller, 268, 269, 294 Ortiz de Urbina, 294 Ouahalla, 132 Pan -Dindelgan, 132 Partee, 216, 225 Pesetski, 29, 47 Picallo, 151, 174 Prrvulescu, 2, 3, 31, 295, 325 Poesio, 226 Poletto, 104, 105, 107, 111, 132 Pollock, 15, 22, 46, 48, 74, 79, 80, 85, 89, 133, 318, 319 Postal, 69, 80 Prince, 221, 226, 229, 234, 235, 240, 241, 263 Progovac, 106, 107, 133 Ramchand, 133 Raposo, 133 Raymond, 171, 263 Reinhart, 49, 53, 65, 66, 80
Index
333
Ritter, 145, 147, 151, 174, 187, 195, 210, 226, 307, 312 Rivero, 9, 22, 27, 28, 30, 31, 35-37, 47, 48, 85, 87, 133, 227, 229, 231, 232, 247, 252, 253, 263, 290, 294, 300, 301, 312, 319, 323, 326 Rizzi, 27, 28, 32, 33, 46, 48, 50, 78-80, 96, 104-108, 116-118, 121, 122, 133, 226, 259, 263, 269,283,294,319,326 Roberge, 3, 31, 295, 304, 308, 310, 312, 325 Roberts, 83, 93, 111, 112, 131, 133 Rochemont, 265, 268, 269, 280, 284, 294 Rooryck, 2, 135, 252, 256, 263 Rooth, 80, 216, 225 Rosetti,299, 313 Ross, 61, 63, 80 Rothstein, 64, 81, 174 Rouveret, 145, 174, 195, 226 Rudin, 2, 29, 48, 265, 274, 286, 290, 294, 326 Sadock, 254, 263 Safir, 45, 53, 81, 104, 132, 133 Samek-Lodovici, 235, 241, 262-264 Schroten, 164, 165, 173 Siloni, 2, 145, 174, 177, 187, 190, 193, 195, 206, 207, 212, 226 Smolensky, 227, 229, 234, 240, 241, 254, 263, 264 Sportiche, 22, 48, 88, 92, 133 Starke, 98, 104, 131, 133 Stechow, 54, 81, 225 §tefinescu, 4, 5, 7, 17, 22-24, 42, 48, 326 Stemefeld, 268, 269, 294 Stowell, 58, 59, 67, 79 Sufier, 133 Szabolcsi, 58, 152, 174, 320 Tasmowski, 3, 6, 135, 141, 174, 186, 187, 224, 316, 323-325, 327 Terzi, 33, 35-37, 48, 229, 252, 253, 263, 327 Tomi<5, 227, 238, 246, 264 Tsai, 66, 81 Uriagereka, 133 Valois, 61, 80, 147, 174 Vergnaud, 156, 165, 175, 218, 226 Wilson, 263, 324 Woisetschlaeger, 204, 226
334 Comparative studies in Romanian syntax Zanuttini, 46, 79, 252, 253, 257, 261, 264, 302, 303, 310, 313, 319 Zubizarreta, 156, 165, 175, 218, 226 Zwicky, 234, 245, 254, 263, 264
Index
335
SUBJECT INDEX Accusative, 2, 4, 8, 140, 144, 149, 157, 158, 160-162, 164, 165, 238, 242, 244, 248, 258 Agreement, 3, 5, 6, 8, 15, 17-19, 22-24, 35, 38, 43, 46-48, 52, 57, 58, 67, 72, 83-86, 99, 105, 106, 108, 111-113, 129, 133, 142, 144, 146, 157, 160, 171, 177, 180, 195, 212, 214, 219, 234, 253, 293, 302, 310, 312, 320, 321, 324, 326 Agreement-Object, 23 Agreement-Subject, 23 al, 4-6, 46, 73, 111, 135, 136, 142-150, 154-162, 164-166, 172, 173, 178, 179, 184-190, 193-198, 201, 208, 209, 212, 215, 218, 220, 237, 249, 254, 297, 299, 311-313, 316,317,323 Albanian, 256, 261 alignment, 10, 228, 229, 234-237, 239-242, 246, 248, 250, 252, 257, 260-264 argumental position, 25, 92, 116, 126, 130, 274 article reduplication, 3, 135, 163, 165, 166, 170-172, 224 auxiliary verb, 24, 240 avea, 14, 15, 17, 18, 21, 23, 39, 305 A-chain, 324 Balkan, 27, 33, 34, 36, 37, 39, 45, 48, 85, 227, 230, 232, 234, 235, 238-240, 246, 248, 253, 254, 256, 264, 274, 293, 316, 319, 323-325, 327 Bare Quantifier, 102, 106, 126, 274 Bulgarian, 29, 227, 230, 232, 235-239, 246-248, 251, 252, 254, 260-262, 269, 270, 290, 293, 294 ca, 34-38, 109, 174, 289, 292, 324 ca, 9, 29, 32-34, 37, 44, 266, 272, 278, 289, 292 Case assigning preposition, 4 categorial feature, 27, 57, 64, 72, 149 eel, 5, 46, 166, 190, 316, 317, 321 Celtic, 93 clitic climbing, 17, 19, 132, 260 clitic doubling, 11-13, 29, 33, 44, 46, 47, 97, 104, 234, 272, 275, 278, 279, 293, 318, 321, 326 Clitic Phrase, 87, 88
336
Comparative Studies in Romanian Syntax
clitic pronoun, 11, 21, 101, 244, 260 clitic subject, 98, 103, 104, 106, 107, 110, 112, 122, 123 Clitics, 3, 9, 10, 12, 16, 19, 21, 25, 28, 30, 31, 33, 46, 88, 89, 98, 104-108, 110, 114, 118, 132, 133, 225, 227-230, 232-240, 242-248, 250-252, 256, 257, 259-264, 278, 289, 293, 301-303, 305, 312, 315, 316, 319, 321, 323, 324 CLLD, 12, 26, 33, 118, 119, 123, 125, 265, 266 Complementizer, 32, 34-37, 39, 40, 272, 278, 292, 303 conditional, 14-17, 21, 27, 232, 234, 244, 295 constraint ranking, 229, 237, 240, 241, 261 Construct State hypothesis, 144 Control, 18-20, 34-36, 40, 47, 225, 317, 320, 327 Czech, 248, 261 daca, 21, 267, 279, 291, 292 Dative, 2, 7, 8, 12, 61, 84, 140, 143, 178, 204, 205, 238, 244, 248, 316, 317, 326 de, 5, 6, 10, 11, 15, 16, 19, 21, 22, 26, 34, 35, 38, 39, 41-44, 46-48, 55, 68, 69, 72-74, 90, 91, 93-97, 100, 109, 124, 131, 132, 144, 148, 156, 169-171, 173, 174, 177, 178, 183, 191-193, 195, 198-201, 204, 206, 207, 209, 221-225, 231-233, 244, 248, 262, 263, 273, 278, 291-294, 300, 312, 315-327 declarative clause, 26 default marking, 305 definite article, 2, 5-7, 41, 63, 68, 138-141, 143, 144, 147-149, 151, 162, 163, 166, 168, 170, 171, 177-180, 183-185, 187, 188, 190, 192-194, 199, 204, 205, 209, 214, 215, 217-219,222,223,227 demonstrative pronoun, 273 deontic modal, 19 Determiner, 3-5, 7, 46, 54, 135, 136, 148, 155, 156, 165, 166, 171, 172, 175, 180, 184, 188, 192, 193, 195, 199, 208, 214, 215, 219, 226, 304, 317 Dutch, 16, 97, 169 edge, 228, 229, 232, 234-236, 241, 243, 245, 250, 251, 260 English, 3, 13, 26-29, 32, 34, 43, 45, 51, 52, 55, 56, 60, 63, 65, 67, 70, 77, 121, 122, 125, 127, 138, 150, 169, 172, 175, 180, 181, 185, 188, 195, 198, 201, 209, 212, 218, 221-224, 226, 228, 249, 264-269, 271-273, 277, 279-283, 285-287, 294, 309, 312,315,317,322 EPP, 23, 276 expletive determiner, 4 external argument, 308 feature checking, 3, 6, 42, 49, 57, 137, 146, 268, 273, 277, 281, 285 FF, 51, 60, 64, 65, 75, 76, 265, 285, 286
Index
337
FF raising, 285, 286 fi, 15, 18, 19, 21, 22, 33, 34, 38, 40, 96, 103, 124, 275, 278, 289, 290, 292, 305, 306 finite verb, 21, 23, 25, 27, 28, 88, 91, 242, 244-246 finiteness, 43, 117, 230, 238-240, 248, 254 Fiorentino, 105-108 floating quantifier, 94 Focus, 2, 3, 13, 20, 25-27, 32, 33, 46, 51, 62, 64, 65, 68, 72-74, 79, 80, 84, 96, 116-124, 127-131, 133, 165, 174, 227, 230, 234, 235, 248, 264-285, 287, 290, 291, 293, 294,302,315,321,324 French, 3, 6, 15-17, 38, 46, 50, 51, 55-57, 61, 63, 67, 68, 70, 72, 74, 75, 77, 83, 84, 86, 90, 91, 98, 104, 106-110, 112, 122, 125, 126, 135, 136, 148, 156, 165-172, 175, 183, 190-193, 195, 198, 200, 204, 206, 207, 209, 212, 221-224, 226, 234, 245, 248, 252,262,318,322,323 Friulian, 302 functional category, 3, 5, 6, 57, 64, 181, 302-304 functional features, 10, 228, 229, 233, 234, 237, 239, 240, 252, 260, 267 fUture, 14-17, 27, 232, 234-236, 248 Genitive, 2, 4-7, 47, 135, 136, 138, 140, 143, 144, 146, 148, 151-153, 155, 156, 158, 166, 172-174, 177-181, 184-193, 195-201, 204-207, 209, 213-221, 223-225, 316, 317,321 German, 16, 44, 45, 51, 52, 54-57, 60, 61, 63, 67, 77, 97, 169, 209, 217 gerund, 85, 86, 239, 242, 253, 296, 306 Greek, 131,252,256,312 head-to-head movement, 4, 7, 46 Hebrew, 3, 6, 45, 144, 145, 173, 174, 177, 184, 188-190, 193-197, 199-203, 206-211, 213, 215, 217, 218, 220, 221, 223, 224, 226 illocutionary force, 28, 117, 252, 300, 302-304 imperative, 30-32, 229, 244, 246, 252-254, 256, 259, 261, 295-308, 310-312 imperatives, 3, 30, 31, 33, 228-230, 237, 244, 248, 250, 252-259, 261, 263, 295, 300-308, 310, 312,313,322,325 imperfect, 19, 85, 86, 306, 308 indefinite article, 8, 170, 178, 194, 196, 198, 200-202, 207, 208 indicative, 20, 23, 30-34, 85-87, 254, 295-299, 302, 305-308, 310 infinitive, 14, 19, 20, 30, 35, 37-40, 43, 44, 85, 86, 229, 253, 254, 261, 296, 297, 299, 302, 305, 306,310,317,323,324 Inflection, 2, 8, 18, 35, 61, 83-85, 92, 105, 114, 233, 253, 254, 264, 296, 305 internal argument, 32, 190, 206, 210, 308, 311 Italian, 10, 16, 19, 24, 46, 48, 51, 55, 56, 63, 67, 84, 91, 94, 96, 98, 104-107, 110, 112, 113,
338 Comparative studies in Romanian syntax 119, 121, 122, 125, 126, 129-133, 148, 150, 169, 230, 245, 253, 254, 257-261, 302,318 Last Resort, 3, 135, 136, 138, 156, 158, 165, 168, 172, 173, 224, 316 LHM, 9, 28, 87, 88, 110 Macedonian, 227, 230, 234, 238, 239, 246-248, 251-253, 260, 261, 263 mai, 10, 21, 22, 30, 33, 38, 42, 44, 60, 63, 67, 69, 70, 74, 75, 88, 89, 95, 99, 101, 102, 124, 169, 231-233, 235, 244, 245, 288-290, 292 Mirror Principle, 87, 88 modal, 19, 39, 227, 230, 232, 249, 325 Mood, 22, 23, 30, 31, 33, 36, 38-40, 46, 84, 85, 87, 88, 238, 253, 263, 302-306, 320, 322, 323 morphological Case, 2, 4, 12, 138, 140, 160, 209 morphological feature, 307 multiple wh, 29, 30, 47, 48, 270, 287, 294, 317, 326 negated imperatives, 302, 305 Negation, 21, 33, 42, 47, 54, 79, 88, 264, 293, 302, 303, 310, 313, 322 negative adverb, 91 Nominative, 2, 4, 8, 24, 25, 39, 40, 83, 84, 95, 98, 105, 107, 110-114, 117, 121, 140, 144, 149, 152, 157, 158, 160-162, 164, 165, 178, 248, 317, 323, 326 non-argumental position, 130 non-finite verbs, 239, 247 Noun, 2-4, 6-8, 12, 46, 57, 63, 101, 132, 135, 138, 139, 145-147, 149-151, 153, 157, 160, 162-168, 171, 172, 174, 178, 189, 193, 196, 202, 203, 205, 210, 211, 213, 214, 219, 220, 222, 223, 225-227 NP-movement, 25, 37, 38, 43, 324 nu, 11, 12, 15, 19, 21, 22, 26, 30, 32-34, 38, 40, 42, 60, 63, 88, 89, 96, 103, 109, 113, 115, 119, 120, 123, 124, 126, 129, 187, 229, 231-233, 242, 253, 254, 257, 266, 271, 272, 275, 278, 279, 284, 286, 292, 296, 306, 310, 324 Null Operator, 43, 268, 269, 272, 273, 283 Number, 2-4, 8, 11, 15, 20, 22, 24, 26, 29, 30, 35, 36, 45, 50, 52, 55, 56, 59, 61, 62, 67, 72, 75, 85, 88, 90, 99, 112, 113, 117, 135, 138, 142, 143, 147-149, 151, 157, 173, 178, 188, 191, 212, 228, 244, 253, 256, 260, 296, 297, 307, 310, 312 oare, 26, 27, 267, 270, 271, 274, 276, 277, 279, 280, 282, 285, 288-293 object position, 32, 55, 118, 222, 249 oblique Case, 2, 4, 7, 61 Old Romanian, 38, 299 only-clauses, 44 operator feature, 44, 50 operator-variable chain, 29
Index
339
Optimality, 10, 227, 229, 237, 262-264, 323 optimization, 230, 234, 240, 241, 246, 257, 259, 260, 262 parea, 39 participle, 16, 18, 40, 85, 90, 91, 228, 229, 233, 245, 251, 296, 306 past, 14-16, 18, 40, 41, 53, 85-87, 90-92, 95, 96, 114, 228, 229, 233, 245, 251, 296, 306, 308 past perfect, 15, 85, 86, 306, 308 pe, 11-13, 15, 16, 26, 29, 31, 35, 39, 56, 60, 63, 73, 88, 89, 91-93, 95, 97, 99, 100, 109, 111, 115-117, 119, 120, 123, 124, 127-129, 187, 205, 256, 270, 272, 275, 278, 285, 288-290 Person, 8, 15, 23, 24, 30, 33, 85, 87, 88, 90, 99, 103, 104, 111-113, 212, 248, 253, 296-299, 301,304-308,311,312 Pied-Piping, 44, 49, 52-54, 59, 65, 66, 80 Possessor, 4-6, 135, 136, 142, 144-146, 152-154, 162, 168, 172, 179, 190, 193, 195, 205, 206, 209,214,221-223,316 predication, 84, 132, 269, 277, 291, 308 present, 1, 6, 8, 14-16, 18, 19, 22-28, 31-33, 36, 37, 39, 40, 44, 45, 63, 76, 78, 85-87, 90, 114, 117, 119, 122, 145, 149, 158, 173, 184, 187, 223, 228, 230, 232-234, 237, 240-242, 247, 249, 251, 256, 259-261, 265, 273, 274, 279, 285, 286, 289, 296-298, 305-308 present perfect, 14-16, 18 pro, 18, 21, 25, 35-37, 40, 48, 84, 92, 98, 106, 110, 113, 114, 130, 133, 186, 188, 243, 317, 327 putea, 19, 20, 39, 45, 117, 230-234, 266, 325, 327 QR, 49, 51, 59, 65-67, 75-78, 80, 107, 116 quantificational feature, 43 quantifier, 13, 28, 29, 43, 44, 49, 54, 79, 80, 94-96, 102, 106-108, 118, 124, 126-128, 130, 170, 182, 216, 272-274, 278 raising verb, 19, 39 reflexive pronoun, 101 relative clause, 61, 124, 207, 273 Renaissance Veneto, 111 reranking, 230, 237, 245, 256-260 resumptive pronoun, 20 Romance, 3, 5, 6, 15, 19, 26, 27, 32, 33, 36, 37, 44-47, 49, 79, 83-86, 105, 112, 118, 131-133, 135, 136, 138, 140, 145, 147, 148, 165, 169, 173, 174, 227, 230, 234, 236, 253, 254, 260, 262-265, 293, 302, 312, 316, 318-326 Romanian, 1-3, 1, 3, 5-40, 43-48, 51, 55, 56, 59, 62, 63, 67, 68, 70, 72, 74, 75, 83-85, 87-100, 104, 108-119, 121, 122, 125-127, 130-132, 135, 136, 138, 140-142, 144, 145,
340
Comparative Studies in Romanian Syntax 147-152, 156-158, 163, 165, 166, 168, 169, 172-174, 177-181, 183-185, 188-191, 193-201, 204-206, 208, 209, 212, 215, 218-220, 222-224, 227-245,
247-263, 265-279, 281-287, 293-296, 298-300, 302, 309-312, 315-318, 320-326 sa, 17-19, 30, 31, 33-40, 43, 44, 73, 74, 85, 87, 88, 156, 245, 275, 289-292 SAI, 16, 28, 120, 228, 232-234, 249, 261, 267, 271, 282, 284 Saxon, 6, 177, 178, 180-182, 184, 185, 188, 197, 199, 204, 212, 214, 215, 217-221 scope position, 49, 56, 58, 59, 64, 65, 73, 75-78, 266, 268, 272, 274, 280, 281 se, 10, 11, 15, 37, 46, 60, 63, 86, 87, 90, 98, 100, 109, 132, 156, 184, 196, 288, 291, 306, 318, 319 Serbo-Croatian, 37, 232, 234, 248, 252, 261, 294 si, 10, 46, 51, 55, 68, 75, 318, 319 simple past, 85, 86, 306, 308 sine, 11 Slavic, 28, 30, 47, 238, 248, 262-264, 293, 294 Slovak, 261 Spanish, 11, 24, 36, 91, 96, 133, 148-150, 169, 174, 206, 225, 252, 253, 261, 320 specificity, 12, 46, 47, 293, 321 specifier position, 89, 149, 153, 157 structural Case, 200, 201, 209, 210, 293 structural quantifier, 13, 28, 29, 43, 44, 272, 273, 278 structural quantifier parameter, 13, 29, 43, 44, 272, 273, 278 subject, 3, 10, 15, 16, 18, 19, 21, 23-28, 30, 32-40, 43, 45, 47, 55, 58, 59, 63, 67, 69, 72, 75-77, 83-86, 89, 92-117, 121-133, 137, 152, 164, 173, 181, 185, 192, 193, 195, 204-206, 222, 228, 230, 233-235, 241, 243, 245, 247-250, 252, 253, 259-262, 266, 267, 269, 271, 274-277, 281, 291, 308-310, 315, 317, 318, 324, 325, 330 subject obviation, 35-37 subject position, 25, 36, 38-40, 59, 72, 75-77, 83, 84, 93-95, 99, 107, 108, 110, 114, 117, 121-123, 126, 129, 130, 222, 271, 274, 275 subjunctive, 17, 19, 20, 23, 30, 31, 33-41, 46, 47, 85, 87, 88, 254, 295, 297, 302, 305, 306, 318, 320, 323, 324, 326 subjunctive clause, 17, 35, 36 subjunctive morpheme, 17, 33, 35, 36 superlative, 166, 167, 169-172 supine, 40-43, 296, 322 surrogate imperatives, 30, 31, 33, 301, 302, 306 Swedish, 169 syntactic clitics, 108 Template DP hypothesis, 154, 157, 158, 167, 168, 172
Index
341
Tense, 4, 19, 22, 23, 30, 32, 83, 85, 86, 113, 114, 227, 228, 232, 238, 241, 253, 270, 295, 296, 301-304, 306-308, 310, 311, 322 Topic, 1, 20, 25-28, 33, 72, 83, 84, 90, 110, 114, 117-120, 122-131, 133, 259, 265, 266, 269, 270, 274, 275, 293, 294 Torinese, 107, 108 tough-constructions, 43, 44, 325 transitive verb, 16 trebui, 117 Trentino, 104-109 true imperatives, 30, 252, 301-303 Veneto, 106, 111 verb movement, 45, 48, 80, 84, 85, 89, 131-133, 228-230, 241, 248, 251-253, 255, 259, 261, 276, 283, 284, 301-303, 322 verb restructuring, 19 Verb Second, 48, 80, 294, 326 voi, 9, 14, 15,28,30 voice, 90 vrea, 14, 36, 37, 110, 266, 290, 327 Walpiri, 240, 248 Welsh, 93, 112, 113, 195 wh-island, 30, 138