Subject Positions And Interfaces: The Case Of European Portuguese (Studies in Generative Grammar)

Subject Positions and Interfaces: The Case of European Portuguese

≥

Studies in Generative Grammar 73

Editors

Jan Koster Harry van der Hulst Henk van Riemsdijk

Mouton de Gruyter Berlin · New York

Subject Positions and Interfaces: The Case of European Portuguese

by

Joa˜o Costa

Mouton de Gruyter Berlin · New York

Mouton de Gruyter (formerly Mouton, The Hague) is a Division of Walter de Gruyter GmbH & Co. KG, Berlin.

The series Studies in Generative Grammar was formerly published by Foris Publications Holland.

앝 Printed on acid-free paper which falls within the guidelines 앪 of the ANSI to ensure permanence and durability.

ISBN 3-11-018112-6 Bibliographic information published by Die Deutsche Bibliothek Die Deutsche Bibliothek lists this publication in the Deutsche Nationalbibliografie; detailed bibliographic data is available in the Internet at ⬍http://dnb.ddb.de⬎.

쑔 Copyright 2004 by Walter de Gruyter GmbH & Co. KG, D-10785 Berlin. All rights reserved, including those of translation into foreign languages. No part of this book may be reproduced in any form or by any means, electronic or mechanical, including photocopy, recording, or any information storage and retrieval system, without permission in writing from the publisher. Cover design: Christopher Schneider, Berlin. Printed in Germany.

Contents

Foreword . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1

1.1. Preview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2. Grammar model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.3. Assumptions on V-movement and adverbs . . . . . . . . . . . . . . . .

3 4 6

2. Preverbal subjects: Spec,IP or left-dislocated? . . . . . . . . . . . . . . . 11 A. B. C. D. E. F. G. H. I.

Multiple preposing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Unstressed negative QPs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . No minimality effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Reconstruction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Unmarkedness of SVO . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Raising constructions and definiteness . . . . . . . . . . . . . . . . . . . Pronominal doubling contextually limited (Costa 2000) . . . . . . Subject positions in C-less if-clauses (Costa and Galves 2002) . . Preverbal and postverbal pronominal doubling . . . . . . . . . . . . .

13 14 14 15 15 17 18 19 19

3. Postverbal subjects: syntax and discourse – VSO and VOS orders . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23 3.1. Post-verbal subjects in VSO context . . . . . . . . . . . . . . . . . . . . . 23 3.2. Postverbal subjects in VOS contexts . . . . . . . . . . . . . . . . . . . . . 30 3.3. Objects in VOS: Scrambling in European Portuguese . . . . . . . . 35 4. Inversion and information structure . . . . . . . . . . . . . . . . . . . . . . . 71 4.1. 4.2. 4.3. 4.4. 4.5. 4.6. 4.7. 4.8.

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71 Focus-movement? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72 Word order and focus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78 Mapping syntax-discourse . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81 No focus-movement in Portuguese . . . . . . . . . . . . . . . . . . . . . . 91 Phases, locality and subjects in Spec, VP . . . . . . . . . . . . . . . . . . 97 Related evidence: agreement in copular constructions . . . . . . . 100 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105

vi

Contents

5. Optionality and left-dislocated subjects: semantic and discourse properties . . . . . . . . . . . . . . . . . . . . . . . . . 107 5.1. SV and VS in unaccusative contexts . . . . . . . . . . . . . . . . . . . . . 107 5.1.1. Lack of agreement in postverbal position . . . . . . . . . . . . 109 5.1.2. Hypothesis and arguments . . . . . . . . . . . . . . . . . . . . . . . 112 5.1.3. Summing up: is there optionality? . . . . . . . . . . . . . . . . . 118 5.2. VSO and SVO in answers to multiple wh-questions . . . . . . . . . 119 5.2.1. Semantic properties of multiple-wh questions . . . . . . . . 120 5.2.2. Syntactic consequences . . . . . . . . . . . . . . . . . . . . . . . . . . 125 6. Subjects in Spec,TP and the interface with morphology . . . . . . . 129 6.1. Spec,TP available in I-to-C contexts only . . . . . . . . . . . . . . . . . 129 6.2. The interface with morphology . . . . . . . . . . . . . . . . . . . . . . . . . 131 6.3. The non-parametric availability of Spec,TP . . . . . . . . . . . . . . . 138 7. Syntactic outputs and the interfaces . . . . . . . . . . . . . . . . . . . . . . . . 141 7.1. The behaviour of ditransitives in English and in European Portuguese . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142 7.2. The behaviour of possessives in Portuguese and Italian . . . . . . 153 7.3. The differences between subject-verb inversion in Brazilian and European Portuguese . . . . . . . . . . . . . . . . . . . . 155 7.4. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157 8. Summary and conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159 Appendix: On the nature of agreement in European Portuguese . . . 163 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163 1. The facts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165 1.1. DP-internal number agreement . . . . . . . . . . . . . . . . . . . . 165 1.2. Subject-verb agreement . . . . . . . . . . . . . . . . . . . . . . . . . . 167 1.3. Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173 2. Proposal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174 3. Further predictions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182 4. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195 Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209

Foreword

This book is a much revised version of my 1998 dissertation, integrating some of the work developed afterwards. The book shares with the original work the plan of analyzing patterns of word order variation. However, it does not focus only on the interface with discourse, but also on the interfaces between syntax and semantics and syntax and prosody. By doing so, it is possible to discuss and analyze more data, explaining not only the differences between subjects in Spec,Agr and Spec,VP, but also accommodating data involving Spec,TP and left-dislocated subjects. I would like to thank all colleagues who have contributed to this research with comments, questions and suggestions. Special thanks to my family and friends for making life a really nice thing!

1. Introduction

European Portuguese, like other Romance languages, displays a great amount of word order variation. Out of the six logically possible permutations between Subject, Verb and Complement in a transitive sentence, five are possible: SVO, VSO, VOS, OVS and OSV. The primary goal of this book is to provide an analysis of the several positions where the subject may surface in European Portuguese. Departing from an architecture of the clause as sketched in early minimalist work, containing two subject-related functional categories above VP (AgrP and TP), it is shown that the subject may surface in all potential landing sites: Spec,AgrP, Spec,TP and Spec,VP. Moreover, just like any other argument of the clause, it is claimed that subjects also have the possibility of surfacing in a left-dislocated position, arguably adjoining to the clause’s left periphery. It is shown that there is no free variation. Each of these positions may be occupied by the subject, only if two requirements are met: i) The position is made available by syntax; ii) The position does not violate any interface condition. In other words, the following model is argued for: syntax generates legitimate outputs. At the interface levels, each output may be selected or filtred out, according to requirements of the interface. These interface licensing conditions operate in the following way for each of the identified surface positions: 1. Spec,VP – The subject may surface in Spec, VP, because it is able to check Case under Agree (Chomsky 2000). Likewise, Case may be checked under Move. The consequence is that both SVO and VSO outputs are equally well-formed from a syntactic point of view. It is argued that Information Structure constraints and their interplay with prosody may choose a VSO ouput over an SVO order, when the subject is the focus of the sentence and must receive the sentence’s nuclear stress.

2

Introduction

2. Spec,TP – This position provides an interesting puzzle. Looking at adverb positions, it appears that Spec,TP is an available surface position for subjects in I-to-C contexts. This observation is explained if the interface with morphology is taken into account. It is argued that the subject cannot be stranded in Spec,TP when the subject is in T, since it blocks the merger of the heads Agr and T. 3. Left-dislocation and Spec,AgrP – Non-focused preverbal subjects are shown to occupy the specifier of the topmost functional category of the inflectional domain. This goes against recent claims in the literature that preverbal subjects in null subject languages are left-dislocated. Nevertheless, it is shown that the fact that preverbal subjects occupy an Aposition does not imply that they necessarily must occupy an A-position. Looking at contexts of apparent optionality in answers to multiple whquestions, it may be shown that in the appropriate context subjects may be left-dislocated. For a subject to appear in adjunction to the clause, it must meet semantic requirements such as non-exhaustivity. The picture emerging from the proposal made in this book is the following: syntax proper does not need to refer to conditions best placed at the interface. All that is needed from syntax is that it generates an array of wellformed outputs. Such outputs may be evaluated a posteriori by each of the interfaces. If they meet requirements of the interface, they are selected as legitimate. If, on the contrary, some interface condition is violated, they are ruled out. Under this approach, three independent results are derived: i) an explanation is found for the patterns of word order variation; ii) syntax proper may be reduced to its own tools, not having to manipulate semantic, discourse or prosodic variables; iii) the intuition that European Portuguese is an SVO language is derived: this word order corresponds to the one in which the subject occupies the only specifier position in which the other interfaces play no role. In this first chapter, a preview of the book is presented. Also, some general assumptions necessary for the analyses to be developed are spelt out and motivated.

Preview

3

1.1. Preview In the past few years, some authors have proposed that there is no preverbal A-position for subjects in Null Subject languages (Barbosa 1995; Alexiadou and Anagnostopoulou 1998; among many others). If this proposal is right, it is pointless to investigate the difference between subjects in Spec,IP (=Spec,AgrP) and subjects stranded in Spec, VP. Chapter 2 tests the proposal that preverbal subjects are left-dislocated in European Portuguese, providing evidence showing that this analysis makes wrong predictions in several respects. The tests presented involve a comparison with the syntactic behavior of left-dislocated (non-subject) constituents, an evaluation of definiteness effects, and the observation of agreement facts. The conclusion drawn is that preverbal subjects may occupy an A-position in European Portuguese, which precludes a generalization stating that this is impossible in Null Subject Languages. The third chapter investigates the position of the subject in VSO and VOS orders. Arguments are presented showing that postverbal subjects may be analyzed as stranded in Spec, VP, which does not exclude an I-to-C analysis for different sorts of inversion. It is further contended that the object in VOS orders displays properties reminiscent of the ones found in Germanic shortdistance-scrambling. It is therefore claimed that VOS is analyzed as a involving a scrambled complement and a subject in Spec, VP. This type of analysis is compared to alternatives, in particular to remnant-movement proposals. A first approximation to the contexts in which inversion is legitimate is made, suggesting that the interface between syntax and discourse may be at play. The contexts in which inversion orders are produced is investigated, and the conclusion reached is that subjects are in Spec, VP only if they are focused. This observation will favor an implementation of Reinhart’s (1995) analysis of scrambling in Germanic for these data in terms of an interface between syntax, information structure and prosody. The type of interface between Information Structure, syntax and prosody is further explored in chapter 4. It is argued that there is no need to assume that focused constituents move to discourse-related functional categories, and that the relevant data follow from an algorithm for sentence-stress assignment. The relations found between word order and discourse context allow for clearly saying that there basically is no free variation. In the final section of this chapter, it is argued that syntax generates multiple outputs that can be used for different discourse purposes, because Case can be licensed under Move or Agree.

4

Introduction

Locality effects on inversion and sensitivity to phase-boundaries supporting this analysis are presented. Chapter 5 investigates a case of apparent optionality, which would contradict the conclusion of the previous chapter. In answers to multiple whquestions of the type who did what?, both SVO and VSO are legitimate. The conclusions drawn from a closer look at the two word orders are the following: i) they are not semantically equivalent: the postverbal subject is not a true focus, and it must be non-exhaustive; ii) the postverbal subject behaves as if it is left-dislocated. These two conclusions combined show that the interface with semantics is also relevant for deriving all word order patterns found, and that, as defended in chapter 2, the claim that preverbal subjects are in Spec,AgrP and the claim that preverbal subjects may be leftdislocated do not exclude each other, provided that appropriate semantic and discourse conditions are met. Chapter 6 deals with the availability of Spec,TP as a surface position for the subject. It is shown that, in declarative SVO sentences, this is not a legitimate landing site. However, if there is I-to-C movement, as in wh-questions, the subject may surface in Spec,TP. The explanation for this puzzle relates the fact that there is short V-to-I movement and explores the interface between syntax and morphology, adapting Bobaljik’s (1995) solutions for the availability of Spec,TP in Germanic languages. It is argued that, since the verb does not move to Agr, for Agr and T morphology to merge, these two heads must be adjacent. For the adjacency requirement to be met, Spec,TP must be empty. In I-to-C contexts, the verb must move from T-to-Agr for locality reasons, and the adjacency is no longer relevant, which explains the restricted availability of this position. In the last chapter, I summarize the results obtained throughout the book, discussing in further detail the relation between the syntactic component and the interfaces. In particular, it is argued that the interfaces act as filters or as selectors of multiple outputs generated in the syntactic component. According to this view, syntax proper excludes interface considerations, and does not have to integrate discourse, prosodic or morphological notions.

1.2. Grammar model Since this book deals with interface issues, it is important to spell out the basic assumptions concerning the relation established between the several grammar components. I will be basically assuming a model of grammar, as

Grammar model

5

depicted in Chomksy (1995), combining it with proposals made in Reinhart (1995) for the interplay between syntax and discourse, and in Halle and Marantz (1993) for the interface between syntax and morphology. Accordingly, I assume that the output of syntax after Spell-Out feeds a morphological component of the grammar. I further assume that the level of Information Structure, in which notions such as topic and focus play a role, has access to the syntactic outputs. Obviously, these two considerations make it necessary to free syntax from discourse-related rules or operations as well as from deriving all morphological aspects. These two consequences will be explored in chapter 4 and 6, respectively. The architecture assumed is summarized in the schema in (1): (1)

Lexicon Spell-Out

Morphology

PF

Information Structure LF As emphasized by many authors, in particular in the last decade (Adger 1994; Reinhart 1995; Zubizarreta 1998; Costa 1998), this type of articulated model of the grammar opens up possibilities for dealing with apparently free word order. If syntax generates multiple converging outputs, each one of such outputs may be used in compliance with different requirements of one of the interfaces. Alternatively, syntax may generate multiple outputs, and some of them may be ruled out at one of the interface levels. The former situation will be argued for to explain the alternation between SV and VS orders in Portuguese. As mentioned above, I will argue that a subject may be licensed both in Spec, VP and in Spec,IP. These two outputs may be used for satisfying different requirements of the interfaces with prosody and information structure. The latter case will also be explored. It will be argued that some subject positions are not available, because the presence of the subject in that specific position violates some interface requirement. This will be the argument built for subjects in Spec,TP. The task I will be doing in this book is therefore the following. Let us assume a clause structure as the one proposed in Pollock (1989) and revised in Belletti (1990): (2)

[ AgrP [ TP [ VP ]]]

6

Introduction

According to this structure, in principle, there are three A-positions where the subject can surface. Moreover, if there are processes of left-dislocation, the subject can also surface in the left periphery. Since it is known that there is no free variation, it must be established if all positions must be used, and under what conditions.

1.3. Assumptions on V-movement and adverbs In most cases to be dealt with, the position of the subject will be determined by looking also at the position of the verb. This is very obvious for the alternation between SV and VS. As was made clear above, I will explore the patterns of word order variation found assuming that there are several surface positions for the subject. Therefore, it must be previously established where the verb is, in order to make it possible to take its position as a valid diagnostic for detecting subject positions. Likewise, since adverbs are a traditional diagnostic to detect V-movement and different landing sites for arguments, it must be established which adverbs are used as a diagnostic, motivating their usage. The goal of this section is therefore to spell out my assumptions concerning V-movement and adverb placement. In Costa (1996, 1998), it is argued that, in European Portuguese, verbs do move out of VP, but do not target the highest functional head. Assuming the clause articulation in (2), the proposal made is that V moves to T, not reaching Agr. Let us review the evidence in favor of this analysis. The first fact to be mentioned is the following: like French infinitives, verbs in European Portuguese may either precede or follow an adverb: (3)

a. O João ontem leu o livro. João yesterday read the book b. O João leu ontem o livro. João read yesterday the book

In his (1989) paper, for dealing with similar French data, Pollock suggests that the verb may either stay inside VP or move up to Agr. Given the current trend to eliminate optionality from the grammar, Pollock (1994b) makes an attempt to circumvent this problem by saying that when there is optional movement, what is at stake is the occurrence of forms which are morphologically ambiguous. This would be the case for French infinitives, which

Assumptions on V-movement and adverbs

7

would allow an interpretation either as nominal forms or as verbal forms. Since in French, only verbal elements do move to I, only when the infinitival forms are interpreted as verbal will this movement be required. If the infinitival form is interpreted as nominal, movement is not induced. This analysis is however not tenable for European Portuguese, since the case of apparent optional verb movement just presented cannot be analyzed in terms of morphological ambiguity. The form leu/read in (3) is unambiguously a verbal form in the third person singular. The mechanism proposed by Pollock (1994b) for the optional movement of French infinitives may not be involved here, since this form is not subject to be analyzed as a nominal element. An alternative view is to assume that adverbs may be adjoined to different projections. Such an assumption, combined with the possibility of moving the verb to intermediate functional projections (which is also assumed in Pollock 1989), would then derive the word order facts in Portuguese. (4) illustrates the relevant analysis: (4)

[AgrSP O João [TP leu [VP tV o livro]]]

If the adverb ontem is adjoined to TP, we get the ‘English-like order’: S-Adv-V-O: (5)

[AgrSP O João [TP ontem [TP leu [VP tV o livro]]]]

If the adverb is adjoined to TP, the surface word order will match the pattern typical of French: S-V-Adv-O: (6)

[AgrSP O João [TP leu [VP ontem [VP tV o livro]]]]

The same type of analysis is also preferred if one takes into account sentences containing sequences of auxiliaries and in which adverbs may surface in several positions without radical meaning changes. (7) is an example of such a case: (7)

(provavelmente) O Paulo (provavelmente) tinha (provavelmente) probably the Paulo probably had probably lido (provavelmente) o livro (provavelmente) à Maria read probably the book probably to Maria (, provavelmente). probably.

8

Introduction

Adopting the rigid view on adverbs that there is a single position for adverb attachment and trying to derive the multiple possibilities via optional movements of all the other constituents is quite implausible, since it creates the rather unlikely need of making as many optional movements as there are constituents and positions for adverbs. The method of attaching the same adverb to different positions is not new. It has been used in the literature to derive the same effects of Pollock (1989) by Iatridou (1990) and Williams (1994), among others. It has also been used by Zwart (1993) to derive scrambling in Dutch, without resorting to optional object movement.1 Another potential approach to the order S-Adv-V in Portuguese would be to follow Belletti’s (1990) approach to similar facts in Italian. (8)

Gianni probabilmente sbaglierà. Gianni probably fail-FUT-3ps ‘Gianni will probably fail.’

Belletti (1990) argues that the order presented in (8) is to be derived in terms of left-dislocation of the subject, which would explain that the subject and the verb (in I) would not be adjacent. This analysis is confirmed by her observation that the order S-Adv-V is only possible with definite subjects (indefinites may not be left-dislocated in Italian) or with indefinites bearing a heavy stress (which may independently be A-bar moved, in contrastive focus constructions). The contrast between the two types of elements is exemplified in (9): (9)

a. Ognuno probabilmente sbaglierà. everyone probably fail(-future) b. *Nessuno probabilmente sbaglierà. no-one c. NESSUNO probabilmente sbaglierà. ‘No-one will probably make a mistake’

Although, it seems to make the right predictions for Italian, this analysis cannot be extended to Portuguese. The reason is that the counterpart of (9) does not display any asymmetries: both types of subjects may occur in the order S-Adv-V, independently of focal stress. This is exemplified in (10) below:

Assumptions on V-movement and adverbs

(10) a.

9

Todos provavelmente errarão.

b.

Ninguém provavelmente errará.

c.

NINGUÉM provavelmente errará.

Another argument for not adopting Belletti’s analysis for Portuguese comes from the fact that there is no focus-movement of the Italian type in Portuguese, as I will show in chapter 4. An additional argument for the claim that the verb only moves up to T comes from the distribution of floating quantifiers. Since Sportiche (1988), floating quantifiers are taken as a diagnostic for A-movement. More specifically, Sportiche argues that floating quantifiers indicate where a subject has been on its way to Spec,IP. For instance, in the sentence in (11) from Portuguese, the floating quantifier may either move along with the subject to Spec,IP (11a) or be stranded in the base position of the subject in (11b): (11) a.

Todos os miudos foram ao cinema. All the kids went to the cinema ‘All the kids went to the movies.’

b.

Os miudos foram todos ao cinema. The kids went all to the cinema

Note that the distribution of floating quantifiers is one of Pollock’s (1989) arguments for saying that the verb does not move in English. He notes the following contrast between English and French: (12) a. The kids all love Mary. b. *The kids love all Mary. (13) a. Les enfants aiment tous Marie. The kids love all Marie b. *Les enfants tous aiment Marie. In (12), the verb stays in situ, and as a consequence, the stranded floating quantifier precedes it. In (13), the verb moves to I, hence the stranded floating quantifier follows it. Note that if the same argumentation is followed, there is again evidence for short verb movement of the verb in Portuguese, since the two positions for the floating quantifier are possible:

10

Introduction

(14) a.

Os miudos amam todos a Maria. The kids love all Maria

b.

Os miudos todos amam a Maria.

As for adverbs, I will assume that low monosyllabic adverbs mark the leftedge of VP. This assumption is based on the the following arguments presented in Costa (1996, 1998): – Monosyllabic adverbs do not distribute as freely as other adverbs; – Monosyllabic adverbs only surface in clause-final position when it is possible to show that the internal arguments of the verb have moved out of the VP. Summing up, on the basis of the discussion above, I will be assuming that there is short-V-movement in European Portuguese, that is, the verb moves out of VP targeting T and not the highest head of the IP-domain.2 Moreover, it will be assumed that certain adverbs may be taken as diagnostics for movement. In particular, it is argued, following the argumentation developed in Costa (1996, 1998) that low monosyllabic adverbs adjoin to the left of VP.

2. Preverbal subjects: Spec,IP or left-dislocated?

A well known fact about null subject languages is that several word orders are allowed. European Portuguese behaves as expected, in displaying almost all possible alternations between subject, verb and direct object. Considering the possible orderings between Subject, Verb and Object, it can be observed that Portuguese exhibits five out of six logically available orders: (1)

SVO ?*SOV VSO VOS OSV OVS

(2)

a. O Paulo comeu a sopa. Paulo ate the soup b. ?*O Paulo a sopa comeu. c. Comeu o Paulo a sopa. d. Comeu a sopa o Paulo. e. A sopa, o Paulo comeu. f. A sopa, comeu o Paulo.

In this chapter, I will discuss the status of preverbal subjects, providing tests to identify the position it occupies. This is a relevant task, in light of recent proposals suggesting that preverbal subjects in Null Subject Languages are left-dislocated (Barbosa 1995; Alexiadou and Anagnostopoulou 1998). The pattern of word-order variation presented in (2) is well known and has been subject to several analyses in the Generative tradition, among which Ambar (1992) and Duarte (1987) were the first studies considering the relation between these word orders and discourse. Both authors agree that, in spite of the variation, the unmarked/canonical word order of European Portuguese is SVO. It is interesting to investigate why it is the case that SVO is considered unmarked, or putting it in other terms, why an unmarked context does not allow for several word orders to co-occur. Duarte (1987)

12

Preverbal subjects: Spec,IP or left-dislocated?

presents an analysis of sentence-initial objects, showing that they may be either topicalized, as in (3a) or clitic-left-dislocated, as in (3b): (3)

a. O bolo, o Pedro comeu. the cake, the Pedro ate. b. O bolo, o Pedro comeu-o. the cake, Pedro ate itCL-ACC

I will assume her analysis for sentence-initial objects and concentrate on preverbal subjects, and take the properties associated with left-dislocation of objects in order to test the status of preverbal subjects. Sentence-initial preverbal subjects, like in (4), are traditionally assumed to occupy the Spec, IP position (cf. Duarte 1987; Ambar 1992; Martins 1994). (4)

O Paulo comeu a sopa. Paulo ate the soup

This assumption has been challenged by several authors (see e.g. Barbosa 1995, 2000 for Portuguese, Valmala Elguea 1994, Ordoñez and Treviño 1995 for Spanish and Alexiadou and Anagnostopoulou 1995, 1998 for Greek). The assumption shared by these authors is that in Null-SubjectLanguages preverbal subjects are not in Spec,IP, but rather left dislocated. As noted, for instance in Barbosa (1995), this has theoretical advantages in a framework like the one proposed in Chomsky (1993). According to Chomsky, derivations are uniform, and the overt or covert nature of a syntactic operation is determined by the specification of formal features associated either with the categories to be moved or with the functional category where they land. Strong features trigger overt movement and weak features trigger covert movement only. Within such a framework, optionality in word order of the type observed above is a problem, since it is theoretically undesirable to have features that are at the same time weak and strong. Barbosa and the other authors propose that N-features of the subjects in Null Subject languages are uniformly weak. Hence, subjects are only expected to move to Spec,IP in covert syntax or they do not move at all, if agreement is pronominal and sufficient to check EPP-features. Now, nothing prevents them from being topicalized or left-dislocated like other categories are, since overt movement of topics may be formalized by assigning them some [topic] feature requiring checking.

Multiple preposing

13

In this section, I will review this analysis and argue that, although some preverbal subjects may be left-dislocated, the traditional analysis according to which preverbal subjects are in Spec,IP makes right predictions for European Portuguese. This debate regarding the status of the preverbal subject is quite controversial, since some data are not very clear. In any case, it is possible to compare the two analyses, showing that some data must involve subjectleft-dislocation, while other data must involve assuming that the subject is in Spec,IP. In Costa (1998), a more radical position was assumed: it was claimed that the analysis involving subject-left-dislocation could not be right for most cases. However, as it will be shown below, there is no clear reason not to assume that both analyses are compatible.3 In order to decide whether the subject is in Spec,IP or not, it is necessary to find out whether preverbal subjects exhibit A- or A-bar properties. For this reason, it will be necessary to compare preverbal subjects with A-bar left-peripheral constituents. Let us then list some of the arguments in favor of assuming that preverbal subject in European Portuguese are in Spec,IP.

A. Multiple preposing When two complements of the verb are preposed, their relative order is not rigid. This is illustrated in (5): (5)

a. Aos alunos, sobre sintaxe, o Rui falou. to the students, about syntax, Rui talked b. Sobre sintaxe, aos alunos, o Rui falou. about syntax, to the students, Rui talked

Assuming that the subject is left-dislocated, it is expected that the order between a preposed complement and the subject is not rigid. However, as shown in (6), there is a contrast between SOV and OSV: (6)

a. Esse bolo, o Paulo comeu-o. that cake Paulo ate it b. ??O Paulo, esse bolo, comeu-o. Paulo, that cake, ate it

14

Preverbal subjects: Spec,IP or left-dislocated?

Assuming that the preverbal subject is in Spec,IP, it is expected that it will not be able to surface before the complement. The lack of ordering restrictions will only be expected if both sentence-initial elements are adjuncts.

B. Unstressed negative QPs Another argument in favor of the analysis according to which preverbal subjects are in Spec,IP comes from the fact that it is possible for a negative QP to occur in sentence-initial position, non-adjacent to the inflected verb, as in (7): (7)

Ninguém provavelmente leu esse livro. nobody probably read that book

This word order is impossible in Italian. Belletti (1990) shows that a negative QP may only occur before a sentence adverb if it bears heavy stress. According to her analysis, in that case, the negative QP is A-bar moved. The impossibility of non-adjacency between unstressed negative QPs and the inflected verb is explained in terms of the impossibility of left-dislocating negative QPs. (8)

NESSUNO/*Nessuno probabilmente ha sbagliato. nobody probably failed.

The fact that in European Portuguese there is no contrast like the one in Italian argues for an analysis in which the preverbal negative QP is not necessarily left-dislocated. In (8), the negative QP is in Spec,IP, which is not inherently associated with topic properties; it may be unstressed, since it is not an A-bar position.4

C. No minimality effects Like in English, in embedded contexts, wh-movement is not obligatorily followed by I-to-C movement. This fact makes it possible to observe adjacency between the moved wh-phrase and the subject, like in (9a). The analysis assuming that the subject is left-dislocated predicts that the whconstituent may be followed by some other left-dislocated constituent. This prediction is however not borne out.

Reconstruction

(9)

a.

15

Perguntei que livro o Pedro leu. I asked which book Pedro read

b. *Perguntei que livro, à Maria, lhe deram. I asked which book, to Maria, herCL-DAT (they) gave It is important to note that there seems to be variation concerning the grammaticality of (9b).5 In any case, since there are speakers for whom the contrast exists, the difference can only be explained by assuming a different status for the subject and for the left-dislocated constituents. A straightforward way of explaining this contrast is to assume that wh-movement to the left of a left-dislocated constituent yields a problem of minimality. If the subject is in Spec,IP, it is expected that it will not create any problem regarding minimality, since it occupies an A-position.

D. Reconstruction As it is well-known, A-bar movement obligatorily reconstructs, unlike Amovement. If it is assumed that the preverbal subject is in Spec,IP, it is expected that there is no reconstruction of the subject, while a left-dislocated complement may reconstruct. In fact, the preverbal subject in (10a) cannot reconstruct, scoping obligatorily over the agent. In (10b), the left-dislocated complement may reconstruct, and the sentence is ambiguous: (10) a. Três livros foram lidos por dois estudantes. three books were read by two students b. Três livros, dois estudantes leram-nos. three books two students read them

S>Ag; *Ag>S S>O; O>S

The fact that there is no ambiguity in (10a) and the contrast between the two cases seems to indicate that the preverbal subject and the preverbal object occupy different positions.6

E. Unmarkedness of SVO As it will be discussed later in this chapter, unmarkedness may be detected by looking at answers to the question what happened?. This context is

16

Preverbal subjects: Spec,IP or left-dislocated?

unmarked from the point of view of discourse, since all elements will yield new information. In EP, the answer to this question is SVO: (11) A: O que é que aconteceu? what happened B: a. O Pedro partiu o braço. Pedro broke the arm b. #Partiu o Pedro o braço. broke Pedro the arm c. #O braço, o Pedro partiu-o. the arm, Pedro broke it The fact that SVO emerges in this context is problematic for the left-dislocation analysis for two reasons. First, it is possible to observe that left-dislocation is illegitimate in this context (11c), which makes it difficult to explain why the subject can be left-dislocated if other elements cannot. Second, even if one would assume that the subject could be left-dislocated, it would be necessary to explain why it cannot stay in its base position (Spec, VP). Note that the impossibility of (11b) is not due to the exhaustive nature of the subject in inversion constructions (Costa 2000), since the behavior is the same in mono-argumental intransitive contexts:7 (11) O que é que aconteceu? what happened a. O João espirrou. João sneezed b. #Espirrou o João. sneezed João a. O João viajou. João travelled b. #Viajou o João. travelled João Once again, if the only A-position for the subject were Spec, VP, inversion should be found in this context.

Raising constructions and definiteness

17

F. Raising constructions and definiteness In Barbosa (2000), cases of preverbal subjects are presented that can only be analyzed as instances of left-dislocation. For instances, sentences like (12) must involve subject left-dislocation, and not movement to Spec,IP, otherwise they would be instances of super-raising: (12) a. O homem parece que viu um monstro. the man seems that saw a monster b. O João parece que está parvo. João seems that is fool Her analysis is corroborated by the fact that there is a definiteness effect in this type of construction. If the preverbal subject is indefinite, the left-dislocation is impossible: (13) a. *Umas meninas parece que estão doentes. some girls seems that are sick b. *Baleias parece que comem peixe. whales seems that eat fish This definiteness effect is expected, since left-dislocation typically affects definite XPs. Now, the comparison with constructions that are not obligatorily analyzed as instances of left-dislocation becomes crucial at this point. The SVO word orders in (14) do not exhibit any definiteness effect:8 (14) a. O homem foi assassinado. the man was murdered a’. Um homem foi assassinado. a man was murdered b. As meninas estão doentes. the girls are sick b’. Umas meninas estão doentes. some girls are sick c. As baleias comem peixe. the whales eat fish c’. Baleias comem peixe. whales eat fish

18

Preverbal subjects: Spec,IP or left-dislocated?

The assumption that the preverbal subjects in (14) are in Spec,IP derives the lack of definiteness effects associated with left-dislocation.

G. Pronominal doubling contextually limited (Costa 2000) As shown in Barbosa (1995), in some null subject romance languages, preverbal DP subjects are obligatorily doubled by a pronoun, which strongly argues in favor of her analysis. In EP, doubling is possible, for instance in answer to a multiple wh-question (Costa 2002a): (15) A: Quem leu o quê? who read what B: a. O João, ele leu o livro. João he read the book b. O João leu o livro. João read the book The possibility of pronominal doubling in EP suggests that Barbosa’s analysis is right for this language as well. There are, however, contexts in which doubling the subject is ungrammatical. For instances, if the whole sentence is focused, as in (16), doubling is marginal at best: (16) A: O que é que aconteceu? what happened B: a. O João leu o livro. João read the book b. ??*O João, ele leu o livro. João he read the book Barbosa’s analysis, assuming that the preverbal subject is always left-dislocated, predicts that doubling by a pronoun will always be possible, like in other Romance null subject languages, which is not the case. Assuming that, in a context like the one given in (16), the subject is in Spec,IP derives the ungrammaticality of pronominal doubling in this context.

Subject positions in C-less if-clauses

19

H. Subject positions in C-less if-clauses (Costa and Galves 2002) Complementizer-less if-clauses force I-to-C movement in EP, like in English. As illustrated in (17), in these sentences the subject appears right after the inflected verb. The sentence initial position is ungrammatical:9 (17) a. Tivesse o João ido ao Brasil... had João gone to Brazil b. *O João tivesse ido ao Brasil... John had gone to Brazil… If the subject is in Spec,IP, the word order in (17a) is predicted, since I-to-C movement will make the verb cross the subject position. If the subject were right-dislocated, one would expect to find the word order in (17b), since the left-dislocated subject would be adjoined to the left of CP.10

I. Preverbal and postverbal pronominal doubling Returning to the cases in which the preverbal subject can be doubled by a pronoun, it is possible to observe that the pronoun may be preverbal or postverbal: (18) a. O João, leu ele o livro. João, read he the book b. O João, ele leu o livro. João he read the book The postverbal position of the pronoun is not problematic for the analysis according to which the subject is left-dislocated. (18b) is problematic, since it must be analyzed as a case of left-dislocation of both the DP and the pronoun. Note that there is nothing wrong in assuming that a DP and a co-referring pronoun are left-dislocated. Such a case is illustrated in (19), in which the DP complement is doubled by the strong pronoun a ele. The strong pronoun itself is left-dislocated and doubled by a clitic: (19) O João, a ele, vi-o no cinema. João, to him, I saw himCL-ACC in the movie theater

20

Preverbal subjects: Spec,IP or left-dislocated?

Now, if the clitic may cooccur with multiple fronting, as in (19), it is expected that the postverbal pronoun, corresponding to the clitic in (19), may co-occur with the preverbal DP subject and the preverbal pronoun. This prediction is not borne out: (20) O João, ele leu (*ele) o livro. João he read he the book Assuming that the preverbal pronoun is in Spec,IP, it is expected that it will not be possible to find another subject in a lower position, since the subject in Spec,IP was generated in Spec, VP. In Costa (2003) and Costa and Duarte (2003), some additional evidence against the idea that null subject languages are languages that check EPP by means of pronominal Agr was presented. First, there are languages which are only semi-pro-drop. This is the case of Brazilian Portuguese (Coelho, Costa, Figueiredo Silva and Menuzzi 2001) and Cape verdean creole (Pratas 2002). In these two languages, referential null subjects are ungrammatical, but expletive pro, available with weather verbs and unaccusative inversions, is available. This fact, predicted under Rizzi’s (1982) licensing conditions for pro, is illustrated in the examples below: (21) a. Chegou o Pedro. arrived Pedro

BP

b. *Viajou o Pedro. traveled Pedro

BP

c. *pro viajou. traveled

BP

(22)

Está chovendo. is raining

BP

(23) a. Txiga tres pesoa. CVC arrived three persons b. *Papia tres pesoa. talk three persons

CVC

c. *pro papia talks

CVC

Preverbal and postverbal pronominal doubling

(24) Txobi. rains

21

CVC

Inversion is possible in BP and CVC, in contexts in which Spec,IP is occupied by expletive pro. These languages do not allow referential pro. Crucially, if the availability of null subjects were a consequence of the availability of a pronominal Agr responsible for EPP checking, there should be no mixed systems. The existence of mixed systems suggests that what is at stake is an interaction between different constraints, as suggested by Rizzi (1982) and Coelho et alii (2001), among others.11 The relevant constraints are: structural (i.e. is Spec,IP projected?), which has consequences for checking of EPP and Casefeatures; lexical (i.e. does Spec,IP have to be lexically filled?), which is an EPP matter; paradigmatic (i.e. does pro exist in the pronominal system?). A language in which Spec,IP is not projected will be a language with both referential and expletive null subjects. A language in which Spec,IP is projected, but it does not need to be lexically filled will be a language with both types of null subjects or with expletive null subjects only, depending on the availability of pro. Another domain providing counter-evidence for the claim that preverbal subjects are left-dislocated is language acquisition. If preverbal subjects are left-dislocated in null subject languages, it is expected that VSO is unmarked, and that children will only produce SV sentences by the time they master leftdislocation (Adragão and Costa 2003). A study of the acquisition of subjects in a child in his second year of life (Adragão 2001) reveals that inversion is highly marked and rare in the child’s early productions: (25) % of SV/VS utterances in the child’s productions: SV – 93 VS – 7

(out of 1060 sentences)

From these, most VS structures corresponds to passives, unaccusatives (79%) and predicative structures, which are contexts in which inversion is unmarked in the target adult system as well. It is importante to note that, in her study, Adragão was looking for contexts in which the subject should be inverted in the adult counterpart. Notice, as well, that at this stage, there are very few OV sentences, and that there is no evidence for strategies of clitic left-dislocation, since the acquisition of clitics is quite late in EP (cf. Duarte and Matos 2000). These data provide additional evidence for the unmarked status of the pre-

22

Preverbal subjects: Spec,IP or left-dislocated?

verbal position for subjects. The correlation between inversion and unaccusativity make it impossible to claim that children do not know whether Spec,IP is projected. Finally, the existence of inflected gerunds in Dialectal European Portuguese, and their behavior in identifying the reference of null subjects (Ribeiro 2002) casts some doubt on the pronominal status of Agr in this language. The problem identified by Ribeiro is that there is no great difference between dialectal and standard EP, as far as the licensing of null subjects is concerned.12 Independently of the presence of Agr, a null subject in a gerund clause must be controlled by the subject of the matrix clause, as shown in (26): (26) Dialectal EP:

Standard EP:

a. Estandes eci cansado, tui podes ir. being-2sg tired, you may go

a. Estando eci cansado, tui podes ir. being tired, you may go.

b. Estandes tu cansado, eu posso ir. being-2sg you tired, I may go

b. Estando tu cansado, eu posso ir. being you tired, I may go

c. *Estandes eci cansado, euj posso ir. being-2sg tired, I may go

c. *Estando eci cansado, euj posso ir. being tired, I may go

The ungrammaticality of (26c), and the lack of difference between the two dialects casts some doubt on the idea that Agr is pronominal in European Portuguese. From the results of the tests listed above, I conclude that the traditional analysis assuming that preverbal subjects in EP are in Spec,IP (cf. Ambar 1992, Duarte 1992, among others) cannot be dispensed with. It is important to note that the results of the application of the tests above to European Portuguese do not imply that the results may be extended to other Null Subject Languages. Actually, in Costa (1998), I assumed that the descriptions and analyses of Greek by Alexiadou and Anagnostopoulou (1998) and Spanish by Ordoñez and Treviño (1995) are correct, and accounted for the difference between these languages and Portuguese. As implicit in the discussion above, stating that subjects may be in Spec,IP does not mean that they are in Spec,IP in all constructions. As shown in Barbosa (2000), some constructions must be analyzed as instances of left-dislocation of the subject (see for instances the cases of apparent super-raising). In fact, since all constituents may be left-dislocated, subjects do not behave differently: in chapter 4, I will show that they can be left-dislocated in a specific context.

3. Postverbal subjects: syntax and discourse – VSO and VOS orders

In this chapter, the properties of postverbal subjects will be studied. It will be argued that VS word orders may arise not only in contexts of I-to-C movement (Ambar 1992), but also if the subject is stranded in its base position, Spec, VP. Two contexts for post-verbal subjects will be considered: VSO and VOS, both in declarative contexts. For both contexts, the conclusion will be that the subject is in Spec, VP. The argumentation will be the following: first, I will show that the postverbal position of subjects is not necessarily a consequence of V-movement across the subject from I° to another functional projection. This conclusion will be based on similarities between root and embedded contexts. The analysis will also be based on assumptions regarding the reliability of adverb positioning as a diagnosis for tracing the left-edge of VP.

3.1. Post-verbal subjects in VSO context In any theory admitting the existence of functional projections, which are potential landing sites for V, and accepting the VP-Internal Subject Hypothesis (Koopman and Sportiche 1991), there are at least three ways of deriving postverbal subjects in VSO order. Such mechanisms are illustrated in (1), (2) and (3): (1)

[FP Vi [IP Subjectk [I’ ti [VP tk ti Object]]]]

(2)

[IP Vi [VP Subject [V’ ti Object]]]

(3)

[IP Vi [XP Subjectk [VP tk ti Object]]]

(1) illustrates an analysis of VSO, according to which this word order arises in the following way: both Subject and Verb move out of VP; the subject stops in Spec,IP, but the verb is further moved to the head position of a functional projection above IP.13

24

Postverbal subjects: syntax and discourse – VSO and VOS orders

In (2), an alternative is presented: the verb moves up to I° and stops there. The subject does not precede it, because it has never been moved from its base-position Spec, VP. Finally, (3) illustrates another type of analysis, according to which the subject undergoes short-movement out of VP, in the sense that it does not move all the way up to its ‘normal’ landing site, Spec,IP. The verb undergoes movement to I°. It is important to note that these three types of analyses do not exclude each other. In fact, it may be seen that they are all independently instantiated in different types of languages. For instance, (1) is the type of analysis argued for Dutch and German Verb-second phenomenon and for English SubjectAux inversion in questions (cf. 4); (2) has been argued by Ouhalla (1991) to be the correct analysis for Celtic VSO (5); (3) has been argued for Icelandic Transitive Expletive Constructions by Bobaljik and Jonas (1996): (4)

a. Dutch V2: Gisteren heeft Jan het boek gelezen. Yesterday has Jan the book read b. English Subject-Auxiliary Inversion: Who had Mary seen?

(5)

Celtic VSO: Darnellod y plentyn y lyffr. Read the child the book

(6)

Icelandic Transitive Expletive Constructions: fia› hafa margir jólasveinar bor›a› bú›ing There have many Santa Clauses eaten puding

For European Portuguese, all three types of analyses have been proposed: Ambar (1992) has argued that post-verbal subjects in this language are the reflex of movement of I-to-C, along the lines of (1); Martins (1994) and Duarte (1997) have proposed that under certain circumstances subjects may be stranded in the Specifier of a functional projection below IP but higher than VP, following the lines of (3); in this chapter, as in Costa (1997, 1998), I argue for an analysis along the lines of (2). The fact that the three types of analyses exist independently of each other does not mean that they exclude each other. For instance, it is possible to

Post-verbal subjects in VSO context

25

assume that V moves to C in questions, as proposed in Ambar (1992), and that the subject stays in Spec, VP. That is, we may have two analyses for the sentence in (7): (7)

Quem viu o João? Who saw João ‘Who did João see?’

One analysis would assume that the subject is in Spec,IP and the verb has raised up to C, as in (8a); another analysis would claim that the subject is in Spec, VP, as in (8b): (8)

a. [CP Quem [C’ viu [IP o João [I’ tV …. b. [CP Quem [C’ viu [IP tV [VP o João….

That the two analyses are necessary is confirmed by the fact that there are two adverb positions, one before the subject and one after it, which may be explained by assuming that the adverb is adjoined to VP, and adopting the explanation in (9): (9)

a. Quem viu o João ontem? yesterday a’. [CP Quem [C’ viu [IP o João [I’ tV [VP ontem…. b. Quem viu ontem o João? b’. [CP Quem [C’ viu [IP tV [VP ontem [VP o João….

Bearing this idea in mind, I will now show that for simple declarative affirmative sentences, an analysis claiming that post-verbal subjects are in Spec, VP is more attractive. There are four main arguments for not deriving VS order in declarative sentences necessarily in terms of an analysis involving I-to-C movement. The arguments are listed in a) through c):

a) VS is possible in embedded sentences: If postverbal subjects were only derived by V-movement to C, they should not appear in embedded contexts, since C is occupied by the complementizer. This prediction is not borne out by the data: 14

26

Postverbal subjects: syntax and discourse – VSO and VOS orders

(10) O Paulo disse que comeu a Maria a sopa. ‘Paulo said that ate Mary the soup’ Note that this is not problematic for an analysis in which the subject is stranded in the Specifier of a functional projection higher than VP and lower than IP. Such an analysis will be considered below, when this word order is confronted with possible readings of adverbs.

b) Sequences auxiliary-participle(s): The second problem for the type of analysis defending that VS arises necessarily by virtue of movement of the verb to a functional projection above the landing site of the subject is the fact that in more complex verbal constructions (involving an auxiliary verb and one or more non-finite forms of the verb), the subject may follow all of them: (11) a. Tinha comido o Paulo maçãs. ‘had eaten Paulo apples’ b. Tem estado a comer o Paulo maçãs. ‘has been eating Paulo apples’ If inversion could only be derived by moving the verb across the subject to the functional projection above it, we would expect to find the subject obligatorily following the auxiliary verb, in a construction similar to Rizzi’s (1982) Aux-to-Comp. In order to assume that the subject in (31a) is in Spec,IP, one has to postulate at least two heads above IP. If more auxiliaries are present, more heads have to be postulated, as (11b) shows.

c) The distribution of adverbs: The distribution of adverbs also supports the idea that postverbal subjects do not reflect V-movement to a position higher than IP. In Costa (1998), I have proposed that a proper characterization of the distribution of adverbs with a subject-oriented reading requires that these adverbs are adjoined to TP, and that this reading is only triggered whenever the subject moves up to Spec,IP (=Spec,AgrSP). Assuming these results, there is a further diagnostic for identifying the position of the subject: if postverbal subjects were in

Post-verbal subjects in VSO context

27

Spec,IP, adverbs with a subject-oriented reading should be expected to adjoin to TP, thus intervening in between the subject and the object. This prediction is incorrect: (12) *Comeu o Paulo inteligentemente maçãs. ‘ate Paulo cleverly apples’ (12) is grammatical, but only with a manner reading for the adverb. The structure allowing (12) would be the one in (13): (13) [CP Comeu [IP o Paulo [I’ tV [TP inteligentemente [TP …[ VP tSubj tV maçãs]]]]]] On the basis of these arguments, I reject an analysis of inversion in declarative sentences that is solely based on Verb movement across the subject in Spec,IP. I stress nevertheless that such an analysis is still necessary e.g. for interrogative sentences, as shown in Ambar (1992 and subsequent work). Having pointed out some problems for an analysis of inversion solely in terms of verb movement across a subject moved to Spec,IP, it is my task now to develop the alternative analysis for VS orders. In this section I will argue that VSO orders may be analyzed as the result of a representation in which subjects stay in their base-position, Spec, VP, and do not move to Spec,IP. Evidence for this analysis will come from the distribution of adverbs. In Costa (1996, 1998), I argued that the position of monosyllabic adverbs is a good test for determining the left edge of VP. Although I emphasized that the conclusions reached are valid for English, but not necessarily for other languages, it seems that similar adverbs may be used as diagnostic for the left-edge of VP in Portuguese as well. Let us recall the two main properties of these adverbs: a) First, they do not display the properties associated with right-adjunction, which may be seen in the order PP-Adv, which, without any special intonational mark on the adverb is ill-formed both in Portuguese and in English: (14) a. John looked hard at some pictures. b. *John looked at some pictures hard.

28

Postverbal subjects: syntax and discourse – VSO and VOS orders

(15) a. O Paulo olhou bem para alguns daqueles quadros. Paulo looked well at some of those pictures b. *O Paulo olhou para alguns daqueles quadros bem. b) Secondly, monosyllabic adverbs have a more restricted distribution than other types of adverbs: (16) a. John (carefully) has (carefully) looked (carefully) at some pictures (carefully) b. John (*well) has (*well) looked (well) at some pictures (*well). (17) a. O Paulo (cuidadosamente) tinha (cuidadosamente) olhado (cuidadosamente) para aqueles quadros (cuidadosamente). b. O Paulo (*bem) tinha (*bem) olhado (bem) para aqueles quadros (*bem). Given the similarities between English and Portuguese, I will consider the monosyllabic adverb as a reliable test for marking the left-edge of VP also in Portuguese. Let us then consider the possible positions for the monosyllabic adverb bem ‘well’ in a VSO sentence: (18) a. *Bem comeu o Paulo maçãs, ‘well ate Paulo apples’ b. ?*Comeu o Paulo bem maçãs, c. *Comeu o Paulo maçãs bem. d. Comeu bem o Paulo maçãs, Example (18) shows that the only position for the adverb to surface in a VSO sentence is in between the verb and the subject. All other positions are excluded.15 If the position of these adverbs is the same in English and Portuguese, as the similarity of the data makes one suspect, it is legitimate to conclude that in the VSO word order, the subject is in Spec, VP. Note that it is not possible to adopt an analysis of this word order, in which the subject would move out of the VP to some functional projection above it but lower than AgrSP, as has been argued for Icelandic Transitive

Post-verbal subjects in VSO context

29

Expletive Constructions by Bobalijk and Jonas (1996), and for Portuguese by Duarte (1997) and Martins (1994). The problem with such an analysis is the fact that the adverbs considered here obligatorily follow the participial form in a sequence consisting of the inflected auxiliary followed by the participial form. This is exemplified in (19): (19) a. O Paulo tinha lido bem alguns livros. Paulo had read well some books b. *O Paulo tinha bem lido alguns livros. The contrast in (19) not only shows that it is not possible to extend Bobaljik and Jonas’ (1996) analysis to Portuguese, but it also provides further evidence for the claim put forward here regarding the distribution of monosyllabic adverbs. Note that the facts in (19) imply that the participial form moves out of the VP to some functional head. This is possible and confirmed by the distribution of the adverb itself. It should however be noted that the participial form could remain in VP, as it does in French (M.T. Vinet, personal communication): (20) a. Jean avait bien lu les livres. Jean had well read the books b. *Jean avait lu bien les livres. The fact that (20b) is ungrammatical in French excludes an analysis of (19) in terms of adjunction of the adverb to some lower level within VP. Assuming that monosyllabic adverbs are adjoined to VP in the two languages, we may explain the contrast between the order adverb-participle in French and Portuguese with the usual tools deriving differences in head movement: participles obligatorily move out of VP in Portuguese, while they never do so in French. I thus conclude that the distribution of adverbs seem to support the idea that postverbal subjects are in Spec, VP. Further evidence for this claim comes from the study of VOS orders.

30

Postverbal subjects: syntax and discourse – VSO and VOS orders

3.2. Postverbal subjects in VOS contexts In the preceding section, I have investigated the position of postverbal subjects in VSO contexts. In this section, I will look at another possible position for post-verbal subjects, the one in which they follow the object. I will argue for an analysis of this word order similar to that of VSO. In VOS, subjects will be claimed to be in Spec, VP, but, differently from VSO, the object has scrambled out of the VP, crossing the subject. The argumentation in defense of the movement of the object will be presented in the second half of this chapter (cf. Section 3.3). I will therefore assume the results of that section in my discussion of the subject position in VOS, and refer the reader to section 3 for the evidence supporting object movement in European Portuguese. The VOS context is exemplified in (21): (21) Comeu a sopa o Paulo. Ate the soup Paulo. In the traditional analysis for this word order, subjects are claimed to be rightdislocated (see e.g. Rizzi 1982 for Italian). I do not think that the subjectright-dislocation analysis is to be dispensed with, but would nevertheless like to clarify the data a little bit. In this section, I will be concerned with sentence-final subjects that are not right-dislocated. How can the two types of subjects be differentiated? I would like to show that right-dislocated subjects and sentence-final base-generated subjects distribute differently. The first distinguishing factor is intonation: while right-dislocated subjects are preceded by a pause, sentence-final subjects are not: (22) a. Comeu a sopa # o Paulo. Ate the soup Paulo b. Comeu a sopa o Paulo. This observation may seem at first sight rather awkward: it could be that, in both cases, subject right-dislocation is involved, the pause being optionally possible. As a result, different intonations would not be associated with different syntactic structures. However, if the paradigm in (23) is taken into consideration, it is possible to see that, when a pronoun is inserted in Spec,IP, the pause is obligatory:

Post-verbal subjects in VOS contexts

31

(23) a. Ele comeu a sopa # o Paulo. ‘he ate the soup Paulo’ b. *Ele comeu a sopa o Paulo. (24) shows that the pronoun is ungrammatical in a VSO context, which cannot be treated as a case of subject right-dislocation: (24) (*Ele ) comeu o Paulo a sopa. ‘he ate Paulo the soup’ One way to interpret these facts is to assume with Kayne (1994) and Zubizarreta (1998) that the right-dislocated subjects are clause-external (tags in Zubizarreta 1998, the result of clause reduction for Kayne 1994). Further evidence for this claim comes from the interaction between question tags and subjects. Den Dikken (1995) shows that shifted heavy NPs, traditionally analyzed as right dislocated, follow question tags: (25) They have found, haven’t they?, the treasure buried on that island 100 years ago. The same is true in EP: right-dislocated subjects follow question tags while subjects in Spec, VP precede it. This is shown by the interaction between these orderings and pronoun insertion: (26) a.

Comeu a sopa o Paulo, não comeu? ‘ate the soup Paulo not ate’

b. *Ele comeu a sopa o Paulo, não comeu? he c.

Comeu a sopa, não comeu?, o Paulo

d.

Ele comeu a sopa, não comeu?, o Paulo.

The only possibility to obtain the order subject-tag is either to leave out the pronoun, or to introduce a pause before the subject. This just shows that the two sentence adjuncts, tag and right-dislocated subject are interchangeable: (27) (ele) comeu a sopa # o Paulo # não comeu?

32

Postverbal subjects: syntax and discourse – VSO and VOS orders

In the remainder of this section, I will ignore right-dislocated subjects, since they appear to be extra-sentential. I will concentrate on the ones I claim to be in Spec, VP, that is postverbal subjects not involving an intonational break after the object. Having established this difference, let us now look at the tests that enable us to identify the position of the subjects in VOS sentences.

a) The distribution of adverbs: Applying the same test as for VSO, it is now possible to check the paradigm for the several potential positions for the monosyllabic adverb bem ‘well’ in VOS contexts. Keeping in mind that cases of right-dislocation are to be ignored and can be tested by inserting a nominative pronoun, the results are the following: (28) a. Comeu bem a sopa o Paulo ‘ate well the soup Paulo’ b. *Comeu a sopa o Paulo bem. c. Comeu a sopa bem o Paulo. The results in (28) seem to indicate that, independently of what happens to the object, the subject is VP -internal, surfacing obligatorily to the right of the adverb. In section 3, I will return to these examples and explain why the two word orders between the object and the adverb are possible.

b) Binding effects. Binding effects seem to support the analysis of VOS I am presenting here. As it is well know, A-movement feeds binding (cf. 29), while A-bar movement does not (cf. 30). (29) a. *It seems to each other that the boys are tired. b. The boys seem to each other to be tired. (30) a. *Paul introduced each other to Mary and Paul. b. *Who did Paul introduce to each other?

Post-verbal subjects in VOS contexts

33

The same effects may be visible in Portuguese: in (31), the QP only binds the possessive in the passive construction (31b), and not when it is whextracted (cf. the Weak Cross -over effects in 31c): (31) a. *O seui realizador viu todos os filmesi. ‘their director saw all the movies’ b. Todos os filmesi foram vistos pelo seu realizadori. ‘all the movies were seen by their director’ c. *Que filmesi viu o seu realizadori? ‘which movies saw their director’ Now consider the behavior of subjects and objects in a VSO order: if the object is a QP and the subject contains a possessive anaphor, binding is impossible. If object scrambling were A-movement binding would be acceptable, on a par with (31b). This is not the case, as (32b) illustrates: (32) a. *Viu o seui realizador todos os filmesi. ‘saw their director all the movies’ b. *Viu todos os filmesi o seui realizador. ‘saw all the movies their director’ The impossibility of binding becomes even clearer in cases like (33), which are only acceptable if interpreted as involving VSO order: (33) a. *Viu [Obj o Pauloi ] [Subj o seui irmão]. ‘saw Paulo his brother’ b. Viu [Subj o Pauloi ] [Obj o seui irmão]. One could argue that these effects do not constitute conclusive evidence in favor of the analysis of VOS order I am advocating, since binding would be impossible anyway if the subject were moved rightwards and adjoined to a position higher than the object, making it impossible for the object to ccommand it. However, the following examples prove that this is not true: in the cases which were identified as instances of right-dislocation, i.e. those cases in which a tag may intervene between the object and the subject, binding is possible:16

34

Postverbal subjects: syntax and discourse – VSO and VOS orders

(24) a. Viu [Obj o Pauloi ], não viu?, [Subj o seui irmão]. ‘saw Paulo, not saw, poss brother’ b. Viu [obj todos os filmesi], não viu?, [Subj o seui realizador]. ‘saw all the movies, not saw, poss director’

c) The differences between definites and indefinites: If VOS contexts indeed involve subjects in Spec, VP, there should not be any difference between definites and indefinites, since both types of subjects occur in VSO sentences, as illustrated in (25): (25) a. Comeu o Pedro a sopa. ate Pedro the soup b. Comeu um homem a sopa. ate a man the soup In fact, postverbal subjects in VOS may be either definite or indefinite: (26) a. Comeu a sopa o Paulo. ‘ate the soup Paulo’ b. Comeu a sopa um homem. ‘ate the soup a man’ The fact that definiteness does not play any role also confirms that VOS word order is not necessarily derived in terms of right-dislocation of the subject, since in the true cases of subject-right-dislocation, indefinite subjects are marginal, as illustrated in (27), in which the question tag helps to differentiate the two types of constructions: (27) a.

Comeu a sopa, não comeu?, o Paulo. Ate the soup, not ate?, Paulo

b. *Comeu a sopa, não comeu?, um homem qualquer. Ate the soup, not ate?, some man The definiteness effect in (27) is expected, since right-dislocation may be associated with pronominal doubling, as shown above. Since only definite DPs can be doubled, right-dislocated subjects behave as expected.

Objects in VOS: Scrambling in European Portuguese

35

From the tests above, I conclude that in VOS context, subjects occupy the Spec,VP position. The difference between VSO and VOS will be derived in terms of object scrambling in the latter, in accordance with the evidence to be presented in section 3. A comparison between the scrambling analysis and an analysis involving remnant-IP movement, along the lines of Kayne (1998) will provide further evidence in favor of the claim that subjects in VOS are in Spec, VP. At this point, it is possible to tell which position the subject occupies in each of the attested word orders. The conclusions reached so far are the following: 1. Preverbal subjects occupy the Spec,IP position; like all other arguments, it can be left-dislocated in some contexts. 2. Postverbal subjects in VSO context may be in Spec, VP 3. Postverbal subjects in VOS without intonation break between the object and the subject are in Spec, VP In the next section, the position of objects in VOS sentences, which is crucial for the analysis of this word order, will be discussed.

3.3. Objects in VOS: Scrambling in European Portuguese In the preceding section, I have assumed without providing arguments that Portuguese VOS orders involve object scrambling across the subject. In this section, I will motivate that assumption, arguing that it is possible to identify similarities between scrambling in European Portuguese and the well-known scrambling configurations in Dutch and German. In order to achieve this goal, I will first show some properties of scrambling in Dutch and German. This will establish the grounds for a comparison between Portuguese and Dutch/German, and determine how to trace scrambling in a VO language. As mentioned in the previous section, if the scrambling analysis is on the right track, there will be further evidence in favor of the claim that the subject may be stranded in Spec, VP in EP. In order to strengthen this claim, the scrambling analysis will be compared with an analysis in terms of remnant IP-movement, following proposals of Kayne (1998). Tests will be provided for comparing the two analyses. I will further review three of the theories of scrambling: case-driven movement (Zwart 1993 among others), semanti-

36

Postverbal subjects: syntax and discourse – VSO and VOS orders

cally-driven movement (Diesing and Jelinek 1995) and prosodically-driven movement (Reinhart 1995, Zubizarreta 1998), showing that the Portuguese data provide evidence in favor of the latter. The results of this section will challenge the view advocated in Webelhuth (1989) and Neeleman and Reinhart (1999) that scrambling is a property of OV languages only. Moreover, showing that there is scrambling in Portuguese will add one more language to the group of languages in which this phenomenon is visible, which makes it possible to broaden the empirical coverage of the very debated issue of what the exact nature of scrambling is. Scrambling was first discussed by Ross (1967) to refer to the syntactic process that permits breaking the adjacency complement-verb by insertion of an adjunct. The scrambled variant of (38a) is (38b): (38) a. Adverb Object Verb b. Object Adverb Verb

No scrambling Scrambling

In this section I will illustrate some of the properties of scrambling in German and Dutch, just to establish a basis for comparison between these languages and European Portuguese. I will sometimes use the term movement for scrambling, but this does not yet mean that I am adopting a specific theory of scrambling. At this stage, I will remain neutral with respect to the correct analysis of scrambling (movement or base-generation of the complement in the scrambled position). The properties that will be considered are the following:

a) Scrambling moves NPs and PPs17: (39) Dutch: a. dat Jan in Amsterdam zijn vriendin ontmoet. that Jan in Amsterdam his girlfriend meets b. dat Jan zijn vriendin in Amsterdam ontmoet. that Jan his girlfriend in A’dam meets (40) a. dat Jan waak op zijn vriend wacht. that Jan often for his friend waits b. dat Jan op zijn vriend waak wacht. that Jan for his friend often waits

Objects in VOS: Scrambling in European Portuguese

37

b) Differently from the categories in (39) and (40), predicative APs and some Small Clauses cannot be moved, as (41) and (42) show: (41) *dat Jan Marie aardig altijd vindt. that Jan Marie always nice finds (42) *dat Jan ziek altijd is. that Jan sick always is c) Another property of scrambling noted by Bennis and Hoekstra (1984) is that it licenses parasitic gaps: (43) a. Jan heeft die boeken zonder te lezen weggelegd. Jan has these books without read put away b. *Jan heeft zonder te lezen die boeken weggelegd. Jan has without read these books put away With respect to this property, it is important to note that some authors do not consider these facts conclusive for determining the type of movement involved in scrambling (see e.g. de Hoop 1992, Zwart 1993, Neeleman 1994, among others), either because there are constructions in which subjects of passives license parasitic gaps (cf. de Hoop 1992), or because parasitic gaps are marginal in Dutch, or because there are cases in which parasitic gaps appear to be licensed by a constituent in an A-position (other than subjects of passives) (cf. (44) from Webelhuth 1989). The problem with (44), which exemplifies what came to be known as Webelhuth’s paradox, is that the QP jeden/everyone is in a position where it both licenses the parasitic gap and A-binds a pronoun (contained in the same clause that contains the parasitic gap). Note that the fact that the parasitic gap and the pronoun are in the same clause is really paradoxical, and cannot be solved by means of two-step movement, as proposed in Vanden Wyngaerd (1989) and Mahajan (1990). (44) German (Webelhuth’ paradox): weil Maria jedeni ohne pg anzuschauen seinem Nachbarn ti because Maria everyone without to-look-at his neighbor vorgestellt hat introduced has ‘because Maria introduced everyone to his neighbor without looking at him’

38

Postverbal subjects: syntax and discourse – VSO and VOS orders

Here, I will take the opposite approach to this problem, ignoring the A-binding facts. For a solution for this problem, I refer the reader to Lee and Santorini (1994). One of the reasons for this option is that there is not a clear description of what the WCO and WCO repair facts are, and, moreover, pure A-positions (meaning absolute subjects and not subjects derived form object positions never license parasitic gaps as (65) illustrates (see also Diesing 1995, Vikner 1994 and Bobaljik 1995 for similar ideas). (45) *Jan heeft zonder pg te zien het boek gelezen. Jan has without to see the book read I emphasize, though, that this is a purely methodological option, due to the unclarity of binding as a diagnostic. It is very likely that whenever there is a clear description of the facts, the methodology will have to be reviewed.

d) Though this property is not available in Dutch, but only acceptable in German, it has been observed that scrambling may move an object across a subject: (46)

German: weil den Patienten der Arzt besucht hat. because the patient-ACC the doctor-NOM visited has ‘because the doctor visited the patient’

This difference between German and Dutch may be explained assuming with Diesing (1992) that subjects may stay in Spec, VP in German. If this option is not available for Dutch, the contrast between the two languages follows. It is not a consequence of the availability of a different type of scrambling in German, but rather an effect of the availability of Spec, VP as a legitimate position for the subject only in German.

e) Another well-known property of Dutch and German scrambling are the semantic/pragmatic requirements of this operation (the scrambled constituent has to be specific or non-novel, cf. de Hoop 1992; Diesing 1995; among others).

Objects in VOS: Scrambling in European Portuguese

39

Having described some of the properties of Dutch and German scrambling, let me now move to a description of the Portuguese data and check out the similarities with Germanic scrambling. The problem with identifying scrambling in Portuguese is that this language has V-to-I movement, which makes it difficult to identify the position of the object. Since the verb is not in VP anymore, the position of the object is ambiguous between its base position or the scrambled position. That is, an SVO sentence may be analyzed as either (47a) without scrambling, (47b) with the object adjoined to VP, or (47c) with the object moved to Spec,AgrOP: (47) a. [IP S [I’ V [VP tS tV O]]] b. [IP S [I’ V [VP O [VP tS tV tO]]]] c. [IP S [I’ V [AgrOP O [VP tS tV tO]]]] I will assume, once again, that given their properties monosyllabic adverbs provide a good test to check whether the object is inside or outside VP. Recall the distribution of monosyllabic adverbs in English: (48) a. John read the book well. b. *John well read the book. c. John looked hard at those pictures. d. *John looked at those pictures hard. In Costa (1996, 1998), the English pattern was analyzed by assuming, following Pesetsky (1989) and Johnson (1991), that both verbs and nominal objects move out of the VP overtly: NPs move to Spec,AgrOP and the verb to the first functional projection immediately above AgrOP. The sequence PP-Adv is bad because these adverbs do not right-adjoin to VP and because differently from nominal complements, PPs do not need to move to license Case. The behavior of these adverbs is not an idiosyncrasy of English. As already mentioned in the discussion of subject positions, the same basic generalizations hold for Portuguese: a paradigm almost equivalent to (48) may be built for this language:

40

Postverbal subjects: syntax and discourse – VSO and VOS orders

(49) a. O Paulo fala francês bem. Paulo speaks French well b. *O Paulo bem fala francês. Paulo well speaks French c. O Paulo olha bem para aqueles quadros. Paulo looks well at those pictures d. *O Paulo olha para aqueles quadros bem. Paulo looks at those pictures well e. O Paulo olha para aqueles quadros BEM. Paulo looks at those pictures well The crucial difference between English and Portuguese is that only in the latter is the order V-Adv-NP possible: (50) a. O Paulo fala bem francês. Paulo speaks well French b. *Paul speaks well French. The fact that the nominal complement may either precede or follow the adverb that marks the left edge of the VP is quite similar to the distributional pattern of nominal complements in scrambling languages: in Dutch and German, the objects may appear either to the left or to the right of a VPadjunct.18 The only difference is that these languages are V-final, hence the adjacency between complement and verb is visible in the case of non-scrambling. In a language with V-to-I movement, there is only adjacency if there is no other adjunct in between the inflected verb and the object, as in (51): (51) O Paulo fala sempre francês bem. Paulo speaks always French well Given the distribution of monosyllabic adverbs and the possible orderings between the verbal complement and the adverb, I will assume that the word order in (52a) traces a scrambling configuration, while the word order in (52b) indicates that the object occupies its base-generated position: (52) a. [IP V [ scrambled Object [VP Adv [VP tV tObj b. [IP V [VP Adv [VP tV non-scrambled Object

Objects in VOS: Scrambling in European Portuguese

41

Having established that adverb placement may be used as a test to identify the structural position of the complements, let me now go through the properties identified for scrambling in Germanic, and see the similarities between scrambled objects in the latter and pre-adverbial or pre-subject objects in Portuguese.

a) Novelty effects (de Hoop 1992; Büring 1997; Webelhuth 1989; Diesing 1995; among others): The first property of the XP-Adv order in European Portuguese I want to consider is the novelty effects also observed in Dutch/German scrambling. Checking for felicitous answers for questions, we see that in the answer to (53aA) the complement yields new information, and the order of the constituents has to be Adv-NP. On the contrary, if it is the adverb that introduces new information, the NP that is repeated from the question has to be scrambled, as (53b) shows. (53) a. A: Há alguém aqui que fale bem francês ou inglês? Is there anyone here who speaks French or English well? B: #Não, mas o Paulo fala alemão bem. No, but Paulo speaks German well B’: Não, mas o Paulo fala bem alemão. b. A: Como é que o Paulo fala francês? How does Paulo speak French B: O Paulo fala francês bem. the Paulo speaks French well B’: #O Paulo fala bem francês. This patterns like Dutch, where the elements that are new information, appear rightmost. The only difference is that rightmost in Dutch means being to the left of the non-inflected verb:

42

Postverbal subjects: syntax and discourse – VSO and VOS orders

(54) Dutch: a. A: Wanneer heeft Jan Marie gekust? When has Jan Marie kissed B: Jan heeft Marie gisteren gekust. Jan has Maries yesterday kissed. B’: #Jan heeft gisteren Marie gekust. b. A: Wie heeft Jan gisteren gekust? Who has Jan yesterday kissed B: Jan heeft gisteren Marie gekust. Jan has yesterday Marie kissed B’: #Jan heeft Marie gisteren gekust.

b) Parasitic Gap licensing. Another similarity between European Portuguese on the one hand and Dutch and German on the other has to do with the fact that scrambling licenses parasitic gaps. Prima facie, this appears to be hard to test, since it is known that null objects exist in Portuguese (see Raposo 1986). However, Bianchi & Figueiredo (1994) and Menuzzi (1994) have shown that null objects (of the type treated in Raposo’s work) can only occur if the antecedent is inanimate. This is exemplified in (55) and (56) below: (55) (Talking of [the car we have just seen]i):

(from Menuzzi 1994)

a. O José conhece [NP a mulher [CP que comprou eci]. José knows the woman who bought (it) b. O José [VP ficou nervoso] [CP porque a Maria comprou eci] José got nervous because Maria bought (it). (56) (talking of [Paulo]i): a. *O José conhece [NP a mulher [CP que beijou eci]. José knows the woman who kissed (him) b. *O José [VP ficou nervoso] [CP porque a Maria beijou eci] José got nervous because Maria kissed (him)

Objects in VOS: Scrambling in European Portuguese

43

Bianchi and Figueiredo (1994) and Menuzzi (1994) conclude from the sensitivity to islands in (56) that only in these cases is operator movement involved. They disagree with respect to the proper characterization of the empty category in (56). Nevertheless, what is relevant for the present discussion is that this difference allows differentiating between null objects and gaps left by movement.19 A parasitic gap has to be involved when an animate antecedent is used, since null objects may not be animate, as the examples above show. Accepting this, it is now possible to test for parasitic gap licensing, excluding the possibility of being misled by results involving a null object construction. If scrambling in Portuguese is like in Germanic, it is expected that an object be able to license the parasitic gap from the scrambled position, but not from the base-position. This prediction is borne out. Like in German and Dutch, the in-situ object does not license a parasitic gap contained in a left-adjoined PP: (55) a. *O Pedro conhece [mesmo sem nunca ter visto pg ] bem a Maria. b. O Pedro conhece a Maria [mesmo sem nunca ter visto pg ]. The post-adverbial object in (55a) cannot license the parasitic gap, unless it is scrambled.

c) Predicative APs and Small Clauses do not scramble. Like in German and Dutch, Small Clauses and predicative APs do not scramble. This is shown in (56) and (57) below: (56) a. O Paulo acha sempre a Maria simpática. Paulo finds always Maria friendly b. *O Paulo acha a Maria simpática sempre. (57) a. O Paulo é sempre muito simpático. Paulo is always very nice b. *O Paulo é muito simpático sempre. In (56), it can be seen that a small clause (complement of the verb achar ‘to find’) may not be scrambled to the left of the adverb, which modifies the

44

Postverbal subjects: syntax and discourse – VSO and VOS orders

matrix verb.20 In (57), it is shown that a predicative AP may never scramble to the left of an adverb modifying a predicative verb.

d) Scrambling does not feed binding. Although I have not exemplified this property for Dutch and German (see Déprez 1989, Webelhuth 1989, Vanden Wyngaerd 1989, Müller and Sternefeld 1994, Corver and Riemsdijk 1994, among others), in Portuguese, scrambling does not feed binding, patterning like other cases of A-bar movement and unlike A-movement. This is exemplified in (58): the sentence (58a) in which the subject contains a possessive anaphor not bound by the QP in object position does not improve if the object is scrambled across the subject. (58) a. *Viu o seui pai cadai criança. saw his father each child b. *Viu cadai criança o seui pai. saw each child his father ‘His father saw each child’ As mentioned above, I am not taking into account the binding facts, since I do not think that it is possible to establish a parallel with Germanic. I wanted nevertheless to add these data, to show that, in Portuguese, scrambling an object across a subject does not feed binding. Note that (58b) contrasts with (59), where A-movement is involved and the binding relation can be established: (59) Cada criança foi vista pelo seu pai. Each child was seen by his father

e) Scrambling across the subject. This property cannot be really used as an argument in favor of the scrambling analysis, since it corresponds to the phenomenon I want to analyze: the order VOS. Nevertheless, I would like to note that deriving VOS in EP in terms of scrambling of the object across the in-situ subject is not anomalous, since this word order also exists in German, being derived in the same way.

Objects in VOS: Scrambling in European Portuguese

45

From the evidence presented above and the similarities with German and Dutch, I conclude that European Portuguese has scrambling of the Germanic type. Moreover, given its properties, scrambling seems to be A-bar movement, as has been argued for German and Dutch. I am aware that the assumption that scrambling in German and Dutch is not A-bar movement is quite controversial (see de Hoop 1992, Zwart 1993, Bobaljik 1995 for an Amovement analysis). Here, I will follow Vikner’s (1994) conclusion that there is a distinction between scrambling in West Germanic and Object Shift in Scandinavian. Vikner (1994) compares the constructions in both language groups, arguing that scrambling is A-bar movement, while objectshift is A-movement. The table below lists the properties associated with each construction: Table 1. German/Dutch

Scandinavian

EP

Parasitic gap licensing

yes

no

yes

Movement of NPs

yes

yes

yes

(Icelandic and Faroese: NPs; the other languages: pronominal NPs)

Movement of PPs

yes

no

yes

no (?) 21

yes

no 22

Order Adv-NP-Adv?

yes

no

yes 23

Interaction with V-movement?

no

yes

no

FQ-licensing?

Vikner’s (1994) observations summarized in the left-hand columns of (T1) apparently provide robust evidence in favor of the claim that there are two types of object movement out of VP: West Germanic scrambling, sharing properties with O-Adv orders in EP displays A-bar properties, while Scandinavian object shift displays A-properties. I will assume these results, opting for representing scrambling as adjunction to VP, as in (60a), and object-shift of full NPs, like in Icelandic, as movement to a Case A-position (Spec,AgrOP), as in (60b). These representations will account for the A- vs A-bar properties discussed above.

46

Postverbal subjects: syntax and discourse – VSO and VOS orders

(60) a. [IP [VP Obj [VP Adv [VP tObj b. [IP [AgrOP Obj [VP Adv [VP tObj The consequences of establishing the existence of scrambling in European Portuguese are the following: First, scrambling may not be seen as an OV phenomenon anymore (as suggested, for instance, in Vikner 1994). Furthermore, the lack of scrambling in other VO languages such as English becomes an interesting question. Finally, European Portuguese can be considered another language allowing for the investigation of what the reason for scrambling is. Below, an explanation will be offered as to why scrambling does not occur in a language like English. Apart from these considerations, note that I have already provided evidence in favor of the analysis proposed in section 2 for Portuguese VOS. In order to strengthen the analysis, it is important to compare it with an alternative analysis available in the literature, according to which, in VOS sentences, the subject is in Spec,IP. There are two competing analyses for VOS word orders that may be found in recent literature. The Scrambling analysis argues that this word order is derived via movement of the verb to I and scrambling of the object across the subject left in its base-generated position. The configuration arrived at may be as in (61a) or (61b), depending on the assumption regarding the landing site of the object (Spec,AgrOP or adjunction to VP). This is the analysis outlined above. Proponents of this analysis include Zubizarreta (1995), Ordóñez and Treviño (1995), for Spanish, Alexiadou and Anagnostopoulou (1998), for Greek, and Costa (1997, 1998) for Portuguese. (61) a.

IP 2 I AgrOP V 2 Obj AgrO’ 2 AgrO VP 2 Subj V’ 2 V tObj tV

b.

IP 2 I VP V 2 Obj VP 2 Subj V’ 2 V tObj tV

Objects in VOS: Scrambling in European Portuguese

47

This analysis raises the question of knowing how the subject in Spec,VP is assigned or checks Case. So far, I offered no solution for this problem. An alternative analysis for this word order in Romance languages is found in work following Kayne (1998). Let us term this type of analysis Remnant movement. According to the Remnant movement analysis, VOS word orders are arrived at by moving the subject to Spec,IP, or to a leftperipheral functional projection, and moving the remnant constituent TP or VP, containing the trace of the subject to the specifier position of another functional category higher than the one where the subject has moved to. The configurations obtained are as in (62a,b). Proponents of this type of analysis include Kayne and Pollock (1998), for stylistic inversion in French, Ordóñez (1997), Zubizarreta (1998a), and Bok-Benema (1998), for VOS in Spanish, and Ambar and Pollock (1998) for VOS in interrogative contexts in French and Portuguese. (62) a.

FP b. FP1 3 3 TP F’ IP F’1 2 2 5 2 T VP F IP tSubj V O F1 FP2 5 2 2 VO Subj I’ Subj F’2 2 2 I tT/VP F2 tIP

As explicitly emphasized by Zubizarreta (1998a), this type of analysis is advantageous with respect to the scrambling analysis, since the Case problem finds a solution. In either configuration in (62), the subject has moved or landed in Spec,IP where it may be assigned nominative Case. However, this analysis is not exempt of problems. I will now turn to presenting empirical arguments listed in Costa (2002b) that seem to disfavor the remnant movement analysis, comparing it with the scrambling analysis. It will be shown that the problems raised to the remnant movement analysis do not arise under a scrambling analysis for VOS. The argumentation will be based on an examination of the following aspects of VOS sentences:

48

Postverbal subjects: syntax and discourse – VSO and VOS orders

A – it will be tested whether it can be argued independently for the mobility of the remnant VP/TP; B – Adverb positioning; C – Floating quantifiers; D – Pronominal doubling and question tags; E – The discourse function of the subject; F – Scope and c-command; G – Clitics; H – Properties of the object. For each of these properties, it will be shown that the remnant movement analysis either makes wrong predictions or offers no explanation.

A. TP \VP mobility Consider a sentence like (63) below with VOS word order: (63) Leu o livro o Paulo. read the book Paulo According to the remnant movement analysis, the constituent TP or VP leu o livro (read the book) is moved to the left of the DP subject. Arguably, it should be possible to find independent evidence for the mobility of this constituent. Indeed for a sentence like (83), it is possible to find this type of independent motivation. The VP or TP containing the verb and the object may be clefted (cf. 64), or preposed (cf. 65): Cleft: (64) Foi ler o livro o que o Paulo fez. it was read the book what Paulo did Preposing: (65) O Pedro disse que leria o livro o Paulo, e ler o livro o Paulo leu. Pedro said that would-read the book Paulo, and read the book Paulo did However, if the verbal form is complex, the evidence for mobility of the constituent TP or VP is not so compelling. For a sentence like (66), it must be argued that the constituent containing the sequence Auxiliary-Main Verb-Object is moved to the left of the subject:

Objects in VOS: Scrambling in European Portuguese

49

(66) Tinha lido o livro o Paulo. had read the book Paulo However, there is no evidence for movement of this constituent. The constituent containing the auxiliary verb, the main verb and the object cannot be clefted (cf. 67) nor preposed (cf. 68): Cleft: (67) *Era ter lido o livro o que o Paulo tinha feito. it was have read the book what Paulo had done Preposing: (68) ??O Pedro disse que teria lido o livro o Paulo, e ter lido o livro ele tinha. Pedro said that would-have read the book Paulo, and have read the book he had. Thus, there is no clear evidence that the material preceding the subject forms a constituent that can be moved. The fact that some sequences VO or Aux-V-O are not movable is not problematic for a scrambling analysis, since no claims are made concerning the necessity of moving this constituent. Under the scrambling analysis, the only claim that is made is that the object is moved to the left of the subject. It can be argued that an object may be moved independently of the complexity of the verbal form. For a pair of sentences like those in (69), it can be shown that the object may be clefted (cf. 70) or preposed (cf. 71) independently of the material that precedes the subject. (69) a. Leu o livro o Paulo. read the book Paulo b. Tinha lido o livro o Paulo had read the book Paulo Cleft: (70) a. Foi o livro o que o Paulo leu. it was the book that Paulo read b. Era o livro o que o Paulo tinha lido. it was the book that Paulo had read

50

Postverbal subjects: syntax and discourse – VSO and VOS orders

Preposing: (71) a. O livro, leu o Paulo. the book, read Paulo b. O livro, tinha lido o Paulo. the book had read Paulo In short, postulating that VOS word order is derived via movement of the constituent containing V and O is problematic, since evidence for the mobility of this constituent is not very strong.

B. Adverb placement The second problematic set of data for the remnant movement analysis comes from the distribution of adverbs. Let us consider again the placement of monosyllabic adverb in VOS sentences: as shown above, and repeated under (72), the adverb may occur in two positions: either in between the object and the subject or in between the verb and the object: (72) a. Leu aqueles livros mal o Paulo. read those books bad Paulo b. Leu mal aqueles livros o Paulo read bad those books Paulo It was observed that these adverbs have a very restricted distribution. Nevertheless, they appear in two possible positions in a VOS sentence. Under a remnant movement analysis, there is no clear explanation for this pattern. Why should the adverb have a less restricted distribution when there is movement of the constituent containing it? The scrambling analysis offers a natural explanation for this behavior. Under this analysis, the object is scrambled. The list of properties associated with scrambling led us to the conclusion that scrambled objects in EP adjoin to VP. If this is true, it is expected that there is no fixed order between two VP-adjuncts. V-O-Adv-S and V-Adv-O-S are the expected word orders. The two possibilities are illustrated in the structures in (73) and (74):

Objects in VOS: Scrambling in European Portuguese

(73)

IP 2 I VP V 2 Obj VP 2 Adv VP 2 Subj V’ 2 V tObj tV

(74)

IP 2 I VP V 2 Adv VP 2 Obj VP 2 Subj V’ 2 V tObj tV

51

C. Floating quantifiers Consider now the behavior of floating quantifiers. As it is well-known, floating quantifiers are possible after subject movement to Spec,IP (Sportiche 1988, Koopman and Sportiche 1991). In EP, floating quantifiers behave as expected: they may surface in any post-subject position, tracing the path of the subject from Spec, VP to Spec,IP. (75) a. Os meninos todos tinham lido o livro. the children all had read the book b. Os meninos tinham todos lido o livro. the children had all read the book c. Os meninos tinham lido todos o livro. the children had read all the book.

52

Postverbal subjects: syntax and discourse – VSO and VOS orders

Under the proposal made by the remnant movement analysis, the subject moves to Spec,IP, or higher, and the remnant constituent containing its trace moves to its left. Note that there is no problem in moving a VP or TP containing a floating quantifier to the left of a subject. As shown in (76), when a VP is clefted, it may contain a floating quantifier: (76) Foi ler todos o livro o que os meninos fizeram. it was read all the book what the children did Considering the assumption of the remnant movement analysis and the fact that moved VPs or TPs may contain floating quantifiers, it is legitimate to claim that the remnant movement analysis predicts that floating quantifiers surface before the subject in VOS sentences. However, this prediction is not borne out, as shown by the ungrammatical sentences in (77): (77) a. *Tinham todos lido o livro os meninos. had all read the book the children b. *Tinham lido todos o livro os meninos. had all read the book the children. c. Tinham lido o livro todos os meninos. had read the book all the children The only possibility for the floating quantifier is to occur adjacent to the subject, that is, in a non-floating position. If we now check the predictions made by the scrambling analysis, it is possible to see that this problem does not arise. Under the scrambling analysis, the subject stays in Spec, VP, therefore floating quantifiers are not predicted to surface.

D. Topic doubling and question tags At this point, it is important to recall the difference between VOS with flat intonation vs VOS with a right-peripheral subject. Recall that only the latter can be doubled by a pronoun, and that they may surface after a question tag. The behavior of subjects in the latter type of construction is not different from the behavior of other constituents. Topic information appearing sentence-finally after a pause can be doubled by a clitic or a pronoun, as shown in (78). This is true for direct objects, indirect objects and subjects. The only

Objects in VOS: Scrambling in European Portuguese

53

difference between the latter and the internal arguments of the verb is that the absence of subject clitics in Portuguese forces doubling by a strong pronoun. (78) a. Direct object: O Paulo leu-o, esse livro. Paulo read it, that book b. Indirect object: O Paulo deu-lhe o livro, à Maria. Paulo gave her the book to Maria c. Subject: Ele leu o livro, o Paulo. he read the book, Paulo Clause-internal material cannot be doubled. This is true for direct objects, as in (79a), in which the direct object appears right after the doubling clitic without a pause, and left-adjacent to a marker of the right-edge of the sentence. It is also true for the indirect object, as in (79b), in which the indirect object appears before the direct object and adjacent to the doubling clitic, without being preceded by a pause. The fact that doubling is only possible with clause-external material enables a distinction between this pattern of clitic doubling and the pattern of clitic doubling exhibited by other languages, in which the clitic may double clause-internal elements. Based on the contrasts between (78) and (79), it is legitimate to assume that the pronominal elements occupy the thematic positions, and the DPs are peripheral to the sentence. (79) a. Direct object: *O Paulo leu-o esse livro ontem. Paulo read it that book yesterday b. Indirect object: *O Paulo deu-lhe à Maria o livro. Paulo gave her to Maria the book Let us now consider the case of subjects. Like the direct and indirect object, subjects can only be doubled if they follow a question tag or any other marker of the right edge of the sentence, no matter whether the pronoun is preverbal or postverbal. Thus, they behave like the other sentence constituents.

54

Postverbal subjects: syntax and discourse – VSO and VOS orders

(80) a. Ele leu o livro, não leu?, o Paulo he read the book, didn’t he?, Paulo b. Leu o livro ele, não leu?, o Paulo. read the book he, not read, Paulo c. *Ele leu o livro o Paulo, não leu? he read the book Paulo, didn’t he? Let us now consider the relevance of this behavior for the analysis of VOS sentences. In a specific implementation of the remnant movement analysis for accounting for stylistic inversion in French, Kayne and Pollock (1998) argue that, in VOS sentences, the subject is in Spec,TopP. As we have seen above, topic subjects can be doubled. Even if they are sentence-initial, they can be doubled in some contexts (see section 2): (81) O Paulo, ele leu esse livro. Paulo, he read that book The question that arises for Kayne and Pollock’s analysis is why the alleged topic cannot be doubled in VOS order without a break, as shown above and repeated in (82). (82) *Ele leu o livro o Paulo. he read the book Paulo The remnant movement analysis for VOS proposed in Kayne and Pollock (1998) offers no clear answer for the ungrammaticality of (82). Under the scrambling analysis, this pattern is expected. The subject is clause-internal, it is in Spec, VP. Since it is not in a topic position, doubling is not expected.

E. Discourse function of the subject So far, I have not explicitly discussed the relation between the several word orders and discourse. It will be shown, however, that the sentence-final subject in VOS word orders is a focus. This may be tested in the following question-answer pair:

Objects in VOS: Scrambling in European Portuguese

55

(83) A: Quem leu o livro? who read the book B: Leu o livro o Paulo. read the book Paulo In Portuguese, contrastive foci are marked with stress (Frota 1998). Crucially, a constituent may be simultaneously information and contrastive focus. In other words, it may answer a question and be marked with strong stress. The difference between (83) and (84) is that the answer in (84) implies that only Paulo and no-one else read the book. (84) A: Quem leu o livro? who read the book B: Leu o livro O PAULO. read the book Paulo Having noted this discourse function of the subject for Spanish VOS orders, Ordóñez (1997) proposes that, since the subject in VOS is focused, it is moved to a left-peripheral functional projection codifying focus information: Focus Phrase. The remnant constituent yielding old information (in (83) and (84), the constituent containing the verb and the object) is moved to a topic position to the left of the focus position. Although it may account for the discourse function of the subject, this analysis is problematic. In his study of the left-periphery, Rizzi (1997) shows that wh-phrases and displaced focused constituents are in complementary distribution. The prediction made by Ordóñez’s implementation of the remnant movement analysis is then that wh-phrases in VOS word orders should be ungrammatical. As shown below, this prediction is not borne out: (85) A quem deu o livro O PAULO? to whom gave the book Paulo Under Kayne and Pollock’s (1998) analysis, it is proposed that the subject is in Spec,TopP. Assuming this, the problem raised by Ordóñez’s analysis does not arise, but the relation with the actual discourse function of the subject is left unexplained. Ambar and Pollock (1998) provide a good argument in favor of Kayne and Pollock’s analysis. They note that in interrogative contexts with VOS

56

Postverbal subjects: syntax and discourse – VSO and VOS orders

order, the sentence-final subject cannot be a pronoun or an indefinite pronoun. Crucially, these elements cannot be topicalized in Portuguese. The similarity between the type of subject that occurs sentence-finally in VOS interrogatives and the type of subject that can be topicalized provides, thus, clear evidence in favor of the analysis according to which the sentence final subject in VOS is in Spec,TopP. (86) a. A quem deu o livro o João? to whom gave the book João b. *A quem deu o livro ele? to whom gave the book he c. *A quem deu o livro alguém? to whom gave the book someone Although the argument is a good one, the analysis proposed in terms of topicalization and remnant movement is not exempt of problems. First, note that the sentence-final position for the subject is not reserved for topics. As shown in (85), other discourse functions, such as contrastive focus, are available in this position. More problematic for the analysis proposed is that there is no additional independent evidence for the claim that the subject is topicalized. One expects to find properties of topicalization in VOS word orders in interrogative contexts. Duarte (1987, 1996) identifies several properties of topicalization in European Portuguese that may be used to test Ambar and Pollock’s claim. Duarte shows that topicalized elements license parasitic gaps. This is illustrated in (87a). Accordingly, the sentence-final subject in VOS interrogatives should be able to license parasitic gaps, since it is assumed to occupy a topic A-bar position. However, parasitic gaps are not found in this context (cf. 87b). No parasitic gap licensing: (87) a. Esse artigo, o João elogiou sem ter lido. that article, João praised without having read b. *Quando foram criticados sem o Paulo ter lido os artigos? when were criticized without Paulo have read the articles If there is a main clause and an embedded clause, constituents from each of these clauses may be topicalized, as shown in (88a’). The prediction made by

Objects in VOS: Scrambling in European Portuguese

57

Ambar and Pollock’s analysis is thus that both subjects of the main and the embedded clause may be topicalized, as in (89b), and the remnant IPs move to the functional categories to the left of the landing site of the subject, as in (89c). This would derive a word order in which both subjects would appear adjacent at the end of the sentence. However, this word order is ungrammatical (cf. 88b).24 No multiple topicalization: (88) a. O Pedro disse à Maria que o Paulo ia à praia nesse dia. Pedro said to Maria that Paulo would-go to the beach on that day a.’ À Maria, o Pedro disse que, à praia, o Paulo ia nesse dia. to Maria Pedro said that to the beach Paulo would go on that day b. *Quando disse que ia à praia o Paulo o Pedro? When said that would-go to the beach Paulo Pedro (89) a. wh-movement [CP Quando [FP [TopP [IP o Pedro disse [CP que [FP [TopP [IP o Paulo ia à praia twh ]]]]]]]] b. Subject topicalization in both clauses [CP Quando [FP [TopP o Pedro [IP tDP disse [CP que [FP [TopP o Paulo [IP tDP ia à praia twh ]]]]]]]] b. Remnant IP-movement: [CP Quando [FP [IP tDP disse [CP que[FP [IP tDP ia à praia twh ][TopP o Paulo tIP ]]]] [TopP o Pedro tIP ]]] A final problem for Ambar and Pollock’s analysis comes from the lack of parallelism with another property of topicalization. As shown in (90), constituents other than subjects can be moved to the topic position of the main clause. This is an option available for the subject, as shown in (90b). (90) a. O Pedro contou à Maria que o Paulo vai ao cinema. Pedro told Maria that Paulo goes to the movies b. O Paulo, o Pedro contou à Maria que t vai ao cinema. c. À Maria, o Pedro contou t que o Paulo vai ao cinema. d. Ao cinema, o Paulo contou que o Paulo vai t.

58

Postverbal subjects: syntax and discourse – VSO and VOS orders

Since the topicalizations in (90) are possible, the prediction made is that in an interrogative sentence, any of these constituents of the embedded clause may surface in sentence-final position. However, as (91) illustrates, only the (allegedly topicalized) subject may occur in this position. The remnant movement analysis incorrectly predicts the grammaticality of (91b) and (91c). (91) a. Quando t contou à Maria que o Paulo vai ao cinema o Pedro? when told Maria that Paulo goes to the movies Pedro b. *Quando o Pedro contou à Maria que t vai ao cinema o Paulo? c. *Quando o Pedro contou t que o Paulo vai ao cinema à Maria? Under the scrambling analysis, the lack of parallelism with topicalization constructions is not a problem. If the subject stays in Spec, VP, it may be interpreted as the focus of the sentence, since it is in the rightmost/most embedded position, where it may be assigned sentence nuclear stress (Nespor and Vogel 1986, Cinque 1993, Zubizarreta 1998, among others), an issue we will return to in the next chapter. The interpretation of the subject as focus will follow from the interaction between prosody and syntax, rather than from an association with a specific functional projection. The topical properties of sentence-final subjects in interrogative contexts are not entirely unexpected nor constitute a problem for the scrambling analysis: the intonation of interrogative sentences is different; hence no prediction regarding the discourse function of the subject is made. As shown in Mateus et alii (1989), wh-questions have an initial rising tone and lowering at the end, which predicts that constituents that can be downstressed (like topics) will tend to emerge sentence-finally. F. Scope and c-command In VOS sentences, the object c-commands the subject. This statement is confirmed by the fact that a quantified object may scope over the subject, as in the ambiguous sentence in (92), and by the principle-C effects induced in VOS sentence, as in (93) and (94). In both cases, an object preceding the subject seems to c-command it. Quantifier scope: (92) Leram um livro dois alunos. read a book two students

(O>S, S>O)

Objects in VOS: Scrambling in European Portuguese

59

Principle-C effects: (93) SVO: a. O irmão do Pauloi viu-oi. the brother of Paulo saw him VOS: b. *Viu-oi o irmão do Pauloi saw him the brother of Paulo (94) SV IO DO: a. O irmão do Pauloi deu-lhei o livro. the brother of Paulo gave him the book V IO DO S: b. *Deu-lhei o livro o irmão do Pauloi gave him the book the brother of Paulo These facts based on quantifier scope and violation of principle-C are problematic for the remnant movement analysis. According to this analysis, a configuration is obtained in which the object is a constituent of the moved constituent. The configuration obtained is as in (95). Note that, in such a structure, the object does not c-command the subject: (95) [FP [TP\VP V O ][XP S....]] The prediction made by the remnant movement analysis is, thus, that objects in VOS sentences should not take scope over the subject, and that there should be no violation of principle-C, since, in (95), the object does not ccommand the subject. Note that reconstruction would not help, since the object would never c-command the subject. The prediction made by the scrambling analysis is that object-wide scope and violations of principle-C are found. The configuration obtained after scrambling of the object across the subject is like the one in (96). In this configuration, the object c-commands the subject: (96) [FP V [XP O [VP S ]]] The ambiguity of quantifier scope may be explained by the scrambling analysis, assuming that the scrambled object is able to reconstruct into its base generated position, in which it is c-commanded by the subject.

60

Postverbal subjects: syntax and discourse – VSO and VOS orders

G. Clitics: enclisis vs. proclisis As it is well-known, the distribution of clitics in European Portuguese may be dependent on syntactic properties of the sentence (cf. Duarte and Matos 2000, among others). A clitic cooccurring with a preverbal non quantified DP subjects is enclitic, as shown in (97): (97) a. O Paulo viu-o. Paulo saw him b. *O Paulo o viu. If the subject is quantified, as in (98), proclisis is triggered: (98) a. Ontem todos os meninos o viram. yesterday all the children him saw b. *Ontem todos os meninos viram-no. Recall that the remnant movement analysis proposes that, in VOS sentences, the subject is in Spec,IP or at least has passed through this position. The prediction made by this analysis is that a quantified subject in VOS sentences should trigger proclisis. However, this prediction is not borne out. Enclisis is the pattern found in VOS sentences:25 (99) a. *Ontem o deram à Maria todos os meninos. yesterday it gave to Maria all the children b. Ontem deram-no à Maria todos os meninos. Note that the argument based on the distribution of clitics also holds for whphrases that remain in-situ. A moved wh-phrase triggers proclisis, as illustrated below: (100) a. Quem o leu? who it read b. *Quem leu-o? who read it

Objects in VOS: Scrambling in European Portuguese

61

A subject wh-phrase in VOS context does not trigger proclisis: (101) a. *O deu à Maria quem? it gave to Maria who b. Deu-o à Maria quem? gave it to Maria who These patterns of clitic placement, that are problematic for the remnant movement analysis, are expected under the scrambling analysis. According to the latter, subjects in VOS contexts are in their base-generated position, Spec, VP. Since proclisis is only triggered by quantified subjects in Spec,IP, it is predicted that clitics are enclitic in this context. There is an objection that may be raised to this argument. Let us assume with Kayne and Pollock (1998) and Ambar and Pollock (1998) that the subject in VOS is in a topic position. In such case, it could be argued that proclisis is not expected, since topicalization does not trigger proclisis, independently of the position of the subject: (102) a. Esse livro, o Paulo leu-o. that book, Paulo read it b. Esse livro leu-o o Paulo that book read it Paulo However, if the topic is quantified, proclisis is triggered, as argued in Raposo (2000): (103) Muitos livros lhe leu o Paulo. many books him read Paulo We are now able to test whether it may be argued that proclisis is not triggered because the subject is topicalized. The test case is the use of a quantified subject in VOS context. As shown in (104), a quantified subject in VOS similar to the object in (103) does not trigger proclisis: (104) a. Leram-lhe livros muitos meninos. read him book many children b. *Lhe leram livros muitos meninos. him read books many children

62

Postverbal subjects: syntax and discourse – VSO and VOS orders

The fact that enclisis is always obtained independently of the properties of the subject permits discarding the objection to this argument based on the idea that subjects in VOS sentences are topicalized.

H. Properties of the object The acceptability of VOS sentences varies depending on the type of object of the sentence. So far, almost all examples contain definite or strong DPs as objects. However, if the object is indefinite, the acceptability of the sentence is not as good, and it clearly degrades with non-specific indefinite DPs: (105) a. ?Viu um gato o Paulo. saw a cat Paulo b. *Viu um homem qualquer o Paulo. saw some man Paulo c. ??Leu algo o Paulo read something Paulo This relation between the properties of the object and the degree of grammaticality of the sentence is problematic for the remnant movement analysis. According to this analysis, the object is just a part of the moved constituent. There is thus no clear reason for there to be sensitivity to the definiteness of a subconstituent of the moved XP. This problem does not arise under the scrambling analysis. The constituent that is moved in order to yield the VOS order is the object. It is therefore not surprising that there is some sensitivity to the definiteness of the moved constituent itself. That scrambling is better with definite DPs than with indefinite DPs is a well-known fact in Germanic languages (see de Hoop 1992 among others). We thus find just a similar pattern in Romance. Summing up, although it solves the Case problem, the remnant movement analysis faces a high number of empirical problems that do not arise under the scrambling analysis. Incidentally, we ended up finding some additional arguments in favor of the scrambling analysis by comparing the two. Taking into consideration the arguments listed above, I will not consider the remnant movement analysis for the derivation of VOS word orders in EP.

Objects in VOS: Scrambling in European Portuguese

63

So far, nothing has been said concerning the trigger for scrambling configurations. In recent literature, three proposals have been made to account for the nature of scrambling: a) Scrambling is A-movement and it is case-driven b) Scrambling is A-bar movement: i) Semantically-driven ii) Prosodically-driven In what follows, I review these three hypotheses, arguing in favor of the prosodic-driven approach by showing that it is the one that provides the best account of the data. Several authors have suggested that scrambling is movement to Spec,AgrOP, hence driven by the requirement for objects to check accusative features (Vanden Wyngaerd 1989, Mahajan 1990, de Hoop 1992, Zwart 1993, Bobaljik 1995 among others). Under this hypothesis, there should be no difference between Scrambling of the type just investigated and Object-shift in languages like Danish, Faroese, Icelandic, Norwegian and Swedish (Holmberg 1986, Webelhuth 1989, Vikner 1994, Bobaljik 1995, among others).26 However, as mentioned above, object-movement in object-shift languages more clearly displays Amovement properties that scrambled objects in Dutch, German and Portuguese. Two additional arguments may be considered for not considering scrambling Case-driven movement. a) Adverbs that are subcategorized by verbs scramble in Dutch (Costa 1995: 131a) shows that a verb like live in the relevant interpretation is not grammatical if no argument is selected. The argument may be either specific (106b) or non-specific (106c).27 Adding another adverb to check the position of the subcategorized one shows that the adverb can be scrambled (106e,h). If a non-specific adverb is scrambled (106g,j), the sentences become ungrammatical (under a non-specific reading). (106) a. *Ik woon. I live b. Ik woon daar. I live there

64

Postverbal subjects: syntax and discourse – VSO and VOS orders

c. Ik woon ergens. I live somewhere d. Ik woon steeds daar. all-the-time e. Ik woon daar steeds. f. Ik woon steeds ergens. g. *Ik woon ergens steeds. h. dat ik daar steeds woon. i. dat ik steeds daar woon. j. *dat ik ergens steeds woon. k. dat ik steeds ergens woon. This pattern is very similar to the behavior of NPs. The null hypothesis is to subsume both categories under the same phenomenon. Assuming, as usual, that adverbs do not require case, the movement of the adverbs may not be driven by case.

b) Another argument against case-driven scrambling comes from the behavior of subjects of unaccusatives in Portuguese: as hypothesized above, subjects in Portuguese may either stay in Spec, VP or move to Spec,IP. Allegedly, subjects of unaccusatives do not stay in Spec, VP but in their object base position. This is not possible to determine for Portuguese, but it has been noted for Italian by Pinto (1994,1997) that subjects of transitives and intransitives may not be low in wide focus context, while subjects of unaccusatives may be. Crucially, subjects of unaccusatives can also scramble, as shown in (107c) by the order NP-adv. (107) a. O Paulo chegou depressa. Paulo arrived fast b. Chegou depressa o Paulo, não chegou?. arrived fast Paulo, not arrived c. Chegou o Paulo depressa. arrived Paulo fast

Objects in VOS: Scrambling in European Portuguese

65

In case scrambling were to reduce to case-driven movement, this pattern would remain very mysterious: if the central property of unaccusative verbs is that they do not assign accusative case to their objects, scrambling of the object of unaccusatives should always be bad, since there would be objectmovement to an inert case-licensing projection. As (107c) shows, that is not the case: movement of the subject is possible, but case is not licensed, hence the target of movement is not a case-licensing position. Note that it cannot be argued that the object has moved to a potential landing site for Case-licensing, since that yields ungrammatical results in other types of constructions in Portuguese. For instance, in (108), a subject in a raising construction may not be stranded in the embedded infinitival Spec,IP (a potential case-licensing position): (108) *Parecem os meninos comer o bolo. Seem(3ps-pl) the kids eat the cake Given the counter-arguments presented here, and also those presented in Vikner (1994), I conclude that scrambling is not (case-driven) A-movement. If scrambling is not A-movement, the other option is that it is A-bar movement, not considering the existence of mixed positions. Diesing (1995) and Diesing and Jelinek (1995) take the fact that scrambling seems to have semantic effects, more in particular specificity effects, following de Hoop (1992), as evidence for an approach taking semantics to be the motivation for scrambling. According to Diesing and Jelinek (1995), scrambling is a type-mismatch repair operation. Following Diesing’s (1992) Mapping hypothesis, they propose that definite NPs and QPs have to scramble out of VP in order to escape existential closure. That is, VP is a domain for existential material. Since definite NPs and QPs are not existential, they may not be within that domain by the time a derivation reaches LF. Since definites may appear inside or outside VP at surface-structure, they suggest that this operation may apply overtly or covertly, provided that at LF the right mapping may be established: at LF, there may be neither definite NPs nor QPs within VP. A problem they are left with is the fact that indefinites do not fall under the scope of this operation, since there is no problem for them to be within the domain of existential closure. However, indefinites may scramble, as in (109). In order to solve this problem, Diesing and Jelinek (1995) propose that scrambling of indefinites follows from the assumption that the Scope Condition applies at S-structure in German. According to this condition, for an element to have scope over another element, the former must c-command the

66

Postverbal subjects: syntax and discourse – VSO and VOS orders

latter. This condition is semantically defined, and must hold at LF. According to the authors, the specificity of German is that the effects of the Scope Condition are surface-true. This assumption is based on the observation that different semantic representations result, depending on whether the indefinite follows or precedes an adverb. (109) German (from Diesing 1995) a. weil Elly immer Lieder singt. since Elly always songs sings a.’ ALWAYSt [time (t)] ∃x song(x) & sings (Elly,x,t) b. weil Elly Lieder immer singt since Elly songs always sings b.’ ALWAYSx[song(x)] sings(Elly,x) Diesing and Jelinek’s proposal may be falsified if we find a case of scrambling of indefinites across an adverb which does not affect the semantic/ temporal representation of the sentence. Such a case would be problematic for their analysis, since we would have a case in which the indefinite does not need to take scope over the adverb and yet it scrambles. Such examples in which no ambiguity between adverb and indefinite arises, and scrambling of indefinites is still possible, exist and are exemplified in (110). The difference between (109) and (110) lies in the choice of adverb. If we control for an adverb that has no effect on the semantics of the sentence, the prediction would be that scrambling of indefinites becomes impossible. However, that is not true. In (110), there is no necessary scope relation to be established between the indefinite and the manner adverb, and yet the scrambling order is possible: (110) a. O Paulo fala bem uma língua. Paulo speaks well a language b. O Paulo fala francês bem. Paulo speaks a language well An interesting fact about the adverbs that Diesing (1992,1995) and Diesing and Jelinek (1995) use has been brought up to my attention by Danny Fox (p.c.). He notes that their examples involve adverbs like always (cf. 109) whose semantics is sensitive to focus in the sense of Rooth (1985). If these

Objects in VOS: Scrambling in European Portuguese

67

effects are controlled for, scrambling appears to be still possible. Moving a bit ahead, and supposing that scrambling and focus are related, it is not surprising that changing the location of focus will change the semantics of these sentences. Depending on the location of focus, the adverb will associate with the indefinite or not (captured in Diesing and Jelinek’s analysis in terms of scopal relations). If the adverb is sensitive to focus and the indefinite is focused, the adverb will associate with the indefinite. If the focus is on some other constituents, the meaning of the sentence will be different. Under a strictly semantic account, leaving aside discourse effects, this correlation between adverbs, scope and focus would be hard to identify. I take this observation as the first piece of evidence in favor of the idea that scrambling is basically associated with the discourse structure of the sentence, in the sense of Reinhart (1995).28 The last hypothesis concerning the motivation for scrambling and the one I am going to adopt was outlined in Reinhart (1995). Adopting Cinque’s (1993) sentence stress algorithm, which states that the most deeply embedded constituent of a sentence is the one that will receive the most prominent stress, Reinhart observes that in a normal SOV sentence in Dutch, the object is the most embedded constituent; hence it is the element bearing the most prominent stress and gets interpreted as the focus. Obviously, a distinction must be established between default stress and marked stress: the default stress is the one that follows from the stress algorithms proposed in Nespor and Vogel (1986), Cinque (1993) and Nash (1995), and normally associates with clause-final position and correlates with embedding. Heavy prosodic stress is required when the syntactic configuration obviates the application of the default rule (e.g. stress shifting and heavy stress on focused subjects in English, cf. Zubizarreta 1995 among others). Now, where does scrambling come in? According to Reinhart, the motivation for scrambling is to be found at the interface between PF constraints and discourse-structure. The motivation for scrambling the object is to make it escape the default focus stress. Actually, Reinhart proposes that scrambling allows the verb to be focused. However, this description of the facts seems not to be entirely correct, for several reasons: in both Dutch and German, the verb may be focused without resorting to scrambling: it is sufficient to shift the stress; scrambling across an adverb focuses the adverb independently of the stress that the verb receives (see Baart 1987, Büring 1997 among others for the relevant data). The problem with Reinhart’s initial generalization is that the difference between default and marked stress, though acknowledged, is not completely spelled out. Default stress appears as the normal stress in declarative sentences,

68

Postverbal subjects: syntax and discourse – VSO and VOS orders

marked stress requires a marked intonation contour on the focused constituent.29 Once these two are differentiated, it is easy to see what the generalization is like: the complement of the verb scrambles in order to leave the adjunct in the most embedded position where the latter can receive the most prominent stress by default. Of course, for the complement to be able to do this, it must not require stress itself. Scrambling is thus a twofold operation: on the one hand it involves defocusing of the object, on the other hand it is an operation that allows another constituent to be stressed.30 Scrambling is possible with stress on the verb, but then marked stress is needed. Data showing these differences are presented in Baart (1987). Independently of this, Reinhart’s analysis seems correct: scrambling is used for creating appropriate focus configurations, namely to make the element bearing the sentence nuclear stress escape it. Let us see how this hypothesis is corroborated by Portuguese (the data are similar in Dutch and German): in (111), it is expected that the object be the focus of the sentence given that it is the new information requested in the question. Indeed in this case the object cannot be scrambled, as the infelicitous sentence in (111Ba) shows: (111) A: O que é que o Paulo fala bem? what does Paulo speak well? B: a. #O Paulo fala francês bem. Paulo speaks French well b. O Paulo fala bem francês. Paulo speaks well French If, on the other hand, the adverb is questioned, so that the object is old information, it must not be in the position where it gets default stress; hence it must be scrambled, allowing the adverb to receive the default stress: (112) A: Como é que o Paulo fala Francês? how does Paulo speak French? B: O Paulo fala francês bem. Paulo speaks French well. #O Paulo fala bem francês. Paulo speaks well French

Objects in VOS: Scrambling in European Portuguese

69

An interesting piece of data brought to my attention by Martin Honcoop is the case in which a scrambled element is associated with focus, which, in principle, seems to disconfirm Reinhart’s theory: (113) Jan zei dat ik DE KRANT gisteren las, (en het boek vandaag) Jan said that I the newspaper yesterday read, and the book today This example is interesting because the NP is scrambled while it is also part of the focus of the sentence. However, there are two important differences between this case and the cases Reinhart describes: first, the scrambled NP bears heavy stress, and second, the NP is not the focus of the sentence by itself. When this sentence is uttered, the speaker intends to focus both the NP and the adverb gisteren/yesterday. According to Reinhart (1995), the focus set of a sentence is the constituent bearing the most prominent stress plus everything it dominates. That is, in order to build an appropriate focus set for (113), the speaker has to make the object dominate the adverb. The only way to do that is by scrambling it. Now, according to the motivation for scrambling that explored by Reinhart, the scrambled position will not allow the object to receive the default stress of the sentence, since it is not the rightmost constituent anymore. The only way to make it the focus of the sentence is to resort to prosodic stress (see also Nash 1995). Therefore, example (113), rather than disconfirming Reinhart’s theory presents evidence in favor of it: scrambling is related to the need to create appropriate focus configurations: furthermore, it is related to defocusing, since the NP in the scrambled position cannot be the focus of the sentence unless it bears a heavy stress. The only way in which (113) would be problematic would be if the scrambled NP were be the focus and the adverb were not included in the focus set of the sentence. Under that interpretation, (113) is infelicitous. Summing up, in this section, I have shown that European Portuguese has scrambling of the German/Dutch type, by looking at the properties of scrambling in Dutch and German, and by comparing them with the order NP-Adv in European Portuguese. I have argued that scrambling is an A-bar movement operation that adjoins the complement of the verb to VP. Finally, I discussed the motivation for scrambling, arguing in favor of Reinhart’s (1995) prosodic/ discourse explanation of scrambling. I showed that this was the hypothesis facing the least number of empirical problems. Crucial for the discussion in this chapter is the conclusion that subjects in VOS orders may be analyzed as being stranded in Spec, VP.

70

Postverbal subjects: syntax and discourse – VSO and VOS orders

Two questions remain unanswered: in which contexts are in-situ subjects and how is the licensing of in-situ subjects done? These will be the topics of the next chapter.

4. Inversion and information structure

4.1. Introduction In this chapter, it is argued that, in a language like European Portuguese, properties of the information structure partially determine the distribution of arguments. It will be argued that subjects stay in Spec, VP, only if they are focused, and that in-situ subjects are licensed under Agree (cf. Chomsky 2001). Likewise, they may only raise to Spec,IP if they are not focused, and when the entire sentence is focused. The view on focus to be adopted is that information focus is rightmost, for prosodic convergence. This analysis of focus leads to a number of questions regarding my assumptions concerning the exact formulation of this constraint. A very common view on focus is to assume that focused constituents must be licensed at the specifier of a functional projection (Focus Phrase), which in most analyses appears at the left periphery of the sentence (Brody 1990; Kiss 1995; Rizzi 1997). In the analysis to be developed here, the exact reverse effect is obtained: focused constituents appear at the rightmost position of the sentence. As it will be shown later, the two approaches are not incompatible, but they operate on two different kinds of discourse objects. Nevertheless, it is important to be clear about what type of focus I am talking about, and why I am defending an in-situ analysis of focus rather than an analysis that involves raising to a functional projection. Therefore, I will present some arguments against a focus-movement approach. Let me emphasize, for the sake of clarity, that I will be looking at information focus: that is, focus that conveys new information (Dik 1978; Büring 1997; de Hoop and Swart 2000; Kiss 1996 among many others). This remark is important, since the term focus is often used with a very wide variety of meanings, rendering discussion of phenomena and identification of scope of research quite difficult (see Givón 1990 for classification and examples of several constructions involving focus-related aspects). I am interested in this study in the focus that is normally associated with a high pitch accent (Selkirk 1984, among others) and that can be identified in question-answer pairs and correction contexts. This type of focus is often called information focus. That is, it is focus in the sense that it conveys new infor-

72

Inversion and information structure

mation without altering the truth value of the sentence (see Vallduví 1990, among others). This excludes from the discussion contrastive focus and focus involving uniqueness or exhaustive listing in the sense of Szabolcsi (1981). In the last section of this chapter, I will present some tests to distinguish the several types of focus, and show that they do not necessarily intersect. The behavior of contrastive focus will be shown to be different both in distribution and in meaning. Below I provide examples of each of the constructions involving foci, in order to specify which will be the topic of this chapter: (1)

Focus Constructions: a. In situ: Comeu a sopa o Paulo. Ate the soup Paulo b. Syntactically marked focus (cf. Raposo 1994, among others) Muito vinho o João bebeu! a-lot-of wine João drunk c. Focus-preposing (ungrammatical in Portuguese, OK in other languages). *ESSE LIVRO, o João leu. That book João read

(1a) will be the topic of this chapter. I will not talk about (1b), which is discussed in Uriagereka (1995), Raposo (1994), among others, since its discourse characteristics are different from (1a), and I will argue that the ungrammaticality of (1c) in European Portuguese is an additional argument for not considering focus-preposing as an argument for movement of foci.

4.2. Focus-movement? Focus is represented in different ways in different languages. While a language like English displays focus in situ, as in (2), a language like Hungarian seems to require movement for licensing focused constituents: (2)

I saw JOHN.

Focus-movement?

(3)

a. AZ ÚJSÁGOT dobtam el. the newspaper threw-I away

73

(from Horvath 1995)

b. *Eldobtam AZ ÚJSÁGOT. Chomsky (1976) has argued, on the basis of the contrast in (4), that even in English focused constituents need to be moved. Chomsky showed that focus as in (4b) induces weak-cross-over effects just like other operators do e.g. in (5): (4)

a. Hisi mother saw Johni b. *Hisi mother saw JOHNi.

(5)

*Whoi does hisi mother like?

According to Chomsky, the most natural way to explain the parallelism between (4b) and (5) is to assume that, at LF, focused constituents move establishing an operator-variable relation with their traces, yielding a structure like (6b) for a sentence like (6a): (6)

a. Mary loves JOHN. b. [S JOHNi [S Mary loves ti ]]

This accounts for the weak cross over effects in a rather natural way. The LF-representation of (6b) is a violation of Koopman and Sportiche’s (1983) bijection principle in the same way the overt syntax of (5) is: (5)

*[S JOHNi [S hisi mother loves ti ]]

(6)

Bijection Principle: There is a bijective correspondence between variables and A-bar positions.

Given (6), the problem with (5) is that ‘John’ is binding two variables: the pronoun and the trace, yielding a violation of the bijection principle. The weak-cross-over argument together with the existence of overt focus-movement in Hungarian led linguists to assume that languages with focus in-situ like English need to move focused constituents at LF.31

74

Inversion and information structure

Movement theories of focus have been criticized in Anderson (1972), Rooth (1985), von Stechow (1990), among others, on the basis of their failure to account in an appropriate way for lack of ECP and subjacency effects, multiple foci and crossing paths at LF. I will briefly discuss these three problems. First of all, Focus displays a lack of ECP and subjacency effects. If focus involves movement, focusing should be impossible within a barrier for extraction. That this is not true can be seen from the following examples:32 (7)

Focus on an embedded subject (that-trace effects are expected): Mary thinks that JOHN will go to the movies.

(8)

Focus within an adjunct: Mary arrived late because she was SICK.

(9)

Focus within shifted Heavy NPs: I read yesterday all the books MY teacher recommended.

(10) Focus inside Wh-islands: I wonder what to write with THIS PEN. (11) Focus inside a complex NP: John announced a plan to steal FIVE cars tonight. (12) Focus within a coordinate structure: John saw Mary and ALL the other students. Any movement approach to focus predicts that these sentences should be ungrammatical, since ECP effects (and subjacency) are assumed to be operative at all levels of representation (May 1985; Huang 1982; Koster 1987; Bayer 1995). If alternatively, one would assume that ECP is not operative at LF, one would miss the empirical generalization that may be drawn considering LF-movement, namely the similarities with overt movement (cf. the studies cited above). Another problem with assuming that focused constituents move at LF (on a par with wh-movement) is the existence of multiple foci: (13) a. JOHN saw MARY. b. John gave the BOOK to BILL.

Focus-movement?

75

The problem with these sentences is different depending on the theory of focus-movement assumed. If it is assumed, as Brody (1990) does, that focused constituents enter a Spec,Head agreement relation with a Focus head and that Hungarian is the overt counterpart of English, this would imply that a head has multiple specifiers.33 This in turn predicts a lack of adjacency between the foci and the element lexicalizing the head (V in Hungarian, according to Brody (1990)). However, multiple focus movement is not possible in Hungarian: (14) Évat János várta a mozi elótt. Eve-acc John waited the cinem in-front-of Eve, JOHN waited for her in front of the cinema.

(from Kiss 1995)

If multiple specifiers are permitted, sentence (14) might incorrectly yield the English interpretation in (15). An uniformization with wh- in situ is not possible, since the postverbal PP in (14) is not interpreted as focus: (15) JOHN waited for EVE in front of the cinema. The third problem comes from the fact that focus movement may create crossing paths at LF. This argument obviously does not go through in theories allowing for crossing paths (e.g. Chomsky 1995). I will not take a position with respect to whether the theory should or should not allow for crossing paths. In case it should not (Pesetsky 1982, Kayne 1994), these cases remain problematic. The creation of crossing paths at LF arises in a focus movement analysis for a sentence like (16): (16) Who did JOHN wait for? This sentence should be ungrammatical because moving JOHN at LF would induce a crossing path, which is a ill-formed path according to Pesetsky (1982): (16’) JOHN who did’t wait for t? These three types of evidence should be enough to discard an analysis of focus in terms of movement. However, there is still the evidence from WCO-effects and the distribution of focus in Hungarian. I will return to these problems after finishing the discussion of the focus-movement analyses.

76

Inversion and information structure

If focus-movement exists, it is also important to determine the exact nature of this movement: focus movement establishes an operator-variable relation, but why should there be such a relation?, and if there is movement, where do focused constituents move to? Two widely accepted theories of focus movement are the ones advocated by Brody (1990) and Horvath (1986,1995). In this section, I will discuss the major aspects of both theories showing that they are not empirically adequate. Brody (1990) argues that there is a Focus Phrase, where focused constituents move to in order to satisfy the focus-criterion, parallel to Rizzi’s (1991) wh-criterion:34 (17) Focus-criterion: a. At S-structure and LF, the Spec of an FP must contain a +f-phrase. b. At LF, all +f-phrases must be in an FP. The source of cross-linguistic variation is then whether (17a) is satisfied overtly (Hungarian) or covertly (English). As noted by Horvath (1995), this approach to focus is not satisfactory since it does not account for the fact that languages other than English and Hungarian codify focus in positions that are not either the base-position or the leftmost position of the sentence. Horvath (1995) proposes that focus is either assigned like any other grammatical feature (e.g. Case) by a non-lexical head or freely assigned. The need for V-adjacency in Hungarian comes from the need to lexicalize the functional head that assigns the focus feature (which Horvath claims to be I° in Hungarian). Her formulation of the focus parameter is given in (18). The conditions in (133,1– 4) determine what type of manifestation of focus can be found in different languages: (18) Focus parameter: 1. nature of the feature: (i) freely occurring, i.e. not transferred from another category (English)

vs.

(ii) assigned by a specific X° category (Hungarian)

2. what X° functional category of the clause is the assigner, i.e. the source of the feature 3. whether the feature-assigning category needs to be lexicalized

Focus-movement?

77

4. the mode/nature of the process of feature-assignment: (i) feature transfer (ii) Spec-head agreement Horvath’s focus parameter has the advantage of accounting for crosslinguistic differences on the representation of focus without crucially resorting to an additional functional projection, making the set of primitive categories of the theory simpler. However, relying on a specific functional projection for assigning of focus creates problems for cases in which focus surfaces in isolated constituents. Let us consider the case of English. According to (18), focus in English is not codified in the syntax. Therefore, focus is freely assigned. However, as long as the English syntax becomes more flexible, alternations seem to be able to codify different information structures: (19) A: What did you give to Mary? B: a. I gave Mary a book. b. #I gave a book to Mary. (20) A: Who did you give a book to? B: a. #I gave Mary a book. b. I gave a book to Mary. If focus were freely assigned to any category, the contrast between (19) and (20) should not obtain. Any of the alternations should be adequate independently of the context. One of the most widely accepted theories of focus-in-situ is Rooth’s (1985). Rooth proposes a semantic theory for focus, according to which no focus-movement is required: given a sentence where focus can be identified, a set of alternatives is construed. The set of possible alternatives is constrained within a certain contextual domain reminiscent of Jackendoff’s (1972) P(resupposition)-set. Rooth (1985) accounts for Chomsky’s (1976) weak-cross-over independently of focus-movement. Rooth claims that the bound variable readings require λ-abstraction, so that the pronouns or noun phrases can be interpreted as bound variables and not as free variables. A mechanism enabling this to happen is already available in the grammar: Quantifier Raising (May 1985). In other words, it is not necessary to have focus movement as an

78

Inversion and information structure

independent rule of the grammar: like all NPs, focused NPs may be QR-ed but need not. However, if a bound variable reading is intended, QR is obligatory. In that case, LF-evaluation constraints are operative, and weak-crossover configurations are ruled out. Rooth’s approach does not yet explain why deaccenting (that is, removing the stress from the focused constituent) obviates the WCO effects, but it is empirically superior to Chomsky’s since it excludes the obligatoriness of WCO with focus (see also Vallduví 1990 for discussion).35 Adopting Rooth’s approach allows for dismissing the weak-cross-over argument as evidence in favor of focus-movement. At best, we can keep it as evidence for QR. However, Rooth’s theory of interpretation of focus in situ does not say anything concerning languages like Hungarian, in which focused constituents move to a specific position. Also, it does not explain the behavior of languages like Portuguese, in which, as it will be shown, focused constituents seem to stay very low in the structure. Since Rooth’s analysis is not complete enough to take care of the word order facts, I will keep it as a semantic approach to focus. Syntax together with prosody will enable an identification of the focus set of constituents for a given sentence. The identified focus-set will be operated on by semantics. Provided that there is an algorithm permitting a correct identification of focus, semantics may apply over the material identified as focus.

4.3. Word order and focus Having established part of my assumptions concerning the representation of focus, let us look at the relationship between word order and focus, as far as the placement of subjects is concerned. The first important aspect to note is that the attested word order variation is not free: each word order can be used only in given contexts. This has been noted several times in literature on Portuguese (Duarte 1987, 1996, 1997; Ambar 1992, 1994, 1996, 1997, Martins 1994; Costa 1997, 1998). For each of the possible word orders listed in (21), a different felicity context is associated. In (21), I also indicate the position each constituent occupies in the clausal structure, according to the conclusions reached in the previous chapter. (21) SVO: subject in Spec,IP, object in base-position VSO: subject in Spec, VP, object in base-position

Word order and focus

79

VOS: subject in Spec, VP, object adjoined to VP (via scrambling) OSV: subject in Spec,IP, object topicalized OVS: subject in Spec, VP, object topicalized The relation between each word order and context is demonstrated in the following examples. I will proceed by providing an appropriate discourse context and check which possible order is an appropriate continuation for each case. For the proper characterization of these contexts, the notions topic and focus will be relevant. I will assume the following tests to identify topics and focus: a) In a question-answer pair a focused constituent in the answer replaces the wh-word in the question (cf. Dik 1978, Bresnan and Mchombo 1987, Rochemont and Culicover 1990, among others) b) A Topic is information already referred to in the discourse or a subpart of a referent already mentioned (see Büring 1997 for discussion and relevant examples).36 (22) Object focused: A: O que é que o Paulo partiu? what Paulo broke B: O Paulo partiu a janela. Paulo broke the window #Partiu o Paulo a janela. #Partiu a janela o Paulo. #A janela o Paulo partiu. #A janela partiu o Paulo. (23) Sentence-focus: A: O que é que aconteceu? what happened B: O Paulo partiu a janela. Paulo broke the window #Partiu o Paulo a janela. #Partiu a janela o Paulo. #A janela o Paulo partiu. #A janela partiu o Paulo.

80

Inversion and information structure

(24) Subject and object focused: A: Ninguém partiu nada. noone broke anything B: #O Paulo partiu a janela. Paulo broke the window Partiu o Paulo a janela. #Partiu a janela o Paulo. #A janela o Paulo partiu. #A janela partiu o Paulo. (25) Subject is focused: A: Quem é que partiu a janela? who broke the window B: #O Paulo partiu a janela. Paulo broke the window #Partiu o Paulo a janela. Partiu (a janela) o Paulo. A janela o Paulo partiu. A janela partiu o Paulo. I will not discuss in detail the cases in which the object appears in sentenceinitial position. For the moment, it is enough to say that for objects to appear in sentence-initial position, they have to have been referred in previous discourse and/or have some contrastive force: (26) A: A Ana viu o Paulo? Ana saw Paulo B: O Paulo, ela viu. Paulo she saw (27) A: Quem é que partiu as janelas? who broke the windows B: Esta janela partiu o Paulo this window broke Paulo. Summarizing, in terms of information structure, all constituents that convey new information appear to the right, and constituents that convey information

Mapping syntax-discourse

81

previously referred in the discourse appear at the left periphery of the sentence. The following question remains to be answered: why should these relations between positions and discourse information exist? What does each of the identified syntactic positions have to do with the correspondent discourse function?

4.4. Mapping syntax-discourse I will be assuming the following: a) Old information has to be either topicalized or defocused, while new information is the Focus of a sentence; b) I will follow proposals concerning the correlation between syntax and phonological sentence stress which suggest that sentential stress falls on the rightmost constituent of a sentence. Thus, in a normal SVO sequence, stress will fall on the object. Assuming with Jackendoff (1972) that the focused element in a sentence is the one bearing the most prominent stress, the proposals above capture the fact that in an SVO sentence with unmarked intonation the object is interpreted as the focus (cf. Lambrecht 1994 who shows that objects tend to be focus). Still concerning focus, I will follow Reinhart’s (1995) suggestion that XPs may be marked as foci with a heavy stress. This happens to any XP that does not get default stress by virtue of not being rightmost.37 c) I will assume with Zubizarreta (1995) and Reinhart (1995) that some syntactic operations are prosodically motivated. d) I will assume that the following tendencies hold, as observed by Lambrecht (1994): subjects tend to be topics; objects tend to be foci; definites tend to be old information (topic); indefinites tend to be new information (focus). It is important to emphasize that these are just tendencies and not absolute statements. Optimally, it would be possible to derive these tendencies from some structural property of subjects, objects, definites and indefinites. Let us now come back to the issue under investigation and see how these assumptions derive the facts described in the previous sections. From the line of inquiry I am pursuing, it is obvious that some new facts have to be

82

Inversion and information structure

added to the paradigms described. More specifically, since the notion of focus is crucial, and I am assuming with Nespor and Vogel (1986), Cinque (1993), Zubizarreta (1998) and Reinhart (1995) that focus is partly a prosodic phenomenon,38 it now becomes important to see how subjects behave with respect to intonation in each of the positions identified above. It is a well-known fact that focus is normally associated with high stress. In order to get a clear idea of the correct way to interpret focus, it is necessary to investigate where prominence appears in the sentences. By doing so, I will also be able to derive the generalization of the preceding section. Frota (1994,1995) has argued in favor of representing focus in Portuguese as a phonological category that is freely assigned. Part of her arguments is based on the behavior of clitics (see Frota and Vigário 1996 for complete argumentation). The domain of Frota’s argumentation is narrow (or contrastive) focus, and not information focus (though of course, they may coincide). Actually, Frota (1997) suggests that two types of prominence may be necessary to describe the two types of focus marking. What is relevant for the discussion is that independently of the type of focus looked at, Frota’s claim appears to be correct: focus (contrastive and information) is marked phonologically in Portuguese. If the distribution of stress in each of the word orders discussed above is considered, evidence may be found in favor of Frota’s claim: there is a oneto-one correspondence between the first focused constituent of the sentence and its most prominent stress.39 Consider (28) below, where capital letters indicate high stress: (28) a. Partiu o PAULO a janela. broke Paulo the window b. *PARTIU o Paulo a janela. c. *Partiu o Paulo A JANELA. (28) is a VSO sentence in which subject and object are in focus. In that case, both have to appear in the right periphery of the sentence. The subject is the first focus that appears and it bears the most prominent stress of the sentence. Conveying the same information by means of the same word order and with other stress patterns is not possible (28 b,c). In (29), only the subject is in focus (VOS). In that case, the subject bears the heaviest stress, though it is not heavier than the neutral stress present in unmarked SVO sentences.

Mapping syntax-discourse

83

(29) a. Comeu a sopa o Paulo. ate the soup Paulo b. *Comeu a sopa o PAULO. c. *Comeu a SOPA o Paulo. d. *COMEU a sopa o Paulo. In SVO sentences, the most natural intonation is the rightmost prominence, without a very heavy stress: (30) O Paulo partiu a janela. Paulo broke the window The literature on prosody often distinguishes between neutral stress and heavy stress (see e.g. Chomsky and Halle 1968). This distinction has been criticized by Selkirk (1984), among others, who claims that there is no empirical advantage in proposing such a distinction. Selkirk’s argumentation is based on focus projection. Focus projection is the term given to the fact that neutral stress on e.g. a sentence-final object may yield an interpretation in which only the object is focused, or the VP is focused or the whole sentence is focused. Selkirk claims that not only neutral sentence-final stress projects allowing an interpretation in which there is focus on sentence-final element only, on the VP or on the whole sentence. Selkirk claims that heavy stress in a non-final constituent also permits focus-projection in the sense just explained. If Selkirk’s observations are applied to the Portuguese case, it is expected that sentence (30) be an appropriate answer to any of the questions in (31): (31) a. What happened? (Sentence-focus) b. What did Paulo do? (VP-focus) c. What did Paulo break? (Object-focus) This is indeed correct. However, differently from Selkirk’s claim for English, a distinction between neutral stress and marked stress is relevant for the identification of focus in Portuguese. Crucially, every time there is a high stress on a non-final constituent, there is no projection of focus in this language. (30) is not a legitimate sentence for expressing VP-focus. Since high stress is only necessary on constituents that are not in absolute sentence-

84

Inversion and information structure

final position, this invalidates Selkirk’s claim: marked stress does not project. (32) Partiu O PAULO a janela. broke Paulo the window Selkirk claims that stress on XP, which is a constituent of YP, will enable percolation of stress to YP. If focus projection would apply in these terms, (32) could be a felicitous answer to (31b), with focus on the whole VP, since the subject is VP-internal. In other words, the subject corresponds to XP and the VP to YP in Selkirk’s algorithm. Hence, stress on the subject should project to VP. Now, this is not true, though it is predicted by Selkirk’s theory. Note that even neutral stress does not project if there is a change in the unmarked word order of the language. Hence, a VOS sentence is not a legitimate answer to a question that requires something else than the subject to be in focus: (33) What did Paulo do? a. O Paulo partiu a janela. Paulo broke the window b. #Partiu a janela o Paulo. Although the sentence’s main stress fall on the rightmost constituent in both cases, and in both cases the rightmost constituent is VP-internal, focus projection is not allowed. Actually, even the English cases Selkirk presents as possible cases of focus-projection without rightmost prominence are difficult to evaluate. Selkirk claims that (34) may have VP-focus: (34) (from Selkirk 1984): Did John give a BOOK to Bill? Selkirk claims that (35) is an appropriate answer for this sentence: (35) No, he grew a pot of NARCISSUS for him. In (35), the verb and the NP contrast with give a book in (36). Selkirk concludes from this that prominence on the NP may give VP focus.40 It seems,

Mapping syntax-discourse

85

though, that this is not a very accurate conclusion, since if there were VPfocus in (36), the sentences in (36) might as well be appropriate answers, since (36a) and (36b) involve alternatives to the focused VP: (36) a. No, he grew a pot of NARCISSUS for Mary. b. No, he killed Mary. Now, (36a,b) are not appropriate answers to (34), presumably because in (35) there is no VP-focus. Actually, Selkirk acknowledges that for (35) to involve VP-focus, “John, Bill, and, say, Bill’s recent birthday are old information in the discourse”(p.216, my emphasis). If Bill is old information, it is difficult to understand how it can be maintained that the whole VP is in focus. In other words, I am suggesting that the idea that VP-focus is involved comes from the fact that grow X for Y and give X to Y are minimally different in the relevant context. No new information is added by replacing give with grow. In this sense, the only new information is the NP, and VPfocus is only apparent. The use of a proper name in the question may also determine the impression that there might be VP-focus even without Bill being focused. Proper names never convey absolute new information. If instead of a definite, someone is used, a VP-focus interpretation never arises unless there is rightward prominence: (37) A: What did John do? B: #John gave a BOOK to someone. John gave a book to someone. Speakers who accept (37B) report to me that the DP a book clearly must have contrastive force. We have thus a case of overlap of two types of focus: a contrastive focus on the DP and information focus on the VP. Since these two types of focus may be distinguished in semantic terms, the behavior of one should not be used as evidence for the other. Summarizing, it is thus possible to interpret the stress pattern of (35) in different terms: since the pronoun is old information (it refers to Bill), it must not be heavily stressed. In the context given, grew a pot of narcissus for him and give a pot of narcissus to him are equivalent. Hence, the only new information which is contrasted is the DP object. The stress pattern emerges as a consequence of shifting the stress from the sentence-final PP to the object DP. The VP-focus effects arise only if the identity between

86

Inversion and information structure

grow X for Y and give X to Y in the ‘birthday’-context are not acknowledged. The fact that not any VP may replace the one in the question confirms that in this case there is no VP-focus. It seems thus that whenever there is a change either in the unmarked word order (SVO for Portuguese) or in the normal intonational pattern (rightmost prominence), there is no projection of focus. The intonation of information focus in Portuguese can then be summarized as follows: Focused constituents are prosodically prominent. If they are rightmost they bear neutral stress, if they are not rightmost they are assigned a high pitch accent. If there is more than one focus, the leftmost bears the heavy stress: all constituents following the heavy stress are interpreted as focus. Information focus is not incompatible with other types of contrast. Hence, any constituent may bear heavy stress, independently of its being the focus of the sentence for contrast purposes. This will, however, make projection of focus more difficult for the reasons pointed out above. Under the review of Selkirk’s discussion of projection of focus made above, I reach conclusions similar to hers regarding the status of projection as not being exceptional, although in the exact opposite sense. She concluded that focus-projection is not exceptional, since it may happen almost everywhere. Differently, I concluded that focus-projection is unexceptional, since it does not need to be postulated. The reasoning goes as follows: all that is needed is rightmost prominence. The effects of projection are a consequence of coincidence of rightmost borders of constituents (NP,VP,IP). Any other stress pattern will preclude projection, since projection does not exist as an independent phenomenon. It is just the effect of the ambiguity of several rightward constituent borders. Combining the distribution of focus and the prosodic facts, the following generalizations are obtained: a) Focused constituents are rightmost in the sentence; b) Focused elements bear high stress (neutral or marked); c) If there are multiple foci, they appear all to the right of the non-foci elements. d) If there are multiple foci, the first in a left-to-right fashion bears heavy stress. These observations may serve as the cues to the formulation of an algorithm to identify information focus in European Portuguese, which is given in (38).

Mapping syntax-discourse

87

Note that (38) is meant as an algorithm to identify information focus, not the other types of focus, excluding thus the cleft constructions discussed above, which were nevertheless useful for the study of focus projection. This formulation is partially based on Reinhart’s (1995) observations concerning the distribution of focus and on her own formulation of what a focus-set of constituents is: (38) The focus set of constituents of a sentence is the prosodically most prominent constituent plus everything it c-commands. Prosodic prominence is defined as in (39), adapted from Nespor and Vogel (1986), Cinque (1993), Zubizarreta (1995) and Nash (1995): (39) The prosodically unmarked most prominent constituent is the rightmost one, following the recursion pattern of a language. (39) states that in VO languages, the most prominent constituent is the rightmost one to the right of the verb, while in OV languages, the most prominent constituent is the rightmost to the left of the verb (the XP in bold in (40) below): (40) V O XP XP O XP XP V Let me now return to the algorithm presented above. When the conditions for a constituent which has to be interpreted as focus to receive the sentence (neutral) most prominent stress are not met, a heavy stress has to be assigned. This happens e.g. in the case of multiple foci, in which two constituents cannot occupy the rightmost position at the same time. Let us see how the algorithm in (39) allows for identifying focus in the cases discussed above. In the case of a VSO sentence with high stress on the subject, the set of focused constituents will be the subject and the object. The subject is interpreted as focus, because it is the most prominent constituent. The object is interpreted as focus, because it is c-commanded by the subject. Since the verb is out of the c-command domain of the subject, it is not interpreted as focus. In the case of VOS sentences, the subject bears the main neutral stress and does not c-command anything, hence it is the only constituent interpreted as focus. The focus on the subject may not project, because the unmarked word order is changed. The impossibility for rightmost focus to project when

88

Inversion and information structure

there are changes at the unmarked word order will not follow from any considerations in this chapter. A solution for this problem will be proposed later. As for SVO, two situations are possible: the most natural is that the object is the rightmost element, bearing sentence neutral stress. In that case, since it does not c-command anything, only the object is interpreted as new information. The other two possibilities are interpretation of these sentences as VP-focus and everything in focus. These interpretations also follow from the algorithm in (39), since the object is also the rightmost part of the constituent IP or VP that one wants to focus in cases of sentence-focus and VPfocus respectively. Focus projection is thus interpreted here as a natural consequence of the general neutral stress rule: assign the most prominent stress to the rightmost element. This rule is general to any constituent, independently of its label, as observed by Cinque (1993) among others41: (41) a. [NP the good man with the red shirt] b. [AP much more beautiful] c. [PP before midnight] d. [VP give something to someone] e. [IP John gave a book to Mary] Now, why is focus projection impossible in a VSO sentence? The explanation is simple: since the subject is the most prominent constituent, it cannot be the case that the whole sentence is in focus, since the stress does not fall on its rightmost element (see also Cinque 1993 and Reinhart 1995 for a similar reasoning). Should it be possible then that stress on a sentence-final subject would yield sentence-focus? In principle, nothing precludes it, since the subject is both the rightmost element and the most prominent one. There is however a reason for excluding this possibility: if one compares a VOS sentence with a SVO sentence, they are equal in terms of prosody. However, the former is more marked, since the subject does not appear in its canonical position: Spec,IP. Hence, since there is no difference in status in terms of possible interpretations, the least marked SVO sentence is grammatical.42 Let us now see why some sentences are impossible when associated to some interpretations, and how that follows from the algorithm in (39). Why can’t a VSO sentence be interpreted with focus on the subject only? That is, why is (42B) not a felicitous answer to (42A):

Mapping syntax-discourse

89

(42) A: Who broke the window? B: #Partiu o PAULO a janela. broke Paulo the window The problem with this answer is that it forces the object to be interpreted as new information, since it is within the c-command domain of the subject which bears the most prominent stress. Since window has been referred to in the discourse, this interpretation is not felicitous. Another case excluded by the algorithm in (38) is the interpretation for a VSO sentence in which the subject is not stressed: (43) *Partiu o Paulo a JANELA. broke Paulo the window The problem with this sentence is that, according to the definitions in (39), only the object can be interpreted as focus. Now, if that is the case, the sentence is ruled out for either of the two following reasons: if the subject is to be interpreted as new information, the sentence is out since the subject is not included in the focus set of constituents. This is because the subject is not c-commanded by the most prominent constituent. Alternatively, if the interpretation required is one in which the subject is not to be interpreted as focus, and then there is no reason for it to stay in a low position.43 Also, heavy stress should never appear in sentence-final position, unless for contrast purposes, since it is not necessary in order to assign prominence to a constituent. It seems thus that the principles given above make a large number of correct predictions, concerning the possible intonations and mappings between word orders, intonations and discourse functions. Summarizing the results of this section, I present the following table, containing the several possible word orders with most prominent element marked in bold. The second column of the table indicates the focus set of constituents given by each word order, and the third column indicates the reason why some pairs word order/intonation and word order/focus-set are impossible:

90

Inversion and information structure

Table 1. Word Order

Focus-set

Reason for ungrammaticality

SVO

O, VP or IP

VSO

S and O

VOS

S

*SVO

S

S is not c-commanded by the most prominent element

*SVO

S

V and O are c-commanded by the most prominent element; they should be part of the focus-set

*SVO

IP

The sequence with unmarked stress blocks the more marked one

*VSO

IP

Stress is not in the rightmost constituent of IP

*VSO

IP

SVO is preferable, since the word order is less marked (cf. Costa 1998, and section 3.4)

*VSO

O

Stress falls on the subject

*VSO

O

Since the subject is not in focus, there is no reason for it to stay low

*VOS

VP, IP

SVO is preferable, since the word order is less marked (cf. 3.4)

*VOS

S and O

O has been scrambled for escaping focus, so it may not be the focus.

Table 1 illustrates the relevance of two aspects for the identification of focus: stress assignment and c-command by the most prominent element. It is important to note at this stage that the algorithm in (39) just represents the instructions hearers have to follow in order to identify the focus of a sentence. As such, these principles are quite descriptive in nature. They are explanatory only to the extent that they follow from independent principles:

No focus-movement in Portuguese

91

in order to get to (39), it is necessary to have a theory of phrasal phonology which relates sentence stress assignment to the directionality parameter (Nash 1995; Nespor, Guasti and Cristophe 1995). In addition, a theory of sentence structure making sure the c-command requirement is met in the relevant configurations is required. Finally, there has to be an independent explanation for the part of the principles that require focus to be prosodically prominent (see Givón 1984 for an explanation in terms of attention span). It is a combination of these three aspects that makes it possible to arrive at a formulation of an algorithm for sentence stress identification.

4.5. No focus-movement in Portuguese Concerning the debate focus-in-situ vs. focus-movement, from the Portuguese data discussed above, one may conclude that several rearrangements of the sentence word order must be made for the sake of identifying focus, although there is not one specific position where focused elements move to on the left periphery of the sentence, as is the case in Hungarian. Instead, focused elements seem to stay in the rightmost position, and in most cases they do not undergo any movement in order to reach this position. Given this behavior, the conclusion obtains that Portuguese is not of the Hungarian type, and it can be treated as a language with focus in situ.44 In Costa (1998, 2000b), empirical arguments are presented showing that a focus-movement approach would make wrong predictions for dealing with these data. For completeness, let me just point out that the syntactically marked focus constructions that are found in Portuguese, described by Raposo (1994), among others, involve quantified DPs, and is exemplified in (44) (Raposo, p.c.): (44) Muito vinho o João bebeu! Much wine João drank This type of construction is much more natural if the sentence is exclamative, which casts some doubt on their status as instances of focus-movement. Comparing this construction with the Italian case, they appear to be alike. Naturally, not all properties may be checked, since the construction is possible only with quantified DPs: i) As far as combination with clitics is concerned, the preposed quantified DP may not be doubled:

92

Inversion and information structure

(45) a. Portuguese: Muito vinho, o João (*o) bebeu(*-o)! Much wine, João it drank it b. Italian: IL TUO LIBRO, (*lo) ho comprato your book, it have-I bought ii) There may only be one quantified DP in this preposing construction: (46) Portuguese: a. Muitas vezes o João bebeu vinho! Many times João drank wine b. Muito vinho o João bebeu muitas vezes. Much wine João drank many times c. *Muitas vezes, muito vinho bebeu. Many times much wine drank Italian: d. *IL LIBRO, A GIANNI, darò. the book to Gianni, (I-will-)give (46c) shows that two quantified DPs may not be preposed. This sentence is only acceptable if the first preposed element is interpreted as a topic. Under the exclamative interpretation that these preposed DPs receive (see below), the sentence is ungrammatical. iii) Preposed quantified DPs in the relevant interpretation are incompatible with wh-elements: (47) Portuguese: a. *Muito vinho, com quem é que o João bebeu? Much wine, with whom João drank b. *Com quem, muito vinho o João bebeu? With whom much wine João drank Italian: c. *A GIANNI, che cosa hai detto? to Gianni, what did you tell

No focus-movement in Portuguese

93

Again, (47a) is only possible under the topic interpretation for the DP muito vinho. With the exclamative interpretation, the sentence is ungrammatical. This is not surprising, since preposing is more natural in exclamative sentences, and a sentence may not be simultaneously interrogative and exclamative. Given the similarity of behavior with the Italian construction, when comparing in-situ with moved foci, I will use quantified DPs to be sure that we are looking at syntactically marked foci. The two types of constructions can be fruitfully compared contrasting the function of preposing and focus-in-situ respectively (see also Raposo 1994; Duarte 1987; Casielles 1996). It will be argued that for a constituent to be preposed, it has to yield given information, even in the constructions involving quantified DPs only. I have been considering throughout this chapter that the focus that appears in situ is used to introduce new information; hence it is felicitous as an answer to a wh-question: (48) A: Quem é que o Paulo viu? who did Paulo see B: O Paulo viu muita gente. Paulo saw many people Now, if the difference between preposing and focus-in-situ is just a difference in terms of the locus of application of the operation focus-movement in the derivational history of the sentence, both cases are expected to yield a felicitous answer to a wh-question. However, that is not true, as the inappropriateness of (49) attests:45 (49) A: Quem é que o Paulo viu? who did Paulo see B: #Muita gente, o Paulo viu. many people, Paulo saw Note that the facts in (48) and (49) are true for any type of preposing. If we prepose a PP, the same effects obtain: (50) A: Com quem é que o Paulo falou? With whom Paulo talked B: O Paulo falou com {a Maria/muita gente} Paulo talked with Maria / many people

94

Inversion and information structure

(51) A: Com quem é que o Paulo falou? With whom Paulo talked B: #Com {a Maria/muita gente} o Paulo falou With Maria /many people Paulo talked What is then a felicitous context for preposing? Consider the following fragments of discourse, in which preposing is felicitous. As mentioned before, preposing is possible if the preposed constituent is given in the discourse (or in the context). These are cases in which preposing is undistinguishable from topicalization. (52) and (53) exemplify such cases: (52) A: Quem é que comeu muita sopa? who ate much soup B: MUITA SOPA, ninguém comeu, (muito pão comeu o Paulo). much soup noone ate (much bread Paulo ate) (53) A: Alguém leu o teu livro? someone read your book B: O MEU LIVRO, o Paulo leu (não o teu). my book Paulo read (not yours) Another context for preposing (now for the case considered in Raposo’s work cited above) is exclamative sentences: (54) Muito vinho o Paulo bebeu! Much wine Paulo drank (55) Muita gente tu encontraste! Many people you met In these cases there is no implicit contrast with anything else. Note however that, even in an exclamative answer to a question, this word order is not licit if the quantified DP is new information (the information focus of the sentence). The relevant contrasts are given in (56) and (57): (56) A: O que é que o Paulo bebeu? What Paulo drank B: a. O Paulo bebeu muito vinho! Paulo drank much wine b. #Muito vinho o Paulo bebeu! Much wine Paulo bebeu

No focus-movement in Portuguese

95

(57) A: Quem chegou? Who arrived B: a. Chegou muita gente! Arrived many people b. #Muita gente chegou! Many people arrived A rather convincing confirmation of the exclamative force of the sentences discussed in Raposo’s work was brought to my attention by Inês Duarte: it is impossible to add a question tag to a sentence with a preposed quantified DP, which shows that the sentence may not be interpreted as declarative: (58) Muito vinho bebeu o João, (*não bebeu?) much wine drank João, not drank It is thus possible to conclude that the focus-preposing construction may not serve to introduce new information. It is possible to obtain this word order if the quantified DP may be deduced from the context. (59) A: A festa foi óptima! The party was great B: Muito vinho o Paulo bebeu! As the examples above show, in a felicitous context for preposing, the constituent that is in the left periphery of the sentence has to have been previously referred to in the discourse or be deduced from the context. Moreover, it never introduces new information. Rather, it contrasts some piece of old information with something else. If preposing would convey new information, it might be used for answering questions. As shown above, this is not true. As also mentioned in work by Inês Duarte (Duarte 1987, 1996), anytime there is preposing of a constituent in Portuguese, the information conveyed by that constituent is given (independently of whether it is further contrastive or topical). Actually, the part that answers the question can never be fronted as (60) and (61) illustrate, independently of whether the constituent conveying new information is contrasted to another one or not: (60) A: Quem é que viste ontem? who did you see yesterday B: ONTEM vi [muita gente]F, (hoje pouca) yesterday I saw many people, (today few)

96

Inversion and information structure

An answer in which the focus is fronted and the contrasted temporal adverb is left in situ is simply infelicitous: (61) A: Quem é que viste ontem? who did you see yesterday B: #[Muita gente]F vi ontem. Many people I saw yesterday This is true even if the focus of the sentence (i.e. the part that replaces the wh-phrase) is itself contrasted: (62) A: O que é que puseste na prateleira? What did you put on the shelf B: #[MUITA COISA]F pus na prateleira, [POUCA COISA]F, na cadeira Many things I put on the shelf, few things on the chair A felicitous answer for this question would involve preposing the PP: (63) B: NA PRATELEIRA, pus [muita coisa]F, NA CADEIRA [pouca coisa]F on the shelf I put many things, on the chair few things It seems thus, that in spite of the need to be contrasted, a constituent that constitutes absolutely new information can never appear in fronted position. On the other hand, topics may be contrasted. This conclusion suggests that taking the constructions of preposing as the overt counterpart of focus in situ is erroneous, and that the label focus-preposing is often used without looking at the specific properties of the language involved and at the contexts in which each construction may be used, as also pointed out by Büring (1997). For the case of Portuguese, the so-called focus-preposing construction is restricted to quantified DPs, and even in those cases, it is not an alternative to focus-in-situ, since the discourse function is different. Hence the construction involving preposing must not be taken as an argument for focusmovement.46 Summing up, it was shown in the previous sections that, for a subject to occur postverbally in its base position, it must be focused. This observation supports the view that interface conditions play a role in determining the potential surface positions of subjects. The remaining question is how the in-situ subjects are licensed. This is the topic of the next section.

Phases, locality and subjects in Spec, VP

97

4.6. Phases, locality and subjects in Spec, VP Following recent work by Chomsky (1998, 2000), it is legitimate to suppose that the Case-features of the head attracting the subject may be licensed under two mechanisms: Move or Agree. If the former is chosen, the subject is attracted to Spec,IP. If the latter option is at stake, the subject stays in its base-position, and Case-features are licensed under Agree. Note that, as proposed above, if the subject is in Spec, VP, Spec,IP is empty: (64) a. [IP S [CASE] I[CASE] [VP t t b. [IP I[CASE] [VP S [CASE] t At first sight, this gives rise to optionality: there are two converging derivations for licensing subjects. Assuming with Adger (1994), that the choice between two converging outputs is made post-syntactically, taking into accout the information structure of the clause, this is a desirable result. Given a pair of converging outputs SVO and VSO, the latter will only be picked if the subject is focused.47 If Agree is the relevant mechanism for licensing in-situ subjects, it is predicted that locality effects should arise.48 This prediction is borne out. As the following data show, an inverted subject may occur within a non-finite clause: (65) a. Decidiram ler todos os alunos esse livro. Decided-3pl read all the students that book b. Querem ler todos os alunos esse livro. Want-3pl read all the students that book The surprising aspect of the examples in (65) is that the empty subject of the matrix verb is correferential with the subject of the non-finite verb, and yet no principle-C effects arise. This follows from the proposal that Spec,IP is empty in inversion contexts.49 It must be noted that this type of in-situ subject within an infinitival clause is restricted: it only happens if the matrix verbs do not select a CP. Let us consider clitic-climbing, as in (66), or weakadverb climbing50, as in (67), as a signal that the complement of the matrix verb is defective. The contrasts in the examples below between finite and non-finite complements show that these two phenomena are dependent on the absence of a CP-node (cf. Gonçalves 1999, among many others):

98

Inversion and information structure

(66) a. Eu só o quero ler amanhã. I only it-acc want read tomorrow “I only want to read it tomorrow” b. *Eu só o quero que leias amanhã. I only it-acc want that (you) read tomorrow “I only want that you read it tomorrow” c. Eu só quero que o leias amanhã. I only want that (you) it- acc read tomorrow “I only want that you read it tomorrow” (67) a. Eu só lá decidi ir ontem. I only there decided go yesterday “I only decided to go there yesterday.” b. *Eu só lá decidi que ia ontem. I only there decided that (I) would-go yesterday “I only decided that I would go there yesterday” c. Eu só decidi que ia lá ontem. I only decided that (I) would-go there yesterday “I only decided that I would go there yesterday” Accordingly, if there is no evidence for the absence of CP, either because the matrix verb is not a restructuring verb, or if an embedded negation is inserted blocking restructuring, or because there is no independent evidence for any type of transparency relation between the matrix verb and the embedded complement (independently of restructuring), the sentences become ungrammatical, contrasting with their non-inverted counterparts: (68) Embedded negation: a. ??Decidiram não ler todos os alunos esse livro. Decided-3pl read all the students that book b. Todos os alunos decidiram não ler esse livro. All the students decided not to read that book (69) Non-restructuring verbs (projecting CP): a. *Recusaram ler todos os alunos esse livro. Refused-3pl read all the students that book

Phases, locality and subjects in Spec, VP

99

b. Todos os alunos recusaram ler esse livro. All the students refused to read that book c. *Negaram ler todos os alunos esse livro. Denied to read all the students that book d. Todos alunos negaram ler esse livro. All the students denied to read that book These data provide the necessary evidence for assuming that Agree is at stake. Assuming with Chomsky (1998, 2000) that the scope of Agree is the strong phase, and accepting the standard view that not all non-finite complements project CP (Boskovic 1997), these data follow. For instances, in the restructuring contexts, CP is not projected, there is no strong phase boundary, and Agree may operate. A consequence of the present proposal is that in VSO contexts, there is no strong phase boundary intervening between T and Spec,VP. In other words, it must be assumed that either vP does not exist in European Portuguese, or it is a weak phase. In Costa (1998), following claims on the nonuniversality of functional categories (e.g. Bobaljik and Thrainssón 1996), it is suggested that AgrO does not project in languages in which scrambled objects behave like adjuncts. If v is taken as the category providing the landing site for objects in Object-shift languages, the assumption that AgrO does not project may be adapted to v. If this is the case, there is indeed no strong phase boundary between T and Spec,VP, and Agree may take place. Alternatively, it is legitimate to assume that the subject is generated in Spec,vP, and the results achieved for inverted subjects may be translated, assuming that inverted subjects are stranded in the specifier of VP. In that case, the relevant configurations are: (70) a. [IP S [CASE] I[CASE] [vP t t b. [IP I[CASE] [vP S [CASE] t Since the subject is at the edge of the strong phase (vP), Agree can probe this position and the licensing of the relevant features takes place.

100

Inversion and information structure

4.7. Related evidence: agreement in copular constructions Before concluding, let us consider a set of data involving rightward agreement, which may provide additional evidence for the proposal made above that subjects in-situ are licensed under Agree. The data to be considered comes from inverted copular constructions, in which the verb agrees to the right. We will add data from Brazilian Portuguese, since this language differs from European Portuguese in two interesting ways: first, it does not have subject-verb inversion of the type discussed above (cf. Figueiredo Silva 1996, among others);51 second, it does not have clitic climbing (cf. Duarte and Gonçalves 2000, among others). These two differences and the agreement facts will permit checking whether there is any correlation between locality, agreement and the possibility for subjects to occur in a postverbal position. Let us then consider the facts. Both in Brazilian and in European Portuguese, in specificational sentences, the verb agrees to the right, if there is a pronominal form on the right, as shown in (71): (71) O assassino sou eu. The murderer am I If the pronominal form is to the left of the verb, the verb agrees to the left, as in (72): (72) Eu sou o assassino. I am the murderer The interesting difference between the two languages emerges when there is a modal verb involved. If the specificational sentence occurs with a modal verb, the same agreement pattern is reproduced in European Portuguese. The modal verb agrees with the pronominal form to its right. European Portuguese: (73) a. O assassino devo ser eu. the murderer must-1sg be I b. O assassino posso ser eu. the murderer may-1sg be I In Brazilian Portuguese, however, the agreement pattern with modal verbs is different. If the specificational sentence contains a modal, the modal verb agrees with the preverbal element:

Related evidence: agreement in copular constructions

101

Brazilian Portuguese: (74) a. O assassino deve ser eu. the murderer must-3sg be I b. O assassino pode ser eu. the murderer may-3sg be I Note that, in (74), Brazilian Portuguese is behaving like English in normal specificational sentences in not allowing agreement to the right: (75) a. The murderer is me. b. *The murderer am I. These agreement patterns raise the following questions: i) What is the difference between Brazilian and European Portuguese modal verbs underlying the differences between the agreement pattern in the two languages? ii) What underlies the difference between Brazilian Portuguese and English simple cases, making agreement to the right possible only in the former language? We assume that the agreement pattern found in specificational sentences is the result of there being inversion (cf. Tavares (in preparation)). Let us further assume that the subject does not reach Spec,IP in this type of inversion, as illustrated in (76), and remain agnostic as far as the target of the fronted predicate is:52 (76) [XP O assassino [I’ sou [vP eu In this sense, the structure in (76) is similar to other cases of predicate inversion, as in (77): (77) Inteligente sou eu. clever am I The assumption that the case in which there is agreement to the right is a (special) case of subject-verb inversion implies assuming an analysis similar

102

Inversion and information structure

to the one proposed above for other cases of subject-verb inversion. In other words, extending the analysis proposed in the previous section to these structures, one can assume that the inverted subject eu in (76) is licensed under Agree. This assumption straightforwardly explains why there is agreement to the right in Portuguese, German or Icelandic, but not in English (Tony Kroch, p.c.). Only in the latter is there no subject-verb inversion, which blocks the possibility of tracing rightward agreement as an effect of subject-verb inversion. In our terms, English does not allow for any subject to be licensed under Agree, which has reflexes on the lack of agreement to the right. The option of licensing subjects to the right under Agree is restricted in Brazilian Portuguese to some verb classes. Let us now come back to the difference between the two varieties of Portuguese. As illustrated above, in European Portuguese, a modal verb agrees to the right with the subject embedded in the infinitival construction, which is not true for Brazilian Portuguese. These agreement facts with modals may be shown to follow from locality and the presence or absence of CP in a way similar to the one outlined in the previous section. As shown in (78), modal verbs in European Portuguese are restructuring verbs (Gonçalves 1999), allowing clitic climbing: (78)

Eu não lhe devo dar o livro. I not him must give the book “I must not give him the book”

Unlike in European Portuguese, restructuring is not productive in Brazilian Portuguese. A well known fact about the latter is that it does not permit clitic climbing (cf. Duarte and Gonçalves 2000): (79)

Brazilian Portuguese:

a. Eu não devo lhe dar o livro. I not must him give the book b. *Eu não lhe devo dar o livro. I not him must give the book This difference in the behavior of modal verbs in the two languages gives us the ingredients for understanding the agreement patterns. If in (73) there is restructuring, the modal and the main verb work as a single phase domain,

Related evidence: agreement in copular constructions

103

since CP is not projected. Because there is no phase-boundary between the matrix Infl and the subject, the latter is licensed under Agree. The agreement pattern is therefore expected: we find no difference between the construction in (73) and the constructions in which there is subject-verb inversion within a non-finite clause. Now, let us consider the case of Brazilian Portuguese. If Brazilian Portuguese modal verbs do not restructure, as shown in (75), a locality constraint makes it impossible to derive the configuration with agreement to the right. The reason for a non-restructuring context to block agreement to the right follows from the proposal made in the preceding section. CP is projected, inducing a phase-boundary, and blocking the Agree relation between the matrix Infl and the embedded subject.53 An obvious difference in terms of results must be pointed out. A failure to establish Agree induced ungrammaticality in European Portuguese, while it induces a failure in morphological agreement in Brazilian Portuguese. I do not have a solution to this problem. It may be the case that the embedded Spec,TP plays a role in licensing the embedded subject. Negation is assumed to be associated to TP in Romance languages (Zanuttini 1996). It may therefore be taken as a diagnostic to know whether Spec,TP is projected. As shown in (80), for European Portuguese, the presence of an embedded negation blocks restructuring, since it makes clitic climbing impossible. The data in (80a,b) show that there is no problem with having a negation embedded under the modal. The sentences in (80c,d) show that clitics can not climb in the presence of the embedded negation: (80) a. Eu não devo ver o Pedro I not must see Pedro b. Eu devo não ver o Pedro. I must not see Pedro c. Eu devo não o ver. I must not him see d. *Eu devo-o não ver. I must him not see Combining these facts with the analysis of agreement to the right, it is predicted that the latter should not occur in Brazilian Portuguese. In this language, there is no restructuring, thus, moving the predicate to Spec,IP above the modal will violate locality, since there is another Spec,IP closer to it.

104

Inversion and information structure

This is, however, not a satisfactory answer to the problem, since it predicts that the embedded subject should be licensed by the embedded Spec,TP. I will therefore leave the issue of how the subject in Brazilian Portuguese is licensed for further research. Interestingly, evidence for this analysis may be found internally to European Portuguese. If a verb forces the projection of CP, agreement to the right is never possible. This is the case with the modal verb necessitar ‘to need’. The contrast in (81) shows that this verb does not allow clitic climbing; the data in (82) shows the agreement pattern in the specificational context: (81) a. Eu só necessito de lhe dar-lhe um livro. I just need to him give a book b. *Eu só lhe necessito de dar um livro. I just him need to give a book “I just need to give him a book”. (82) a. *O problema necessito de ser eu. The problem need-1sg to be I b. ?O problema necessita de ser eu. The problem needs to be I “I must be the problem”. The sentence in (82b) patterns like the Brazilian Portuguese cases discussed above. This confirms that the projection of CP is the crucial factor to know whether an embedded subject can be licensed to the right, under Agree. Incidentally, note that (82b) is not ungrammatical, but just marginal, which allows us to get back to the issue raised above for Brazilian Portuguese: how is the embedded subject licensed in sentences like (82b)? The fact that the question may be raised for European Portuguese as well, and the contrast between (82b) and the sentences discussed in the previous section in which the subject occurs after a non-finite verb embedded under a control predicate allows for conjecturing that the problem has to do with the presence of the copula rather than with the failure to Agree. In other words, the presence of the copula be seems to enable an additional last resort way of licensing the inverted subject. Obviously, this must be a last resort strategy, since it would otherwise rescue all ungrammatical cases discussed above. Summing up this section, although some issues remain open, its purpose was to show that there is independent evidence for the claim that the presence

Conclusions

105

of a CP-node induces a boundary for licensing an inverted subject under Agree. The contrast between European and Brazilian Portuguese and its relation with the availability of restructuring with modal verbs in the two languages provides the necessary evidence for relating inversion with the amount of projected structure. The special feature of the data discussed in this section is that we see a correlation between morphological agreement and the cases in which the inverted subject is licensed by the matrix Infl. In the cases discussed, failure to Agree corresponds to a failure to morphologically agree.

4.8. Conclusions This chapter provided evidence for the possibility of subjects to surface in Spec,VP. It was defended that this occurs when subjects are focused, in compliance with requirements imposed by the mapping with prosody. It was further shown that the possibility for subjects to emerge in Spec,VP is primarily conditioned by syntax: it reflects the possibility of checking Casefeatures under Move or Agree.

5. Optionality and left-dislocated subjects: semantic and discourse properties

In chapters 2, 3 and 4, it was observed that there is a correlation between the distribution of information focus and the placement of subjects in European Portuguese. In particular, the following two generalizations arose: (i) information focus surfaces at the clause’s rightmost position; (ii) for each context, there is only one word order. The discussion in chapter 2 led to the conclusion that preverbal subjects in European Portuguese occupy an A-position. Yet, as mentioned, nothing precludes left-dislocation of a subject, in the same way nothing precludes left-dislocation of an object. However, nothing was said concerning contexts favoring subject left-dislocation. This chapter examines two contexts challenging the generalization that for each context there is only one order: unaccusative contexts and answers to multiple wh-questions. The latter will turn out to provide a context in which subjects may be left-dislocated. 5.1. SV and VS in unaccusative contexts As mentioned in chapter 3, one of the criteria used for determining the unmarked word order of a language is to look at contexts of sentence-focus, which may be tested in answers to what happened?. In this context, the only felicitous word order in Portuguese is SVO. This is illustrated in (1) for transitive verbs. All other word orders are unfelicitous in this context: (1)

What happened? a. O Paulo comeu o bolo. Paulo ate the cake b. #Comeu o Paulo o bolo. c. #Comeu o bolo o Paulo. d. #O Paulo o bolo comeu e. #O bolo o Paulo comeu. f. #O bolo comeu o Paulo.

108

Optionality and left-dislocated subjects: semantic and discourse properties

If the verb is intransitive, the only felicitous answer to what happened? is the order SV. This is illustrated in (2): (2)

What happened? a. O Paulo cantou. Paulo sang b. #Cantou o Paulo

The first problem raised by the behavior of subjects of unaccusatives has to do with their behavior in sentence-focus contexts. The analysis presented in the preceding chapters makes the prediction that subjects move to Spec,IP in unmarked contexts independently of the verb they appear with. This prediction was confirmed for transitives and intransitives. The same should be true for unaccusatives. The complement of an unaccusative verb should always raise to Spec,IP to receive nominative Case. However, in sentence-focus contexts, movement of the subject to Spec,IP seems to be optional, as illustrated in (3) (3)

What happened? a. O Paulo chegou. Paulo arrived a.’ Chegou o Paulo. arrived Paulo b. A fábrica ardeu. the factory burnt b.’ Ardeu a fábrica.

This behavior raises the following question: a)

Why is inversion a felicitous word order in unmarked contexts with unaccusatives only?

In other words, what is special about unaccusatives that enables their subjects to remain low, differently from what happens with intransitives and transitives? b) Why is inversion optional with unaccusatives?

SV and VS in unaccusative contexts

109

In transitive and intransitive contexts, there is a one-to-one correspondence between word orders and discourse contexts. In this sense, it may be argued that there is no true optionality in the language, since each word order serves different discourse functions. In this case, however, there seems to be true optionality, since two different word orders emerge for one single discourse context. The answer to this question must involve showing that the optionality is just apparent, or providing an analysis predicting two optional outputs. A finer look at the data will reveal that the former approach is right.

5.1.1. Lack of agreement in postverbal position Before examining the context favoring the alternation SV-VS, let us now turn to an apparently unrelated problem: the patterns of verbal agreement with pre- and postverbal subjects. As the following examples show, verbal agreement does not vary with word order alternations. This is true for transitives (4) and for intransitives (5). A plural subject triggers plural agreement in preverbal position and in postverbal position. All sentences in which the default 3rd person singular agreement appears are ungrammatical: (4)

SVO: a. Os meninos comeram o bolo. the kids ate-3pl the cake a.’ *Os meninos comeu o bolo. the kids ate-3sg the cake VSO: b. Comeram os meninos o bolo. ate-3pl the kids the cake b.’ *Comeu os meninos o bolo. ate-3sg the kids the cake VOS: c. Comeram o bolo os meninos. ate-3pl the cake the kids c.’ *Comeu o bolo os meninos. ate-3sg the cake the kids

110

Optionality and left-dislocated subjects: semantic and discourse properties

(5)

SV: a. Os meninos cantaram. the kids sang-3pl a.’ *Os meninos cantou. the kids sang-3sg VS: b. Cantaram os meninos. sang-3pl the kids b.’ *Cantou os meninos. sang-3sg the kids

At least for these verb classes, it seems to be possible to generalize that, independently of the word order, subject and verb obligatorily agree in person and number.54 A problem for this generalization appears when agreement with postverbal subjects of unaccusative verbs is considered.55 In colloquial speech, it is possible for plural subjects of unaccusative verbs to trigger 3rd person singular verbal agreement. As illustrated in (6), lack of agreement is optional. It is also possible for the verb to fully agree with the postverbal subject: (6)

a. Chegaram o Pedro e o Paulo. arrived-3pl Pedro and Paulo a.’ Chegou o Pedro e o Paulo. arrived-3sg o Pedro e o Paulo b. Fecharam muitas fábricas. closed-3pl many factories b.’ Fechou muitas fábricas. closed-3sg many factories c. Chegaram as cadeiras. arrived-3pl the chairs c.’ Chegou as cadeiras. arrived-3sg the chairs

Note that this is not restricted to coordinated subjects, as (6b) illustrates, or to weak quantifiers, as (6c) illustrates. Even a definite DP may occur in this

SV and VS in unaccusative contexts

111

construction. Partial agreement with coordinated subjects is not restricted to a single verb class, as discussed in Colaço (1998): (7)

a. Brincou o Paulo e o Pedro. played Paulo and Pedro

intransitive

b. Chegou o Paulo e o Pedro. arrived Paulo and Pedro

unaccusative

c. Comeu-o o Paulo e o Pedro. transitive ate it Paulo and Pedro The possibility of obtaining partial agreement is restricted to the postverbal position. In preverbal position, unaccusative and intransitive verbs behave alike in disallowing partial agreement between the subject and the verb: (8)

Intransitive verbs: a. O Pedro e o Paulo cantaram. Pedro and Paulo sang-3pl a.’ *O Pedro e o Paulo cantou. Pedro and Paulo sang-3sg b. Muitos meninos brincaram. many children played-3pl b.’ *Muitos meninos brincou. many children played-3sg

(9)

Unaccusative verbs: a. O Pedro e o Paulo chegaram. Pedro and Paulo arrived-3pl a.’ *O Pedro e o Paulo chegou. Pedro and Paulo arrived-3sg b. Muitas fábricas fecharam. many factories closed-3pl b.’ *Muitas fábricas fechou. many factories closed-3sg

112

Optionality and left-dislocated subjects: semantic and discourse properties

This behavior of subjects of unaccusative verbs with respect to agreement raises the following questions: a) What is the relation between the postverbal position and the lack of agreement? In other words, why is lack of agreement possible only in postverbal position? What makes agreement obligatory when the subject is preverbal? b) Why is this option restricted to unaccusatives? Like the inverted word order is only felicitous in sentence-focus contexts with unaccusative verbs, here, it may be observed that the lack of agreement is only possible with unaccusative verbs. An explanation for this restriction to one verb class is needed. c) What is the relation between the unmarkedness of inversion and the agreement patterns discussed? Finally, one would like to know whether there is any relation between the two problems discussed. Arguably, it is not a coincidence that the only verbs that allow for a felicitous inverted word order are the only ones that allow for lack of agreement. Ideally, one single explanation for the two phenomena described could be found.

5.1.2. Hypothesis and arguments Let us make the following assumptions: a) Subject-verb agreement may be taken as a diagnostic to detect whether nominative Case is assigned to the subject;56 This assumption is a common one. For the specific case of European Portuguese, it is particularly straightforward to make this assumption, since agreement is enough to license nominative case even in non-finite contexts (cf. Raposo 1987): (10) O Paulo pensa irem eles à praia. Paulo thinks go-3pl they-Nom to the beach

SV and VS in unaccusative contexts

113

b) Postverbal subjects of unaccusative verbs may be assigned some other case. In analyses of unaccusative verbs, such as Belletti’s (1988), it is proposed that postverbal subjects of unaccusative verbs are assigned partitive Case. Such proposals are crucial for making it possible to postulate that nominative Case is not the only way to license the argument of the unaccusative verb. c) Language-internal variation may be due to specific properties of lexical items. In other words, I will assume that the difference between the construction with or without agreement has to do with the lexical properties of unaccusative verbs. Accepting these assumptions, let us now formulate the hypothesis for explaining the behavior of unaccusative verbs, which will partially follow Belletti’s (1988) analysis of unaccusative verbs in Italian: (11) Arguments of unaccusative verbs in colloquial European Portuguese are not obligatorily assigned nominative Case. This hypothesis basically argues in favor of an ambiguous status for unaccusative verbs. Either their complements must receive nominative Case, hence must move to the preverbal position Spec,IP, or they may be assigned partitive Case in situ, as in Belletti (1988). In the latter case, there is no need for them to move, and Spec,IP is occupied by an expletive, which brings the unaccusative inversion close to locative inversion constructions. In Costa and Figueiredo Silva (2003), evidence is presented in favour of this analysis. In Coelho et alii (2001) and Costa et alii (2002), it is argued that inversion with unaccusative verbs in Brazilian Portuguese is an instance of locative inversion. This proposal is based on the following observations. First, there is no true optionality. It is not the case that in sentence-focus contexts, SV and VS alternate freely, as shown in the examples in (12)-(14): (12) O que aconteceu? What happened a. Caiu um avião. Fell an airplane b. ??Um avião caiu. An airplane fell

114

Optionality and left-dislocated subjects: semantic and discourse properties

(13) O que aconteceu? What happened a. Nasceram 93 bebés. Were born 93 babies b. ??93 bebés nasceram. 93 babies were born (14) O que aconteceu? What happened a. Um frigorífico descongelou. A freezer unfroze b. ??Descongelou um frigorífico. In fact, both definiteness effects and subtypes of unaccusative predicates seem to favour or disfavour the VS order in this context. The second argument for analyzing VS as locative inversion comes from the comparison with English. Just like in English, VS in Brazilian Portuguese does not provide information-focus on the subject. The same typically holds for English locative inversion and expletive constructions: (15) Brazilian Portuguese: Quem morreu? Who died a. #Morreu o PM. Died the PM. b. O PM morreu. The PM died. (16) English: Who comes? a. A man comes. b. #There comes a man. Based on these two observations, Costa and Figueiredo Silva (2003) hypothesize that unaccusative VS sentences are locative inversions in Brazilian

SV and VS in unaccusative contexts

115

Portuguese (cf. Pinto 1997; Ambar 1998; Cornish 2002 for other languages). Assuming with Levin and Rappaport Hovav (1995) that unaccusatives entering locative inversion constructions are associated to (potentially unrealized) locative or temporal argument (cf. Pustejovsky (1995)), and that expletive pro may correspond to the temporal / locative argument (Pinto 1997), the Brazilian Portuguese facts follow straightforwardly. Since there is expletive pro in Brazilian Portuguese, we expect this language to instantiate VS only in the context in which locative inversion is adequate: the context of presentational focus in which locative inversions are uttered. This analysis makes some further predictions. First, it is expected that not all unaccusatives invert, since according to Levin and Rappaport Hovav (1995), some unaccusative predicates are not associated to locative or temporal arguments. This would be the case for the predicate descongelar/unfreeze, as illustrated in (17). Second, it is predicted that a few verbs other than unaccusatives, selecting the same kind of argument, invert in Brazilian Portuguese. This would be the case for the verb telefonar (to call), as also argued in Pinto (1997), and Cornish (2002). (17) O que é que aconteceu? What happened Telefonou a Maria. Called Maria Finally, since information focus and presentational focus are not identical, it is correctly predicted that they may be linked to different types of syntactic constructions. In fact, English provides independent evidence for distinguishing the two types of VS order in Portuguese. This hypothesis straightforwardly explains the agreement patterns described. When the subject is preverbal, agreement is obligatory, as in (18). This is because the subject is only moved to this position when it is looking for nominative Case, and when expletive pro is not in Spec,IP: (18) a

Muitas fábricas fecharam many factories closed-3pl

b. *Muitas fábricas fechou. many factories closed-3sg

116

Optionality and left-dislocated subjects: semantic and discourse properties

In postverbal position, both full and partial agreement may be found, as in (19): (19) a. Fecharam muitas fábricas. closed-3pl many factories b. Fechou muitas fábricas. closed-3sg many factories The two options are predicted under (11). (19a) corresponds to the construction in which the argument needs nominative Case, although it has not moved to Spec,IP. In this sense, it is just like all other cases of subject-verb inversion, independently of the verb class. (19b) corresponds to the locative inversion construction in which the argument of the unaccusative verb is assigned partitive Case. In that case, the verb does not agree with the subject. Let us consider three arguments that favor the hypothesis formulated in (11), and the analysis just developed. Argument A: The first argument in favor of the hypothesis formulated above comes from the interaction between the patterns of verbal agreement and the distribution of nominative pronouns. Pronouns are the only forms of the language that are morphologically specified for Case. As expected, in preverbal position, nominative pronouns obligatorily agree with the verb: (20) a. Eles chegaram. they-NOM arrived-3pl b. *Eles chegou. they-NOM arrived-3sg Interestingly, in postverbal position, full agreement is also obligatory, as illustrated in (21): (21) a. Chegaram eles. arrived-3pl they-NOM b. *Chegou eles. arrived-3sg they-NOM

SV and VS in unaccusative contexts

117

This result is expected under the hypothesis in (11). Since the pronoun is morphologically specified for nominative Case, it must occur in the context in which it is assigned case. It is therefore impossible to obtain the nominative pronoun with partial agreement, since in that Case there would be no matching between the morphological Case and the structural Case assigned (partitive). Since the pronominal paradigm of European Portuguese does not include partitive subject pronouns, a nominative pronoun may not co-occur with a non-agreeing verb. Argument B: According to Raposo and Uriagereka (1990), in dialectal European Portuguese, it is possible to find overt expletives in constructions involving postverbal subjects. Crucially, this possibility is restricted to unaccusative constructions. It is thus possible to find an overt expletive cooccurring with a postverbal subject in a sentence like (22): (22) Ele morreu muitas pessoas naquele acidente. he died-3sg many people in that accident It is however not possible to obtain overt expletives in intransitive (23) or transitive (24) contexts, independently of the agreement exhibited by the verb: (23) a. *Ele cantaram os meninos. he sang-3pl the kids b. *Ele cantou os meninos he sang-3sg the kids (24) a. *Ele comeram os meninos o bolo. he ate-3pl the kids the cake b. *Ele comeu os meninos o bolo he ate-3sg the kids the cake This behavior is expected under the hypothesis formulated above. Since the Spec,IP position is not necessarily involved in nominative Case licensing with unaccusatives, this position may host an expletive with nominative Case morphology only in unaccusative contexts. With other verbs, nominative Case associated with this position must be assigned to the postverbal subject.

118

Optionality and left-dislocated subjects: semantic and discourse properties

Before closing this section, let me point out that this behavior of unaccusatives provides evidence in favor of the traditional analysis of preverbal subjects in Portuguese (Duarte 1987; Ambar 1992). According to these analyses, preverbal subjects are in Spec,IP. As mentioned before, this hypothesis is challenged in Barbosa’s (1995,1996) work, who suggests that preverbal subjects in null subject languages are instances of left-dislocation. As noted above, preverbal subjects of unaccusative verbs agree obligatorily with the verb. This may be explained if preverbal subjects are in Spec,IP. In this position, they are assigned nominative Case, which is reflected by verbal agreement. If preverbal subjects were left-dislocated, obligatory agreement might not be found. Note that left-dislocation of arguments does not trigger changes in agreement: (25) a. Comeu os bolos. ate-3sg the cakes b. Os bolos, comeu-os. the cakes ate-3sg them Since, as we have seen, postverbal subjects of unaccusatives do not obligatorily agree with the verb, it could be predicted that they might be left-dislocated, as any other subject, according to Barbosa’s proposal. In that case, no change in agreement would be predicted. However, such prediction is not confirmed by the data.

5.1.3. Summing up: is there optionality? The analysis proposed above, based on the proposal of Costa (2001), basically denies the existence of true optionality. The data examined in Coelho et alii (2001, 2002) and in Costa and Figueiredo Silva (2003) suggests that there is no true optionality in the behavior of unaccusative contexts. Taking the fact that under certain circumstances, as shown above, either VS or SV are preferred, and given the comparison with Brazilian Portuguese, Coelho et alii (2001, 2002) suggest that the SV-VS alternation is only apparent, and that the cases of VS are instances of identificational focus. Assuming the analysis of identificational focus put forward in Cornish (2002), the authors claim that the inverted structures correspond to instances of locative inversion, in which Spec,IP is occupied by an expletive:

VSO and SVO in answers to multiple wh-questions

119

(26) [IP pro [ V [VP t DP ]]] As mentioned above, this analysis derives four independent facts: (i) this word order is restricted to unaccusatives, since this verb class favors locative inversion (Levin and Rappoport-Hovav 1995); (ii) agreement may be partial if the inflectional head enters a Spec,head relation with the expletive; (iii) this construction is possible in Brazilian Portuguese, a language in which Spec,IP cannot be empty and only expletive pro is available; (iv) it is expected that not all unaccusative verbs behave alike, since lexical restrictions determine which types of unaccusatives may enter locative inversion constructions. It is crucial for the purposes of this chapter to note that the optionality found in unaccusatives is just apparent, and that a closer look at morphological and semantic properties of the inverted construction was important to determine the differences between SV and VS.

5.2. VSO and SVO in answers to multiple wh-questions Another case of apparent optionality may be found in answers to multiple whquestions. As shown in (27) and (28) both VSO and SVO constitute legitimate answers to a multiple wh-question: (27) Quem leu o quê? who read what (28) a. Leu o João o livro. read João the book b. O João leu o livro. João read the book These data, like the data involving unaccusative verbs, challenges the generalization that, for a single context, there is only one word order available. Moreover, the data exemplified in (28) is problematic for the generalization made in the previous chapter that all focus material is rightmost, since in the answer in (28b) we observe that the focused subject surfaces to the left of the verb, which is not focused. In Costa (2002a), I argued that this optionality is just apparent. As it will be shown, the semantic properties of the two answers are different. Interest-

120

Optionality and left-dislocated subjects: semantic and discourse properties

ingly, we will conclude that the semantic aspects relevant for making the SVO answer legitimate favor the left-dislocation of the subject.

5.2.1. Semantic properties of multiple-wh questions According to Hornstein (1994) and Chierchia (1991), multiple wh-questions have the following properties: (29) A: They can be answered with a pair-list; B: The subject is D-linked (Pesetsky 1987); C: The answer must be exhaustive. Let us then check whether these properties differentiate the two word orders. It may be observed that property (30A) does not distinguish the two word orders: (30) A: Quem leu o quê? who read what B: Leu o João o livro, leu a Maria o jornal e leu o Pedro a revista. read João the book, read Maria the newspaper and read Pedro the magazine (31) A: Quem leu o quê? who read what B: O João leu o livro, a Maria leu o jornal e o Pedro leu a revista. João read the book, Maria read the newspaper and Pedro read the magazine Both word orders may be used with lists of pairs. The other two properties (exhaustivity and D-linking) differentiate the two types of answers. In this section, arguments are presented in favor of the following claims: A. SVO is not exhaustive. When a multiple wh-question is answered with SVO, it is not necessarily a complete answer.

VSO and SVO in answers to multiple wh-questions

121

B. In VSO, the answer and the pair Subject-object are exhaustive. When a multiple wh-question is answered with VSO, there are no other possible answers, and there is a uniqueness relation between the subject and the object. The following arguments provide evidence in favor of the two claims formulated above:

A – Continuation without assertion: In spite of the fact that focused elements are rightmost, there is a type of answer described in Ambar (1998), in which a focused subject appears preverbally, as in (32): (32) A: Quem comeu o bolo? who ate the cake B: O João...comeu. João...ate This type of answer is possible with an intonational break between the subject and the verb, and it means that the speaker does not provide an exhaustive answer. In the case of (32), there may be other persons who ate the cake, but the speaker is not sure about it. Taking this type of non-assertive answer into consideration, it may be tested whether something like this is behind the difference between VSO and SVO in answers to multiple wh-questions. A way of testing this type of answer is resorting to a continuation without assertion. As (33) and (34) illustrate, this type of continuation is possible with SVO but not with VSO: (33) A: Quem leu o quê? who read what B: O João leu o jornal e a Maria leu a revista…os outros não sei o que leram. João read the newspaper and Maria read the magazine…the others, I don’t know what they read (34) A: Quem leu o quê? who read what

122

Optionality and left-dislocated subjects: semantic and discourse properties

B: Leu o João o jornal e leu a Maria a revista...(*os outros não sei o que leram.) Read João the newspaper and read Maria the magazine...the others, I don’t know what they read

B – VP-ellipsis: A similar argument is provided VP-ellipsis. This will only be possible if the preceding sentence is SVO: (35) A: Quem comeu o quê? who ate what B: O João comeu a sopa, e a Maria também comeu. João ate the soup and Maria also ate (36) A: Quem comeu o quê? who ate what B: Comeu o João a sopa (*e a Maria também comeu) ate João the soup and Maria also ate This behavior is expected, considering the proposal formulated above. Since a VSO answer is exhaustive, there cannot be a copy of the VP stating that there is a relation between the same object and a different subject.

C – N-words: Like other arguments, argumental N-words may surface either in preverbal or postverbal position. This option is conditioned by their discourse function. If they are focused, they are postverbal. If they are given, they are preverbal: (37) Quem chegou? who arrived? a. Não chegou ninguém. not arrive no-one b. *Ninguém chegou. no-one arrived.

VSO and SVO in answers to multiple wh-questions

123

(38) O que é que ninguém fez? what did no-one do a. Ninguém chegou. no-one arrived b. *Não chegou ninguém. not arrived no-one According to this behavior, it would be expected that subject N-words would pattern like other subject DPs in answers to multiple wh-questions. At first sight, this appears to be the case: (39) A: Quem comeu o quê? who ate what B: Ninguém comeu a sopa. Noone ate the soup (40) A: Quem comeu o quê? who ate what B: Não comeu ninguém a sopa. not ate noone the soup It is asserted that noone ate the soup, but the hypothesis that someone else ate something else is not excluded. Thus, the two word orders are possible. A different pattern emerges when both the subject and the object are N-words. In this case, only SVO is possible: (41) A: Quem comeu o quê? who ate what B: Ninguém comeu nada. noone ate anything (42) A: Quem comeu o quê? who ate what B: *Não comeu ninguém nada. not ate noone anything

124

Optionality and left-dislocated subjects: semantic and discourse properties

The answer in this case is necessarily exhaustive, since no subjects or objects are left out of the answer. So far, exhaustivity has been linked to VSO word orders, which might lead to the expectation that VSO should be the answer in this case. Note, however, that there is only one possible answer in this context, and that there is a difference in terms of syntactic markedness between SVO and VSO. Only the former involves movement of the subject to Spec,IP, subjects in VSO being licensed under Agree. Since there are not two options available, there is no reason to use the most marked option. D – VSO is preferred in correction contexts. Another context in which VSO can be used is in correction contexts, as in (43): (43) A: Ninguém comeu nada. noone ate anything B: Comeu o João a sopa. ate João the soup In this case, the SVO option is dispreferred. At first sight, there should be no difference between the context illustrated in (43) and the multiple whquestion, since in both cases the focus of the sentence will be constituted by the subject and the object. There is however a crucial difference between the two contexts. In the case of correction, a contrast is being established. In the case of answer to a wh-question, the focus is merely informational. A well-known property of contrastive focus is that it involves exhaustivity and uniqueness (Szabolcsi 1981, Kiss 1996). Since, as mentioned above, VSO involves exhaustivity, this word order is preferred in contrastive contexts. E – Generic contexts: The final piece of argumentation in favor of a semantic difference between SVO and VSO answers comes from the emerging pattern in generic contexts, as in (44): (44) Quem come o quê? who eats what A: As baleias comem peixes. whales eat fish B: ??*Comem as baleias peixes. eat whales fish

VSO and SVO in answers to multiple wh-questions

125

As the example above illustrated, in generic contexts, only SVO is available. A VSO answer is not possible. This follows straightforwardly from the properties of generic contexts and from the properties claimed to be associated with SVO answers to multiple-wh questions. Genericity implies non-exhaustivity (see Krifka et alii 1995). As the continuation test and the VP-ellipsis test show, SVO answers are non-exhaustive, hence it is expected that the word order involving exhaustivity may not be used in generic contexts.

5.2.2. Syntactic consequences This study of word order in the context of answer to multiple wh-questions has some theoretical consequences for the syntactic analysis of the SVO-VSO alternation: a) The generalizations concerning apparent optionality and the distribution of focus may be maintained. First, the two generalizations concerning the non-existence of optionality and the distribution of focused constituents may be maintained. SVO and VSO are not free variants. Instead, they correspond to different semantic options. Second, the generalization that new information is rightmost is not disconfirmed by a focused subject appearing in the leftmost position, since these subjects are always D-linked, which means that they share some properties with topics. b) Additional argument against FocP for contrastive focus: It is traditionally considered that FocP is at the left periphery of the clause, taken into consideration what happens in languages like Hungarian. Above, some arguments were presented against the idea that foci occupy a discourserelated functional category. Yet, one might suppose that a left-peripheral Focus Phrase would be reserved for contrastive focus. However, the context discussed in this paper shows that a contrastive and information focus occupies a position to the right (VSO), not emerging in a left-peripheral position. c) Left-dislocated subjects: The context discussed in this section makes it possible for a subject to be leftdislocated. As discussed in the previous chapter, while preverbal subjects in

126

Optionality and left-dislocated subjects: semantic and discourse properties

sentence-focus contexts exhibit properties that lead to the conclusion that they are in Spec,IP, preverbal subjects in answers to multiple wh-questions provide evidence in favor of the analysis claiming that they are left-dislocated. This reinforces the conclusion stating that an analysis of preverbal subjects in Spec,IP does not obviate their left-dislocation. The observation made above was that when there is no assertion, the subject is clause-initial. This behavior is expected, since the subject is D-linked, and topical or given information tends to be sentence initial. Let us now observe the syntactic properties of the initial subject: A – It may be doubled by a pronoun (45) A: Quem comeu o quê? who ate what B: A Maria…ela comeu a sopa. Maria…she ate the soup Pronominal doubling is typical of left-dislocation constructions, which favors the hypothesis that preverbal subjects in this context are left-dislocated. This is unlike what happens in cases of sentence-focus: (46) A: O que é que aconteceu? what happened B: *A Maria…ela comeu a sopa. Maria…she ate the soup Since doubling in this context is ungrammatical, it may be defended that subjects are not left-dislocated in sentence-focus contexts. B – Indefinite subject are specific: (47) A: Quem comeu o quê? who ate what B: Um cão…comeu o osso. a dog…ate the bone

(= one of the dogs)

VSO and SVO in answers to multiple wh-questions

127

(48) A: Quem comeu o quê? who ate what B: */??Um cão qualquer comeu o osso. some dog ate the bone As shown in (47) and (48), an indefinite preverbal subject in answer to a multiple wh-question is necessarily specific. If a non-specific indefinite is used, as in (48), the sentence is ungrammatical. This behavior is not reproduced in cases of sentence-focus: (49) A: O que é que aconteceu? what happened B: Um cão comeu o osso. a dog ate the bone B’: Um cão qualquer comeu o osso. some dog ate the bone Non-specific indefinites cannot be left-dislocated, which favors the analysis proposed here for preverbal subjects in answers to multiple wh-questions. As for Spec,IP, what triggers movement to this position is EPP and Caselicensing, which is not related with specificity or definiteness. Hence, it is expected that there is no semantic restriction regarding the type of subject that can appear in preverbal position in sentence-focus contexts. The observation that preverbal subjects in two different contexts leads to the following consequence: the two analyses available for preverbal subjects in European Portuguese do not contradict each other. They are both necessary to account for different facts (see Raposo 2000 for a similar conclusion). Again, as stated at the end of chapter 3, syntax may produce two converging outputs: one with a left-dislocated subject, and another one with the subject in Spec,IP. The choice between the two is made by the appropriateness of each one of them, taken into consideration their semantic properties and the discourse context.

6. Subjects in Spec,TP and the interface with morphology

6.1. Spec,TP available in I-to-C contexts only Among the several possible landing sites for subjects in the functional domain, nothing has been said so far concerning the possibility for subjects to be stranded in Spec,TP. In the previous chapters, it was argued that preverbal subjects in European Portuguese move to the specifier of AgrSP, while the verb undergoes shortV-movement from V-to-T. This analysis derives the non-adjacency between subject and verb in sentences like (1): (1)

Ninguém provavelmente leu bem o livro. no-one probably read well the book

The fact that the verb appears in between two adverbs in (1a,b) shows that it is not the case that there is no V-movement at all in European Portuguese. As mentioned in chapter 1, I assume that subject-oriented adverbs must be TP-adjuncts, since they only appear in the position in between the subject and the verb. All other positions for the adverb in (2) yield a manner reading: (2)

a. O João estupidamente entornou o café. João stupidly spilled the coffee

Subj-Or.

b. O João entornou estupidamente o café João stupidly spilled the coffee

Manner/*Subj-Or.

c. O João entornou o café estupidamente. João stupidly spilled the coffee

Manner/*Subj-Or.

In this kind of context, it is possible to show that Spec,TP is not an available position for the subject, in spite of the fact that European Portuguese allows for subject-verb inversion (Ambar 1992). In other words, subject-verb inversion in declarative contexts is not to be analyzed as a case of subject in

130

Subjects in Spec,TP and the interface with morphology

Spec,TP and verb in AgrS. The unavailability of Spec,TP is attested in example (3), in which the subject is doubled by a pronoun, blocking the topic reading for the adverb, and the position for the pronoun in between the subject-oriented adverb and the verb is ungrammatical:57 (3)

a. O João…ele estupidamente entornou o café. João…he stupidly spilled the coffee b. *O João…estupidamente ele entornou o café. João…stupidly he spilled the coffee

So far, the evidence shows that the subject cannot stay in Spec,TP. However, if one looks at wh-questions involving I-to-C movement, it is possible for the subject to surface right after the subject-oriented adverb:58 (4)

a. O que tinha estupidamente o João entornado? what had stupidly João spilled b. Quando tinha cautelosamente o João lido o livro? when had carefully João read the book

There is thus an apparent contradiction: while the data in (3) show that Spec,TP is not an available position for the subject, the data in (4) show that Spec,TP is an available position for the subject. This puzzle becomes more evident in the next example: (5)

a. O João estupidamente tinha já entornado o café. João stupidly had already spilled the coffee b. O João…(ele) estupidamente (*ele) tinha (*ele) já (??ele) entornado (ele) o café. João…(he) stupidly (he) had (he) already (he) spilled the (he) the coffee c. O que tinha (ele) estupidamente (ele) já (??ele) entornado? what had (he) stupidly (he) already (he) spilled

The crucial contrast is the one between the pronouns in bold in (5b) and (5c). (5b) shows that, in the declarative sentence, the pronouns cannot occur in any position between the subject-oriented adverb and the adverb já ‘already’. In the interrogative context (5c), however, the position in between the two adverbs is an available position.

The interface with morphology

131

The analysis outlined above for subject non-adjacency and the puzzle regarding Spec,TP raise at least the following two questions: (6)

a. If V does not raise to AgrS, how do Agr morphemes merge with V? b. Why is Spec,TP an available position for subjects in I-to-C contexts only?

6.2. The interface with morphology The suggestion I would like to make is that the availability of Spec,TP is a consequence of morphological merger of AgrS to V. Let me start by providing some background on how morphological merger operates. According to some works in the framework of Distributed Morphology (Halle & Marantz 1993, Bobaljik 1995), affixation takes place in the Morphological component of the grammar. The fusion of heads is possible under syntactic adjacency, and lexical insertion is made in single slots. Bobaljik (1995) provides two potential scenarios illustrating how affixation may operate. Suppose there is cyclic head-movement, creating the syntactic unit in (7): Head-Movement A: (7)

Agr fy T Agr fy AgrO T fy V AgrO

As defended by Bobaljik, this type of object has consequences for morphological fusion, since there must be two independent morphemes for T and Agr. This is because in a first moment V merges with AgrO, under syntactic adjacency, while in a second step the unit V/AgrO would merge with T. Now, this second step is impossible, since it would predict that the tense morpheme and the verbal root would be competing for insertion in the same slot. As a consequence, fusion does not take place, and both T and Agr morphemes may cooccur.

132

Subjects in Spec,TP and the interface with morphology

The second type of scenario discussed by Bobaljik (1995) is the one in (8): Head Movement B: (8)

Agr fo AgrO AgrS fy fy V AgrO T AgrS

According to Bobaljik, this is the type of head created if T has weak N-features, not being able to attract AgrO. Under such circumstances, AgrS attracts T, and the complex V+AgrO. Unlike in (7), T and AgrS are syntactically adjacent, therefore the two heads may undergo fusion. The consequence for morphology is that T and AgrS morphemes will now compete for insertion in the same slot. As mentioned, for the configuration in (8) to be obtained, T must have weak-N features, hence Spec,TP is unavailable. The big consequence from this type of analysis is that by looking at the verbal morphology, one may know whether Spec,TP is projected. In other words, if T and AgrS morphemes cooccur in a language, then Spec,TP is projected. This analysis is the basis for Bobaljik and Jonas’ 1996 [Spec,TP parameter]. They claim that this is evidence that morphology may act as a filter on syntactic derivations, and that transitive expletive constructions provide the syntactic evidence for knowing whether Spec,TP is used as a landing site for the subject or not. Let us consider some examples discussed by Bobaljik: Icelandic and English contrast in that the latter only provides evidence for Agr or T morphology but not for both, while in the former Agr and T morphemes cooccur: Icelandic: kasta ‘throw’

English: tremble

Present

Past

Present

kasta kasta-r kasta-r

kasta-δi kasta-δi-r kasta-δi

tremble tremble tremble-s

tremble-d tremble-d tremble-d

köst-um kast-iδ kasta

köstu-δu-m köstu-δu-δ köstu-δu

tremble tremble tremble

tremble-d tremble-d tremble-d

Past

The interface with morphology

133

The translation of these facts into distributed morphology is the following: T and AgrS are in complementary distribution in English, competing for insertion in the same slot. The syntactic correlation is the expected one: English lacks Icelandic-like transitive expletive constructions. Note that, since there is no V-to-I in English, affixation must be made under adjacency, a matter I will return to below. In languages with V-to-I or V2, the same type of distinctions may be found: German (Dutch is similar): sagen ‘say’ Present

Past

sag-e sag-st sag-t

sag-te sag-te-st sag-te

sag-en sag-t sag-en

sag-te-n sag-te-t sag-te-n

Like in Icelandic, in German, T and Agr are not in complementary distribution, and there is evidence for the availability of two subject positions in the IP-domain. Bobaljik (1995) crucially presents the contrast between Swedish and Afrikaans. In both languages, the present tense has no distinctive morphology, and there is no evidence for competition between T and Agr. In Swedish, the two types of morphology do not compete, since there is no Agr morphology, and in Afrikaans, the past tense is formed with periphrastic constructions. Swedish: smaka ‘to taste’ Present

Past

smaka-r smaka-r smaka-r

smaka-de smaka-de smaka-de

smaka-r smaka-r smaka-r

smaka-de smaka-de smaka-de

134

Subjects in Spec,TP and the interface with morphology

Afrikaans: werk ‘to work’ Present

Past

werk werk werk

Past formed with auxiliary verbs

werk werk werk Yet, there is syntactic evidence that, in Afrikaans, Spec,TP is available, since there are transitive expletive constructions. Based on this contrast, Bobaljik (1995) suggests that the crucial morphological evidence to know whether T and Agr are in competition for insertion in the same slot in the Germanic languages comes from the Past tense morphology. Before getting back to the Portuguese data, let me just address the issue of what counts for morphological adjacency, when there is no head-to-head movement. Halle & Marantz (1993) and Bobaljik (1995) suggest that doinsertion is used when V and I are not adjacent, blocking morphological merger. The main idea, schematized in (9) is that any lexical material, except for adverbs, blocks the adjacency requirement, and force do-insertion.59 (9)

a. [IP Subj I [VP V I and V are adjacent, morphological merger is possible b. [IP Subj I [VP Adv [VP V I and V are adjacent, morphological merger is possible c. [IP Subj I [NegP not [VP V I and V are not adjacent, morphological merger is not possible (do-insertion) d. [CP wh I+C [IP Subj tI [VP V I and V are not adjacent, morphological merger is not possible (do-insertion)

With the background given above, we now have the necessary tools to address the problems regarding the availability of Spec,TP in EP.

The interface with morphology

135

Let us start with the first question raised above, namely how does AgrS merge with the verbal root, if there is no T-to-Agr movement? It is legitimate to assume that the analysis proposed for verbal morphology in English, a context in which there is no V-to-I movement, applies in EP: since there is no movement from T to AgrS, there must be morphological merger under adjacency. If the subject is in Spec,AgrS, there is adjacency between AgrS and T, independently of the presence of an adverb adjoined to TP.60 This is illustrated in (10a). If the subject would stay in Spec,TP like in (10b), there would be no adjacency and morphological merger would be impossible: (10) a. [AgrSP Subj AgrS [TP (Adv) T+V [VP tV I and V are adjacent, morphological merger is possible b. [AgrSP AgrS [TP Suj T+V [VP tV I and V are not adjacent, morphological merger is not possible So far, this straightforwardly explains how AgrS is merged with the verbal root, and why Spec,TP is not an available position for subjects in declarative contexts. Recall that, according to Bobaljik (1995), there is a correlation between the availability of Spec,TP and the existence of two slots for Agr and T. His comparison between the several germanic languages also shows that the past tense paradigms are the crucial ones. Crucially, the past tense in European Portuguese only displays evidence for a single slot. It is not possible to distinguish independent T and Agr morphemes in the past tense:61 (11) Past tense

falar ‘to speak’

fale-i fala-ste falo-u falá-mos fala-stes fala-ram It remains to be explained why Spec,TP is available for subjects in interrogative contexts. As argued in Ambar (1992), wh-questions with bare wh-forms involve I-to-C movement. European Portuguese does not have any strategy like do-support. This implies that in order for the verb to go to I, T must raise to AgrS, in compliance to the Head Movement Constraint:

136

Subjects in Spec,TP and the interface with morphology

(12) a. [CP C+T+AgrS [AgrSP t [TP t The head created in C has the shape in (13): (13)

C 1 Agr C 1 T Agr 1 V T

Recall from Bobaljik (1995) and Bobaljik and Jonas (1996) that, in a head like this, there may be no fusion of nuclei, otherwise the verbal root and T would be competing for the same slot. If neither fusion nor morphological merger can apply, adjacency between Agr and T is no longer relevant. The syntactic consequence is that nothing prevents using Spec,TP as a position for the subject. This explains why Spec,TP is only available when there is Ito-C movement. This analysis might make a different prediction, as pointed out to me by J. Bobaljik (p.c.). It might be the case that when the verb moves to C, through cyclic head movement, a different morphology would show up. This is however not the case. I will leave this issue unsolved here. Two additional arguments show that the availability of Spec,TP depends on the existence of I-to-C movement. First, one may consider other adverbs that may only be adjoined to TP. Such a case is the adverb sempre ‘always’, which occurring preverbally means something like ‘after all’ (cf. Gonzaga 1997). This adverb provides more robust evidence for the unavailability of Spec,TP in declarative contexts, since, unlike subject-oriented adverbs, it cannot be topicalized: (14) a. O João sempre tinha feito o trabalho. João after all had done the work b. *Sempre o João tinha feito o trabalho. after all João had done the work As predicted by the analysis, in questions involving I-to-C movement, the subject may occur after this adverb:

The interface with morphology

137

(15) a. ?Que trabalho tinha sempre o João feito? which work had after all João done b. Esses trabalho todos, tinha sempre o João feito?, não tinha? all those works, had always João done?, hadn’t he? Second, as shown in Raposo (1987), inflected infinitives in European Portuguese often involve I-to-C movement. Raposo (1987) shows that in factive contexts I-to-C movement is not obligatory, as attested by the grammatical word orders in (16): (16) a. Os alunos lamentam os deputados terem votado a proposta. the students regret the deputies have-3pl voted the proposal b. Os alunos lamentam terem os deputados votado a proposta. the students regret have-3pl the deputies voted the proposal Inflected infinitives provide a good testing ground for the proposal made in this paper for two reasons. First, topicalization is impossible in this context (Barbosa 2000, Costa and Gonçalves 1999): (17) a. *Eu lamento, esse livro, terem eles lido. I regret that book have-3pl they read b. *Eu lamento terem, esse livro, eles lido. I regret have-3pl that book they read Therefore, a pre-subject position for a subject-oriented adverb may not be taken as an instance of topicalization of the adverb. Second, since I-to-C movement is optional, the prediction is that the subject will be occurring in the post-adverbial position (Spec,TP), if the verb is clause-initial, which indicates that there is I-to-C movement. This prediction is borne out. (17a) shows that the subject-oriented adverb may occur in between the subject and the auxiliary verb. (17b) shows that if there is I-to-C movement, the subject may remain in Spec,AgrSP, in the pre-adverbial position. The crucial contrast is the one between (17c) and (17d). (17c) shows that, if there is no I-to-C movement, the subject cannot occur in between the adverb and the auxiliary verb. In other words, if there is no I-to-C movement the subject cannot stay in Spec,TP. In (17d), I-to-C movement occurred, and the subject may surface after the adverb, in Spec,TP.

138

Subjects in Spec,TP and the interface with morphology

(17) a. Os meninos lamentam os deputados cautelosamente terem votado a proposta. the children regret the deputies carefully have-3pl voted the proposal b. Os meninos lamentam terem os deputados cautelosamente votado a proposta. the children regret have-3pl the deputies carefully voted the proposal c. *Os meninos lamentam cautelosamente os deputados terem votado a proposta. the children regret carefully the deputies have-3pl voted the proposal d. Os meninos lamentam terem cautelosamente os deputados votado a proposta. the children regret have-3pl carefully the deputies voted the proposal

6.3. The non-parametric availability of Spec,TP The study developed in this chapter permits drawing the conclusion that the availability of a specific A-position for subjects may derive from specific constructions rather than from a global parameter. This is theoretically interesting, since it raises a number of issues regarding the format of the [Spec,TP parameter] and the null subject parameter. The data from EP regarding the availability of Spec,TP suggest that there is not really a parameter dividing languages as far as Spec,TP is concerned but much more local constraints determining whether this syntactic position may be used language-internally. Another conclusion that may be drawn is that the comparison between Romance and Germanic languages regarding structure of the clause may be established through different paths. The unavailability of transitive expletive constructions in Romance does not obviate taking into consideration these languages for the discussion of the spec-TP parameter. Finally, and getting back to the general question concerning the types of constraints that affect the availability of A-positions as landing sites for subjects, this discussion further confirms that there is not a single answer that

The non-parametric availability of Spec,TP

139

may be given to this question, since completely different factors seem to determine the usage of different A-positions: while the availability of Spec,AgrSP seems to be constrained by the syntax-discourse interface, Spec,TP appears to be constrained by the syntax-morphology interface.

7. Syntactic outputs and the interfaces

The discussion carried out in the preceding chapters enables us to address the issue of how the syntactic component interacts with interface issues. I argued that syntax proper does not need to encode discourse notions. The behaviour of subjects appears to indicate that, whenever two options can be generated in the syntactic component, discourse constraints may choose the one that best suits its purposes. However, in some circumstances, as discussed in chapter 4, the two generations cannot be generated in the syntactic component, for instance, for locality reasons. In that case, even if a subject is focused, it will not be able to occur in the inverted position. This type of data leads to the formulation of the following hypothesis: (1)

When the computational system generates multiple convergent outputs, interface constraints may filter or select them.

The immediate consequence of this hypothesis is twofold: first, the interface conditions do not act as syntactic triggers; second, the interface conditions act on outputs either as selectors of one of multiple options, or as filters. Another consequence of this proposal is that whenever only one output is generated by the syntactic component, ambiguity and/or different types of interface disambiguation strategies may take place. Considering the interface with prosody and discourse, mentioned above, it appears that if the syntax of a language allows multiple outputs in which different constituents are placed at the position in which they bear sentence nuclear stress, this will be the preferred option. Otherwise, stress shift is used as a marked option. Under this alternative view, the syntactic rearrangement is always preferred over stress-shift strategies. The goal of this chapter is to provide further empirical argumentation for this view on the relation between interface conditions and the syntactic component. The arguments to be presented will favour Chomsky’s (2001) proposal that the computational system is ‘blind’ to most interface considerations. In section 1, I will discusse the the behaviour of ditransitives in English (Brandt 1999), and in European Portuguese; section 2 will address the behaviour of possessives in Portuguese and Italian (Cardinaletti 1998; Castro and

142

Syntactic outputs and the interfaces

Costa 2003); in section 3, the differences between subject-verb inversion in Brazilian and European Portuguese (Costa and Figueiredo Silva 2003). These three types of evidence appear to lead to the following conclusion: stress shift is used only when, for syntactic reasons, a constituent cannot appear at the position in which it is assigned nuclear stress.

7.1. The behaviour of ditransitives in English and in European Portuguese The ditransitive alternation in English between V-DO-PP and V-IO-DO has been argued to correlate with focus effects: if the direct object is the focus of the sentence it appears rightmost, if the indirect object is the focus, the order in which it is rightmost is used (Brandt 1999): (1)

Who did you give the book to? I gave the book to Mary.

(2)

What did you give to Mary? I gave her a book.

Not all ditransitive verbs exhibit these two possibilities (e.g. Jackendoff 1990). In those cases, if the non-final constituent is focalized, a marked stress must be used: (3)

What did you donate to the library? a. I donated the BOOKS to the library. b. *I donated the library the books.

These facts indicate that word order alternations are used for discourse purposes, only if the syntax generates two options. When only one option is created, stress-shift strategies are used. Note that, according to most authors, the verb in (4) does not alternate due to lexical semantic reasons (e.g. Marantz 1984). In European Portuguese, there is also a relationship between word order and focus in ditransitive contexts, as illustrated in (4) and (5): (4)

A: A quem é que deste o livro? to whom did you give the book

The behaviour of ditransitives in English and in European Portuguese

143

B: Dei o livro [F ao Paulo] (I)gave the book to Paulo. #Dei ao Paulo o livro. (5)

A: O que é que deste ao Paulo? what did you give to Paulo B: Dei ao Paulo [F o livro]. (I) gave to Paulo the book #Dei o livro ao Paulo.

Interestingly, if the focused argument must bind an anaphor contained within the non-focused argument, the former cannot be rightmost. In that case, the focused argument bears a heavy stress: (6)

A: A quem é que deste os livros? to whom did you give the books? B: Dei [F A CADA AUTOR] o seu livro. (I) gave to each author his book. ?*Dei o seu livro a cada autor. (I) gave his book to each author

(7)

A: O que é que deste aos autores? what did you give to the authors B: Dei [F CADA LIVRO] ao seu autor. (I) gave each book to its author ?*Dei ao seu autor cada livro. (I) gave to its author each book

In Costa (1998a), it is argued that this behavior of focus binders provides evidence for an Optimality-theoretical approach to focus, since the constraint forcing foci to be rightmost can be violated when it conflicts with binding requirements. However, this account failed to accommodate the stress-shift effects. Under the view advocated here, the output in which the focus ccommands the anaphor is the only legitimate ouput. Accordingly, syntax is not generating two convergent ouputs, and stress-shif acts as a last resort strategy. Incidentally, this type of data not only sheds light on the issue of interfaces but also on the structure of ditransitive contexts. So far, we have considered

144

Syntactic outputs and the interfaces

cases in which syntactic transformations generate multiple converging ouputs. The binding facts, and an asymmetry with VOS shows that multiple converging outputs can also be the result of optional base-generations. Let us first consider the debate considering the structure of VP in ditransitive contexts (Kayne 1984; Larson 1988; Pesetsky 1995; Phillips 1996; among others). As noted in Phillips (1996), part of the debated problems derives from the fact that constituency tests yield contradictory results, when applied to ditransitive VPs. For instances, NPI-licensing in (8) provides evidence for a shell structure like in (10), while the fronting tests provide evidence for a layered structure like in (11): (8)

a. John gave nothing to any of the children on his birthday. b. *John gave anything to none of the children on his birthday.

(9)

John intended to give candy to children on his birthday… a. …and [give candy to children on his birthday] he did. b. …and [give candy to children] he did on his birthday. c. …and [give candy] he did to children on his birthday.

(10) VP 2 V’ 2 V VP 2 DO V’ 2 V IO

(11)

VP 2 V’ IO 2 V DO

An asymmetry in binding may shed light on this debate. In Costa (1996, 1998), and in chapter 4, it was argued that in SVO orders the subject is in Spec,IP. VSO and VOS orders instantiate subjects stranded in Spec,VP. According to these results, the difference between VSO and VOS lies on the status of the object. In VOS orders, the objects scrambles out of VP, creating an adjunction configuration. Following this analysis, depicted in (12), scrambling is an instance of A-bar-movement:

The behaviour of ditransitives in English and in European Portuguese

145

(12) Viu a Maria o João. [IP V [VP DO [VP Subj t t ]]] One of the arguments for the A-bar nature of object scrambling came from binding facts. As illustrated in (13) scrambling of an object to the left of the subject does not feed binding, unlike what happens in cases of A-movement (13c). (13) a. *Viu o seu filho cada mãe. saw her son each mother

VSO

b. *Viu cada mãe o seu filho.

VOS

c. Cada mãe foi vista pelo seu filho.

Passive

The binding facts in ditransitive contexts illustrated above pose a challenge to this generalization. In particular, these facts raise the following two questions: A.

What is the nature of indirect object scrambling?

B.

Why doesn’t indirect object scrambling across a direct object yield the same results as direct object scrambling across a subject?

Rephrasing the problem, the binding facts suggest that an object moved across a subject seems to be A-bar moved, while an indirect object moved across the direct object seems to occupy an A-position. The solution I will offer to this puzzle is based on the assumption that syntactic structure is strictly binary, and generated from left-to-right, as proposed in Phillips (1996). I follow Belletti and Shlonsky (1995) in their claim that the order direct object-indirect object in Romance is unmarked. Evidence in favor of this comes from the emerging word order in sentence-focus contexts (14), and from the order found in idioms formed with ditransitive verbs (15–16):62 (14) O que é que aconteceu? What happened a. O João deu uma prenda à Maria. João gave a gift to Maria b. #O João deu à Maria uma prenda. João gave to Maria a gift

146

Syntactic outputs and the interfaces

(15) Dar pérolas a porcos. To give pearls to pigs *Dar a porcos pérolas. To give to pigs pearls (16) Pôr mais lenha na fogueira. To put more wood in the fire *Pôr na fogueira mais lenha to put in the fire more wood However, I deviate from Belletti and Shlonsky (1995), since I assume that the fact that this word order is unmarked does not necessarily mean that it correspond to a base-generated word order. Note that constituency tests also yield contradictory results in EP: on the one hand, binding (18-19) and scopal facts (20) provide evidence for a shellstructure: (18) a. Entreguei cada livro ao seu autor. (I) gave each book to his author b. Entreguei a cada autor o seu livro. (I) gave to each author his book (19) a. *Apresentei a Mariai à filha da Mariai. (I) introduced Maria to the daughter of Maria b. Apresentei à filha da Mariai a Mariai. (I) introduced to the daughter of Maria Maria (20) a. Apresentei uma mulher a todos os homens. (I) introduced a woman to all men

∃ > ∀ (preferred)

b. Apresentei a todos os homens uma mulher. (I) introduced to all men a woman

∀ > ∃ (preferred)

On the other hand, gapping (21) and fronting (22) provide evidence in favor of a layered structure:

The behaviour of ditransitives in English and in European Portuguese

147

(21) a. Eu dei livros à Maria e o Pedro [ deu livros] à Ana. I gave books to Maria and Pedro to Ana b. Eu pus os livros na prateleira e tu [puseste os livros] na gaveta. I put the books on the shelf and you in the drawer. (22) O Pedro queria dar os livros ao Rui ontem. Pedro wanted to give the books to Rui yesterday a. …e [dar os livros] ele deu ao Rui ontem. and give the books he gave to Rui yesterday b. …e [dar os livros ao Rui] ele deu ontem. and give the books to Rui he gave yesterday c. …e [dar os livros ao Rui ontem] ele deu. and give the books to Rui yesterday he gave It is important to note that the evidence for layered structures is not crystalclear, since with other types of ditransitives, stranding one of the arguments is not good, a problem I will not address here: (23) O Pedro queria pôr os livros na prateleira ontem. Pedro wanted to put the books on the shelf yesterday a. ??*…e [pôr os livros] ele pôs na prateleira ontem. and put the books he put on the shelf yesterday b. …e [pôr os livros na prateleira ] ele pôs ontem. and put the books on the shelf he put yesterday c. …e [pôr os livros na prateleira ontem] ele pôs. and put the books on the shelf yesterday he put The contradictory tests suggest that an analysis along the lines of Phillips’ (1996) may be on the right track. The consequence of this type of analysis is that there are two basic ways of generating the two word orders, even if only one of them is unmarked. The conclusion from this type of approach is that there is no direct relation between unmarkedness and base-generation. Phillips’ analysis is based on the following two principles:

148

Syntactic outputs and the interfaces

(24) MERGE RIGHT New items must be introduced at the right edge of a structure. (25) BRANCH RIGHT Where a terminal can be attached to more than one position in the existing structure with no effect on interpretation, the attachment that results in the more right-branching structure must be chosen. The interaction between these two principles operates in the following way for the generation of V-DO-IO and V-IO-DO orders: (26) V-DO-IO: Step 1: merge V with Direct Object, discharging theta-role V

DO VP 2 V > DO θ

Step 2: Create a copy of V (reanalyzing the direct object as a specifier), and merge a PP as its sister, discharging the goal theta-role. VP 2 V DO a.

VP 2 V VP 2 DO V

PP b.

VP 2 V VP 2 DO V’ 2 V > PP θ

Phillips’ analysis predicts that constituency tests may target a step of the derivation in which V and DO form a VP, or the final structure, in which there

The behaviour of ditransitives in English and in European Portuguese

149

is c-command of the DO over the PP. For generating V-IO-DO order, (24) and (25) operate in the following way: (27) V-IO-DO: Step 1: merge V with Indirect Object, discharging goal theta-role V

PP VP 2 V > PP θ

Step 2: Create a copy of V (reanalyzing the indirect object as a specifier), and merge a DP as its sister, discharging the theme theta-role. VP 2 V PP a.

VP 2 V VP 2 PP V

DO b.

VP 2 V VP 2 PP V’ 2 V > DO θ

Note that, as mentioned above, claiming that both orders are base-generated does not entail that they should be optional. The structure built in (27) is only generated if necessary for satisfying binding requirements (or any other constraint forcing the order IO-DO, such as heaviness). Suggesting that V-IO-DO can be base-generated for binding purposes makes two interesting predictions. First, as it is also pointed out in Phillips (1996), it is expected that no case is found in which the verb and the indirect object may be fronted stranding the direct object. This is confirmed by the data in (28):

150

Syntactic outputs and the interfaces

(28) O Pedro queria entregar os prémios aos vencedores... Pedro wanted to give the prizes to the winners …*e entregar aos vencedores ele entregou os prémios. and give to the winners he gave the prizes The only circumstances under which V-IO fronting is possible with DO stranding are cases in which binding is involved. Although the contrast is subtle, (29) is better than (28): (29) O Pedro queria entregar a cada vencedor o seu prémio... Pedro wanted to give to each winner his prize …?e entregar a cada vencedor ele entregou o seu prémio. and give to each winner he gave his prize Second, this type of approach straightforwardly predicts well-known cases of indirect object control. In (30), PRO contained in the direct object is controlled by the indirect object, and it is not necessary to assume that the clause is right-dislocated, which would be problematic. The analysis I am suggesting predicts that both DO and IO occupy A-positions, and the control pattern follows. (30) Eui pedi aos meninosj para PROi/j dançar. I asked to the children to dance Let us finally turn to the solution of the puzzle presented: why it is not possible for a scrambled direct object in VOS to bind into the subject. Recall that the binding possibilities for V-IO-DO and V-IO-DO were derived from the fact that both word orders are base-generated. Therefore, no A-bar-movement takes place. The problem must now be stated in different terms: the issue now is why VOS cannot be base generated. In both cases discussed above, the theta-criterion is satisfied: either the verb or its copy assigns the theta-roles to each one of the arguments. This entails that the theta-criterion is satisfied in the course of the clause’s generation (and not pre-syntactically). This consequence strongly clashes with Baker’s (1988) UTAH. In particular, there is no need to derive one of the orders in ditransitive contexts from the other, since the theta-criterion may be satisfied in both cases.

The behaviour of ditransitives in English and in European Portuguese

151

Independent evidence for the claim that theta-roles may be assigned in the course of the derivation, and that the realization of the theta-role may depend on other factors comes from cases of complex predicate formation (Wurmbrand 1998; Gonçalves 1999). As discussed in Gonçalves (1999), the realization of an embedded agent as a DP or as a PP depends on the transitivity of the embedded verb, on the type of main verb, and on the process of complex predicate formation. This is illustrated in (31): (31) a. Eu mandei correr os meninos. I made run the children b. Eu mandei ler o livro aos meninos. I made read the book to the children. Crucially, if all agents were DPs generated in Spec,VP, the pattern in (31) would be unexpected. The fact that the realization of the agent as a PP in (31b) is forced by constraints on complex predicate formation provides independent evidence for a theory of theta-role assignment that takes into consideration the sentence as a whole rather than each of the theta-assigning heads independently. The relevance of the type of main verb for determining the shape of the embedded agent becomes even more obvious, in the cases discussed in Gonçalves (1999), in which the matrix verb does not make available a PP position, and, as a consequence, the agent theta-role may not be discharged: (32) a. Eu vi correr os meninos. I saw run the children b. *Eu vi ler o livro aos meninos. I saw read the book to the children Returning to the binding difference between V-IO-DO and V-O-S, recall that the object surfacing to the left of the subject reconstructs, while the IO across DO is able to bind. As mentioned above, interpreting reconstruction as a consequence of A-bar movement, and the binding relation between IO and DO as a consequence of base-generation, the question to be answered is why the object cannot be base-generated in VOS. In what follows, I will show that it is not possible to satisfy left-to-right merge and the theta-criterion, creating a base-generated VOS. Step 1 in (33) represents the merging of the verb with the direct object, which like in the case of ditransitive contexts is unproblematic.

152

Syntactic outputs and the interfaces

(33) V-O-S: Step 1: merge V with Object, discharging theme theta-role V

DO VP 2 V > DO θ

Step 2: Create a copy of V (reanalyzing the object as a specifier), and merge a DP as its sister, discharging the agent theta-role. VP 2 V DO a.

VP 2 V VP 2 DO V

Subj b.

VP 2 V VP 2 DO V’ 2 V _=> Subj θ

As shown above, the problem comes about when the subject is merged with the copy of the verb, in compliance with MERGE RIGHT and BRANCH RIGHT. The resulting configuration, illustrated in (33b) does not yield a proper context for assignment of the external theta-role. In other words, the problem is that the subject is not a sister to V’, and is therefore unable to receive theta-role, according to Koopman and Sportiche’s (1991) proposal.63 As a consequence, unlike V-IO-DO, VOS cannot be base-generated. Instead, VOS must be an instance of movement. If it is A-bar movement, the structure in (34) obtains and the binding possibilities follow. (34) [IP V [VP O [VP S t t ]]]

The behaviour of possessives in Portuguese and Italian

153

Crucially, for the discussion in this chapter, we observe a case in which two syntactic outputs are base-generated, and are filtered out at the interface with prosody. One of the two may be filtered out in the syntax, if binding requirements are not met. In that case, stress shift applies.

7.2. The behaviour of possessives in Portuguese and Italian Cardinalletti (1998) shows that pre-nominal and post-nominal possessives are XPs in Italian, and that post-nominal possessives are used in definite contexts, only if they are focussed: (35) a. *la SUA casa, non tua the her house, not yours

Cardinaletti (1998: 19 –20)

b. la casa SUA, non tua the house hers, not yours “Her house, not yours” There is evidence to argue that Portuguese pre-nominal possessives are heads (cf. Castro and Costa 2003). This is confirmed by the following facts: – they cannot be modified (36a);64 – they can refer to entities with the feature [-human] (37a); – they can be phonetically reduced in some dialects of Portuguese (38a). (36) a. *O só meu problema é que não percebo nada disto. The only my problem b. Um problema só meu é que não percebo nada disto. a problem only my (37) a. A suai tampa é muito prática. her lid is very practical “Its/her lid is very practical” b. Encontrei uma tampa suai. I found a lid her “I found a lid of hers”

i = Maria/frying pan

i = Maria/*frying pan

154

Syntactic outputs and the interfaces

(38) a. O m[e] livro the my book “My book” b. Um livro *[me]/[mew] a book my “A book of mine” In spite of there being an association between definiteness and the use of pre- vs postnominal possessives, what is relevant for the purposes of this chapter is the asymmetry between European Portuguese and Italian. While in the latter, the strong postnominal form is used, even in definite contexts, for focalization, in European Portuguese, this option is not available, hence the prenominal possessive must be used even in such contexts, associated with a heavy stress. In Castro and Costa (2003), it is proposed that pronominal possessives are weak (non-clitic) heads. This proposal makes the following predictions: first, it is predicted that the forms under analysis cannot display clitic-like behavior, since they are not clitics;65 second, the asymmetry between the pre- and postnominal possessives is explained. Since the prenominal weak forms are heads, they cannot be modified, they cannot be coordinated without focalization, and they can be reduced. This hypothesis also explains the focalization facts. Unlike in Italian, a focalized possessive is not post-nominal: (39) a. Esse é o MEU problema, não o teu. That is the MY problem, not the yours b. *Esse é o problema meu, não o teu. That is the problem my “That is MY problem, not yours” This asymmetry follows from the categorial status of these forms in the two languages. Since pre-nominal possessives are heads in Portuguese, they do not enter into configurations typical of XPs, which permit the post-nominal order. In Italian, because of their XP-status, possessives may surface in two different configurations. The conclusion that can be drawn from these facts is that the position of the possessive and its relation with its discourse function solely derives from its categorial status. If the DP-final position observed for Italian were a consequence of focus-movement to the right, the asymmetry between DP-internal behaviour and the availability of a final position for foci at the clause level would become difficult to explain.

The behaviour of possessives in Portuguese and Italian

155

In other words, the case of possessives when contrasted to the general availability of clause-final position for focused items indicates that the correlation between domain-final position and focus is to be evaluated in each construction, depending on its specific properties, rather than generalized for a specific language. In the case of possessives, we observe that European Portuguese behaves like a stress-shift language, just because there is no legitimate syntactic ouput placing the possessive on the DP-final position in definite contexts.

7.3. The differences between subject-verb inversion in Brazilian and European Portuguese As mentioned above, in Brazilian Portuguese, unlike in European Portuguese, a subject which is focused does not invert, remaining in preverbal position (Costa and Figueiredo Silva 2003). This contrast between the two languages is illustrated in (40): (40) A: Quem comeu o bolo? Who ate the cake B: a. Comeu o João. Ate João

(EP/*BP)

a.’ O JOÃO comeu. (*EP/BP) João ate As indicated in (40a’), the non-inverted subject bears a heavy stress. This contrast between the two languages is interesting for the present discussion, since the two languages diverge only when the behaviour of subjects (of non unaccusative verbs) is concerned. In other instances of focalization, Brazilian Portuguese behaves like other languages in which focus occurs clause-finally, as illustrated in the following examples for ditransitive contexts: (41) Brazilian Portuguese: A: O que o João deu pra Maria? What João gave to Maria B: O João deu pra Maria um CD. João gave to Maria a CD

156

Syntactic outputs and the interfaces

(42) Brazilian Portuguese: A: Pra quem o João deu o CD? To whom João gave the CD B: O João deu o CD pra Maria. João gave the CD to Maria (43) European Portuguese: A: O que é que o João deu à Maria? What João gave to Maria B: O João deu à Maria um CD. João gave to Maria a CD (44) European Portuguese: A: A quem é que o João deu o CD? To whom João gave the CD B: O João deu o CD à Maria. João gave the CD to Maria As argued in Costa and Figueiredo Silva (2003), the data in (41)–(44) make two different points. First, both languages allow for either complement of the verb to surface clause-finally. Second, as shown by the question-answer pairs, the choice between order V-DO-IO and V-IO-DO may be made in compliance with discourse requirements. This case is of particular interest for the present discussion, since we observe that both languages satisfy the requirement that the focus of the sentence surfaces rightmost. Therefore, positing that the subject in Brazilian Portuguese does not invert in (40) because Brazilian Portuguese does not codify focus in the syntax would not account for the behaviour of complements of ditransitive verbs. What appears to be at stake that differentiates the two languages is that, since Brazilian Portuguese is loosing referential pro (Figueiredo Silva 1996, among others), it is impossible to leave Spec,IP empty, which makes the inverted construction ungrammatical. Therefore, we observe another case in which a syntactic output is not generated for a purely syntactic reason, and therefore the best word order for prosodic purposes does not arise. Again, under such circumstances, stress shift is used as a last resort.

Conclusions

157

7.4. Conclusions The three cases presented instantiate mixed systems, in which languages resort both to word order rearrangements and to stress shift in very similar contexts. In all cases under consideration, it was shown that syntax proper does not need to encode or refer to interface considerations. In other words, the arguments put forward in this chapter provide evidence for a more autonomous syntax, not making reference to interface conditions, and argue for an articulated view of the grammar, in which the interfaces read the outputs of syntax, but do not interfere in syntactic derivations.

8. Summary and conclusions

The primary goal of this book was to provide an analysis of the several positions where the subject may surface in European Portuguese. Departing from an architecture of the clause as sketched in early minimalist work, containing two subject-related functional categories above VP (AgrP and TP), it was shown that the subject may surface in all potential landing sites: Spec,AgrP, Spec,TP and Spec,VP. Moreover, just like any other argument of the clause, it was claimed that subjects also have the possibility of surfacing in a left-dislocated position, arguably adjoining to the clause’s left periphery. It was shown that there is no free variation. Each of these positions may be occupied by the subject, only if two requirements are met: i) The position is made available by syntax; ii) The position does not violate any interface condition. In other words, the following model was argued for: syntax generates legitimate outputs. At the interface levels, each output may be selected or filtred out, according to requirements of the interface. This interface licensing conditions operate in the following way for each of the identified surface positions: 1. Spec, VP – The subject may surface in Spec,VP, because it is able to check Case under Agree (Chomsky 2000). Likewise, Case may be checked under Move. The consequence is that both SVO and VSO outputs are equally well-formed from a syntactic point of view. It is argued that Information Structure and their interplay with prosody may choose a VSO ouput over an SVO order, when the subject is the focus of the sentence and must receive the sentence’s nuclear stress. The idea that Agree is a legitimate way of licensing an in-situ subject was motivated by the observation that in-situ subjects are constrained by locality conditions, definable in terms of phase-boundaries. 2. Spec,TP – This position provided an interesting puzzle. Looking at adverb positions, it appears that Spec,TP is an available surface position

160

Summary and conclusions

for subjects in I-to-C contexts. This observation was explained taking the interface with morphology into account. It was argued that the subject cannot be stranded in Spec,TP when the subject is in T, since it blocks the merger of the heads Agr and T. 3. Left-dislocation and Spec,AgrP – Non-focused preverbal subjects were shown to occupy the specifier of the topmost functional category of the inflectional domain. This goes against recent claims in the literature that preverbal subjects in null subject languages are left-dislocated. Nevertheless, it was shown that the fact that preverbal subjects occupy an A-position does not imply that they necessarily must occupy an A-position. Looking at contexts of apparent optionality in answers to multiple wh-questions, it was shown that in the appropriate context subjects may be left-dislocated. For a subject to appear in adjunction to the clause, it must meet semantic requirements such as non-exhaustivity. The picture emerging from the proposal made in this book is the following: syntax proper does not need to refer to conditions best placed at the interface. All that is needed from syntax is that it generates an array of wellformed outputs. Such outputs may be evaluated a posteriori by each of the interfaces. If they meet requirements of the interface, they are selected as legitimate. If, on the contrary, some interface condition is violated, they are ruled out. Under this approach, three independent results are derived: i) an explanation is found for the patterns of word order variation; ii) syntax proper may be reduced to its own tools, not having to manipulate semantic, discourse or prosodic variables; iii) the intuition that European Portuguese is an SVO language is derived: this word order corresponds to the one in which the subject occupies the only specifier position in which the other interfaces play no role. The generalization obtained is that when the computational system generates multiple convergent outputs, interface constraints may filter or select them. The immediate consequence of this generalization is twofold: first, the interface conditions do not act as syntactic triggers; second, the interface conditions act on outputs either as selectors of one of multiple options, or as filters. Another consequence of this proposal is that whenever only one output is generated by the syntactic component, ambiguity and/or different types of interface disambiguation strategies, such as heavy stress marking, may take place.

Summary and conclusions

161

As mentioned in chapter 7, the arguments put forward throughout the book provide evidence for a more autonomous syntax, not making reference to interface conditions, and argue for an articulated view of the grammar, in which the interfaces read the outputs of syntax, but do not interfere in syntactic derivations.

Appendix: On the nature of agreement in European Portuguese

Introduction The discussion carried out in chapter 6 strongly bears on the need for Agr inflection to merge with the verb in T. Since Agr typically encodes both person and number inflection, it is important to clarify which type of agreement is at stake here. I therefore dedicate this appendix to report on the results achieved in Costa and Figueiredo Silva (2003a), who claim that number morphology is, in most instances, a post-syntactic matter, based on a comparison between three register of Portuguese. If these results are on the right track, the relevant feature for triggering V-movement or morphological merger is person. Since we contend that the most substancial difference in the inflectional systems of European and Brazilian Portuguese relate to the presence of person features, two predictions are made: (i) Number agreement morphology does not necessarily trace the presence of a Spec,head configuration; (ii) There are no crucial differences in terms of V-to-I movement between the two languages, since in both, there is person morphology triggering V-to-I movement (Costa and Galves 2002; Vikner 1997). The main goal Costa and Figueiredo Silva’s (2003a) work is to discuss an additional argument against the view that verbal morphology should be explained in terms of verb movement alone. The following facts from three registers of Portuguese are discussed, some of which were previously noted in Galves (1993) and Figueiredo Silva (1996): 1. In European Portuguese, and in two registers of Brazilian Portuguese, there is no evidence to posit different landing sites for the verb. According to the traditional tests (adverb placement and floating quantifiers), in all of them, the verb seems to have moved from V to T, without reaching the topmost functional head (Costa 1996; Costa and Galves 2002).

164

Appendix: On the nature of agreement in European Portuguese

2. Yet, the agreement patterns in these dialects are different. As shown below, the following generalizations may be drawn: – In European Portuguese (EP), all elements able to bear plural morphology do so: (1)

Os carros são lindos. the-pl car-pl are beautiful-pl ‘The cars are beautiful’

– In one of the registers of Brazilian Portuguese (BP1), there is no DPinternal agreement, although there is subject-verb agreement: (2)

Os carro são lindo. the-pl car-sg are beautiful-sg ‘The cars are beautiful’

– In the other register of Brazilian Portuguese (BP2), there is neither DPinternal agreement, nor subject-verb agreement: (3)

Os carro é lindo. the-pl car-sg is beautiful-sg. ‘The cars are beautiful’

These facts raise at least the following questions: a) If there is robust evidence that in all these language variants, the target of verb movement is the same, can it be maintained that the morphological differences follow from verb movement? b) What type of micro-variation is at play in Portuguese that derives these differences? c) Is there any relation between the DP-internal facts and subject-verb agreement facts? The hypothesis developed in Costa and Figueiredo Silva (2003a) is that the effects of visible agreement must be interpreted as a non-unitary phenomenon. They derive from i) the type of morpheme (singleton or dissociated) (Embick and Noyer 2001), and ii) whether Spec-head configurations trigger

The facts

165

visible agreement. The possible combinations of these factors derive the variation found across the three variants of Portuguese. In section 1of this appendix, we will take a closer look at the data, in order to provide a clearer understanding of the type of morphological variation at stake; in section 2, the assumptions and proposal for explaining the crosslinguistic variation are presented; the predictions of the analysis are explored in section 3.

1. The facts The facts under scrutiny stem from two different domains: the number agreement patterns within the DP (section 1.1), and the patterns of subject-verb agreement (section 1.2). As mentioned, the data comes from three registers of Portuguese: European Portuguese (henceforth EP), and two varieties of Brazilian Portuguese (BP1 and BP2). An important remark must be made concerning the Brazilian Portuguese data: we are somehow idealizing the distinction between two dialects, since the same speaker may use BP1 and BP2. This idealization is however confirmed by sociolinguistic research showing that in designated situations, speakers opt for one of the two varieties.66 If this observation is correct, we are dealing with a case of competing grammars, each one used in different situations, in the sense of Kroch (1994, 1997), and it is possible to tease the two apart.

1.1. DP-internal number agreement DP-internally, European Portuguese displays the pattern of number agreement found in most Romance languages: plurality is expressed in all categories able to bear this type of morphology (noun, determiner, quantifiers, adjectives, possessives and demonstratives). This is shown in (4): (4)

a. Os/estes/alguns/uns livros muito bonitos The-pl / these-pl / some-pl /a-pl book-pl very pretty-pl ‘The/these/some/ books very pretty’ b. Os primeiros livros da biblioteca the-pl first-pl books-pl of-the library ‘The first books of the library’

166

Appendix: On the nature of agreement in European Portuguese

c. Os meus livros the-pl my-pl books-pl ‘My books’ d. Todos os meus primeiros livros bonitos all-pl the-pl my-pl first-pl book-pl pretty-pl ‘All my first pretty books’ Both dialects of Brazilian Portuguese behave alike and differ from European Portuguese. In general, plurality is marked just on the determiner. Nouns and post-nominal adjectives are not marked for plurality: (5)

Os/estes/alguns/uns livro muito bonito The-pl / these-pl / some-pl / a-pl book-sg very pretty-sg ‘The/these/some/ books very pretty’

The opposition between prenominal and postnominal positions within the DP is crucial for establishing the agreement patterns. As noted by Menuzzi (1994), plural markers may optionally surface on other prenominal elements, but if the noun is not marked as plural, no post-nominal element may bear a plural morpheme. This derives some variation in the prenominal domain. Prenominal adjectives may or may not bear plural morphology. A pattern that is not found is agreement on the adjective with a non-agreeing determiner: (6)

a. Os primeiros livro da biblioteca The-pl first-pl book-sg of-the library ‘The first books of the library’ b. Os primeiro livro da biblioteca The-pl first-sg book-sg of the library ‘The first books of the library’ c. *O primeiros livro da biblioteca The-sg first-pl book-sg of the library ‘The first books of the library’

Prenominal possessives differ from prenominal adjectives in an interesting way: just like in the case of adjectives, both the determiner and the possessive may bear the plural morpheme. However, if only one of them is to agree, the possessive, and not the determiner will bear the plural marker, which distinguishes this case from what was observed for adjectives:

The facts

(7)

167

a. Os meus livro The-pl my-pl book-sg ‘My books’ b. *Os meu livro The-pl my-sg book-sg ‘My books’ c. O meus livro The-sg my-pl book-sg ‘My books’

The emerging generalization from these data is the following: either the determiner head or all the prenominal elements bear plural morphology in Brazilian Portuguese. The interesting difference between adjectives and possessives must be accounted for, since the latter is the only case in which the plural-bearing element is not the determiner.

1.2. Subject-verb agreement As shown in the preceding chapters, in European Portuguese, subjects and verb agree independently of the position of the subject. The only exception to this is the possibility for inverted subjects of unaccusative verbs not to agree in colloquial speech (Costa 2000): (8)

a. Os meninos comeram o doce. The children ate-pl the candy ‘The children ate the candy’ b. *Os meninos comeu o doce. The children ate-sg the candy ‘The children ate the candy’ c. Comeram os meninos o doce. Ate-pl the children the candy ‘The children ate the candy’ d. *Comeu os meninos o doce. Ate-sg the children the candy ‘The children ate the candy’

168

Appendix: On the nature of agreement in European Portuguese

e. Comeram o doce os meninos. Ate-pl the candy the children ‘The children ate the candy’ f. *Comeu o doce os meninos. Ate-sg the candy the children ‘The children ate the candy’ (9)

a. Muitas florestas arderam. Many forests burnt-pl ‘Many forests burnt’ b. *Muitas florestas ardeu. Many forests burnt-sg ‘Many forests burnt’ c. Arderam muitas florestas. Burnt-pl many forests ‘Many forests burnt’ d. Ardeu muitas florestas. Burnt-sg many forests ‘Many forests burnt’

(colloquial)

Small clause predicates and passive participles also display obligatory number agreement:67 (10) a. As casas parecem bonitas. The houses seem-pl pretty-pl ‘The houses seem pretty’ b. *As casas parecem bonita. The houses seem-pl pretty-sg ‘The houses seem pretty’ c. As casas foram destruídas. The houses were destroyed-pl ‘The houses were destroyed’ d. *As casas foram destruída. The houses were destroyed-sg ‘The houses were destroyed’

The facts

169

Before presenting the subject-verb agreement pattern in the two dialects of Brazilian Portuguese, it is important to recall that the verbal paradigms are different, as extensively discussed in the literature, in particular in Galves (1993) and Figueiredo Silva (1996). In standard European Portuguese, there are five different combinations of the person and number features: (11) Verb cantar ‘to sing’ – present tense singular I canto II cantas III canta

plural cantamos cantam cantam

In BP1, the verbal paradigm consists of four different combinations of the same features, as shown in (12): (12) Verb cantar ‘to sing’ – present tense singular I canto II canta III canta

plural canta / cantamos cantam cantam

Analyzing these differences, Galves (1993) proposes that this dialect of Brazilian Portuguese has lost a semantic distinction for person, but not its syntactic feature. In fact, the author proposes that the pattern in (12) derives from a combination of binary features for person and number: (13) [+person, -number]

-o

[+person, +number]

-mos

[-person, -number]

-a

[-person, +number]

-m

Obviously, for deriving the European Portuguese pattern, a binary feature for person will not do, since it cannot account for the three-way distinction found in the singular. Independently of the details of Galves’ proposal, the crucial aspect to be kept in mind for this paper is that number is a distinctive feature in the verbal morphology of BP1.

170

Appendix: On the nature of agreement in European Portuguese

In BP2, the verbal paradigm is much more simplified. Taking the same case presented above for the other two variants, it is possible to observe, that there is just a distinction between first person and the rest: (14) Verb cantar ‘to sing’ – present tense singular I canto II canta III canta

plural canta canta canta

For the purposes of this appendix, the relevant aspect is that the difference between BP1 and BP2 may be linked to the role played by number. This is a relevant feature for distinguishing verbal forms in BP1, but not in BP2. In the latter, only person plays a role. Bearing this description in mind, let us now look at the subject-verb agreement patterns in the two dialects of Brazilian Portuguese. In BP1, there is number agreement between the subject and the verb: (15) a. Os menino comeram o doce. The-pl child-sg ate-pl the candy ‘The children ate the candy’ b. *Os menino comeu o doce. The-pl child-sg ate-sg the candy ‘The children ate the candy’ c. Os menino tossiram. The-pl child-sg coughed-pl ‘The children coughed’ d. *Os menino tossiu. The-pl child-sg coughed-sg ‘The children coughed’ Unlike what we did above for European Portuguese, for the sentences in (15), it is not possible to test whether the position of the subject is relevant for the pattern of agreement, since inversion is impossible with transitive and inergative verbs. However, if we test unaccusative verbs, it is possible to detect an interesting difference with respect to European Portuguese. Inverted

The facts

171

subjects do not agree. Agreement with an inverted subject is felt by speakers as posh, as if one is trying to mimic European Portuguese: (16) a. ??Queimaram muitas floresta. Burnt-pl many-pl forest-sg ‘Many forests burnt’ b. Queimou muitas floresta. Burnt-sg many-pl forest-sg ‘Many forests burnt’ Another interesting difference with respect to European Portuguese comes from predicative and passive constructions: as shown in (17), unlike in EP, there is subject-verb agreement, but the adjectival or participial form does not display number agreement with the subject: (17) a. ??As casa parecem bonitas. The-pl house-sg seem-pl pretty-pl ‘The houses seem pretty’ b. As casa parecem bonita. The-pl house-sg seem-pl pretty-sg ‘The houses seem pretty’ c. ??As casa foram destruídas. The-pl house-sg were-pl destroyed-pl ‘The houses were destroyed’ d. As casa foram destruída. The-pl house-sg were-pl destroyed-sg ‘The houses were destroyed’ BP2 significantly differs from BP1. There is no subject-verb number agreement in any context, neither with transitive and inergative verbs in SV order ,nor with unaccusative verbs in any order: (18) a. *Os menino comeram o doce. The-pl child-sg ate-pl the candy ‘The children ate the candy’ b. Os menino comeu o doce. The-pl child-sg ate-sg the candy ‘The children ate the candy’

172

Appendix: On the nature of agreement in European Portuguese

c. *Os menino tossiram. The-pl child-sg coughed-pl ‘The children coughed’ d. Os menino tossiu. The-pl child-sg coughed-sg ‘The children coughed’ (19) a. *Queimaram muitas floresta. Burnt-pl many-pl forest-sg ‘Many forests burnt’ b. Queimou muitas floresta. Burnt-sg many-pl forest-sg ‘Many forests burnt’ c. *Muitas floresta queimaram. Many-pl forest-sg burnt-pl ‘Many forests burnt’ d. Muitas floresta queimou. Many-pl forest-sg burnt-sg ‘Many forests burnt’ Predicative and passive constructions do not display any plural morphology either. The plurality of the sentence is just marked on the subject’s determiner. (20) a. *As casas parecem bonitas. The-pl house-pl seem-pl pretty-pl ‘The houses seem pretty’ b. *As casas parecem bonita. The-pl house-pl seem-pl pretty-sg ‘The houses seem pretty’ c. As casa parece bonita. The-pl house-sg seems-sg pretty-sg ‘The houses seem pretty’ d. *As casas foram destruídas. The-pl house-pl were-pl destroyed-pl ‘The houses were destroyed’

The facts

173

e. *As casa foram destruída. The-pl house-sg were-pl destroyed-sg ‘The houses were destroyed’ f. As casa foi destruída. The-pl house-sg was-sg destroyed-sg ‘The houses were destroyed’

1.3. Summary Before turning to the proposal made in Costa and Figueiredo Silva (2003a), let me sum up the descriptive conclusions obtained, and the issues to be addressed: i)

In European Portuguese, there is full DP-internal agreement;

ii)

In the two dialects of Brazilian Portuguese, number within the DP is marked either on the D head or in all prenominal elements;

iii) Prenominal adjectives and possessives differ in Brazilian Portuguese, in the sense that only the latter may carry number marking in the absence of plurality on the definite article; iv) In European Portuguese, there is full subject-verb agreement, independently of the position of the subject, except in the case of unaccusative verbs in which agreement is optional with inverted subjects in colloquial speech; v)

In European Portuguese, there is full number agreement with passive participles and small clause predicates;

vi) In BP1, there is full subject-verb agreement, except for the case of inverted subjects; vii) In BP1, there is no number agreement with passive participles and small clause predicates; viii) In BP2, there is no number subject-verb agreement; ix) In BP2, there is no number agreement with passive participles and small clause predicates.

174

Appendix: On the nature of agreement in European Portuguese

These descriptive statements raise at least the following questions: a) What is the difference between the plural marker in European and Brazilian Portuguese, allowing it to spread over all elements able to bear it only in the former? b) Why is it possible to find multiple agreeing elements in BP in the prenominal domain only? c) What is conditioning the difference of behavior between prenominal possessives and adjectives in BP? d) What is the difference between subject-verb agreement and DP-internal agreement, differentiating the two dialects of BP? e) What is the difference between subject-verb agreement and agreement with passive participles and small clause predicates, differentiating BP1 from EP? f) Why does inversion have an effect on subject-verb agreement in BP1, but not in EP?

2. Proposal The hypothesis put forward in Costa and Figueiredo Silva (2003a) is that the type of agreement morphology variation found is not a consequence of different landing sites for verb movement. In other words, given the evidence regarding verb movement, it is assumed that the number agreement facts are independent from the locus of verb movement. In fact, taking into account the traditional tests for tracing V-to-I movement, there are no significant differences between European and Brazilian Portuguese, that might be provide evidence for postulating different landing sites for the verb. As shown in Costa and Galves (2002), in both languages, the verb appears to move from V-to-T, not targeting the highest functional head of the IP-domain. This derives the fact that in both languages the verb may surface in between adverbs, and precede or follow floating quantifiers: (21) Ninguém provavelmente leu bem o poema. Nobody probably read well the poem ‘Nobody probably read well the poem’

(EP/BP)

Proposal

(22) a. Os menino(s) todos beijaram a Maria. The-pl child(ren) all kissed-pl Maria ‘The children all kissed Maria’

175

(EP/BP)

b. Os menino(s) beijaram todos a Maria. The-pl child(ren) kissed-pl all Maria ‘The children all kissed Maria’ Contenders of the hypothesis that Brazilian Portuguese has less verb movement than European Portuguese often base their argumentation on the morphological facts (cf. e.g. Ambar, Gonzaga and Negrão 2002). The data in (21) and (22) suggest that there is no positional evidence for this claim 68. The second assumption needed is to assume, following Halle and Marantz (1993), Bobaljik (1995), and Embick and Noyer (2001), that morphemes may attach to heads independently of movement. In other words, contra Belletti (1990), among others, we contend that a specific inflectional morpheme may surface on the verb as a consequence of syntactic head-movement, or as the result of a morphological process. We will further follow the tenets of Distributed Morphology in assuming that a morpheme may be realized in two ways: as a singleton or as a dissociated morpheme. According to Embick (1997) and Embick and Noyer (2001), a dissociated morpheme does not figure in syntax proper. It is inserted after Spell-Out, only indirectly reflecting syntactic structures. An application of this idea is proposed in Embick and Noyer (2001) in order to explain the distribution of definiteness markers in the Swedish DP. These markers attach to the nominal root, when there is no other material, but, if there is a prenominal adjective, they also obligatorily surface on the determiner: (23) a. Mus-en mouse-df ‘the mouse’

b. Den gamla mus-en the old mouse-df ‘the old mouse’

Embick and Noyer (2001) explain this behavior by postulating the following PF-requirements on the realization of N and D[+df]: (24) a. N must be marked for definiteness when D is [+def]. b. D[+def] must have a host. The compliance with these constraints may be established in the syntax. Nto-D movement satisfies both and derives the pattern in (23a). However, if

176

Appendix: On the nature of agreement in European Portuguese

an adjective is present, blocking N-raising for some reason, a determiner must be inserted in D for satisfying (24b). However, this is not sufficient for satisfying (24a). Since [+def] is a dissociated morpheme in Swedish, it can be inserted post-syntactically, attaching to the nominal head, and the constraint in (24a) is satisfied. This analysis of the Swedish facts enables envisaging the agreement between the determiner and the noun as not reflecting a specific syntactic configuration. Adopting for EP the assumption that the proliferation of a given morpheme may be interpreted as a consequence of it being dissociated, we make the following hypothesis: (25) Type of [plural] morphology in Portuguese: a. [plural] is a singleton in Brazilian Portuguese. b. [plural] is a dissociate morpheme in European Portuguese. A corollary of the hypothesis that [plural] may be a dissociated morpheme able to attach to roots independently of their position is that, for agreement to obtain, a Spec-head configuration may be required, although that is not obligatory. This observation conforms to Chomsky’s (2001) proposal that agreement may be triggered under different types of operation, and is empirically founded on the observation that identical spec-head configurations do not trigger overt agreement in all languages. A clear example is past-participle agreement with displaced objects in French, and the lack of it in other Romance languages (Kayne 1989): (26) a. Je les avait faites. (French) I them-acus-fem-pl had done-fem-pl ‘I have done them’ b. Eu tinha-as feito. (European Portuguese) I had them-acus-fem-pl done-masc-sg ‘I have done them’ According to these assumptions, the observed crosslinguistic variation follows straightforwardly. The difference between European and Brazilian Portuguese derives from the type of morpheme associated with plurality. We contend that [plural] is a dissociated morpheme only in EP. This proposal derives the fact that all elements able to bear (subject-agreeing) plural morphology will actually carry such markers, independently of the existence

Proposal

177

of movement and spec-head configurations. In BP, on the other hand, [plural] is not a dissociated morpheme. Therefore, it will attach to the element anchoring the information concerning number. Following Enç (1991), D is the head linking the DP to its LF-interpretation. Accordingly, the plural morpheme is realized on this head, which is able to carry this marker. Since [plural] is not a dissociated morpheme in BP, it will not surface in other categories. This derives the behavior displayed in BP2: there is a single marking of plural. As noticed above, under this type of approach, overt reflexes of Spechead configurations are possible, although not obligatory. This allows for deriving the difference between BP1 and BP2: in the former, but not in the latter, a Spec-head configuration between the subject and an inflectional head yields visible agreement, just like in the case of past participle agreement in French. The fact that the same effect does not obtain in BP2 is not surprising, given the independent evidence for parametrizing the overt effects of spec-head relations. In short, the account for the variation in the morphological expression of plurality may be summarized in the following schema: (27) pl > dissociate morpheme? > Yes (EP) > No (BP) Does Spec-head in I trigger overt agreement? > Yes (BP1, EP, French) > No (BP2) Does Spec-head in AgrO trigger overt agreement? > Yes (French) > No (Portuguese) With these ingredients, it is now possible to readdress the list of questions listed at the end of the previous section: a) What is the difference between the plural marker in European and Brazilian Portuguese, allowing it to spread over all elements able to bear it only in the former? Since [plural] is a singleton morpheme in Brazilian Portuguese, it will only surface in a single head. On the other hand, the fact that it is a dissociated morpheme in European Portuguese enables its postsyntactic association with all elements able to bear plural marking. b) Why is it possible to find multiple agreeing elements in BP in the prenominal domain only?

178

Appendix: On the nature of agreement in European Portuguese

The answer to this question does not entirely follow from the proposed analysis, since it involves doing some additional research on the internal structure of the DP. In any case, it is interesting to note that there is a correlation between the agreement facts in BP and the well-known, though poorly understood, asymmetry between prenominal and postnominal material as far as phrase structure is concerned. As discussed in DiSciullo and Williams (1987), among many others, prenominal material displays some properties typical of heads, resisting modification and not taking complements, while postnominal elements display properties typical of phrases. Independently of the precise implementation of these ideas, if morphology regards all prenominal material as a complex array of heads, it is expected that the only elements marked with plurality are the prenominal ones. By hypothesis, the plural morpheme could then be spreading from D onto the other units of the complex head. c) What is conditioning the difference of behavior between possessives and adjectives in BP? As shown above, if only one of the prenominal elements is marked with plurality, it must be the determiner in D-Adjective-N sequences, which follows from the hypothesis that [plural] attaches to the D head, which serves as an anchor to LF. However, if the sequence is D-Possessive-N, the element bearing the plural morpheme is the determiner, as repeated in (28): (28) A minhas casa the-sg my-pl house ‘My house’ This difference between possessives and adjectives follows from the assumptions concerning the locus of attachment of the plural morpheme. As just mentioned, it must be attached to the head responsible for establishing the link with semantic interpretation. As extensively argued in Castro (2001), two facts must be taken into account in order to understand the behavior of possessives in Portuguese: i) if they occur prenominally, the DP is definite, if they occur postnominally, the DP is indefinite (cf. 29); ii) the definite article in examples like (28) is expletive. Castro (2001) shows that the latter fact becomes very obvious, when it is taken into account that Brazilian dialects omitting the expletive determiner before proper names, also do so before prenominal possessives, as shown in (30) and (31):

Proposal

179

(29) a. O meu livro the my book ‘My book’ b. *O livro meu the book my ‘My book’ c. Um livro meu a book my ‘A book of mine’ d. *Um meu livro a my book ‘A book of mine’ (30) European Portuguese: a. *(O) João (the) João ‘John’ b. *(o) meu livro é azul. (the) my book is blue ‘My book is blue’ (31) Dialectal Brazilian Portuguese: a. (O) João (the) João ‘John’ b. (O) meu livro é azul. (the) my book is blue ‘My book is blue’ Since the definite article in definite possessive constructions is not the marker of definiteness, and is just an expletive, and since it is assumed that the [plural] morpheme anchors on the category codifying the information that is relevant for the interface with the interpretational component, it is expected that the number morphology surfaces on the possessive rather than on the expletive. d) What is the difference between subject-verb agreement and DP-internal agreement, differentiating the two dialects of BP?

180

Appendix: On the nature of agreement in European Portuguese

As spelled out above, DP-internal agreement and the visibility of subject-verb agreement stem from two different conditions: no dialect of BP is expected to exhibit D-N agreement, since the number morpheme is a singleton, and there is no Spec,head relation between D and N. On the other hand, the parametric specification schematized in (27) stipulates that the Spec,head relation between the subject and I triggers visible effects in BP1 only. These two sources for the visibility of agreement combined derive the fact that the two dialects of BP behave alike in the DP-internal domain, but differ in the subject-verb agreement patterns. e) What is the difference between subject-verb agreement and agreement with passive participles and small clause predicates, differentiating BP1 from EP? It was observed that, while in EP, number agreement surfaces everywhere, including in passive participles and small clause predicates, in BP1 there is only subject-verb agreement, but not agreement with the passive participles and small clause predicates. This difference in behavior straightforwardly follows from our analysis: since subject-verb agreement in BP1, unlike in EP, is a consequence of Spec,head agreement, we expect to find it only whenever the subjects stands in a Spec,head configuration with the agreeing head. Since within small clauses, there is no Spec,head relation between the subject and the predicate, and there is a head-complement relation between passive participles and the subject, the configurations needed for agreement to arise do not exist in these two constructions. In EP, on the contrary, since agreement is established via the postsyntactic insertion of the dissociated morpheme, number agreement arises independently of the type of configurational relation between the subject and the agreeing head. f) Why does inversion have an effect on subject-verb agreement in BP1, but not in EP? The answer to this last question is related to the previous one. Assuming with Belletti (1988) that inverted arguments of unaccusative verbs stay in their base-generated position, there is no Spec,head relation between the inflectional head and the subject, hence no visible agreement arises. In EP, on the contrary, since number agreement is a dissociated morpheme, independently of there being a Spec,head configuration, the verb and the subject may agree.

Proposal

181

At this point, it is important to understand what goes on in colloquial European Portuguese, in which, as mentioned above, the inverted subject of an unaccusative may not agree with the verb: (32) Ardeu muitas florestas. Burnt-sg many-pl forests ‘Many forests burnt’ Costa et al. (2002) argue that (32) is an instance of locative inversion. In that case, (32) exhibits Spec,head agreement with an expletive subject. In support of their claim, they show that lack of agreement shows up in cases of identificational focus, but not in cases of information focus: (33) A: Quem é que chegou? (EP) who is that arrived? ‘Who is arrived?’ B: a. ??Chegou os alunos. Arrived-sg the students ‘The students arrived’ b. Chegaram os alunos. Arrived-pl the students ‘The students arrived’ Looking at non-null-subject languages, like English, it is possible to know that expletive constructions and locative inversions are not used in information focus contexts: (34) A: Who is coming? B: a. John is coming. b. #There comes John. The interpretation Costa et al. (2002) make of these facts is the following: in information focus contexts, there is no expletive involved, which derives the obligatory agreement with the argument. In locative inversion contexts, there are two candidates for agreement, the expletive and the argument, and variation is found.69

182

Appendix: On the nature of agreement in European Portuguese

3. Further predictions Besides deriving the agreement facts presented in section 1, the present analysis makes the following interesting predictions with consequences for the treatment of other phenomena that we would like to emphasize: a) In EP, independently of the word order found, there is always full number agreement (Costa 1998), since agreement is not dependent on a specific syntactic configuration. This prediction has consequences for analyses of inversion. If this analysis is on the right track, the visibility of agreement must not be taken as a sign that at some point of the derivation the inverted subject stood in a Spec,head relation with the subject in the inflectional domain. b) Returning to the behavior of possessives in Brazilian Portuguese, an additional fact is correctly predicted by this analysis. Prenominal possessives display number agreement in BP1, while postnominal possessives do not, as shown in (35): (35) a. o meus livro the-sg my-pl book-sg ‘my books’

b. uns livro meu. some-pl book-sg my-sg ‘some books of mine’

Assuming with Schoorlemmer (1998), Castro (2001), among others, that only the prenominal possessive is a head related to D, while the postnominal possessive is an XP, it is expected that the singleton [plural] morpheme only attaches to the former. c) Scherre (1994) shows that in the variation found for the agreement patterns in the Brazilian dialects there are very many variables that do not fit well within a purely syntactic approach for deriving agreement patterns. Such variables include aspects such as phonological salience of the number marker and linear position. It is interesting to note that locating the analysis of agreement patterns at the interface between syntax and morphology, at the PF wing of the grammar, opens up a way for accommodating variables that are morphophonological in nature. d) Finally, this analysis makes an interesting prediction concerning the variation found within Romance languages as to what the target of V-to-I

Conclusion

183

movement is and its correlation with differences in number agreement inflection. One of the departure points leading this proposal was the observation that the differences between the three variants of Portuguese found in the verbal agreement pattern do not correlate with different landing sites for the verb. It was further shown that the differences between the three dialects were linked to the marking of number, since they all encode person differences, even if in different degrees. Let us consider Vikner’s (1997) generalization concerning the morphological evidence for there being V-to-I movement: (36) Vikner’s generalization: A language has V-to-I movement if there is Inflection for Person in all tenses. Since all three variants of Portuguese encode inflection for Person, Vikner’s generalization and this analysis correctly predict that there is no difference as far as the target of verb movement is concerned. It is further predicted that a given language may move the verb higher than the other and yet exhibit a weaker number morphology. Assuming with Costa (1996) and Costa and Galves (2001) that the targets of V-to-I movement are AgrS in French and T in European Portuguese, the prediction is correctly borne out. French is a language moving the verb higher than European Portuguese, and exhibiting less number agreement.

4. Conclusion The proposed analysis not only accounts for the variation described between the three varieties of Portuguese, but it also provides clear evidence for an autonomous morphological component, deriving morphological aspects of language in a way that is partly independent from syntax. Interestingly, it was crucial for these conclusions to be reached to relate the DP-internal agreement facts with subject-verb agreement. Assuming these results for number agreement, it may be hypothesized that the relevant part of inflection for the discussion carried out in chapter 6 concerns person inflection.

Notes

1. I am not adopting the radical position that all adverbs are adjoinable to different projections. Rather, I will follow a mixed approach, accepting that some adverbs have fixed positions and others may be adjoined to different projections. See Costa (1998, 2002) for a discussion of the relevant data. 2. In Costa and Galves (2002), it is argued that the same holds for Brazilian Portuguese. 3. See Costa (2000a) and Costa and Duarte (2002) for a comparison between the two analyses. 4. The non-adjacency between the subject and the verb is explained in terms of partial-V-movement. See chapter 1. 5. Duarte (1996) mentions that those speakers who accept (9b) only accept it if the preposed constituent is not doubled by a clitic, which favors an analysis of this construction in terms of topicalization, rather than clitic-left-dislocation. 6. Note that this case is different from the type of example discussed by Barbosa (2000), presented in (i): (i) a. Cresceu uma flor em todos os vasos. grew a flow in every pot b. ???Uma flor cresceu em todos os vasos. a flower grew in every pot Barbosa (2000) presents this contrast as an argument in favor of her hypothesis, since there is no a priori reason to consider that Spec,IP is incompatible with wide scope reading of the subject. However, the contrast in (i) just leads to the conclusion that, whenever there are two available positions for the subject, they can be used to codify different readings (see also Adger 1994; Costa 1998; among others), independently of the A- vs A-bar status of the positions. 7. As pointed out to me by P. Barbosa, for some reason, the verb telefonar ‘to call’ behaves differently: (i) O que é que aconteceu? a. O João telefonou. b. Telefonou o João. This is the pattern found for unaccusative verbs (Costa 2001). Assuming with Pinto (1997) that the verb telefonar behaves like unaccusatives in being able to select a temporal operator, enabling locative inversion, this behavior follows (see Costa and Figueiredo Silva 2003 for further arguments for this hypothesis).

186

Notes

8. In Costa (1997, 1998), a discussion of differences between definite and indefinite preverbal subjects is presented. Since the data turn out to be not very clear, I will not reproduce the discussion here. 9. See Lobo (2003) for an argument for not establishing a full parallel between these I-to-C contexts and conditional clauses headed by a complementizer. 10. Note that the ungrammaticality of (17b) is problematic not only for Barbosa’s analysis, but also to the claim put forward in this chapter that the two analyses are not incompatible. 11. As noted by a reviewer of Costa (2003), the existence of mixed systems is not enough to rule out the possibility that Barbosa’s analysis holds for languages in which it is not possible to distinguish between different types of pro. The point I am trying to make is that her analysis cannot be generalized to all types of null subject languages. 12. For a difference between the judgements by Ribeiro (2002) and corpus data, see Lobo (2003). 13. I am ignoring here the exact label of I. This may cause some confusion, when this discussion is confronted with the assumptions of chapter 1, since I argued for different landing sites for verbs in English, French and Portuguese. I decided to do so to make the discussions clearer. For clarity: I is the functional category where subjects land, that is, AgrSP in chapter 1. For the present discussion, the crucial fact is whether we need or do not need a landing site for the verb above the category where subjects move for case reasons. 14. VS order in embedded context is possible with all verbs and all types of main verbs, which makes it different from the contextually dependent embedded V2 phenomena in languages such as Icelandic and Yiddish (see McCloskey 1992 for contexts of CP-recursion and Rizzi 1997 for a specific proposal on split-CP). 15. Actually (18a) is grammatical, but the adverb gets a different (aspectual) meaning. For controlling for that factor, one may have another monosyllabic adverb, in the ruled in position, with a contradictory meaning and the sentence would remain grammatical: (i) Bem comeu mal o Paulo maçãs. 16. I am not going to speculate here on the mechanism which renders binding possible in this context. I would like to note anyway that it is very likely that the object QP is bound not by the right-dislocated subject but by the pro in subject position. 17. Speakers of Dutch disagree with respect to the possibility of scrambling PPs. There seems to be some variability with respect to the adverbs that intervene between the PP and the verb. I keep these examples in the text, since they are fully acceptable for some speakers, but I am aware that a more precise collection of data is in order. Thanks to Martin Honcoop, Ruben van de Vijver and Marga Petter for pointing this problem out to me.

Notes

187

18. Monosyllabic adverbs seem to be a reliable test for the distribution of arguments in Portuguese and in English, but not in Dutch. For this reason, I will contrast the sentences in Portuguese with Dutch sentences involving other types of adverbs. I have no explanation for why monosyllabic adverbs differ in behavior cross-linguistically. 19. See also Costa and Duarte (2003) for an alternative proposal regarding the variable status of the null object. 20. (55b) is possible if the adverb is modifying the adjective simpática ‘friendly’. 21. Deprez (1989) presents some data suggesting that scrambled DPs in German license parasitic gaps. See Doetjes (1997) and Costa (1998) for discussion of the status of FQs associated with objects. 22. Note the ungrammaticality of (i): (i) *O Pedro leu os livros bem todos. Pedro read the books well all 23. See example (71). 24. This argument only goes through under the assumption that remnant movement is not a root phenomenon, which must be the authors’ assumption since VOS interrogatives are possible in embedded contexts: (i) Perguntei a quem deu o livro o Pedro. (I) asked to whom gave the book Pedro 25. This argument only holds if the distribution of clitics is indeed syntactically triggered. Barbosa (2000) has suggested that enclisis is triggered any time a clitic is initial in a prosodic constituent, which would be the case in the sentences in (99). If the prosodic analysis proves true, this argument against the remnant movement analysis does not hold. 26. As mentioned above, I will call the languages in which the shift Adv-complement affects only NPs object-shift languages. These include Danish, Norwegian and Swedish, which move only pronouns and Icelandic and Faroese, which move NPs. When unnecessary, I will not distinguish the pronoun-shifting languages from the NP-shifting languages. 27. I am using the alternation specific/non-specific here for ease of explanation. It may be seen below that I will not consider specificity the trigger for scrambling, but for now it is enough to note that similar effects arise with NPs and adverbs. 28. A further problem for Diesing’s approach, pointed out to me by Sjef Barbiers, is that in a sentence like (i), the object must be analyzed as being outside VP, and the subject inside VP, which is obviously not feasible: (i) Er heeft iemand het pakje op tafel laten ligeen. there has someone the pack on the table let lay. 29. See Ladd (1996) for the several types of marked stress. 30. As extensively discussed in Reinhart (1995), this analysis captures the so-called specificity effects.

188

Notes

31. Note, however, that the combination of the two arguments leads to a surprising contradiction, since Hungarian focus movement does not yield WCO-effects, as noted by Brody (1995). 32. The relational nature of focus prevents an analysis in which the constituent containing the focused part is pied-piped. In most analyses involving focus movement (e.g. Ambar 1998), focused constituents move in order to form a domain separate from the part of the sentence that yields given information. If the two types of information occur in the same position because of pied-piping, this type of relation is lost. 33. Such a hypothesis is entertained in Chomsky (1995). It has been pointed out to me that the lack of multiple foci overtly moved in Hungarian may follow from a feature-checking approach. If there is only one strong focus feature in the F position in Hungarian, the other focused elements may stay in situ, on a pair with multiple questions in English. I am not sure whether this analysis may work, since as we will see later, Hungarian focus entails uniqueness and is not recursive in the same way English cleft constructions are not recursive: (i) *It was in front of the cinema that it was John that Eve waited for. If, as it will be discussed below and is argued in É. Kiss (1996), clefts are indeed the counterpart of Hungarian focus movement, it has to be explained why multiple-clefting and multiple focused elements are impossible, if multiple in-situ foci are possible. Furthermore, note that in Hungarian contrastive foci in situ are ruled out, even if there is a focused constituent moved (K. Polgardi (personal communication)). Thus, the parallel with wh-movement does not hold since: (i) there does not seem to be a language with multiple focus movement or with cooccuring moved contrastive focus and in-situ contrastive focus. 34. A related functional head has been proposed for Portuguese by Martins (1994), Uriagereka (1995), Raposo (2000), among others. 35. Note that it is not my goal here to present a solution to Rooth’s problems. The important point of Rooth’s theory which I will follow here is to eliminate Focusmovement from the set of LF-operations. That is, when foci move, they move independently of their being foci: the creation of bound variable readings forces them to be QR-ed. One may wonder what is the advantage of trading focus movement for QR. The advantage is: this QR is independently necessary, and applies to Quantifiers, not to foci without quantificational force. One of my goals in this study is to try derive as many word orders as possible without resorting to covert operations. The best evidence for covert operations comes from the behavior of quantifiers (May 1985), hence, it should be possible to reduce the evidence of some of the apparent focus movement constructions to independently necessary QR operations. 36. It is important to note that this does not mean that topic is the complement of focus. I will assume with Jäger (1996) and Büring (1995) among others a tri-

Notes

37.

38. 39.

40.

41.

42.

43.

189

partite structure for information: topic, focus and background. These analyses make the right prediction that a sentence may have only a focus, or no topic. See Büring and references therein for discussion. This implies that marked stress and rightmost placement are not competing strategies for marking focus. Instead, stress appears as a last-resort mechanism. See Costa and Figueiredo Silva (2003) for further arguments. See arguments in favor of interpreting focus also as a prosodic phenomenon in EP in Frota (1992). Keep in mind that this type of prominence and the one described in Frota’s work are distinct. In this section, no prosodic analysis of the prominence will be made. I will just propose an algorithm enabling the hearer to identify the locus of the main prominence. It is by no means my goal here to make any specific claims about the nature of the stress assigned to focused constituents. Her conclusions and my discussion are independent of the actual structure of VP, since she claims that stress on any constituent of VP is sufficient for enabling projection. Cinque’s algorithm is defined in terms of embedding, while here I use linear order. I will leave aside the discussion between embedding and linearity for the moment, returning to it when it will be relevant. A reviewer to Costa (2002a) points out to me that the algorithm in (49) does not apply to these cases, and that it should be defined in terms of rightmost word rather than constituent. I will stick to the definition in terms of constituent following the tradition in phrasal phonology and in prosody, since the predictions are exactly the same. The same reviewer argues that marked and unmarked stress assignment follow from different mapping conditions. This is not true, as discussed in this paragraph: in case of VSO with main stress on the subject, the focus set is the subject and everything it c-commands (the object), in case of SVO with main stress on the object, the focus set is this element (which is ambiguously the rightmost part of the NP, VP and IP as a consequence of iambic stress) and everything it c-commands (nothing). The crucial difference between the two cases is whether or not the constituent bearing main stress c-commands another constituent. When that is not the case, the effects of focus-projection arise. Crucially, the algorithm in (38) predicts both cases. The same reasoning applies to exclude a sentence-focus interpretation for a VSO sentence with unmarked stress. See the next two chapters for a formalization of this analysis in terms of markedness, and for an explanation of cross-linguistic variation in sentence-focus context. It has been pointed out to me that subject movement in this case corresponds to a case of defocusing yielding an unmarked word order, which is not common to defocusing operations. Note however, that defocusing in these cases also has the function of topic-promotion which very often is associated with subjects

190

44.

45.

46. 47.

48.

49.

50. 51.

52.

Notes being in their canonical Spec,IP position, given the natural tendency for subjects to be the topic of the sentence (cf Lambrecht 1994, Li 1976). Under the assumption that markedness arises as a consequence of a natural tendency not being met, subject topic-promotion should not yield a marked word order. Szendrói (2001) convincingly argues that, even for Hungarian, it is not necessary to assume a focus position in the left periphery. According to this author, the so-called focus movement is an instance of prosodic movement in the sense of Reinhart (1995). More subtle differences concerning the discourse function of preposed elements are given in Ambar (1998), who also shows that preposed elements may not be absolute new information, but argues for the existence of what she calls topic/focus in which there is fronting and some function overlap. Although such cases do not disconfirm the main observation that the function of focus-in-situ is different from the function of preposing, I think it is important to refer to them for completeness. See also Quer (2002) for arguments from Catalan to distinguish focus-movement from QP-fronting. It must be known why SVO is the only felicitous order when the whole sentence is focused. In Costa (1998, 2000c), this fact is analyzed within Optimality Theory as a case of Emergence of the Unmarked. In this work, it must be stipulated that, under equal circumstances, Move takes precedence over Agree. See Wurmbrand (2001) for a discussion of the choice between Move and Agree. These facts are crucial for deciding between this analysis and the proposal made in Costa (1998), in which inverted subjects were violating their licensing requirements, for satisfying a constraint on focus. Incidentally, these data provide an argument in favor of Hornstein’s (1999, 2000) hypothesis that control may be considered movement. I will not take a strong position regarding this issue, since the sentences in (182b,d) must involve PRO. For the status of these adverbs, see Castro and Costa (2003). As mentioned above, inversion n Brazilian Portuguese is restricted to unaccusative contexts. Crucially, it is important to note for the purposes of this section that inversion with copula is possible in Brazilian Portuguese. The lack of minimality effects in embedded contexts, as shown in (i) might lead to the conclusion that inversion in this type of context is an instance of movement of the predicate to Spec,IP rather than some type of topicalization: (i) a. Eu pergunto-me em que ano o rei serei eu. I wonder in which year the king will-be-1sg I b. *Eu pergunto-me que ano o rei verei eu. I wonder in which year the king will-see-1sg I

Notes

53.

54. 55.

56.

57. 58.

59.

60.

191

However, Tavares (in preparation) shows that a comparison between left-dislocated elements and inverted predicates reveals that they share some properties, permitting a treatment of inverted copular constructions as left-dislocation. The following problem arises: Brazilian Portuguese has raising verbs of the seem-type, allowing raising of the subject to the matrix IP, which provides evidence for the lack of CP. However, contrary to the prediction made here, if a copular verb is embedded under seem, the agreement still goes with the matrix verb: (i) O assassino parece ser eu. The murderer seems to be I Although the problem is obvious, it does not necessarily constitute a problem for the analysis, since this type of verbs allow super-raising in Brazilian Portuguese, even when CP is projected. Therefore, the raising of the subject in non-finite contexts is not compelling evidence for saying that CP is not projected in this language. For an analysis of agreement in Portuguese, see Costa and Figueiredo Silva (2003a). The data involving lack of agreement with postverbal subjects of unaccusative verbs come from colloquial speech. These constructions, although produced quite often by the speakers, are not accepted as grammatical in prescriptive grammars. As discussed in Costa and Figueiredo Silva (2003a), this assumption is only valid in a language or dialect in which nominative subjects trigger agreement independently of their placement in transitive and intransitive contexts. In Brazilian Portuguese, this assumption does not hold. (3b) is only legitimate with a topic intonation for the adverb. In most examples involving I-to-C movement, I will use auxiliary verbs so that I am able to control the position of the inverted subject. Leaving the participle behind allows for making sure that the subject surfaces to its left, and not in Spec,vP. The special status of adverbs remains unaccounted for. It is not clear why adverbs should not count as interveners. As Bobaljik (1995) discusses, this is a more general issue, however, since in other domains, adverbs seem to behave in the same way, in not disrupting adjacency relations. For a possible solution to this problem, see Bobaljik (2002). In Costa and Figueiredo Silva (2003a), it is argued that the number agreement morphology in Portuguese has the status of dissociated morphemes, in the sense of Embick and Noyer (2001). This assumption does not compromise the results of this chapter, since the scope of the discussion is both person and number agreement. Cf. the appendix to this chapter for discussion on the status of agreement in Portuguese.

192

Notes

61. The morpheme -va- of the imperfect in forms like fala-va-mos ‘we talked’ may be an aspectual morpheme rather than a tense morpheme, since the past of an imperfect may be expressed with an auxiliary construction, in which tense is expressed by the auxiliary verb. According to some authors, aspectual heads surface below T, and cyclic head-movement predicts that the aspectual morphology surfaces as an independent morpheme. It is likely that the same holds for future, an issue to be further explored. 62. The only cases in which other orders are found in idioms involve Heavy complements: (i) Dar a César o que é de César. Give to Caesar what belongs to Caesar 63. Since this is not a restructuring context, isomorphism with a matrix verb will not play any role. 64. The only cases in which there is apparent modification of a prenominal possessive can be argued not to instantiate of modification of an XP by another XP, but rather head adjunction (Castro and Costa 2003). 65. A question to be addressed is what the distinction is between Xº weak forms and Xº clitics. Note that the same question arises for the distinction between strong XPs and weak XPs. As already noticed by Cardinaletti and Starke (1999), the distinctions do not follow from categorial status alone. In the case at stake, we can hypothesize that the difference between clitics and weak Xº is prosodic. 66. Pereira (1993), cited by Scherre (1994), reports the following rates of agreement for one single literate speaker: talking to his boss

talking to his family

talking to his employees

verbal agreement

98%

91%

24%

DP-internal agreement

90%

52%

42%

This case shows that the same speaker masters the three dialects under discussion in this paper, using each one of them in different conversational situations, confirming our hypothesis that BP1 and BP2 are to be treated separately. 67. Costa and Pereira (2003) present evidence from agreement with the pronominal form a gente (lit.the people), which is grammatically specified for 3rd person singular, and referentially specified as a 1st person plural, showing that the number marker that is at stake is semantic number marking. 68. Even in BP2, a dialect using tudo, a non-inflected form of the quantifier all, we see the same possibilities of word order: (i) Os menino tudo beijou a Maria The-pl boy-sg all kissed-sg the Mary ‘The boys all kissed Mary’

Notes

193

(ii) Os menino beijou tudo a Maria The-pl boy-sg kissed-sg all the Mary ‘The boys all kissed Mary’ 69. Matters may turn out to be more complicated. Based on facts like (i), Tavares (in preparation) claims that in the presence of two candidates for agreement, the one entering in a Spec,head relation with I categorically agrees: (i) Neste jogo, tu {és/*sou} eu e eu {*és/sou} tu. In this game, you {are/am} I and I {are/am} you ‘In this game, you are me and I am you’ At this stage, it is however difficult to incorporate these results in the discussion of the patterns of agreement in unaccusative contexts, since in the latter, there is variation only in colloquial speech.

References

Adger, David 1994 Functional Heads and Interpretation. Doctoral dissertation, University of Edinburgh. Adragão, Maria do Mar 2001 Aquisição da inversão numa criança entre os dois e os três anos. Ms, Universidade Nova de Lisboa. Adragão, Maria do Mar and João Costa 2003 On the status of preverbal subjects in null-subject languages: evidence from acquisition. To appear in GALA 2003 Proceedings. Alexiadou, Artemis and Elena Anagnostoupoulou 1995 SVO and EPP in Null Subject Languages and Germanic. In: FAS Papers in Linguistics 4: 1–21, Potsdam. 1998 Parametrizing Agr: word order, verb- movement and EPP-checking. Natural Language and Linguistic Theory 16.3: 491–539. Ambar, Manuela 1992 Para uma Sintaxe da Inversão Sujeito-Verbo em Português. Doctoral dissertation, University of Lisbon. 1996 Aspects of Focus in Portuguese. paper presented at the International Workshop on Focus, Univ. Paris X. 1998 The syntax of focus in Portuguese: a unified approach. Ms, University of Lisbon. Ambar, Manuela and Jean-Yves Pollock 1998 Topique et Commentaire dans quelques constructions à inversion du sujet en français et portugais. Ms, University of Lisbon & CNRS, Lyon. Anderson, Steven 1972 How to get even. In Language, 48, 893–905. Baart, Joan 1987 Focus, Syntax and Accent Placement. Doctoral dissertation, Leiden University. Barbosa, Pilar 1995 Null Subjects. Doctoral dissertation, MIT. 2000 Clitics: a window into the null subject property. In: João Costa (ed.) Portuguese Syntax. New comparative studies. Oxford University Press, 31–92.

196

References

Bayer, Josef 1995 Directionality and Logica Form. On the scope of focussing particles and wh-in-situ. Kluwer Academic Publisher. Belletti, Adriana 1988 The Case of Unaccusatives. Linguistic Inquiry, 19: 1–34. 1990 Generalized Verb Movement: Aspects of Verb Syntax. Turin: Rosenberg and Sellier. Belletti, Adriana and Ur Shlonsky 1995 The order of verbal complements: A comparative study. Natural Language and Linguistic Theory 13: 489–526, Kluwer Academic Publishers. Bennis, Hans and Teun Hoekstra 1984 Gaps and Parasitic Gaps. The Linguistic Review, 29–87. Bianchi, Valentina & Silva, Maria Cristina Figueiredo 1994 On some properties of agreement object in Italian and Brazilian Portuguese. In: Michael Mazzola (ed.) Issues and Theory in Romance Linguistics. Selected papers from the Linguistic Symposium on Romance Languages XXIII, 181–197. Washington, D.C., Georgetown University Press. Bobaljik, Jonathan 1995 Morphosyntax: the syntax of verbal inflection. Ph.D. dissertation, MIT, Cambridge, Massachusetts. 2002 A-Chains at the PF-Interface: Copies and Covert Movement, in Natural Language and Linguistic Theory 20: 197–267. Bobaljik, J. D. and D. Jonas 1996 Subject Positions and the Roles of TP. In Linguistic Inquiry, 27: 195–236. Bobaljik, Jonathan David and Höskuldur Thrainsson 1998 Two heads aren’t always better than one. Syntax 1.1: 37–71. Bok-Benema, Reineke 1998 Remnant VP movement in Spanish. Paper presented at Going Romance, Utrecht. Boskovic, Zeliko 1997 The syntax of non-finite complementation. Cambridge, Mass.: MIT Press. Brandt, Patrick 1999 Scope, topichood, and double objects. In: Todirascu, Amalia (ed.) ESSLI Student Session 1999, Utrecht. Bresnan, Joan and Sam Mchombo 1987 Topic, pronoun and agreement in Chichewa. In: Language 63.4. 741–82.

References

197

Brody, Michael 1990 Remarks on the Order of Elements in the Hungarian Focus Field. In: I. Kenesei (ed.) Approaches to Hungarian, vol. 3, Szeged: JATE. 1995 Focus and Checking Theory. In: Approaches to Hungarian. Vol. 5, JATE. Büring, Daniel 1997 The 59th Bridge Accent: on the meaning of topic and focus. Routledge Studies in German Linguistics 3, Routledge. Cardinaletti, Anna 1998 On the Deficient/Strong Opposition in Possessive Systems. In: Possessors, Predicates, and Movement in the Determiner Phrase. Artemis Alexiadou & Chris Wider (eds.), 17–53. Amsterdam/Philadelphia: John Benjamins Publishing Company. Casielles, Eugenia 1996 Focus Preposing (it is called). Ms, UMass at Amherst. Castro, Ana 2001 Os possessivos em português europeu e português brasileiro: unidade e diversidade, Actas do XVI Encontro Nacional da Associação Portuguesa de Lingüística, Associação Portuguesa de Lingüística, Colibri, Portugal, 599–613. Castro, Ana and João Costa 2003 Weak forms as Xº: prenominal possessives and preverbal adverbs in Portuguese, in: A. T. Pérez-Leroux & Y. Roberge (eds.), Romance Linguistics: theory and acquisition, Amsterdam & Philadelphia, John Benjamins, 95–110. Chierchia, Gennaro 1991 Functional WH and weak crossover. In Proceedings of the 10th West Coast Conference on Formal Linguistics, Dawn Bates (ed.), 75–90. Stanford, CA: Center for the Study of Language and Information. Chomsky, Noam 1976 Conditions on Rules of Grammar. Linguistic Analysis, 2: 303–351. 1993 A Minimalist Program for Linguistic Theory. In: Ken Hale and Samuel J. Keyser (eds.) The View from Building 20. Cambridge, Massachusetts, MIT Press. 1995 The Minimalist Program. MIT Press, Cambridge, Mass. 1998 Minimalist Inquiries. MIT Occasional papers in Linguistics, MIT. 2000 Derivation by phase. MIT Occasional Papers in Linguistics, MIT. 2001 Beyond explanatory adequacy. Ms, MIT. Chomsky, Noam and Morris Halle 1968 The Sound Pattern of English. MIT Press, Cambridge, Mass.

198

References

Cinque, Guglielmo 1993 A Null Theory of Phrase and Compound Stress. In: Linguistic Inquiry, 24: 239–298. Colaço, Madalena 1998 Concordância parcial em estruturas coordenadas em Português europeu. In Actas do XIV Encontro Nacional da Associação Portuguesa de Linguística, Aveiro. Coelho, Izete, João Costa, Maria Cristina Figueiredo Silva and Sérgio Menuzzi 2001 Ordem VS e sujeito nulo em PE e PB. paper presented at 2º Colóquio do Projecto PE/PB, Universidade do Ceará, Fortaleza. Cornish, F. 2002 Locative inversion in 8 languages: syntax, semantics, discourse-pragmatics, and functional position. Paper presented at 10th International conference on functional grammar, University of Amsterdam. Corver, Norbert and Henk van Riemsdijk 1994 Studies on Scrambling. Mouton de Gruyter. Costa, João 1995 Adverbs as an Argument for OV in Dutch. Ms, HIL\Leiden University [presented at OV\VO meeting, Univ. of Amsterdam]. 1996 Adverb Positioning and V-movement in English: some more evidence. Studia Linguistica 50: 1, 22–34. 1997 Positions for Subjects in European Portuguese. In: Proceedings of WCCFL XV, CSLI, Stanford. 1998 Word order variation. A constraint-based approach. Doctoral dissertation, Leiden University (published by Holland Academic Graphics). 2000a Spec,IP vs. Deslocado: prós e contras das duas análises dos sujeitos pré-verbais. In D.E.L.T.A. 17.1: 283–303. 2000b Focus in situ: evidence from Portuguese. In Probus, 12–2, 187–228. 2000c The Emergence of unmarked word order. In: Géraldine Legendre, Jane Grimshaw and Sten Vikner (eds.), Optimality Theoretic Syntax, 171–203. Cambridge: MIT Press. 2001 Postverbal subjects and agreement in unaccusative contexts in European Portuguese, in: The Linguistic Review 18.1, 1–17. 2002 A multifactorial approach to adverb placement: assumptions, facts and problems. In: A. Alexiadou (ed.), Lingua: special issue on adverbs. 2002a Multiple focus in European Portuguese: apparent optionality and subject positions. In: C. Beyssade et al. Romance Languages and Linguistic Theory 2000, Amsterdam: Johns Benjamins, 93–108. 2002b VOS in Portuguese: Arguments against an analysis in terms of remnant movement. In: A. Alexiadou et al. (eds.). Dimensions of Movement. From features to remnants. Amsterdam: John Benjamins: 69–89.

References 2003

199

Null vs overt Spec,TP in European Portuguese. In: J. Quer et al. (eds.), Romance Languages and Linguistic Theory 2001, Amsterdam: John Benjamins, 31–47. Costa, João, Izete Coelho, Maria Cristina Figueiredo Silva, and Fátima de Oliveira 2002 Considerações sobre a Ordem VS e Sujeito Nulo em Português Europeu e Português Brasileiro. Paper presented at the PEPB Third Colloquium, University of Lisbon, Lisbon. Costa, João and Sandra Pereira 2003 Phases and Autonomous Features: a Case of Mixed Agreement in European Portuguese. Paper presented at the MIT Workshop on EPP and Phases, MIT, Boston. Costa, João and Charlotte Galves 2002 External subjects in two varieties of Portuguese: evidence for a nonunified analysis. In: C. Beyssade et al. Romance Languages and Linguistic Theory 2000, Amsterdam: Johns Benjamins, 109–125. Costa, João and Inês Duarte 2002 Preverbal subjects in null-subject languages are not necessarily leftdislocated. In Journal of Portuguese Linguistics, 159–176. 2003 Objectos nulos em debate. In: I. Castro and I. Duarte (eds.). Razões e emoção. Miscelânea de estudos em homenagem a Maria Helena Mira Mateus. Lisboa: Imprensa Nacional – Casa da Moeda. Costa, João and Maria Cristina Figueiredo Silva 2003 On the (in)dependence relations between syntax and pragmatics. Paper presented at GLOW Workshop on Information Structure, Lund. 2003a Nominal and verbal agreement in Portuguese: an argument for Distributed Morphology. Ms, Universidade Nova de Lisboa. Costa, J. & A. Gonçalves 1999 Minimal Projections: Evidence from Portuguese. Catalan Working Papers in Linguistics, 7. 59–69. Déprez, Viviane 1989 On the Typology of Syntactic Positions and the Nature of Chains. Doctoral dissertation, MIT. Diesing, Molly 1992 Indefinites. Cambridge, Massachusetts, MIT Press 1995 Yiddish VP Order and the Typology of Object Movement in Germanic. Ms, Cornell University. Diesing, Molly and Eloise Jelinek 1995 Distributing Arguments. In: Natural Language Semantics 3.2, 123–176. Dik, Simon C. 1978 Functional Grammar. Amsterdam: North-Holland.

200

References

DiSciullo, Anna-Maria and Edwin Williams 1987 On the definition of word, MIT Press, Cambridge, Mass. Den Dikken, Marcel 1995 Extraposition as intraposition, and the syntax of English. Tag questions. Ms, HIL\Vrije Universiteit Amsterdam. Doetjes, Jenny 1997 Quantifiers and Selection. Doctoral diss., HIL/Leiden University. Duarte, Inês 1987 A Construção de Topicalização na Gramática do Português. Doctoral Dissertation, University of Lisbon. 1996 A Topicalização no Português Europeu: uma análise comparativa. In: I. Duarte and I. Leiria (eds.), Actas do Congresso Internacional sobre o Português, APL/Colibri, Lisbon. 1997 Ordem de palavras: sintaxe e estrutura discursiva. In: Ana Maria Brito et alii (eds.), Sentido que a vida faz – estudos para Óscar Lopes. Campo das Letras. Duarte, Inês and Gabriela Matos 2000 Romance clitics and the Minimalist Program. In: João Costa (ed.). Portuguese Syntax. New Comparative Studies. Oxford University Press, 116–142. Duarte, Inês and Anabela Gonçalves 2000 Construções Causativas em Português Europeu e Português Brasileiro. Paper presented at PEPB – 2000, Universidade de Coimbra. Embick, David 1997 Voice and the interfaces of syntax, unpublished Ph.D. dissertation, University of Pennsylvania, Philadelphia. Embick, David and Rolf Noyer 2001 Movement Operations after Syntax. Linguistic Inquiry 32, 555–595. Enç, Muvet 1991 The Semantics of Specificity. Linguistic Inquiry 22, 1–25. Figueiredo Silva, Maria Cristina. 1996. A posição do sujeito em português brasileiro – frases finitas e infinitivas, Editora da Unicamp, Campinas, Brazil. Frank, R., Y-S. Lee and O. Rambow 1992 Scrambling as non-operator movement and the special status of subjects. In: S. Barbiers et al (eds.) Proceedings of LCJL 3, Leiden University. Frota, Sónia 1992 A Prosódia do Advérbio na Frase: interacção e convergência. In: Actas do VIII Encontro da Associação Portuguesa de Linguística – Lisboa 1992. APL, Colibri, Lisbon.

References 1994

201

Is Focus a Phonological Category in Portuguese. In: M. Schoorlemmer and P. Ackema (eds.), Console I Proceedings, Holland Academic Graphics. 1995 Focus, Phrasing, Stress and Accent in European Portuguese. Paper presented at the International Conference on Interfaces in Linguistics, Oporto. 1997 Focus in the Prosodic Interface: evidence from European Portuguese. talk presented at the University of Lisbon. 1998 Prosody and Focus in European Portuguese. Doctoral dissertation, University of Lisbon. Frota, Sónia and Marina Vigário 2002 Efeitos de peso no Português Europeu». In: Mateus, M. Helena; Correia, Clara N. (eds.), Saberes no Tempo. Homenagem a Maria Henriqueta Costa Campos. Lisboa: Colibri, 315–333. Galves, Charlotte 1993 O enfraquecimento da concordância no português brasileiro, in: Ian Roberts and Mary Kato (eds.), Português Brasileiro: uma viagem diacrônica, Editora da Unicamp, Campinas, Brazil. Givón, Talmy 1984 Syntax: a functional-typological introduction. Vol 1. John Benjamins. 1990 Syntax: a functional-typological introduction. Vol 2. John Benjamins. Gonçalves, Anabela 1999 Predicados verbais complexos em contexto de infinitivo não preposicionado em português europeu. Doctoral dissertation, University of Lisbon. Gonzaga, Manuela 1997 Aspectos da sintaxe do advérbio em português. Mass. dissertation, Universidade de Lisboa. Halle, Morris and Alec Marantz 1993 Distributed Morphology and the Pieces of Inflection. In: K. Hale and S. J. Keiser (eds.), The View from Building 20: Essays in Linguistics in Honor of Sylvain Bromberger, MIT Press, Cambridge, Mass., 111–176. Holmberg, Anders 1986 Word Order and Syntactic Features. Doctoral dissertation, University of Stockholm. de Hoop, Helen 1992 Case Configuration and Noun Phrase Interpretation. Doctoral dissertation, Groningen. de Hoop, Helen and Henriette de Swart 2000 Topic and Focus, L. Cheng en R. Sybesma (eds.), The First Glot International State-of-the-Article Book, Mouton de Gruyter, Berlin/New York, 105–130.

202

References

Hornstein, Norbert 1994 Logical Form. From GB to Minimalism . Blackwell, Oxford. 1999 Movement and Control, Linguistic Inquiry, 69–96. Horvath, Julia 1986 FOCUS in the Theory of Grammar and the Syntax of Hungarian. Foris, Dordrecht. 1995 Structural Focus, Structural Case, and the Notion of Feature Assignment. In: Kiss (1995). Huang, James 1982 Logical Relations in Chinese and the Theory of Grammar. Doctoral dissertation, MIT. Iatridou, Sabine 1990 About Agr(P). Linguistic Inquiry, 21: 551–577. Jackendoff, Ray 1972 Semantic Interpretation in Generative Grammar. Cambridge, Mass., MIT Press. 1990 Semantic Structures, MIT Press, Cambridge, Mass. Jäger, Gerard 1996 Topic, Scrambling and Aktionsart. In: Laura Brugé et alii (eds.), Proceedings of ConSOLE 3. Leiden: Leiden University. Johnson, Kyle 1991 Object Positions. In Natural Language and Linguistic Theory 9: 577–636. Kayne, Richard 1984 Connectedness and binary branching. Dordrecht: Foris. 1989 Facets of Past Participle Agreement in Romance, in: Paola Benincà (ed.), Dialectal Variation and the Theory of Grammar, Foris, Dordrecht, 85–103. 1994 The Antisymmetry of Syntax. Cambridge, Massachusetts, MIT Press. 1998 Overt vs. Covert Movement. Ms, New York University Kayne, Richard and Jean-Yves Pollock 1998 New Thoughts on Stylistic Inversion. Paper presented at the Workshop on Inversion in Romance, University of Amsterdam. Kiss, Katalin 1995 Discourse Configurational Languages. New York: Oxford University Press. 1996 Focus Operator and Information Focus. Working Papers in the Theory of Grammar, Hungarian Academy of Sciences . Koopman, Hilda and Dominique Sportiche 1983 Variables and the Bijection Principle, in TheLinguistic Review 2.3, 139–160. 1991 The position of subjects in Lingua, 85.1, 211–258.

References

203

Koster, Jan 1987 Domains and dynasties. Foris, Dordrecht. Krifka, Manfred et alii 1995 Introduction to Genericity. In The Generic Book, 1–124. Kroch, Anthony 1994 Morphosyntactic Variation. In: K. Beals (ed.), Proceedings of the Thirtieth Annual Meeting of the Chicago Linguistics Society 2, Chicago Linguistics Society, Chicago, 180–201. 2001 Syntactic Change. In: Mark Baltin and Chris Collins (eds.), The Handbook of Contemporary Syntactic Theory, Blackwell, UK, 699–729. Ladd, Robert 1996 Intonational Phonology. Cambridge University Press. Lambrecht, Knud 1994 Information Structure and Sentence Form. Cambridge University Press. Larson, R. K. 1988 On the double object construction. Linguistic Inquiry 1/9: 335–391. Lee, Young-Suk and Beatrice Santorini 1994 Towards resolving Webelhuth’s paradox: evidence from German and Korean. In: Norbert Corver and Henk van Riemsdijk (eds.), Studies on scrambling. Movement and non-movement approaches to free wordorder phenomena (Studies in generative grammar 41), 257–300. Berlin: Mouton de Gruyter. Levin, B. e M. Rappaport Hovav 1995 Unaccusatives. Cambridge, Mass.: MIT Press. Li, Charles 1976 Subject and Topic. Academic Press, New York. Lobo, Maria 2003 Aspectos da Sintaxe das Orações Subordinadas Adverbiais do Português, Ph.D. Dissertation, Universidade Nova de Lisboa. Mahajan, Anoop 1990 The A/A-bar distinction and Movement theory. Ph.D. dissertation, MIT. Marantz, A. P. 1984 On the Nature of Grammatical Relations, MIT Press, Cambridge, Mass. Martins, Ana Maria 1994 Os clíticos na história do português. Doctoral dissertation, University of Lisbon. Mateus, Maria Helena et alii 1989 Gramática da Língua Portuguesa. Lisboa: Editorial Caminho. May, Robert 1985 Logical Form. MIT Press, Cambridge, Mass.

204

References

McCloskey, James 1992 Adjunction, Selection and Embedded Verb Second. Ms, University of Santa Cruz. 1992a On the Scope of Verb Movement in Irish. Ms, UCSC. Menuzzi, Sérgio 1994 Adjectival positions inside DPs. In: Crit Cremers and Reineke BokBenema (eds.), Linguistics in the Netherlands, John Benjamins, Amsterdam, 127–138. 1994 On the Role of F-features: empty categories, binding and the pronominal system in Brazilian Portuguese. Ms, HIL\Leiden University. Müller, Gereon and Wolfgang Sternefeld 1994 Scrambling as A-bar Movement. In: Corver & Riemsdjik 1994. Nash, Lea 1995 Argument scope and Case Marking in SOV and in Ergative Languages: the case of Georgian. Doctoral dissertation, Univ. Paris 8. Neeleman, Ad 1994 Complex Predicates. Ph.D. dissertation, Utrecht University. Neeleman, Ad and Tanya Reinhart 1999 Scrambling and the PF-interface. In: W. Geuder and M. Butt (eds.). The projection of arguments: lexical and compositional factors. CSLI Publications, Stanford, 309–353. Nespor, Marina and Irene Vogel 1986 Prosodic Phonology. Foris, Dordrecht. Nespor, Marina, Teresa Guasti and Anne Cristophe 1995 What can Infants Learn from Prosodic Constituents? In GLOW Newsletter, 34. Ordoñez, Francisco and Esthela Treviño 1995 Los sujetos y objetos preverbales en español. Paper presented at the 5th Colloquium on Generative Grammar, Coruña, Spain. Ordoñez, Francisco 1997 Word Order and Clause Structure in Spanish and other Romance languages. Doctoral dissertation, CUNY. Ouhalla, Jamal 1991 Functional Categories and Parametrization. Oxford: Routledge. Pesetsky, David 1982 Paths and Categories. Doctoral dissertation, MIT. 1987 Wh-in-Situ: Movement and Unselective Binding. In: E. Reuland and A. ter Meulen (eds.). The Linguistic Representation of (In)definiteness. 98–129, Cambridge, MIT Press. 1989 Language Particular Processes and the Earliness Principle. Ms, MIT, Cambridge, Mass.

References

205

Pesetsky, David 1995 Zero syntax. MIT Press, Cambridge, Mass. Phillips, C. 1996 Order and Structure. Doctoral dissertation, MIT. Pinto, Manuela 1994 Subjects in Italian: Distribution and Interpretation. In: Reineke BokBennema and Crit Cremers (eds.), Linguistics in the Netherlands 1994, 175–187, Amsterdam, John Benjamins Publishing Company. 1997 Licensing and Interpretation of Inverted Subjects in Italian. Doctoral dissertation, OTS/Utrecht University. Pollock, Jean-Yves 1989 Verb Movement, Universal Grammar and the structure of IP. Linguistic Inquiry, 20: 365–424. 1994b Notes on Checking Theory, pro-drop, Free Variation and Economy. Paper presented at the Going Romance VIII, Utrecht. Pratas, Fernanda 2002 O sistema pronominal do caboverdiano. Questões de gramática. Mass. dissertation, Universidade Nova de Lisboa. Pustejovsky J. 1995 The Generative Lexicon. Cambridge: MIT Press Quer, Josep 2002 Edging Quantifiers: On QP-Fronting in Western Romance. In: C. Beyssade et al., Romance Languages and Linguistic Theory 2000, Amsterdam: Johns Benjamins, 253–270. Raposo, Eduardo 1986 On the Null Object in European Portuguese. In: O. Jaegli and SilvaCorvalan (eds.), Studies in Romance Linguistics, 373–390. 1987 Case Theory and Infl-to-Comp: the Inflected Infinitive in European Portuguese. Linguistic Inquiry, 20: 85–110. 1994 Affective Operators and Clausal Structure in European Portuguese and European Spanish. Paper presented at 24th Linguistic Symposium on Romance Languages, UCLA/USC. 2000 Clitic Position and Verb Movement in European Portuguese. In: João Costa (ed.), Portuguese Syntax. New Comparative Studies. Oxford University Press, 266–298. Reinhart, Tanya 1995 Interface Strategies. Ms, OTS/Utrecht University. Ribeiro, Raquel 2002 As ocorrências da forma de gerúndio na variedade padrão e numa variedade dialectal do português europeu. Mass. Dissertation, Universidade Nova de Lisboa.

206

References

Rizzi, Luigi 1982 Issues in Italian Syntax. Foris, Dordrecht. 1991 Residual Verb Secondand the Wh-criterion. Technical Reports in Formal and Computational Linguistics, 2. 1997 The Fine Structure of the Left Periphery. In: Haegeman (ed.), Elements of Grammar. Handbook in generative syntax. Kluwer, Dordrecht, 281–337. Rochemont, Michael and Peter Culicover 1990 English Focus Construction and the Theory of Grammar. Cambridge, England: Cambridge University Press. Rohrbacher, Bernard 1994 The Germanic Languages and the Full Paradigm: A Theory of V to I Raising, unpublished Ph.D. dissertation, University of Massachusetts, Amsherst, Mass. Rooth, Mats 1985 Association with Focus. Doctoral dissertation, University of Massachusetts, Amsherst, Mass. Ross, John 1967 Constraints on variables in syntax. Doctoral dissertation, MIT. Santorini, B. 1990 ÎNFL and scrambling in German. Ms, University of Pennsylvania. Scherre, Marta 1994 Aspectos da concordância de número no português do Brasil. Revista Internacional de Língua Portuguesa – Norma e Variação do Português 12, 37–49. Schoorlemmer, Martin 1998 Possessors, articles, and definiteness’. In: Artemis Alexiadou and Chris Wilder (eds.), Possessors, predicates and movement in the Determiner Phrase. John Benjamins, Amsterdam, 55–86. Selkirk, Lisa 1984 Phonology and Syntax: the relation between sound and structure. MIT Press, Cambridge, Mass. Sportiche, Dominique 1988 A Theory of Floating Quantifiers and Its Corollaries For Constituent Structure. Linguistic Inquiry, 19: 425–449. von Stechow, Arnim 1990 Current Issues in the Theory of Focus. In: Semantics. International Hanbook of Contemporary Research. Mouton de Gruyter, Berlin, 804–825. Szabolcsi, Anna 1981 The Semantics of Topic-Focus Articulation. In Formal Methods in the Study of Language.

References

207

Szendrói, Kriszta 2001 Focus and the Syntax-Phonology Interface. Ph.D. dissertation, University College, London. Uriagereka, Juan 1995 An F Position in Western Romance. In: Kiss (1995). Vallduví, Enric 1990 The Informational Component. Doctoral dissertation, University of Pennsylvania. Valmala Elguea, Vidal 1994 Spanish Word-Order and Checking of Morphological Features. Talk given at Going Romance 8, Utrecht. Vanden Wyngaerd, Guido 1990 Object Shift as an A-movement Rule. In MIT Working papers in linguistics. Vikner, Sten 1994 Scandinavian Object Shift and West Germanic Scrambling. In: Corver & Riemsdijk (1994). 1997 Vº-to-Iº movement and inflection for person in all tenses. In: Liliane Haegeman (ed.): The New Comparative Syntax. Longman Linguistics Library, Edinbourgh, UK, 189–213. Webelhuth, Gert 1989 Syntactic Saturation Phenomena and the Germanic Languages. Doctoral dissertation, UMass, Amherst Williams, Edwin 1994 A reinterpretation of the evidence for verb movement in French. In: D. Lightfoot and N. Hornstein (eds.), Verb Movement. Cambridge University Press, 189–206. Wurmbrand, Susi 2001 Infinitives. Mouton de Gruyter, Berlin. Zubizarreta, Maria Luisa 1998 Word Order, Prosody, and Focus. MIT Press, Cambridge. 1998a The Structure of the Higher Middle Field: the position of the verb and the subject. Paper presented at the Workshop on Inversion in Romance, University of Amsterdam. Zwart, Jan-Wouter 1993 Dutch Syntax. A Minimalist Approach. Ph.D. dissertation, University of Groningen.

Index

A-bar, 8, 13–15, 32, 44–45, 56, 63, 65, 69, 73, 144–145, 150–152, 185, adjacency, 4, 8, 14, 36, 40, 52, 53, 57, 75–76, 129, 131, 133–136, 185, 191 adjunction, 2, 14, 27, 29, 31, 36, 40, 45–46, 50, 68, 74, 99, 129, 144, 160, 192 adverbs, 2, 6–10, 14, 23, 25–29, 32, 36, 39–41, 43–44, 48, 50, 63–64, 66–69, 96–97, 129–130, 134–137, 159, 163, 174, 185, 186, 187, 190, 191 Agree, 1, 3, 71, 97–100, 102–105, 124, 159, 190 agreement, 3, 12, 75, 77, 100–105, 109– 119, 163–183, 191, 192, 193 AgrP, 1–5, 6, 20–22, 39, 45–46, 63, 99, 131–139, 159–160 anaphor, 33, 44, 143 A-position, 2–3, 6, 15–16, 37–38, 45, 107, 138–139, 145, 150, 160 auxiliary verbs, 7, 24, 26, 29, 48, 49, 134, 137, 191, 192 binding, 32–33, 37–38, 44, 73, 143–146, 149–153, 186 Brazilian Portuguese, 20, 100–105, 113–115, 118–119, 142, 155–156, 163–183, 185, 190, 191 Capeverdean, 20–21 Case, 1, 3, 21, 39, 45, 47, 62–65, 76, 97, 99, 105, 108, 112–118, 127, 159, 186 c-command, 58–59, 65, 87–91, 149, 189 Celtic, 24 cleft, 48–49, 52, 87, 188

climbing, 97, 100, 102–104 clitic left dislocation, 12, 21, 185 complementizer, 19, 26, 186 contrastive, 8, 55–56, 72–74, 80–82, 85–86, 89, 94–97, 124–125, 188 control, 22, 104, 150, 190 definite, 3, 8, 17–18, 34, 62, 65, 81, 85, 110, 114, 127, 153–155, 173, 175, 178–179, 186 Distributed Morphology, 131–133, 175 ditransitive, 141–151, 155–156 D-linking, 120, 125–126 doubling, 18–21, 34, 48, 52–54, 126 Dutch, 8, 24, 35–45, 63, 67–69, 133, 186, 187 ellipsis, 122, 125 English, 7, 9, 14, 19, 24, 27–28, 39–41, 46, 67, 72–73, 75–77, 83–84, 101–102, 114–115, 132–133, 135, 141–142, 181, 186, 187, 188 EPP, 12, 20–21, 127 exclamative, 91–95 exhaustivity, 2, 4, 16, 72, 120–125, 160 expletive, 20–21, 24, 29, 113–115, 117– 119, 132–134, 138, 178–179, 181 floating quantifiers, 9, 45, 48, 51–52, 163, 174, 187 focus, 1–5, 8, 18, 54–56, 58, 64–69, 74– 105, 107–108, 112–115, 119–121, 124–127, 141–145, 153–155, 159– 160, 181, 188, 189, 190 focus-movement, 9, 71–78, 91–96, 154, 190 French, 6–7, 9, 29, 47, 54, 176, 177, 183, 186

210

Index

fusion, 131–132, 136 generic, 124–125 German, 24, 35–45, 63, 65–69, 102, 133, 187 Germanic, 3, 4, 39–45, 62, 134–135, 138 Greek, 12, 22, 46 Head Movement Constraint, 135 head-movement, 29, 131–136, 175, 192 Heavy NP shift, 31, 74 Hungarian, 72–76, 78, 91, 125, 188, 190 Icelandic, 24, 28, 45, 63, 102, 132–133, 186, 187 if-clauses, 19 indefinite, 8, 17, 34, 56, 62, 65–67, 81, 126–127, 178, 186 indirect object, 52–53, 142, 145, 149–150 infinitives, 6–7, 137 inflected gerunds, 22 Information Structure, 1, 3, 5, 71–105, 159 intransitive, 16, 64, 108–109, 111, 117, 191 inversion, 3, 4, 16, 20–27, 47, 54, 71–105, 108, 111–119, 129, 142, 155, 170, 174, 180–182, 185, 190 Italian, 8–9, 14, 30, 64, 91–93, 113, 153–155 I-to-C, 2–4, 14, 19, 23–25, 129–131, 135–137, 186, 191 language acquisition, 21 left-dislocation, 1–4, 6, 8,11–22, 35,107, 118, 120–127, 159–160, 185, 191 locality, 4, 97–103, 141, 159 locative inversion, 113–119, 181, 185 manner, 27, 66, 129, merger, 2, 131–135, 160, 163 minimality, 14–15, 190

morphology, 2, 4–5, 117, 129–139, 160, 163–183, 191 Move, 1, 3, 97, 105, 159, 190 negation, 14, 98, 103, 122–123 nominative, 32, 47, 108, 112–113, 115– 118, 191 nuclear stress, 1, 58, 67–69, 81, 90, 141– 142, 159, 189 null object, 42–43, 187 null subject, 2–4, 11–22, 118, 138, 160, 181, 186 object shift, 45, 63, 99, 187 optionality, 2, 4, 6, 12, 97, 107–127, 160 OSV, 1, 11, 13, 79 OVS, 1, 11, 79 parasitic gap, 37–38, 42–43, 45, 56, 187 participle, 26, 29, 168, 171, 173–174, 176–177, 180, 191 partitive, 113, 116–117 passive, 21, 33, 37, 145, 168, 171–174, 180 phase, 4, 97–102, 159 possessive, 33, 44, 141, 153–155, 165– 167, 173–174, 178–179, 182, 192 postverbal, 3–4, 19–21, 23–70, 75, 96, 100, 109–118, 122, 191 PP, 27, 36, 39, 43, 45, 75, 85, 88, 93, 96, 142, 148–149, 151, 186 pragmatic, 38 predicative, 21, 37, 43–44, 171–172 preposing, 13, 48–50, 72, 92–96, 190 preverbal, 2–4, 11–22, 35, 53, 60, 100, 107–118, 121, 122, 125–127, 129, 136, 155, 160, 186 pro, 20–21, 115, 119, 156, 186 PRO, 150, 190 pro-drop, 20 pronominal, 12, 18–22, 30–32, 34, 37, 45, 48, 52–53, 56, 73, 77, 85, 100, 116–117, 126, 130, 154, 187, 192

Index prosody, 1, 3, 5, 58, 78, 83, 88, 105, 141, 153, 159, 189

211

SVO, 1–4, 11–22, 39, 59, 78, 81–83, 86–90, 97, 107, 109, 119–127, 144, 159–160, 189, 190

QP, 14, 33, 37, 44, 65, 186, 190 raising, 17, 22, 65, 191 reconstruction, 15, 59, 151 remnant movement, 3, 35, 47–62, 187 right-adjunction, 27 Scandinavian, 45, 63 scope, 15, 48, 58–59, 65–69, 146, 185 scrambling, 3, 8, 30, 33, 35–69, 79, 90, 99, 144–145, 150, 186, 187 small clause, 37, 43, 168, 173–174, 180 Spanish, 12, 22, 46–47, 55 Spec,IP, 3, 5, 9, 11–22, 23–27, 35, 46– 47, 51–52, 60–61, 64–65, 71, 78–79, 88, 97, 101, 103, 108, 113, 116–119, 124, 126–127, 144, 156, 185, 190 Spec,VP, 1, 3, 5, 16, 20, 23–35, 38, 47–67, 71, 78–79, 97–99, 105, 144, 151, 159, 191 specificity, 38, 62–63, 65, 126–127, 187 Spell-Out, 5, 175 stress, 1, 3, 8, 14, 55, 58, 67–69, 78, 81–91, 141–143, 153–157, 159– 160, 187, 188, 189 subject-oriented, 26–27, 129–130, 136– 137

tag, 31, 33, 34, 48, 52–53, 95 topicalization, 12, 56–58, 61–62, 79, 81, 94–65, 126, 136–137, 185, 190 TP, 1–5, 26–27, 47–48, 52, 59, 103–104, 129–139, 159–160 Transitive Expletive Constructions, 24, 132–134, 138 unaccusative, 20–22, 64–65, 107–119, 155, 167, 170–173, 180–181, 185, 190, 191, 193 unmarked, 11, 15–16, 21, 81–90, 107– 108, 112, 145–147, 189, 190 verb movement, 4, 7–9, 27, 39–40, 133, 135, 163–164, 174–175, 182–183 Verb-second, 24, 133, 186 VOS, 1, 3, 11, 23, 30–70, 79, 82, 84, 87–88, 90, 109, 144–145, 150– 152, 187 VSO, 1, 3, 4, 11, 21, 23–35, 78, 82, 87– 90, 97, 99, 109, 119–127, 144– 145, 159, 189 wh-questions, 4, 18, 27, 47, 55–56, 58, 93, 107, 119–127, 130, 135, 160, 187