Parametric Variation: Null Subjects in Minimalist Theory

This page intentionally left blank

Parametric Variation

Parametric variation in linguistic theory refers to the systematic grammatical variation permitted by the human language faculty. Although still widely assumed, the parametric theory of variation has in recent years been subject to re-evaluation and critique. The Null Subject Parameter, which determines among other things whether or not a language allows the suppression of subject pronouns, is one of the best-known and most widely discussed examples of a parameter. Nevertheless its status in current syntactic theory is highly controversial. This book is a defence of the parametric approach to linguistic variation, set within the framework of the Minimalist Program. It discusses syntactic variation in the light of recent developments in linguistic theory, focusing in particular on issues such as the formal nature of minimalist parameters, the typology of null-subject language systems and the way in which parametric choices can be seen to underlie the synchronic and diachronic patterns observed in natural languages. theresa biberauer is Senior Research Associate in the Department of Linguistics at the University of Cambridge and Senior Lecturer Extraordinary at Stellenbosch University, South Africa. anders holmberg is Professor in the School of English Literature, Language and Linguistics at Newcastle University. ian roberts is Professor in the Department of Linguistics at the University of Cambridge. michelle sheehan is Research Associate in the Centre for Research in Linguistics and Language Sciences at Newcastle University.

Parametric Variation: Null Subjects in Minimalist Theory Theresa Biberauer University of Cambridge

Anders Holmberg Newcastle University

Ian Roberts University of Cambridge

and

Michelle Sheehan Newcastle University

CAMBRIDGE UNIVERSITY PRESS

Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, São Paulo, Delhi, Dubai, Tokyo Cambridge University Press The Edinburgh Building, Cambridge CB2 8RU, UK Published in the United States of America by Cambridge University Press, New York www.cambridge.org Information on this title: www.cambridge.org/9780521886956 © Cambridge University Press 2010 This publication is in copyright. Subject to statutory exception and to the provision of relevant collective licensing agreements, no reproduction of any part may take place without the written permission of Cambridge University Press. First published in print format 2009 ISBN-13

978-0-511-77011-1

eBook (NetLibrary)

ISBN-13

978-0-521-88695-6

Hardback

Cambridge University Press has no responsibility for the persistence or accuracy of urls for external or third-party internet websites referred to in this publication, and does not guarantee that any content on such websites is, or will remain, accurate or appropriate.

Contents

Acknowledgements Introduction: parameters in minimalist theory ian roberts and anders holmberg

page vi 1

1

A deletion analysis of null subjects ian roberts

58

2

Null Subject Parameters anders holmberg

88

3

Control into finite clauses in partial null-subject languages anders holmberg and michelle sheehan

125

4

Semi null-subject languages, expletives and expletive pro reconsidered theresa biberauer

153

The null generic subject pronoun in Finnish: a case of incorporation in T anders holmberg

200

6

‘Free’ inversion in Romance and the Null Subject Parameter michelle sheehan

231

7

Subjects, Tense and verb-movement theresa biberauer and ian roberts

263

8

Varieties of French and the Null Subject Parameter ian roberts

303

5

Bibliography Index

328 357

v

Acknowledgements

The research for this book was carried out under the auspices of the project ‘Null subjects and the structure of parametric theory’, 2002–2007, funded by a grant from the AHRC. We’d like, first and foremost, to thank the two other members of the project team, Chris Johns and David Willis, for their contribution to this research. A very large number of people have contributed in various ways, by comments on earlier versions of the papers, or by taking part in workshops we have organised, or by providing data, or organising events where our work has been disseminated, and so on. Thanks to Matthew Whelpton and Jo´hannes Gisli Jo´nsson who held the Null Subjects and Parameters workshop in Reykjavik in 2003, and Peter Svenonius and the CASTL linguists, who hosted another workshop in Troms in the dark of winter 2004. The following people were guest speakers at workshops that we organised on the project theme in Cambridge, Durham or Newcastle: Pilar Barbosa, Adriana Belletti, Cristina Guardiano, Anna Cardinaletti, Paola Crisma, Roberta D’Alessandro, Bob Freidin, Tomo Fujii, Chiara Gianollo, Elena Herburger, Adam Ledgeway, Pascual Masullo, Marjorie McShane, Gereon Mu¨ller, Aarti Nayudu, Marco Nicolis, Sandra Paoli, Mamoru Saito, Ur Shlonsky, Halldor Sigurðsson, loanna Sitaridou, Tatsuya Suzuki, George Tsoulas and Juan Uriagereka. Some other colleagues we would like to mention specially for their insightful comments and the interest they showed in our work are Melvyn Cole, Idan Landau, Pino Longobardi, Marc Richards, Luigi Rizzi, Dominique Sportiche and Maria-Luisa Zubizarreta. Among the ˆ ngela, Jose´ Anto´nio, many people who helped us with data are Jose´ Cruz da A Marlyse Baptista, Eugenio Barrio, Modhumita Bora, Mary Fender, Carmen Rı´os Garcı´a, Anto´nio Guimara˜es, Mikash Hassamal, Ana Carrera Herna´ndez, Incanha Intumbo, Alain Kihm, Marcello Modesto, Urpo Nikanne, De´borah Oliveira, Tatiana Ortolan, Hannu Reime, Cilene Rodrigues and Gerardo Ferna´ndez Salgueiro. And finally, thanks to Iain Mobbs and Neil Myler for the index.

vi

Introduction: parameters in minimalist theory Ian Roberts and Anders Holmberg*

This book represents some of the work carried out in the period 2002–2007 by the group working on the project ‘Null Subjects and the Structure of Parametric Theory’, funded by the Arts and Humanities Research Council, Great Britain (Grant No. APN14458). The group consisted of Theresa Biberauer, Anders Holmberg, Chris Johns, Ian Roberts, Michelle Sheehan and David Willis. The central goal of that project was to investigate and, if possible, refine the notion of parameter of Universal Grammar (UG), as it has been understood in generative theory since roughly 1980, by looking carefully at the phenomena associated with one of the best-known and most widely discussed examples of a parameter: the Null Subject Parameter (NSP). This volume brings together a number of articles focusing on the nature of null subjects in a range of languages; Biberauer (2008b) is a sister volume arising from the same project, which focuses more on parameter theory than on null subjects, while Holmberg (2009) focuses on partial null-subject languages.1 In this Introduction, we would like to set the chapters in context. Accordingly, we first discuss the phenomena from English and various Romance languages which originally motivated the postulation of the NSP. Next, we summarise the main kinds of null-subject system that have been identified in the comparative-syntax literature. We complete Section 1 by summarising the two principal approaches to the analysis of null subjects, whose classical exponents are Rizzi (1986a) and Borer (1986). *

1

We’d like to thank the other members of the null-subject project group for their comments on various earlier versions of this work. We’d also like to thank Hans Broekhuis, Norbert Corver, Riny Huybregts, Ursula Kleinhenz and Jan Koster, the editors of Organising Grammar: Linguistic Studies in Honor of Henk van Riemsdijk (2005) for their comments on the Roberts and Holmberg contribution to that collection, parts of which are taken up again in Sections 2.3 and 2.4. Our thanks also to Fritz Newmeyer, for criticising us (see Newmeyer 2006), and to Pino Longobardi, Cristina Guardiano and Chiara Gianollo for being such stimulating interlocutors. Last but not least, Roberts would like to thank Alastair Appleton, Bob Davies, Luke Donnan and Iain Mobbs, all students on the Cambridge Linguistics MPhil 2007–2008, for listening to and reacting to some of the ideas discussed in Sections 2 and 3 so assiduously and so intelligently. Other material, published and unpublished, generated by the project is detailed on the project website (http://people.pwf.cam.ac.uk/mtb23/NSP/Nullsubjectsprojecthome.html).

1

Ian Roberts and Anders Holmberg

2

Section 2 focuses on the debates surrounding the classical formulations of the NSP, and in particular the ‘typological’ predictions that were initially made, beginning with Rizzi (1982). Here we consider the systematic crosslinguistic investigation whose results apparently indicate that certain predicted correlations do not hold (Gilligan 1987), and the far-reaching and negative conclusions for parameter theory drawn partly on the basis of this by Newmeyer (2004; 2005). We once again attempt to defend a version of Principles-and-Parameters (P&P) Theory against Newmeyer’s critique (see also Roberts & Holmberg 2005; Newmeyer 2006; Biberauer 2008a). Section 3 takes up the wider question of the nature of parameters; here we observe certain problems with the original view, which associated parametric variation closely with a rich, domain-specific array of UG principles. This view can no longer be maintained in full in the context of the Minimalist Program, which undertakes to eliminate as many UG principles as possible, and which calls into question the domain-specificity of what principles we are required to postulate. Ongoing comparative work over the past twenty years or so has tended by and large to favour the postulation of a large number of microparameters (see Kayne (2005a) for discussion, and Baker (1996; 2008b) for a different view); this view of parameters is readily compatible with minimalist assumptions, arguably more so than a ‘macroparametric’ approach. We suggest that the proliferation of microparameters is an instance of the familiar tension between descriptive and explanatory adequacy, which at earlier stages in the development of the theory motivated the simplification of rule systems and led to the development of the Principles-and-Parameters approach itself;2 as before, what seems to be required is a radical increase in theoretical abstraction. We make some tentative suggestions in this direction in Sections 3.4 and 3.5, developing ideas in Gianollo, Guardiano & Longobardi (2008), Roberts (2007a: 443; 2008: 17; Chapter 1), Roberts & Roussou (2003: chapter 5) and Mobbs (2008). This leads us to briefly propose a hierarchical model of parameter schemata which combines the notion of micro- and macroparameters, and to speculate on the shape of comparative syntax beyond explanatory adequacy. 1

Null subjects: the basics

1.1

The observation

Traditional grammars of many languages, for example Latin, observe that a pronominal subject is marked ‘in the verb’, i.e. by the person–number agreement inflection on a finite verb, and as such is not in need of expression by an 2

A similar point is made in Roberts (2001: 90) in relation to parametric accounts of syntactic change.

Introduction: parameters in minimalist theory

3

independent pronoun. The following comment from a well-known traditional grammar of Latin is representative: “Here [in the finite verb, IR/AH] the form contains in itself all the necessary elements . . . the persons being indicated by the endings” (Gildersleeve & Lodge 1895/1997: 144). Jespersen takes up this idea in the following remark: In many languages the distinction between the three persons is found not only in pronouns, but in verbs as well . . . in Latin . . . Italian, Hebrew, Finnish, etc. In such languages many sentences have no explicit indication of the subject, and ego amo, tu amas is at first said only when it is necessary or desirable to lay special stress on the idea ‘I, thou.’ (Jespersen 1924: 213)

This idea has an ancient pedigree, as indicated by the following remark by Apollonius Dyscolus on Ancient Greek: The nominative [subject] is implicitly present in [finite] verbs, and it is definite (i.e. has definite reference) in the first and second persons, but indefinite in the third because of the unlimited number of possible referents. (On Syntax, Book 1, }17; Householder 1981: 25)

What is being alluded to here is that, since a pronominal subject can be expressed ‘in the verb’ in languages such as Greek and Latin, there is no general requirement to pronounce the subject separately as a nominative pronoun. The initial observation behind the NSP, in all its formulations, has to do with this second point: an overt pronominal subject is not required in finite clauses, and, when such a pronoun does not appear, there is no nominal element which overtly realises the subject function in the clause. This fact may reflect a trivial feature of morphology, person-marking on the verb, but the possibility of not expressing the nominal bearing the subject function is of obvious importance for syntax. The first generative study of these matters was Perlmutter (1971). Perlmutter (pp. 100ff.) distinguished languages with the surface filter in (1), which he called Type A languages, from those lacking it, Type B languages: (1)

Any sentence other than an Imperative in which there is an S that does not contain a subject in surface structure is ungrammatical. (Perlmutter’s (9), p. 100)

Perlmutter relates the presence of the surface filter in (1) to the possibility of null subjects and of wh-movement of the subject from a finite embedded clause across a complementiser (this observation has since become known as ‘Perlmutter’s generalisation’), to the presence of obligatory expletives in the relevant kind of impersonal constructions, and to the existence of an arbitrary subject pronoun as a true subject (as opposed, for example, to an arbitrary subject clitic pronoun which surfaces as part of the object clitic cluster). French and English are examples of Type A languages, while “Spanish, Italian, Serbo-Croatian, Arabic, Hebrew, Warlbiri and Basque” (Perlmutter 1971: 115), as well as

Ian Roberts and Anders Holmberg

4

numerous others are Type B languages. This ‘typological distinction’, as Perlmutter referred to it, is not, however, connected to the nature of agreement inflection in the Type B languages in Perlmutter’s treatment. With this exception, and with the important omission of any discussion of ‘free inversion’ (see below), Perlmutter’s discussion identifies the NSP in all but name.3 As Perlmutter pointed out, the basic fact motivating the postulation of this parameter is that certain languages require finite clauses to overtly express a definite, referential, pronominal subject, while others do not. The contrast is illustrated by the following Italian and English examples: (2)

a. Parla italiano. b. *Speaks English.

Spanish and Greek, among many other languages, pattern like Italian, while, as Perlmutter pointed out, French appears to pattern like English ((3c) is ungrammatical as a declarative, although it would be a well-formed imperative): (3)

a. Habla espan˜ol. b. Mila ellinika. c. *Parle franc¸ais.

Thus Italian, Spanish and Greek are null-subject languages, while English and French are non-null-subject languages. The NSP relates, as stated above, to finite, discourse-neutral clauses, and canonically involves the interpretation of the null subject as a definite, referential pronoun. Many non-null-subject languages, including English, allow null subjects under other conditions. For example, both English and French extensively allow or require the subject of non-finite clauses to be null:4 (4)

a. [(Him) smoking] bothers me. b. John expects [(Mary) to leave soon]. c. Jean a essaye´ [de – partir]. John has tried [– to leave].

Such subjects have somewhat different properties from the null subjects of (2) and (3), in that in (4b,c) the empty subject of the infinitive must be coreferent with the subject of the main clause (this is subject control) and 3

4

One might also observe, with hindsight, that the surface filter in (1) is very close to the original formulation of the Extended Projection Principle put forward in Chomsky (1982: 10), i.e. the requirement that every clause must have a subject. For Perlmutter, (1) is parametrised (although, again, this terminology postdates his insights); see Section 1.3 on proposals that the NSP involves parametrising the requirement for a structural subject. Examples of this type do not fall under Perlmutter’s surface filter in (1) since it was generally assumed that such infinitives were VPs (resulting from raising in (4b) where Mary is present, and Equi-NP Deletion where it is not, and in (4c), combined with tree-pruning in the sense of Ross (1967: chapter 3)). In cases like (4a), the S-node of the gerund is pruned when the subject is not present.

Introduction: parameters in minimalist theory

5

that in (4a) must be arbitrary. Accordingly, they have generally been analysed in a different way from those of (2) and (3).5 The initial observation, then, is that some languages allow a definite pronominal subject of a finite clause to remain unexpressed as a nominal bearing the subject function, while others do not. Traditional grammars of languages such as Latin and Greek relate this to the fact that personal endings on the verb distinguish person and number, thereby making a subject pronoun redundant. Languages which allow null subjects are very common: most of the older Indo-European languages fall into this category, as do most of the Modern Romance languages (with the exception of some varieties of French and some varieties of Rhaeto-Romansch; see Roberts (Chapter 8)), the Celtic languages, with certain restrictions in the case of Modern Irish (see McCloskey & Hale (1984), and, for arguments that Colloquial Welsh is not a nullsubject language, Tallerman (1987)), West and South Slavic, but probably not East Slavic (these appear to be ‘partial’ null-subject languages in the sense of Section 1.2.4 below and Holmberg & Sheehan (Chapter 3)). Indeed, it seems that languages which allow null subjects are significantly more widespread than those which do not (Gilligan 1987, cited in Newmeyer 2005: 85). According to the Haspelmath, Dryer, Gil & Comrie’s (2005) World Atlas of Language Structures, of 674 languages for which data is available, subject pronouns can be omitted in 409, and cannot be omitted in 77 (the remainder form various kinds of mixed categories involving clitics, displaced pronouns, etc.; see Map 101 ‘Expression of Pronominal Subjects’). So null-subject languages, of one kind or another, are considerably more common than non-null-subject languages. 1.2

Types of null-subject systems

Since Rizzi’s early work on null subjects, it has been observed that there are different types of null-subject language. Rizzi (1982: 143) proposed that the NSP be divided into two subcases, one applying to languages in which the 5

Both English and French also allow null subjects in special discourse environments or registers. Haegeman (2000: 130) gives the following examples from what she calls “written abbreviated registers” (“written registers in which pressures of economy seem to overrule the ‘core’ grammar” (p. 132), including diaries, short notes and some kinds of colloquial speech): (i)

a. – cried yesterday morning. b. Elle est alsacienne. – paraıˆt intelligente. She is Alsatian. Seems intelligent.

(Plath 1983: 288) (Le´autaud 1988: 48)

Such null subjects, in addition to being restricted to certain types of discourse and/or register, have special properties which distinguish them from the canonical null subjects of (2) and (3) (see Haegeman (2000: 138–41) for details). See also Holmberg (Chapter 2, Section 2). We will leave these cases aside here.

Ian Roberts and Anders Holmberg

6

unexpressed pronoun can only be an expletive, and one applying to languages where it is able to be referential. Huang (1984) observed that many East Asian languages show a much more liberal option of non-expression of pronominal elements, and that this could not be related to person agreement, since that kind of inflection is generally absent in these languages. More recently, the existence of ‘partial’ null-subject languages has been observed: languages in which the pronominal subject may remain unexpressed under restricted conditions determined by both the morphological and the syntactic context. We now briefly describe each of these types of null-subject language one by one. 1.2.1 Consistent null-subject languages These languages have been the most discussed and analysed among the various types of null-subject languages and have, mainly for historical reasons, often been taken to be the only kind of null-subject language. In consistent null-subject languages, all persons in all tenses can feature an unexpressed pronoun.6 These languages characteristically show ‘rich’ agreement inflection, i.e. distinct personal endings on the verb, generally in all tenses. The Italian, Greek and Turkish forms in (5) illustrate: (5)

6

7

a. Italian bevo ‘I drink’ (etc.) bevi beve beviamo bevete bevono b. Greek pino ‘I drink’ (etc.) pinis pini pinume pinete pinun c. Turkish7 ic¸iyorum ‘I drink’ (etc.) ic¸iyorsum ic¸iyor ic¸iyoruz ic¸iyorsunuz ic¸iyorlar

There are sometimes limited exceptions to this generalisation. In Italian, for example, the 2SG pronoun tu must appear when the verb is in the subjunctive. In the present subjunctive, the singular forms of the verb are not distinct: che (io) parli, che tu parli, che (lui/lei) parli (‘that I/you/he/she speak’). ¨ ztu¨rk (2001; 2008) for the view that Turkish is a ‘discourse pro-drop’ language in the But see O sense of Section 1.2.3 below. This paradigm is from Csato´ & Johansen (1998: 214).

Introduction: parameters in minimalist theory

7

The Romance null-subject languages and Modern Greek are the paradigm examples of this kind of language, and have been much discussed and exemplified in the literature. These languages also illustrate the properties originally proposed to form a cluster determined by the positive value of the NSP, which we will turn to in Section 2 below. One further property that we can mention here is that, as pointed out by Jespersen in the quotation above, overt subject pronouns are generally allowed in finite clauses in null-subject languages, although they tend to have what we may loosely call an emphatic interpretation (this is indicated by putting the English pronoun in capitals in the translations below). Thus, alongside (2a) and (3a,b) we have: (6)

a. Lui parla italiano. HE speaks Italian. b. E´l habla espan˜ol. HE speaks Spanish. c. Aftos mila ellinika. HE speaks Greek.

(Italian) (Spanish) (Greek)

This aspect of the interpretation of overt pronominal subjects in null-subject languages emerges slightly more clearly in (7). Here the overt pronoun in the adverbial clause does not allow the interpretation in which it corresponds to the subject of the main clause (see Vanelli, Renzi & Beninca` (1986), Samek-Lodovici (1996), and Frascarelli (2007) for discussion): (7)

a. Il professore ha parlato dopo che (lui) e` arrivato. The professor has spoken after that (he) is arrived ‘The professor spoke after he arrived.’ b. I Maria jelase afou (afti) idhe ton Yianni. The Mary laughed after (she) saw the Yiannis. ‘Mary laughed after she saw Yiannis.’

(Italian)

(Greek)

In other words, the overt pronoun of the adjunct does not show the same ambiguity as its English and French counterparts in (8). Instead, it strongly prefers the interpretation which is disjoint from ‘the professor’, while the English and French pronouns are, out of context, ambiguous between this interpretation and the one where they correspond to ‘the professor’: (8)

a. The professor spoke after he arrived. b. Le professeur a parle´ apre`s qu’il est arrive´.

(¼(8a))

These interpretative differences involving the use of an overt pronoun appear to be related to the fact that subject pronouns may be unexpressed, i.e. to the positive value of the NSP. For the moment, we take the two diagnostic features of a consistent nullsubject language to be: (i) the possibility of leaving the definite subject

8

Ian Roberts and Anders Holmberg

pronoun unexpressed in any person–number combination in any tense; and (ii) the rich agreement inflection on the verb. We will suggest other properties as we proceed. 1.2.2 Expletive null subjects Some languages apparently allow expletive null subjects, but not referential ones. German is one such language, as are some varieties of Dutch and Afrikaans, and a range of creoles (Nicolis (2005; 2008) mentions Cape Verdean, Berbice Dutch, Kriyol, Mauritian, Papiamentu and Saramaccan; Roberts (2007a: 413) adds Haitian and Jamaican). In (9a) the expletive pronoun es cannot be expressed, while in (9b) the same pronoun in the same syntactic position, only now with a referential interpretation, must be expressed (examples from Cardinaletti 1990: 5–6): (9)

a. Gestern wurde (*es) getanzt. Yesterday was (it) danced. ‘Yesterday there was dancing.’ b. Gestern war *(es) geschlossen. Yesterday was (it) closed. ‘Yesterday it was closed.’

Owing to this restriction on their null subjects, languages of this type are not regarded as ‘full’ null-subject languages. Rizzi (1982: 143) identifies what he called two “related but autonomous parameters”: one concerns whether an unexpressed pronoun is allowed at all, and the other whether referential pronouns are allowed to be unexpressed. In languages like English, both parameters are negative, while in Italian both are positive. In German and the creoles just mentioned, the first is positive and the second negative. Hence German allows non-referential null subjects, as in (9a), but not referential ones. According to Rizzi, the fourth logical option is impossible (“if an inflection cannot be pronominal, it cannot be referential either”; p. 143). There is thus an implicational relation between the presence of referential null subjects and the presence of expletive null subjects (see Holmberg (Chapter 2; this point is discussed in some detail in Roberts (2007b: 31–38)). For an alternative analysis of the German facts illustrated in (9), and related phenomena elsewhere in Germanic, see Biberauer (Chapter 4), and the references given there. Expletive null-subject languages (sometimes called ‘semi-pro-drop languages’), then, are distinguished from consistent null-subject languages in that non-dummy pronouns cannot be left unexpressed. 1.2.3 ‘Discourse pro-drop’ (also called ‘radical pro-drop’) A good number of languages which are otherwise typologically and genetically distinct (Chinese, Japanese, Korean, Thai, Vietnamese and others) allow null subjects quite freely, but seem to be entirely without agreement marking of any kind.

Introduction: parameters in minimalist theory

9

The case of Chinese is discussed in Huang (1984). Chinese allows both subjects and objects to remain unexpressed and have a definite pronominal interpretation, as illustrated in (10): (10)

a. – kanjian ta le (he) see he ASP b. Ta kanjian – le. He see (him) ASP ‘He saw him.’

Both pronouns can be dropped under the appropriate discourse conditions. It has been suggested since the earliest studies (Huang 1984; Rizzi 1986a) that the total absence of agreement marking may play a role in facilitating the very liberal availability of null subjects in these languages. Recently, three specific hypotheses have been put forward in this connection. First, Tomioka (2003: 336) proposed the “Discourse Pro-Drop Generalisation” (see Jayaseelan (1999) for a similar idea): All languages which allow discourse pro-drop allow (robust) bare NP arguments . . . Null pronouns in Discourse Pro-Drop languages are simply the result of N0 -Deletion/ NP-Ellipsis without determiner stranding.

This idea expresses a relation between discourse pro-drop and the availability of bare NP arguments (i.e. the grammaticality of a sentence such as (I) saw cat, thereby relating discourse pro-drop in an interesting way to Chierchia’s (1998) Nominal Mapping Parameter).8 It also relates discourse pro-drop to ellipsis, as does Saito (2007) (see below and the discussion in Roberts (Chapter 1)). A different proposal is made by Neeleman & Szendro˝i (2007; 2008). These authors treat fully specified nominals as KPs (since they inherently contain a syntactic position for Case) and posit an operation of context-free KP-deletion. In languages with fusional pronoun morphology, this context-free operation is blocked by the principle of disjunctive ordering (the Elsewhere Condition of Kiparsky (1973)), which states that a more specific operation blocks a more general one in the case where both structural descriptions are met. They further adopt a ‘realisational’ approach to the insertion of pronouns into positions created by the syntax; for example, the English pronoun him is the realisation, or ‘spell-out’, of the feature complex [KP þpronoun, –anaphor,

8

Chierchia (1998) formulated the Nominal Mapping Parameter, which distinguishes languages in which bare nouns are able to function as arguments from those in which they are not (more technically, can NP map directly into type < e > ?). If yes, then the language has generalised bare arguments (allows bare singulars and plurals), has a generalised classifier system and lacks plural morphology. Chierchia proposes that Chinese has the positive value for this parameter, while English has the negative value. See Ramchand & Svenonius (2008) for a different view.

10

Ian Roberts and Anders Holmberg

3rd person, Singular, Masculine, Accusative]. The general ‘radical pro-drop’ rule is the context-free zero-realisation rule (11): (11)

[KP þpronoun, –anaphor]

!

The Elsewhere Condition will always block this realisation of pronouns in English, since, given their fusional nature, English pronouns always have more complex spell-out rules whose structural descriptions properly include that of (11). But this is not true in every language: in some languages, e.g. Japanese, regular, agglutinating case-markers are added to the pronominal root (watasi-ga ‘I’; watasi-o ‘me’, etc.). Japanese thus has separate spell-out rules for the Case (K) morpheme and for the pronoun, which is a category distinct from KP (probably NP). And here is the central idea of their analysis: because of the non-fusional make-up of pronominal KPs, neither the radical pro-drop realisation of KP nor the specific rules for NP and K are in an ‘elsewhere’ relation. Hence Japanese pronominal KPs are optionally allowed a zero realisation. The analysis leads to the following generalisation: fusional pronouns block radical pro-drop. Neeleman & Szendro¨i’s analysis entails a clear implicational relation between non-fusional pronoun morphology and discourse pro-drop, which they show holds across a very wide range of languages. The third analysis of discourse pro-drop put forward recently is due to Saito (2007). Saito suggests that a single covert grammatical mechanism allows for radical pro-drop and argument ellipsis. This mechanism involves covert copying of elements into argument positions from a set of discoursegiven entities. Understood pronouns may be included in this set, along with antecedents for the recovery of elided arguments. The precondition for this covert operation is, effectively, the lack of surface agreement triggers. Hence arguments are not required to be overtly present in order to trigger surface agreement on verbs and other categories. Thus the relation between radical pro-drop and absence of agreement marking is established (see also Kuroda 1988). This analysis of discourse pro-drop is discussed in more detail in Roberts (Chapter 1); see also Section 3.4 below. Whatever the correct analysis, the characteristics of discourse pro-drop languages which distinguish them from consistent null-subject languages are: (i) general possibility of non-expression/ellipsis of nominal arguments in various functions in addition to the subject; and (ii) lack of personagreement marking on verbs. 1.2.4 Partial null-subject languages The existence of partial null-subject languages as a separate type of null-subject language has been more difficult to establish. However, Holmberg (2005: 548–50), Holmberg and Sheehan (Chapter 3), Holmberg, Nayudu & Sheehan (2009) and Holmberg (Chapters 2

Introduction: parameters in minimalist theory

11

and 5) identify a number of characteristics which can serve to distinguish languages of this type from languages of the Italian type described in Section 1.2.1 (see also the papers in Holmberg (2009)). These include Finnish, Hebrew, Russian, Icelandic, Marathi and probably several other Indic languages, and Brazilian Portuguese.9 Here we take Finnish as our example of a partial null-subject language. We observe three things which distinguish the null subjects of Finnish from those of a language like Italian. First, only 1st and 2nd person pronouns are freely able to be left unexpressed in any finite context: (12)

(Mina¨) puhun englantia I speak-1SG English ‘I speak English’ (etc.) (Sina¨) puhut englantia You speak-2SG English *(Ha¨n) puhuu englantia He/She speak-3SG English (Me) puhumme englantia We speak-1PL English (Te) puhutte englantia You speak-2PL English *(He) puhuvat englantia They speak-3PL English

(Holmberg 2005: 539)

Second, it is not the case that 3rd person pronouns can never be unexpressed. As Holmberg (p. 539) says: “A 3rd person definite subject pronoun can be null when it is bound by a higher argument, under conditions that are rather poorly understood.” This possibility is illustrated by examples like the following: (13)

Pekkai va¨itta¨a¨ [etta¨ ha¨ni/j/i/*j puhuu englantia hyvin]. Pekka claims that he speaks English well (Holmberg 2005: 539)

This is characteristic of partial null-subject languages. Holmberg & Sheehan (Chapter 3) is an investigation of the conditions under which these null subjects are licensed. Third, in partial null-subject languages “generic pronouns can, and must, be null” (Holmberg 2005: 540):

9

Brazilian Portuguese is an interesting case, as it seems to differ from European Portuguese in being a partial null-subject language, while European Portuguese is a consistent null-subject language. Duarte (1995) traces this development in nineteenth- and twentieth-century Brazilian Portuguese. For some speculation regarding the diachronic development of Brazilian Portuguese in this respect, see Roberts (forthcoming a).

12 (14)

Ian Roberts and Anders Holmberg Ta¨a¨lla¨ ei saa polttaa Here not may smoke ‘One can’t smoke here’

This contrasts with languages like Italian and Greek, where a special clitic (Italian) or verb form (Greek) is required: (15)

a. Qui non si puo` fumare. Here not SI can smoke b. Apoghorevete to kapnisma. prohibit-3SG-.mediopass the smoking ‘One can’t smoke here.’

(Italian) (Greek)

In Italian, omission of the si-clitic gives rise to a string comparable to that in the Finnish example in (14), but the unexpressed pronoun must be understood as definite: ‘Here s/he can’t smoke.’ The Greek example features the mediopassive form of the verb ‘to prohibit’, which gives rise to the same kind of impersonal interpretation as Italian si. The above differences suffice to establish that partial null-subject languages have a range of properties distinguishing them from consistent nullsubject languages. It is very likely that many languages that have been seen as consistent null-subject languages are in fact partial null-subject languages; in this connection, the simple descriptive facts need to be clarified. One should regard categorical statements in descriptive grammars to the effect that a given language is a null-subject language with some scepticism (with the possible exception of the very best-studied languages). The fact that subjects are sometimes unexpressed does not make a language a null-subject language in the technical sense. As is well known, even (spoken) English can drop its subjects in certain contexts (see Footnote 5). Distinguishing a consistent from a partial null-subject language requires consideration of contexts such as (7) and the syntax of impersonal constructions at a level of detail which is, probably, rarely attained in the research behind descriptive grammars. 1.2.5 Conclusion We see that there are four identifiable types of nullsubject language. We can range them along a scale of ‘liberality’ as follows: (16)

expletive null subjects ⊃ partial null subjects ⊃ consistent null subjects ⊃ discourse pro-drop

Placing discourse pro-drop languages at the right edge of the scale is motivated particularly if other arguments (direct and indirect objects) are taken into account; discourse pro-drop languages allow (referential) null objects as well as subjects, which is not the case for consistent or partial null-subject languages, across the board. And of course, we could add non-null-subject languages such as English at the left-hand edge of the scale in (16). All other

Introduction: parameters in minimalist theory

13

things being equal, for each system Si in (16), the set of positions in which a pronoun can remain unexpressed in Si is a proper subset of the set of positions in which a pronoun can remain unexpressed in all systems Sj, where Si is to the left of Sj in (16).10 We will return to the possibility of arranging different kinds of systems into some kind of hierarchy in Sections 3.3 and 3.4. 1.3

Two analyses of null subjects

There have been two main views on the nature of null subjects and the NSP in the literature for some time. One view, most influentially put forward in Rizzi (1986a), is that null subjects are occurrences of a phonologically unrealised, or empty, pronoun pro in the subject position (see Roberts (Chapter 1, Section 1) and Holmberg (Chapter 2, Section 9) for more details of Rizzi’s account). The other view, which has its origins in Borer (1986), is that there is no overarching requirement for a subject position as such (i.e. that the Extended Projection Principle (EPP) of Chomsky (1982: 10) does not hold, or at least does not hold universally). The null subject may then be directly expressed by the rich verbal agreement inflection, and there is therefore no need for a distinct empty pronoun to realise the subject function. This view implicitly accepts that agreement inflection can function like a pronoun, in that it can bear a grammatical function, and a thematic relation, in the way that nominal expressions typically do. Since agreement is located in the Infl, or I, position, this view is known as the ‘I-subject’ view. This view articulates the intuition behind comments in traditional grammars of the kind exemplified in Section 1.1 above. The two views just sketched survive in current work. Developing Borer’s (1986) I-subject idea, it has been suggested by various authors that, since person–number specification of the subject can be exhaustively computed from the verbal inflection, the preverbal subject is effectively optional and when it appears it acts as a left-dislocated (or more precisely ‘clitic leftdislocated’) element occupying a position peripheral to the clause (i.e. one not associated with a grammatical function) and with the verbal inflection functioning analogously to a clitic pronoun. In differing ways, this view is put forward by Alexiadou & Anagnostopoulou (1998), Barbosa (1995; 2009), Fassi Fehri (1993), Manzini & Savoia (2005), Nash & Rouveret (1997), Ordo´n˜ez (1997), Platzack (2004) and Pollock (1997). Holmberg (Chapter 2) articulates what might be viewed as a version of this hypothesis.

10

As it stands, this is an idealisation. There is, in fact, considerable variation among the discourse pro-drop languages as regards the use of pro-drop. Chinese is apparently more restricted in this respect, making more use of overt pronouns, than, for example, Japanese, and possibly more than many consistent null-subject languages: see Cole (forthcoming).

14

Ian Roberts and Anders Holmberg

On the other hand, Cardinaletti (1997; 2004), Holmberg (2005) and Sheehan (2006) have argued that the subject position is present, at least in some nullsubject languages, and hence it is filled by the null pronoun pro (see for example Sheehan (Chapter 6) on this). Holmberg (2005) and Roberts (Chapter 1) follow Cardinaletti & Starke (1999) in taking pro to be a weak pronoun, that is a ‘deficient’ pronoun whose distribution is restricted to certain designated positions. Furthermore, following a long line of work going back at least to Rizzi (1982), they take pro to be licensed by a special pronominal feature (usually termed a D-feature) associated with the head bearing the features realised as person agreement on the verb. The two views can be taken to converge on the idea that the inflectional head must be pronoun-like in a null-subject language (they may converge in other respects too; see Roberts (Chapter 1, Section 2.5)). This appears to be the core of the Null Subject Parameter, whatever the further details. Let us encode this property formally as the presence or absence of a D-feature associated with T. Following the yes/no-question format for parameters adopted in Roberts (2007a), let us then state the NSP as follows: (17)

The Null Subject Parameter Does T bear a D-feature?

T clearly does not bear a D-feature in non-null-subject languages, while it clearly does in consistent null-subject languages. Partial null-subject languages and languages only allowing null expletives do not have a D-feature in T, which is why they allow referential null subjects only under very restricted circumstances (partial null-subject languages) or not at all; see Holmberg (Chapter 2), Holmberg & Sheehan (Chapter 3) and Roberts (Chapter 1) for refinements. We leave open for now the status of discourse pro-drop languages, although it is worth noting that if, following the Nominal Mapping Parameter, non-D elements can function as arguments, perhaps the requirement for a D-feature on T is waived in such languages. We will explore this a little more in Section 3.4 below (and see Roberts (Chapter 1, Section 2.5)). 2

The NSP in the context of P&P theory

2.1

Rizzi (1982): clustering properties in Romance and English

The NSP has played a prominent role in the theoretical study of comparative syntax in recent years, not just because of the characterisation it gives us of languages like Italian, and how they differ from English, but primarily because it has been seen as a good example of the way in which rather abstract grammatical properties (such as that given in (17)) may have proliferating effects, unifying apparently unrelated surface phenomena.

Introduction: parameters in minimalist theory

15

To see the full importance of this idea, we need to consider Chomsky’s (1964: 28f.) definitions of levels of adequacy for linguistic theory. These were observational, descriptive and explanatory adequacy. An observationally adequate grammar presents the data correctly, while a descriptively adequate grammar “specifies the observed data . . . in terms of significant generalizations that express underlying regularities in the language” (p. 28). Explanatory adequacy “can be interpreted as asserting that data of the observed kind will enable a speaker whose intrinsic capacities are as represented in th[e] general theory to construct for himself a grammar that characterizes exactly this intuition”; in other words, attaining explanatory adequacy involves showing how a given empirical phenomenon can be deduced from UG. The postulation of parametric variation in UG principles was a very large step in the direction of explanatory adequacy, since, one could assume, if we can say that this syntactic feature of this language is due to setting that parameter to that value, we have provided an explanatorily adequate account of the syntactic feature in question in that we have related it directly to UG, as well as a descriptively adequate account to the extent that the analysis of the relevant property of the language is correct. Moreover, the parametric account has immediate cross-linguistic implications, since it implies that another language lacking the property in question will set the parameter in question to a different value. Now, if each parameter value determines a cluster of disparate syntactic features, then explanatory adequacy is enhanced, especially if certain features are readily accessible to acquirers on the basis of impoverished evidence while others are hardly likely to be easily accessible. In this case, arriving at a parameter value determining both the accessible and relatively inaccessible feature gives us a simple account of how the inaccessible feature can be acquired, thus accounting for an aspect of the poverty of the stimulus to language acquisition and thereby, again, reaching explanatory adequacy. At the same time, other things being equal, a ‘typological’ prediction is made: the inaccessible feature will be acquired whenever the acquired one is, since both reflect the same abstract property of Universal Grammar, the setting of a given parameter to a given value. Let us state the following conjecture in relation the ‘clustering effect’ associated with parameters:11 11

(19) is simplified in various ways. First, we are assuming that parameters always have binary values, although of course one could in principle add to (19) a clause specifying how values vi, i þ 1 . . . n each entail distinct clusters of P-expressions C1, . . . Cn, each of which may be partially disjoint from the others. For simplicity, and following general practice in discussions of parameters, we assume that all parameters are in fact, or can be formulated as if they were, binary. Second, taking C0 to be the negative correlate of C is clearly the simplest assumption we can make, although not required. We will nevertheless make it.

16

Ian Roberts and Anders Holmberg

(18)

A substring of the input text S expresses a parameter P just in case a grammar must have P set to a definite value in order to assign a well-formed representation to S. (Roberts 2007a: 133)

(19)

For any UG parameter P, fixing P at value vi entails a cluster C of grammatical expressions of P(vi), while setting P to value vj6¼i entails a disjoint cluster C0 of expressions of P (vj6¼i).

(20)

A cluster of P-expressions is a set of properties of a surface morphosyntactic form of cardinality equal to or greater than 2, which are reflexes of P’s setting to a determinate value vi.

(21)

A substring of the input text S is a trigger for parameter P if S expresses (Roberts 2007a: 133) a determinate value vi of P.

It follows from these definitions that a trigger for a given parameter value is included in the cluster of expressions of that parameter value, and indeed properly included in that set if triggers must be unique (if there is to be a one-toone mapping from triggers to parameter values). A trigger must be accessible in the primary linguistic data (PLD), while the other expressions in a given cluster C may be relatively inaccessible, along the lines described above. The properties connected to the NSP by Perlmutter (1971) and, in particular, Rizzi (1982), can be seen as a cluster. We take the NSP to be a UG parameter, and fixing the NSP at value þ entails the cluster C of grammatical expressions of NSP(þ) in (22): (22)

a. b. c. d.

The possibility of a silent, referential, definite subject of finite clauses. ‘Free subject inversion.’ The apparent absence of complementiser-trace effects. Rich agreement inflection on finite verbs.

(22b) refers to the general possibility of expressing an overt subject, usually with a focus interpretation, in post-verbal position: (23)

a. Hanno telefonato molti studenti. b. *Ont te´le´phone´ beaucoup d’e´tudiants. Have telephoned many students. ‘Many students have telephoned.’

‘Free inversion’ is in fact subject to slightly differing constraints in different languages, being more freely available in Spanish and Greek than in Italian, for example; see Sheehan (2006; Chapter 6) and the references given there. (22c) relates to Perlmutter’s generalisation, since it originates in Perlmutter’s (1971) pioneering work. Perlmutter’s generalisation expresses the fact that in non-null-subject languages the subject of a finite clause cannot undergo wh-movement if the complementiser introducing the clause is present. This constraint holds of English and French, as the following examples show:

Introduction: parameters in minimalist theory (24)

17

a. *Who did you say that – wrote this book? b. *Qui as-tu dit qu’ – a e´crit ce livre? (¼(24a))

Here the questioned constituent (who/qui) corresponds to the subject of the subordinate clause, so there is a ‘gap’ in that position. The ungrammaticality of (24a) is known as a ‘complementiser-trace effect’, since in many versions of the theory of movement it is held that the empty subject position at the movement site in the complement clause contains a trace of the moved wh-element. The idea that the presence of the complementiser determines the ungrammaticality of such examples is supported by the fact that (24a) becomes grammatical if that is omitted. In French, (24b) can be rendered grammatical by altering the form of the complementiser from que to qui. These points are illustrated in (26): (25)

a. Who did you say – wrote this book? b. Qui as-tu dit qui – a e´crit ce livre? (¼(25a))

In null-subject languages, as Perlmutter observed, it appears that complementisertrace effects are not found (Rizzi (1982) argued that in fact this is not true if certain structures covertly derived at the level of Logical Form are taken into consideration). The subject of a finite clause introduced by a complementiser can readily be questioned in these languages: (26)

a. Chi hai detto che – ha scritto questo libro? Who have-2SG said that – has written this book b. Pjos ipes oti – egrapse afto to vivlio? who said-2SG that – wrote this the book ‘Who did you say wrote this book?’

(Italian) (Greek)

This feature of the null-subject cluster can be reasonably thought of as relatively inaccessible in the PLD, while rich agreement inflection is presumably very accessible (especially given the known sensitivity of acquirers to inflections; see Wexler 1998) and the other two properties may be somewhat accessible. In terms of the definitions above, setting the NSP to the value – entails the disjoint cluster C0 of expressions of NSP(–): no null subjects, no free subject inversion, complementiser-trace effects and ‘poor’ agreement inflection. English, French and the Mainland Scandinavian languages have the cluster C0 , as do many creoles. The above discussion relates primarily to language acquisition, and shows how parametric clusters take us towards explanatory adequacy. But it is also clear that this approach defines language types. In this way, typology, in the sense of the establishment of cross-linguistic relations and of a structure to cross-linguistic variation, and acquisition become intrinsically related. This is a very positive development as it clearly opens the way to a two-pronged empirical approach to understanding the nature of UG.

Ian Roberts and Anders Holmberg

18

It is also worth noting that the definitions given in (19–21) provide very strong definitions of types, far stronger than what is usually found in the Greenbergian typological tradition. According to these definitions, all the properties in C are biconditionally related to one another. In other words, all things being equal, we can derive the following set of implicational statements: (27)

a. A language has null subjects iff it has rich agreement. b. A language has null subjects iff it has free subject inversion. c. A language has null subjects iff it does not show complementiser-trace effects. d. A language has rich agreement iff it has free subject inversion. e. A language has rich agreement iff it does not show complementiser-trace effects. f. A language has free subject inversion iff it does not show complementiser-trace effects.

In Rizzi’s (1982) terms, all of these properties are connected by the presence of the silent pronoun pro in the subject position. This element is licensed by rich agreement inflection, and can satisfy the general requirement for a subject position (the Extended Projection Principle of Chomsky (1982: 10)), allowing an overt subject to appear in the ‘freely inverted’ position (see Sheehan (Chapter 6) for more on this), and indeed to be wh-moved from this position (as argued by Rizzi (1982)). Thus the formal property which underlies the NSP, on this analysis, is the availability of pro subjects. Once acquirers deduce this (on the basis of the universal principles determining the availability of null pronouns), they will immediately deduce the existence of the other properties in the cluster, and the very strong implicational links among the properties in (27) follow. Hence we expect strong typological correlations to support parametric clusters, and thereby to motivate analyses of the general type instantiated by Rizzi’s account of the cluster associated with the NSP. One might also, at least initially, think that all languages must fall on one side or the other of a parametric divide that distinguishes cluster NSP þ from cluster NSP–, and that perhaps this is true of all clusters defined by all parameters. 2.2

Gilligan (1987): universal clusters

Building on Rizzi (1982), then, it was possible to take the NSP to define two disjoint language types, in which the four properties in (22) are biconditionally related to one another. Putting aside the partial null-subject language Brazilian Portuguese, the validity of this typology across Romance looks reasonable: all the Romance languages except French (and some varieties of Rhaeto-Romansch) appear to show the properties in (22).12 12

See Holmberg & Sheehan (Chapter 3), and especially Sheehan (2006: chapter 6) on Brazilian Portuguese.

Introduction: parameters in minimalist theory

19

But of course if this is a valid linguistic typology, it should extend further. Consider, then, the Celtic languages Welsh and Irish. (Literary) Welsh and Irish are both null-subject languages (on Irish, see McCloskey & Hale (1984); on Welsh, see Roberts (2005: chapter 2), Borsley, Tallerman & Willis (2007: 34)). In both systems, the presence of null subjects appears to be fairly closely tied to rich agreement inflection on the finite verb. However, it is very hard to evaluate the status of the correlation with the absence of complementiser-trace effects, since subjects never appear adjacent to finite complementisers in these languages owing to the fact that the basic word order in finite clauses is VSO. This highlights a major difficulty in trying to generalise the correlations in (22): in general, such tightly connected clusters can only hold among languages where a range of further conditions are held constant, and this in practice often means rather closely related languages. Even in the case of closely related languages there may be other, independent differences which disguise the effect of a parameter. Among the Germanic languages, English and the Mainland Scandinavian languages show none of the properties of (22) (except for the dialects which allow that-trace extractions; see Sobin (1987); Lohndal (2007)). But the status of Afrikaans, Dutch and German is less clear, since their SOV order makes it difficult to determine whether free inversion occurs. As for Icelandic, it shares with Brazilian Portuguese the property of being a partial null-subject language, which therefore does not conform to the classical NSP (see Holmberg (Chapter 2)). Simple, surface evaluation of correlations of the kind in (22) is not possible, without making further assumptions. In this respect, such correlations differ from Greenbergian correlations such as “If a language is OV, then it has postpositions” in that more assumptions are required in order to test them. This, of course, is connected to the fact that the correlations arose from a deeper syntactic analysis of the languages in question than that underlying the simpler Greenberg-style correlations. Our intention is not to criticise either Rizzi’s or Greenberg’s approach to establishing typological correlations; both may be able to reveal previously unsuspected cross-linguistic correlations. The point is simply that correlations of the type in (22) may be very difficult to establish and maintain as more and more extra variables are brought into play. For example, what do we conclude from the fact that Welsh and Irish are null-subject languages in which the presence of complementiser-trace effects cannot be determined in finite clauses owing to the existence of VSO order? It seems reasonable to conclude that VSO order neutralises this property, and so here we do not have a true counterexample to the proposed cluster, but at the same time we do not have the strong biconditional relation of the type in (27). Such biconditional statements must be subject to a general qualification along the lines of ‘all other things being equal’.

20

Ian Roberts and Anders Holmberg

Another potential counterexample is posed by those dialects of Mainland Scandinavian and English which do not show the that-trace effect in finite clauses (cf. Sobin 1987; Lohndal 2007). (28)

Vem tror du att _ a¨r intelligentast? who think you that is most intelligent ‘Who do you think is most intelligent?’

(Fenno-Swedish)

Again, such facts indicate that the correlation between the that-trace effect and the other properties of (27) is not biconditional. Null subjects and free inversion entail the absence of that-trace effects, but not vice versa. This is not necessarily an argument against Rizzi’s theory. Note that the theory has an explanation of the correlation of absence of that-trace effect with the other properties, which is that the presence of pro in SpecTP makes it possible for A0 -movement to by-pass this position, moving directly from a lower subject position to SpecCP. In recent work, Rizzi argues that SpecIP is a ‘criterial position’ in his terms, which means that movement out of this position is universally impossible (cf. Rizzi & Shlonsky 2005). This creates a problem for subject extraction which every language has to deal with somehow. In null-subject languages the problem is solved quite simply by by-passing SpecTP. English and French solve it by operating on C, which in English is morphologically realised as deletion of that, in French as substitution of qui for que, thus modifying the criterial property of SpecIP (see Rizzi & Shlonsky for details). Fenno-Swedish, Ozark English, etc. have to solve this problem, too. The most likely hypothesis is that these varieties do not solve it in the manner of the null-subject languages, but perhaps in a manner akin to that in other varieties of Mainland Scandinavian and English, by operating on C, but without a morphological effect on the complementiser. A further alternative may be that the EPP does not apply at SpecTP here, allowing that position to be by-passed by ‘long’ wh-movement. The point here is that complementiser-trace violations are not predicted not to occur in non-null-subject languages. Again, we see that the correlations are not accurately formulated as biconditionals. Returning to the Celtic languages, it is difficult to be sure of the status of free inversion in these languages. One might claim that the position occupied by the subject in VSO order is in fact that of a freely inverted subject, although the fact that the subject position is interpolated between an auxiliary and non-finite verb in compound tenses in VSO languages, while in cases of free inversion the subject follows both the auxiliary and the non-finite verb (see (23)), argues against this. Nonetheless, we could tentatively conclude for Celtic that complementiser-trace effects cannot be tested owing to effect of VSO order. In both cases, the very strict nature of the correlations is part of the problem. Again, if the correlations were stated as one-way implications, or were hedged by a statement such as ‘all else, particularly basic word order,

Introduction: parameters in minimalist theory

21

being equal’, the situation would be clearer in that the status of these languages as true counterexamples would be easier to determine. The Celtic languages, although typologically quite different in a number of superficial respects from Germanic and Romance (see Haspelmath 2001), are nonetheless Indo-European languages and have been in contact with some Germanic and Romance languages for millennia. If we want to really establish the universal coverage of a cluster like (22) and the typology it implies, we should of course move far beyond merely comparing Germanic, Romance and Celtic. However, we can only expect that the kind of difficulty involved in interpreting the data, sketched above in relation to Celtic, will multiply, perhaps out of control. Again, this stems from the degree of syntactic analysis needed to establish the relevant properties combined with the very strict nature of the clustering we have postulated. Gilligan (1987) tested the correlations put forward by Rizzi (1982) against a 100-language sample, which he attempted to correct for areal and genetic bias. As reported in Newmeyer (2004: 202–206; 2005: 88–92) and Croft (2003: 80–84), Gilligan found just four robust cross-linguistic correlations, taking the properties discussed by Rizzi pairwise: (29)

a. b. c. d.

Free inversion ! expletive null subjects Free inversion ! allow complementiser-trace violations Referential null subjects ! expletive null subjects Allow complementiser-trace violations ! expletive null subjects

Newmeyer (2005: 90–91) concludes that “[t]hese results are not very heartening . . . In three of the four correlations, null non-thematic subjects are entailed, but that is obviously a simple consequence of the virtual nonexistence of languages that manifest overt non-thematic subjects”. In a sense, this situation is exactly what we would expect, given the above considerations: expanding the database from roughly ten to roughly one hundred languages simply multiplies the number of uncontrolled variables to a point where, without further detailed analysis of a wide range of constructions in a wide range of languages, the correlations can no longer be discerned. In a word, then, these results are inconclusive (this is not at all the conclusion Newmeyer draws, as we will see in the next section). However, Gilligan’s results are not quite as inconclusive as they first appear. Consider the only implication which does not involve null expletive pronouns: (29b). This relates to one of the most important conjectures in Rizzi’s (1982) analysis: that true complementiser-trace violations are in fact universally ruled out (due to the ‘criterial’ nature of SpecIP, according to Rizzi & Shlonsky (2005)), and null-subject languages can circumvent them owing to the availability of a different position from which movement can take place in finite clauses introduced by C, the ‘freely inverted’ subject

Ian Roberts and Anders Holmberg

22

position, as discussed above.13 This claim has very clear explanatory force in relation to the poverty of the stimulus: the acquirer encountering the relatively accessible phenomenon of free inversion in the PLD will thereby ‘acquire’ the possibility of the complementiser-trace violations, an otherwise fairly inaccessible aspect of the PLD. Moreover, we can in fact combine the implications in (29) to give a modest implicational scale along the following lines: (30)

Free Inversion ! (allow that-trace violations ! expletive null subjects).

This scale defines three types of language: Type I has all three properties; Type II allows complementiser-trace violations and hence allows expletive null subjects, and Type III only allows expletive null subjects (here ‘allowing complementiser-trace violations’ is intended to mean that no manipulation of C is required for long wh-movement of the lower subject to be possible, hence the varieties of English and Mainland Scandinavian discussed above are not relevant here). It is worth noting that we have weakened the original clustering claim to a series of one-way implicational statements; in other words, the claims in (30) are weaker and therefore easier to support empirically than those in (27). As we have now observed several times, this seems desirable for establishing parameter-based typologies. This seems to be due to the intricate nature of parameter interactions. In his study of the Null Subject Parameter in creoles, Nicolis (2005; 2008) observes that Cape Verdean, Berbice Dutch, Kriyol, Mauritian, Papiamentu and Saramaccan all have expletive null subjects and tolerate that-trace violations, but do not allow referential null subjects (Nicolis 2008: 275); while Haitian and basilectal Jamaican have expletive null subjects but do not tolerate that-trace violations. No creole has free inversion. So we have the following picture:14 (31)

Type I: Italian, Spanish, Greek, etc. Type II: Cape Verdean, Berbice Dutch, Kriyol, Mauritian, Papiamentu, Saramaccan Type III: Haitian, basilectal Jamaican

So we see that a very strict, theory-driven typology such as that which emerges from Rizzi (1982) makes very strong predictions which almost immediately 13

14

On this basis, we might have expected (29b) to go the other way, stating a one-way implicational relation between complementiser-trace violations and free inversion. But as discussed above, languages have different strategies for avoiding complementiser-trace violations, including an operation on C, as in English and French. On the other hand, (29b) makes the highly substantive claim that any language with free inversion will allow complementiser-trace violations. See Roberts (2007b: 31ff.) on the possibility of including referential null subjects in the scale in (31).

Introduction: parameters in minimalist theory

23

become very difficult to evaluate as soon as the cross-linguistic database is extended only modestly. Unsurprisingly, it becomes impossible to evaluate once further languages are taken into consideration, as shown by Gilligan’s survey. When a very large number of genetically and typologically highly diverse languages are compared for the ‘same’ properties, with no control as to the other typological features of these languages, the original correlations were shown not to hold in their original form, although four implicational statements could still be gleaned. To us, this does not seem like a bad or shocking result for parametric theory, but rather a fairly promising result from the admixture of a very large amount of essentially random data into an originally carefully controlled database. The fact that any coherent patterns survived is telling, and a sign that Rizzi’s observations were clearly on the right track. We conclude that the claim that results such as Gilligan’s invalidate the particular parametric cluster proposed is not warranted since: first, we simply do not know enough about dozens of problematic languages in order to be sure whether they are genuinely counterexamples; second, the most intriguing, non-obvious and explanatory conclusion of Rizzi (1982) remains (namely, the implication in (29b)); and third, a more modest implicational hierarchy can be constructed on the basis of Gilligan’s results, one which shows that Rizzi’s cluster has some significance for language typology. In the next section, we review the consequences of drawing the opposite conclusion from this one. 2.3

Newmeyer (2004; 2005; 2006): negative conclusions

We have seen that the kind of clustering of properties predicted by a parameter like the NSP can be construed as making very strong typological predictions, but the exact evaluation of these predictions may in practice be rather difficult. At every step, possible counterexamples and difficult cases proliferate, and, although in our judgement Rizzi’s original proposals certainly have a valid cross-linguistic core, it is of course always possible that a given proposal is wrong. This alone would not, of course, invalidate the approach. In earlier work (Roberts & Holmberg 2005), aside from defending parametric theory in general against a range of specific charges made against it in Newmeyer (2004), we suggested a different example of parametric clustering. This concerns an abstract feature of Agr, or T, which accounts for a range of differences between the Insular Scandinavian languages (ISc) and the Mainland ones (MSc) involving the possibility of null non-referential subjects, non-nominative subjects, stylistic fronting, V-to-T in embedded clauses and relatively rich subject–verb agreement (see Holmberg (Chapter 2)). Newmeyer’s

24

Ian Roberts and Anders Holmberg

(2006) response to this is that the parametric cluster in question lacks sufficient cross-linguistic justification: For the facts in (1a–e) [i.e. the correlating syntactic properties just listed, IR/AH] to support parametric theory, it would be necessary to demonstrate that in language after language the constellation of properties particular to ISC and MSC reappear. But aside from some brief remarks about Middle English, Yiddish, and Old French, two of which, like Scandinavian, are Germanic, and the third of which was heavily influenced by Germanic, they ignore the typological dimension entirely. (Newmeyer (2006:3))

Here it becomes clear that Newmeyer sees it as a requirement that, in order for any observed, or predicted, cluster of properties “to support parametric theory”, what he refers to as “the typological dimension” must be taken into account. As becomes clear from the following discussion of each of the surface properties we proposed to be connected to a single formal grammatical property, what he calls the “typological dimension” means bringing in as much unanalysed data from as many unrelated languages as possible. For the reasons given in the previous section, the likelihood is that this approach will only obscure any patterns that may be observable in more controlled data samples (although, as we have observed, any partial or weakened generalisations will be all the more striking, as is the case for (29b)). But for Newmeyer only the strongest possible justification can satisfy the “typological dimension”: every language must fall on one side or the other of the proposed parametric divide (i.e. every parameter must be shown to define a cluster corresponding to the value Pþ, and the complementary cluster C0 ¼ P– in terms of (18–21) above), and this must be ascertainable on the basis of a superficial survey of the relevant properties. If not, the parameter lacks empirical support. But almost no known distinction among languages will meet this desideratum for empirical support. For example, the correlation between VO vs. OV order and pre- vs. postpositional order has 141 languages defined as not falling into one of the four types, and 48 straight counterexamples, according to Dryer (2005: 386). Taking Newmeyer’s “typological” stricture seriously would presumably cast doubt on this correlation, too. More generally, it will simply result in the abandonment of any attempt to establish cross-linguistic generalisations. Newmeyer (2004; 2005; 2006) concludes from his discussion of the NSP, and equally superficial discussion of a range of other proposed parameters, that parameter-based approaches to cross-linguistic variation “have failed to live up to their promise” (2005: 181) and that “the hopeful vision of UG as providing a small number of principles each admitting of a small number of parameter settings is simply not workable” (p. 185). Instead, he suggests that “language-particular differences are captured by differences in languageparticular rules” (p. 183). The nature of these ‘rules’ is never really clarified,

Introduction: parameters in minimalist theory

25

and so the last claim is hard to evaluate. One case he discusses, though, is that of the head–complement parameter, which, according to Newmeyer (2004; 2005: 74) should be replaced by two rules: (a) Complements are to the right of the head, and (b) Complements are to the left of the head. Some languages apply (a), other languages (b). In Roberts & Holmberg (2005) we pointed out that this is equivalent to a parameter in the sense of P&P theory: UG leaves two options open, and acquisition of head–complement order is a matter of choosing between the two options on the basis of experience. Newmeyer (2006) agrees, but insists that if a set of grammatical facts can be characterised as a rule, then it should be, rather than being characterised as the value of a parameter.15 However, if rules are, at least in certain cases, empirically equivalent to parameters, we can conclude that, in his view, they will fare just as badly, by his criteria, in accounting for cross-linguistic variation. At the very best, the proposed rules are as inadequate as he proposes parameters are. At worst, they may be much worse: the only construal of ‘rule’ in the literature on generative syntax refers to phrase-structure rules and transformational rules. Such rules were thought of as largely language-particular, language acquisition consisting of the selection among the infinite class of such rules compatible with the PLD on the basis of an evaluation metric. If this is what Newmeyer is advocating, it is clearly a retrograde step, and undoubtedly represents a retreat from explanatory adequacy (the same observation is made in Dryer’s (2007) review of Newmeyer (2005): cf. Dryer’s (245–46) remark that “Newmeyer argues for . . . a retreat to a version of C[homskyan]G[enerative]T[heory] from the period of Chomsky (1973) up to, but not including Chomsky (1981)”). If it is not what he is advocating, he owes us an account of his notion of rule. Even then, if he retains the view that rules are largely equivalent to parameters, there is no reason to adopt his approach over a parameter-based one, since, as he has argued, the parameter-based approach is inadequate. What must be shown is that the notion of rule, as opposed to parameter, is both empirically superior to the notion of parameter (i.e. more descriptively adequate) and can take us closer to understanding UG, i.e. to explanatory adequacy, than the notion of parameter. But as far as we can tell, Newmeyer merely gives up the pursuit of explanatory adequacy along with the notion of parameter. For this reason, we cannot accept his conclusions. Implicit in the above discussion has been a distinction between two approaches to universals, or to cross-linguistic correlations of any kind. 15

A possible reason behind Newmeyer’s opposition to characterising the choice between head-precedes-complement and complement-precedes-head as two values of a parameter is that, in this particular case, the values are not given by properties of UG, but rather follow from the ultimately physical fact that words must be linearly ordered, allowing exactly those two options. We return briefly to this point in Section 3.6.

26

Ian Roberts and Anders Holmberg

These are, roughly speaking, the Chomskyan approach and the Greenbergian approach. The Chomskyan approach is based on the pursuit of explanatory adequacy. Hence the central idea is that a given piece of grammatical knowledge hardly accessible through experience given normal assumptions about the nature of the PLD can be acquired ‘for free’ given the existence of a correlating, more accessible piece of evidence. Rizzi’s discussion of the NSP meets this desideratum, and the survival of implication (29b) under Gilligan’s empirical scrutiny is highly significant in this connection. The other approach involves the observation of properties which covary on the surface, without prejudice as to any deeper correlations or poverty-of-the-stimulus considerations. The empirical scope of the generalisations unearthed in this tradition is impressive (see in particular Haspelmath, Dryer, Gil & Comrie 2005), and seems to set an important research agenda, and a series of empirical challenges, to generative approaches committed to more abstract analyses and, in particular, to achieving explanatory adequacy. As Gianollo, Guardiano & Longobardi (henceforth GGL, 2008) point out, both approaches to typology have flaws, particularly if they are seen as heuristics for the accumulation of information regarding which properties (co-)vary and which do not. As they point out, the Chomsky–Rizzi approach, in which poverty-of-the-stimulus considerations remain paramount, “provides deep and often correct grammatical insights, but, alone, it may lead little further than the study of single parameters in a few languages” (p. 113). We have seen above the reasons why this may be so. The Greenbergian approach, they say, is subject to two objections. First, “it is practically hardly usable for the relevant purposes, because of the depth of grammatical investigation required by sound parametric analysis” (again, we have made this point in the foregoing). Second, and perhaps more challengingly, “it is anyway likely to be insufficient if investigation is not guided by strong abductive and theory-oriented considerations, even if hypothetically extended to all languages, because the cardinality of the universal parameter set is so wide as to generate a number of possible languages of which the actual existing or known ones represent an infinitesimal sample” (p. 113). In other words, typological observations, however well-supported by data from the currently available set of languages, may nevertheless represent accidental correlations coexisting contingently at this historical moment, and not reflect the true nature of a UG which generates a far larger set of languages than those currently extant. This second point, in fact, undermines Newmeyer’s objection to parameters such as the one put forward by Holmberg & Roberts (2005), whose effects are most visible in Germanic languages, on the grounds that “it leaves open the possibility of historical accident (language contact, drift due to some non-parametric cause)” (Newmeyer 2006: 5); so does the Greenbergian approach, only on a wider historical and geographical scale.

Introduction: parameters in minimalist theory

27

GGL go on to advocate a very interesting kind of ‘halfway’ approach, based on their notion of Modularised Global Parametrisation, which they argue is largely immune to both types of objection. Whatever the merits of GGL’s approach, we suggest that the fundamental error in Newmeyer’s critique of parameters is that he conflates the two original approaches. He therefore judges parameters proposed in the Chomskyan tradition, and which have something to offer from the point of view of explanatory adequacy, by the extensional, taxonomic, surface-oriented criteria of the Greenbergian approach. This inevitably leads to his highly negative assessment of work such as Rizzi’s on the NSP. Essentially, any parameter would have to show clustering effects which exhaustively partition all languages into two classes, with this partitioning immediately accessible to superficial scrutiny in all cases, i.e. with no masking effects due to extraneous properties or extensional overlaps in coverage. Ideally, some of the clustering properties would be relatively inaccessible in the PLD, giving the parameter explanatory depth. Of course, no proposed parameter (or Greenbergian universal) meets such stringent requirements. In the case of the NSP, for example, any and every kind of phonologically non-realised subject ought to fall under its purview, and this mistake underlies much of the criticism of the nature of the parameters and the alleged failure of the correlations. But it has been known since the earliest work that there are at least two, if not three, different kinds of null-subject systems: the Italianstyle ‘consistent’ system, those allowing only non-referential null subjects and the East Asian type of ‘discourse pro-drop’ (more recently, ‘partial’ systems have been added to this list, as we have seen). Newmeyer arrives at his position because his critique of the parametric approach conflates and confuses the aims and methods of two distinct traditions. Given this, it is no surprise that he advocates a retreat from the goal of explanatory adequacy. And it is fundamentally for this reason that we cannot accept his conclusions. 2.4

A performance-efficiency based alternative

So, after setting the bar for descriptive adequacy for any proposed parameter impossibly high with his allusion to the “typological dimension”, Newmeyer proposes a retreat from explanatory adequacy in the form of a return to the rule-based systems of the 1960s and 1970s, thereby dissolving the link between acquisition and typology discussed in Section 2.1 above. Nonetheless, he does believe that a number of typological observations (mostly made in the Greenbergian tradition) reflect genuine cross-linguistic generalisations. He suggests, however, that these are amenable to “performance explanations” of the type advocated by Hawkins (1994; 2004).

Ian Roberts and Anders Holmberg

28

Hawkins’ central idea is the Performance-Grammar Correpondence Hypothesis (PGCH), which we state as follows: (32)

Grammars have conventionalised syntactic structures in proportion to their degree of preference in performance, as evidenced by patterns of selection in corpora and ease of processing in psycholinguistic experiments. (Hawkins 2004: 3, cited in Newmeyer 2005: 119)

Preferences are the reflection of efficiency: “speakers attempt to increase efficiency by reducing structural complexity” (Newmeyer 2005: 122–23). Hawkins proposes three efficiency principles, one of which is Minimise Domains (MiD). This efficiency principle is said to be able to account for the fact that OV and postpositional orders tend to cooccur and VO and prepositional orders tend to cooccur, but OV tends not to cooccur with PO and VO tends not to cooccur with OP. The idea is that in the consistently ‘head-initial’ (VO and PO) and ‘head-final’ (OV and OP) orders, the ‘distance’ from V to P or P to V is shorter than in the inconsistent orders. The structures in (33) and (34), adapted from Newmeyer (2005: 124), illustrate: (33)

a. [VP V NP [PP P NP]] b. [VP [PP NP P] NP V]

(34)

a. [VP V NP [PP NP P]] b. [VP [PP P NP] NP V]

VO and PO (not far from V to P) OV and OP (not far from P to V) VO and OP (too far from V to P) OV and PO (too far from P to V)

According to these schematised structures, in the frequently occurring orders of (33), only one NP intervenes between the two heads V and P, favouring these structures on the grounds that they are easy to process. The rarer orders in (34) are disfavoured because two NPs intervene. Newmeyer concludes that MiD and the similar efficiency principles can do the work that parameters have been proposed to do, only better. Although there are a number of obvious objections that can be raised against this rather simplified account,16 we will concentrate our discussion here on the conceptual issues. The key point, we submit, is the notion of 16

These include: (i) Why are all verbs with prepositional complements assumed to also have a direct object? English and most other languages have verbs like rely, hope, depend, etc., which can take PP but not NP complements. Do we therefore expect a different typological generalisation for this type of verb? If not, why not? (ii) Why is V’s PP complement assumed to pattern with its NP complement such that both always systematically precede/follow the verb? In other words, why are orders such as the following not considered? (i) (ii)

[VP NP V [PP P NP]] (OV and PO) [VP [PP NP P] V NP] (VO and OP)

Introduction: parameters in minimalist theory

29

‘efficiency’. This is left implicit in Newmeyer’s account, although some general notion of ‘ease’ of processing is clearly intended. Since we are dealing with a cognitive capacity, processing as an aspect of performance, it is reasonable to think of efficiency in terms of reduction in computational effort. If that is so, then, following Mobbs (2008), we suspect that there may indeed be a case to made for viewing typological skewings as the consequence of general computational principles. Rather than applying at the level of performance and processing, we suspect, like Mobbs, that they may be more deeply implicated in the organisation of the language faculty and the learning device. We will return to these points in Section 3.5 below, and look a little more at the other efficiency principles proposed by Hawkins, in addition to MiD.

2.5

Conclusion

Here we have described and illustrated the idea of parametric clustering in relation to Rizzi’s classic (1982) work on the NSP. We have observed the difficulties in straightforwardly generalising the predictions without further

Here the ‘distance’ – however that is computed – between V and P is smaller than in the crosslinguistically common cases in (33). Therefore, all other things being equal, we expect to find these more frequently than (33), which is not the case. ‘Mixed’ orderings of complements like these do exist: depending on one’s analysis of elements such as Mandarin and other Chinese varieties may have [VP [PP P NP] V NP] (see Li’s (1990) analysis of the ba-construction), which is predicted to be as common as (33) by MiD, but is cross-linguistically rare. German and Dutch allow the order in (i) in subordinate clauses with ‘PP-extraposition’; again, this pattern is predicted to be more frequent than (33) and (34), which is very probably not the case. These objections concern the presence of an ‘extra’ NP in the representations given by Newmeyer and schematised in (33). Hawkins (1990: 238–39) leaves the extraneous NP out, and makes essentially the same argument. Here, too, some objections can be made. First, what about adjunct PPs, which presumably occur in a different configuration from those in (33) and (34); do we predict a different typological generalisation for these? Second, Dryer’s (2005: 386) figures show that (34b) is four times as rare as (34a). Why? Third, Dryer states that in 141 of the 1,033 languages he surveys, the word order cannot be determined. It may be that some of these languages have ‘free’ word order on the surface. For Newmeyer/Hawkins, such orders, if truly free, would presumably be the hardest of all to process, since there is no consistent ‘distance’ in the ‘domain’ (or perhaps the easiest, for the same reason). But actually they are neither the rarest type, being much commoner than the inconsistent types in (34), nor particularly common, being much rarer than the consistent types. Of course, it would be possible to assume an underlying fixed order as in generative work, but such a move will make a processing account hard to maintain and raise the possibility of movement operations perturbing underlying order in the other cases. Whichever way things are construed, then, more than 10 per cent of the languages in Dryer’s sample pose an insuperable problem for Hawkins’ approach. Newmeyer’s strictures regarding the ‘typological dimension’, so ruthlessly applied to generative work, mean that he cannot simply appeal to analyses that have not yet been carried out. Jumping the gun somewhat, then, we find Newmeyer’s preferred alternative as unconvincing as his arguments against the Principles-and-Parameters approach.

30

Ian Roberts and Anders Holmberg

analysis and/or taking into account interfering factors (such as VSO order in the Celtic languages). Newmeyer’s critique of the NSP and of parameterbased approaches to cross-linguistic work completely overlooks this point and conflates the two distinct approaches to comparative syntax that have emerged in recent years. This leads to him setting the bar for even descriptive adequacy for the postulation of parameters impossibly high, with the logical but regrettable result that he retreats entirely from the goal of explanatory adequacy in comparative syntax. We briefly illustrated his preferred approach, based on Hawkins’ Performance-Grammar Correpondence Hypothesis and associated efficiency principles, which we return to in Section 3.5. By and large, though, we reject Newmeyer’s conclusions regarding the Principles and Parameters approach to comparative syntax. At the same time, we recognise that the kind of clustering effect discussed in Sections 2.1 and 2.2 is of limited typological interest, and that extending the typological purview of parameters like the traditional NSP may be problematic (although the brief discussion in Section 2.2 indicates that weaker clusterings and correlations can be envisaged). What is needed, as also pointed out by Biberauer (2008a), is a more systematic basis for extending the typological domain in which theoretically inspired parameters may be more readily evaluated. We now turn to this point. 3

Parametric theory

3.1

Problems with the ‘classical’ formulation

Although we believe that aspects of Newmeyer’s critique of parameter-based comparative syntax are seriously misguided, and, for the reasons given above, we cannot accept his conclusions, it is nonetheless true that there are problems with the original conception of parameters. In our view, this warrants a refinement of the idea, rather than its abandonment (see also Biberauer 2008a). One valid criticism is raised by Newmeyer (2005: 83). A consequence of the large amount of comparative work that has been done since the 1980s is that there has been a proliferation of parameters as descriptive devices. This has been particularly apparent in the ‘microparametric’ work on closely related languages and dialects, typified by the papers collected in Kayne (2000; 2005b). Given this proliferation of parameters, a natural question which arises is, quite simply, how many parameters there are. It is very likely that the number of parameters is in the hundreds, and at least possible that it is the thousands. For example, Roberts (2007b) discusses five well-known parameters (null subjects, V-to-T, T-to-C, negative concord, wh-movement). The ‘head parameter’ is arguably non-unitary, and breaks up into several sub-parameters, probably at least ten. There is also a parameter governing

Introduction: parameters in minimalist theory

31

subject-raising to SpecTP (whose negative value, along with a positive value for V-to-T raising, gives rise to VSO orders), at least four parameters governing auxiliary selection (and this does not take into account the impressive microvariation found in Central-Southern Italo-Romance), several parameters governing ergativity (to account for varieties of split ergativity), presumably pertaining to the feature content of v, a parameter concerning the ability of C to Agree for Case with the subject of the TP it introduces (related to ‘Exceptional Case-marking’), a parameter determining the availability of the option of preposition-stranding and a parameter concerning whether wh-expressions are DPs. This brings the total to 24. Adding Polysynthesis, Subject Side and Serial Verbs from Baker (2001), the total comes to 27. Baker (2008b) proposes two macroparameters governing agreement (Direction of Agreement and Case-Dependency of Agreement). Furthermore, the nullsubject parameter may break up into at least three parameters (see Holmberg (Chapter 2)). Finally, Longobardi & Guardiano (2008) propose 51 parameters which affect DP-internal syntax only. This brings the number of parameters up to 80, and, there is little doubt many more than this would be needed just to reach descriptive adequacy. Newmeyer concludes that “we are not yet at the point of being able to ‘prove’ that the child is not equipped with 7,846 . . . parameters, each of whose settings is fixed by some relevant triggering experience. I would put my money, however, on the fact that evolution has not endowed human beings in such an exuberant fashion” (2005: 83). Although, as Newmeyer implicitly admits, this is only a plausibility argument, we agree with him. It seems highly implausible that UG should specify detailed microparameters governing the nature of clitic systems or agreement systems (or classifier systems or tone systems) when so many languages lack such systems entirely. Clearly, what is needed is some structure to parameter systems, at the very least along the lines of specifying ‘if L has a clitic/agreement/tone/classifier system, then what particular kind of system does L have?’, where the consequent may break down into a further series of implicational choices. A related point arises from the discussion above of the correlations connected to the NSP in (22) and (27). All other things being equal, this type of clustering, deriving from a parameter closely linked to a UG principle (such as, in this case, licensing of a particular empty category), does yield a series of biconditional relations determining a range of properties that are predicted to be either all present or all absent in any language. We saw above, very briefly, the kinds of analytic difficulty that approach can entail. What is preferable, in fact, is weaker, one-way implications of the kind in (29), or the kind typically put forward in the Greenbergian tradition. But these, too, require a rather different approach to the nature of parameters than the classical one illustrated by Rizzi (1982).

Ian Roberts and Anders Holmberg

32

Arguably the real issue underlying both of these points is the familiar tension between descriptive and explanatory adequacy. Parameters have in recent years to an extent shared the fate of 1960s-style transformations. They are very powerful formal devices that make possible, for the first time ever, the precise, theory-internal description of cross-linguistic relations (and, correspondingly, descriptions of what children must be able to acquire). However, if over-exploited, and especially in the absence of any general restrictions on their form and functioning, these devices become mere facilitators of taxonomies. Newmeyer quite correctly observes that the very large number of parameters that we seem to need, just to approach cross-linguistic descriptive adequacy, reduces their efficacy as explanatory devices. What is required, as took place with the theory of transformations from the late 1960s onwards, is a theory of parameters which places substantive restrictions on their form and function while maintaining their descriptive power.17 3.2

The minimalist view

In the context of the Minimalist Program for linguistic theory, as it has been pursued in relation to syntax for almost two decades, a slightly different view of parameters has been widely accepted. This can be thought of, following Baker (2008a: 3; 2008b: 155f.), as the ‘Borer–Chomsky conjecture’, or BCC: (35)

All parameters of variation are attributable to differences in the features of particular items (e.g. the functional heads) in the Lexicon.

More precisely, we can restrict parameters of variation to a particular class of features, namely formal features in the sense of Chomsky (1995b) (Case, j and categorial features) or, perhaps still more strongly, to attraction/repulsion features (EPP-features, Edge Features, etc.). This view has a number of advantages, especially as compared with the earlier view that parameters were points of variation associated with UG principles, and arguably takes us a little way towards resolving the tension between descriptive and explanatory adequacy at the parametric level as described in the previous section. Let us see why this is so, particularly in relation to the latter point. First, the BCC imposes a strong limit on what can vary. In most versions of syntactic theory informed by minimalist ideas there are a small number of extremely general principles: Merge, Agree, Select, etc. It is highly unlikely 17

A similar point is made in connection with diachronic linguistics by Roberts (2001). GGL (2008: 6) observe a tension between explanatory and what they term ‘evolutionary’ adequacy: “once parameters are included by a theory of UG, the minimization of the genetic endowment (produced by the supposed economy constraint on stored innate knowledge) should probably amount to minimizing the number of parameters as well as resulting, if anything, in a reduction, rather than in the observable extension, of the space of variation.”

Introduction: parameters in minimalist theory

33

that any of these are subject to parametric variation (pace Baker (2008a)). For example, it would be strange to propose a parameter determining whether or not Merge has to be binary, or whether or not internal Merge (i.e. Move) is possible (Kayne (2008b: 1) suggests that internal merge may be a defining feature of human language, i.e. the ‘narrow language faculty’ in the sense of Hauser, Chomsky & Fitch (2002)). Similarly, although languages may vary a fair amount as to which features may be subject to Agree relations, the nature of the operation itself (the locality restrictions operative, the nature of Match, the definition of an active probe, etc.) seems unlikely to vary. Of course, we have no proof that these kinds of variation do not exist, but they seem unlikely, have not – to our knowledge – been postulated and would be excluded by the BCC. Second, as has often been pointed out (initially by Borer 1984: 29), an advantage of the BCC is that associating parameter values with lexical entries reduces them to the one part of a language which clearly must be learned anyway. Ultimately, on this view, parametric variation reduces to the fact that different languages have different lexica, in that sound–meaning pairs vary arbitrarily: the most fundamental and inescapable dimension of cross-linguistic variation. The child acquires the values of the parameters valid for its native language as it acquires the vocabulary (more precisely, as it acquires the formal features associated with the functional categories of its native language). A third advantage of the BCC is that it imposes a restriction on the form of parameters. The attraction property of a functional head, which may be responsible for fairly major aspects of word-order variation, can be formulated in terms of a simple diacritic associated with a given functional head, or a given feature of a functional head. For example, the matrix C of a V2 language may be associated with a movement-triggering diacritic, causing an XP to move into its specifier position; in a ‘residual V2’ language like English or French, the diacritic would be associated with a [wh]-feature of C, giving rise to movement of a wh-XP into SpecCP exactly where C has the [wh]-feature. In a language like Chinese, where there is no movement to SpecCP at all, there would be no such feature associated with C. We will give a more general formulation of possible parametric variation, consistent with the BCC, in Section 3.4 below. Clearly, if parametric variation can be restricted to formulations of this type, this would be a step in the direction of explanation. This simplicity of formulation in turn makes possible a statement of parametric variation at the UG level which relies on the logic of underspecification. It may, for example, be possible to state that parameter P has value vi, perhaps a movement-triggering/attraction property, when this is stated as such, and vj, the absence of this property, elsewhere. This in turn

34

Ian Roberts and Anders Holmberg

raises the prospect of applying disjunctive ordering to parameter setting, and thereby the setting up of markedness relations. For the moment, we merely note that the simplicity of the formulation of parameters, given the BCC and the typical minimalist conception of parameter setting, allows this. In subsequent sections we will explore the possible implications of this idea further. A further point, also made by Roberts (2001) and Roberts & Roussou (2003), is that the BCC, combined with a simple restriction of parameters to formal features of functional heads, allows us in principle to calculate the upper bound to the set of grammars. Suppose we have two potential parameter values per formal feature (i.e. each feature offers a binary parametric choice), then for n ¼ jFj, the cardinality of the set of formal features, the cardinality of the set of parameters jPj is 2n and the cardinality of the set of grammatical systems jGj is 2n. For (rather implausible) illustration, assume 30 formal features, then we have n ¼ 30, then jPj ¼ 60 and jGj ¼ 230, or 1,073,741,824. One further parametrisable formal feature raises jGj to 231, and so on. Since it is likely that UG makes available more than 30 formal parametrised features, the upper bound to the number of possible grammatical systems is likely to be considerably greater than just over a billion, as in the example just given. Of course these are upper bounds; what is not taken account of is interaction among parameters so as to rule out possible combinations of values; as GGL (2008) show, such interactions are likely to be pervasive. The lower bound is therefore very likely to be a great deal lower than the kinds of numbers resulting from the above calculation; at this stage it is hard to determine a general characterisation for this (for a general speculation, derived from Kauffman’s (1995) work on dynamical systems, see Roberts (2001: 91–93)). Although not in itself a step towards greater explanatory adequacy, this feature of the BCC has the heuristic value of allowing, in principle, an exact quantification of the cost of the postulation of a new parametrised formal feature. What has the above to do with the NSP? Rizzi (1982: 143) identified the core property allowing null subjects as a ‘pronominal Agr’ in Infl in terms of the system he was assuming. In more contemporary terms, this can easily be restated as the presence of a D-feature on T (see Section 1.3, (17)), or, following Barbosa (1995) and Alexiadou & Anagnostopoulou (1998), on the verbal inflection itself. These two alternatives correspond to the two basic hypotheses for accounting for (non-discourse-driven) null subjects in Holmberg (2005: 536–37) and discussed in Roberts (Chapter 1). If the requirement for a subject in SpecTP (Chomsky’s (1982: 10) Extended Projection Principle) can be reduced to the realisation of a D-feature on T or its Specifier, then there is no need for the subject to raise to SpecTP in a language with D in T, giving rise to the possibility of free subject inversion and the associated possibility of extraction from the inverted position, evading the ban on extraction from

Introduction: parameters in minimalist theory

35

SpecTP when C is finite and overtly realised (the complementiser-trace effect). Finally, if the D-feature is associated with rich agreement (see Roberts (Chapter 1), for a proposal for this), then we arrive at an account which ties together all four clustering properties associated with the NSP. Hence it is very easy to formulate the NSP in terms of a formal feature of a functional head.18 Finally, although the idea of reducing parameters to formal features of functional heads has largely been associated with ‘microparametric’ approaches, it is not necessarily restricted to that case. In fact, we have just sketched a way to replicate the clustering predictions of Rizzi’s original NSP in these terms. Such clustering is usually seen as the hallmark of a ‘macroparameter’, Baker (2008a: 4) points out that it “is perfectly possible that a lexical parameter consistent with [the BCC, IR/AH] could have a substantial impact on the language generated, particularly if it is concerned with some very prominent item (such as the finite Tense node)”. This is, in fact, exactly what we have just seen in relation to the NSP. In this section we have described the BCC, the idea that parametric variation is associated with formal features of functional heads, and we have indicated what we see as some of its advantages. We have pointed out that, by postulating a restriction on what can be parametrised, on the form of parameters, and, possibly, interactions among parameters determined by disjunctive ordering (and therefore markedness relations), the BCC can take us some way to resolving the tension between descriptive and explanatory adequacy we observed in Section 3.1. However, if we take seriously the idea that parameters are lexically determined, even as properties of functional heads, we run the risk of seeing them as ‘less universal’ in various ways. For example, could there be a language-particular parameter? This presumably depends on whether we are willing to countenance language-particular features of functional heads. A second question is whether a single language, i.e. Lexicon, can tolerate contrary specifications of the same parameter on distinct lexical items. Third, can a parameter simply fail to have a value in a given language, by not having a lexical item which realises this feature? (Thanks to David Willis for raising these issues.) We will return to this last question in Section 3.4 below; GGL, for example, allow for this possibility. But the most important objection to a purely microparametric approach remains the question of the highly specific innate endowment in formal features that the BCC appears to demand; this is clearly a stumbling block to true explanation. We will look at this question more closely in the next two sections. 18

Newmeyer (2005: 208) asserts that the minimalist approach to parameters, which involves seeing them as inherently connected to features of functional heads, “makes it all but impossible to predict any significant degree of clustering”. This assertion is false, as we have just seen.

36

3.3

Ian Roberts and Anders Holmberg

Micro- and macroparameters

In introducing the BCC in the previous section, we have already introduced one way of thinking about the distinction between microparameters and macroparameters. Although it does not require it, as we saw, the BCC favours a microparametric approach. According to this view, cross-linguistic variation consists of variant features of (a subclass of) lexical items which determine a small range of variation, and larger-scale differences among languages represent the accumulation of numerous microvariants of this kind. On the other hand, macroparameters such as the Polysynthesis Parameter of Baker (1996) and, possibly, the Head Parameter determine in one fell swoop a huge range of possibly variant properties. As Baker (2008a: 5) puts it “there are at most a few simple (not composite) parameters that define typologically distinct sorts of languages”. Baker gives interesting arguments for the existence of macroparameters alongside microparameters. In addition to two arguments based on the crosslinguistic distribution of different kinds of agreement marking (these are developed at greater length in Baker (2008b), but we will not go into them here), he gives an interesting statistical argument. Essentially, his point is that if all variation were microvariation, we would not expect to find coarsegrained types of the ‘head-initial’, ‘head-final’ kind. If each category were able to vary freely, independently of all others, for its linear order in relation to its complement (this can of course be phrased in terms of triggering movement of its complement or not), then we would expect there to be a normal distribution of word-order variants across languages. As he says (Baker 2008a: 10), “there should be many mixed languages of different kinds, and relatively few pure languages of one kind or the other”. On the other hand, if there were only macroparameters, we predict, falsely, the kind of situation described above in relation to the NSP: every category in every language should pattern in one way or the other. But if we admit both macroparameters and microparameters, we expect to find a bimodal distribution: languages should tend to cluster around one type or another, with a certain amount of noise and a few outliers from either one of the principal patterns. This, Baker points out (again drawing on the statistics for OV/OP vs. VO/PO order in Dryer (2005: 386)), is essentially what we find. Baker suggests (pp. 11–12, citing his earlier 1996 work) that the same is true regarding polysynthesis. We find Baker’s argument fairly convincing. One might add a similar diachronic argument. A canonical example of microparametric variation comes from the Italo-Romance dialects. Although the variation among these varieties is highly impressive (as the 2,500 pages of Manzini & Savoia (2005) amply attest), a large number of features remain constant: all Italian dialects

Introduction: parameters in minimalist theory

37

are SVO, all are prepositional, none show a systematic ergative case/ agreement pattern (although some ‘split-ergativity’ is attested), none is fully polysynthetic, none shows the Chinese value of Chierchia’s (1998) Nominal Mapping Parameter (i.e. in allowing a singular count noun to stand alone as an argument, giving I saw cat), all have definite and indefinite articles, all have moderately rich agreement systems, all (except a small number of Rhaeto-Romansch varieties; see Beninca` & Poletto (2005)) have complement clitics, none has a full morphological case system, etc. On the other hand, the microparametric variation involving the existence and behaviour of subject clitics, the expression of negation, the position of both finite and non-finite verbs in relation to subject and complement clitics and various classes of adverbs, the nature of object and subject agreement on past participles in compound tenses, the nature and choice of aspectual auxiliaries, the expression of various forms of finite and non-finite complementation, and a range of other properties, is extremely intricate. These are exactly the conditions which favour productive microparametric work, as Kayne has convincingly argued (see in particular Kayne 2005a). But, one could ask, why are certain properties variable in Italo-Romance and others not? The microparametric answer, as it were, is that no theoretical significance should be attached to what varies and what does not in this particular synchronic geographically defined domain; this is attributable to a historical accident, in that the common features are due to a shared inheritance. But if we consider Latin, we find OV order, a full morphological case system, the complete absence of pronominal clitics, no (active) compound tenses, and a system of complementation in which finite clausal subordination was a minority pattern. As has often been observed, the Modern Romance languages (or the Italo-Romance subgroup) are more similar to one another than any of them are to their common ancestor Latin. The microparametric explanation for this observation would presumably appeal to the accumulation of microparametric changes in the common ancestor language before it broke up into the dialects, i.e. in Late or Vulgar Latin. The question here, though, is to what extent Vulgar Latin can be reasonably regarded as a single system; the term is generally used as a cover term for the varieties of nonliterary Latin spoken in Italy and elsewhere in the Roman Empire, whose written records are somewhat uniform but have been argued to form a koine´ (Palmer 1961: 223). In this connection, Clackson (2004: 790) says: “the construction of a uniform ‘Vulgar Latin’ probably oversimplifies a very complex linguistic situation. Different communities of speakers used different varieties.” If there ever was a single ‘Proto(-Italo)-Romance’ variety, it would probably have to be dated rather early, as Hall (1950) suggests on phonological grounds (proposing 250–200bc, exactly the period in which Roman rule was extended to the whole Italian peninsula). Although the Latin of this

38

Ian Roberts and Anders Holmberg

period is known to differ somewhat from Classical Latin, and to have certain ‘Vulgar’ features, it is highly unlikely that it had the syntactic characteristics of Romance rather than Classical Latin (OV rather than VO order, etc.). It seems, then, that the current microparametric variation either derives historically from an archaic, typologically distinct, single ancestor variety of Latin, or there is no ancestor variety common to all the dialects. Either way, the major typological differences between Latin and (Italo)-Romance cannot be traced to a single microparametric change or series of microparametric changes in a single variety; there must have been typological drift across the varieties of Vulgar Latin. This poses a problem for a purely microparametric approach: other things being equal, we might have expected some dialects to have retained a case system, or OV order, or synthetic passive forms and not to have developed clitics, etc., others to have developed in the way we observe, and still others to have developed in a mixed fashion, preserving certain archaic features and innovating others. But what we observe, instead, is typological drift: from OV to VO, and in the general direction of greater analyticity (as elsewhere in Indo-European), allowing for a considerable amount of truly microparametric variation of the kind that we observe to develop. The simplest account of this kind of parallel development involves distinguishing macroparametric from microparametric change: certain macroparameters (OV vs. VO, for example) changed in the transition from Latin to Romance, while much of the synchronically observable variation among the Romance languages, and certainly among Italian dialects, involves microparameters.19 So let us conclude, with Baker, that macroparameters exist alongside microparameters. Then two related questions arise: (i) What are their properties? (ii) How are they distinguished from microparameters? Two rather unsatisfactory and partial answers to these questions are that macroparameters ought to be rather few in number, and they ought to be extremely pervasive in their influence on the grammatical system. The first point holds because, as Baker (2008a: 7) says “[i]f there were many macroparameters and they 19

One could perhaps attempt a contact-based explanation for the parallel developments. The Southern dialects were in contact with Greek and the other Italic varieties of Indo-European: Oscan, Umbrian and related varieties, collectively known as Sabellian. These are broadly similar to Latin in typological terms, being predominantly SOV (Wallace 2004: 832), and so are unlikely to be responsible for the common development of the dialects. The Northern varieties were in contact with forms of Celtic and Venetic. In Tuscany and indeed in the early days of Rome itself, there was contact with Etruscan, a non-Indo-European SOV language (Rix 2004: 961). A very thorough study of contact between Latin and all of these languages, at various times and places, is Adams (2003). One possibility, which could have had far-reaching consequences, is that Latin demonstratives developed into articles partly due to contact with Greek, which had an article system. Adams (p. 518) points to the use of demonstratives as articles in a passage of Plautus. This is significant because it can be traced to Greek influence and because of its early date (see Adams’ discussion for details).

Introduction: parameters in minimalist theory

39

interacted with one another in complex ways, then languages could differ crazily in ways that would be hard to pull apart”. But this does not seem to be the case in practice: universal properties and microparameters account for much that is shared and much that varies. The second point holds because macroparameters, perhaps by definition, can affect large-scale aspects of the grammar such as all headed phrases, or all instances of Agree. In the preceding section we argued that the BCC, which Baker quite reasonably takes to underlie microparametric variation, has a number of desirable consequences, and may even take us some way towards constraining the form of parameters in the way that is required in order to resolve the tension between explanatory and descriptive adequacy which has arguably arisen in this domain. But here we have suggested that Baker is right in suggesting that a small number of macroparameters may also exist. Baker (2008a: 3) explicitly proposes that macroparameters are to be formulated in a manner incompatible with the BCC. So we appear to be in a quandary. We tentatively suggest a way out of this quandary which, we believe, points the way to truly resolving the tension between explanatory and descriptive adequacy in the parametric domain. This involves retaining a formally ‘microparametric’ view of macroparameters, i.e. seeing macroparameters as aggregates of microparametric settings, but as proposing that these aggregate settings are favoured by markedness considerations. This proposal was made in Roberts (2007a: 274) for the Head Parameter (and is suggested as an ‘intermediate’ approach to the question of macro- vs. microparametric variation by Baker (2008b: n. 2)). It has often been noted that the Head Parameter is rather problematic. If it were a single (macro)parameter, determining the order of head and complement across all categories once and for all, it would predict a spectacular clustering of properties, which is not actually attested in the majority of languages. If it is broken down into a series of related microparameters relating to each head–complement pair then, without some further statement, all predictions regarding word-order correlations are lost. The preference for ‘harmonic’ ordering seems to derive from an overriding tendency for independent parameters to conspire to produce a certain type of grammar. To capture this, Roberts (2007a: 194) suggested that a restatement of Hawkins’ (1983) generalisation regarding cross-categorial harmony is needed, along the following lines: (36)

There is a preference for the EPP-feature of a functional head F to generalise to other functional heads G, H . . .

We can think of (36) as an approximation to a markedness convention of the type proposed for phonology by Chomsky & Halle (1968: chapter 9).

Ian Roberts and Anders Holmberg

40

To take a specific example, suppose, following Kayne (1994), that VO is the universal underlying order and, following Biberauer (2003), that OV orders derive from the combination of V-to-v raising and remnant VP-fronting to SpecvP, as illustrated in (37): (37)

[vP [VP O (V)] vþV (VP)]

If movement represents a marked option, as suggested by Roberts & Roussou (2003), then v is set to a marked parameter value here. Following Chomsky & Halle’s notation, let’s write this as the mEPP value for v.20 In rigidly headfinal languages like Malayalam or Japanese, many, perhaps all, functional heads will have at least one EPP-feature of this kind. Such systems will therefore emerge as very marked indeed, in terms of what we have said so far, and yet they are more common than ‘mixed’ types like Latin, German, etc., which would be less marked on this approach. It is here that markedness conventions and the concept of the markedness of a whole system, or subsystem, of parameters comes in. Let us postulate, for concreteness, the following convention: (38)

For a class of heads H, uEPP for H[F:–] 6¼ v ! {[þEPP]/v[þEPP];} {[–EPP] elsewhere}

What (38) says is that the unmarked value of the EPP-feature for some head of a particular type with an unvalued feature (i.e. a probe, capable in principle of triggering movement) is [þEPP], i.e. the presence of an EPP feature, just where v has an EPP-feature, i.e. in an OV system. This has the effect that, for all head–complement pairs, head-final is the unmarked order in an OV system, and head-initial in a VO system. In these terms, rigidly head-final languages are relatively unmarked, as of course are rigidly head-initial languages, while ‘mixed’ languages are relatively marked (and one can in principle quantify exactly how marked different types of mixed systems would be). Furthermore, Dryer’s (1992) observation that VO vs. OV order is the basic determinant of ordering among other head–complement pairs is directly captured. One might speculate that v is the crucial category determining the markedness of the system with respect to word order because it is the head of the phase which determines argument structure and therefore the category whose features are most important for determining the positioning and licensing of arguments. We can understand a markedness convention like (38) in terms of conservatism of the learner, assuming that the learner exploits pieces of input, 20

Here we are only concerned with the feature which attracts VP; let us leave aside whatever it is that attracts V. In Biberauer’s system, VP-movement is triggered by an EPP-feature associated with the v’s property of probing for D, VP-movement representing the ‘pied-piping’ option as compared to object-shift (movement of the object DP).

Introduction: parameters in minimalist theory

41

perhaps marked input, to the full. So we could entertain something like the following: (39)

Generalisation of the input If acquirers assign a marked value to H, they will assign the same value to all comparable heads.

(38) can naturally be understood in terms of (39). Moreover, both (38) and a simple feature-counting simplicity metric like that put forward by Roberts & Roussou (2003: 210) can be seen as different aspects of the overall conservatism of the learner, which essentially tries to set parameters in the most efficient way possible. The Subset Principle can also be seen in this light: one aspect of the learner’s conservatism is to avoid superset traps (see Berwick 1985; Clark & Roberts 1993; Biberauer & Roberts 2009). We will return to the question of why something like (39) should hold in Section 3.5. For the moment we may observe that macroparametric effects arise from aggregations of microparameters acting in concert for markedness reasons. One could perhaps extend this to the Polysynthesis Parameter. According to Baker (1996: 14, 17, 496), the central property of polysynthetic languages is that all argumental DPs must correspond to a morphological expression in the head that y-marks them. The reflexes of this very general condition (which Baker calls the Morphological Visibility Condition) are syntactic noun-incorporation, rich object- and subject agreement marking, ‘free prodrop’ of all arguments, free word order and a range of other major morphosyntactic properties (see Table 11.1 in Baker (1996: 498–9)). Kayne (2005a: 7) observes that there is a similarity between these features of polysynthetic languages such as Mohawk and what is found in clitic doubling/dislocation constructions in Romance. He speculates that “it could alternatively be the case that the systematic obligatoriness of pronominal agreement morphemes in Mohawk is just an extreme example of what is found to a lesser extent in (some) Romance” (p. 7). Suppose, then, that Mohawk and other polysynthetic languages have generalised clitic left-dislocation (CLLD), while Romance languages have partial CLLD and languages such as English lack it altogether. There is a link to the NSP here, in that rich subject agreement on the verb may be thought of as comparable to a clitic and the subject then seen as clitic left-dislocated (this is one variant of the I-subject idea in Section 1.3, (17), pursued by Barbosa (1995; 2009) and Alexiadou & Anagnostopoulou (1998)). Clitics can be thought of as the overt realisation of both the probe’s and the goal’s matching j-features on the probe (see Roberts (Chapter 1; forthcoming b)); this prevents the goal from being realised in its grammaticalfunction position (this applies to non-clitic-doubling; the difference between clitic-doubling and non-clitic-doubling may have to do with the interpretability of the features, but we will leave that aside here). We can thus envisage

Ian Roberts and Anders Holmberg

42

a markedness statement based on (38) which would specify that realising all features on all possible probes is the more marked option than never realising them, but less marked than realising them sporadically. Again, the Polysynthesis Parameter may reduce to an aggregation of microparameters concerning clitics/agreement, governed by a markedness constraint. Are there other macroparameters? One intriguing recent suggestion is due to Huang (2007). He observes that a range of properties, up to twenty or more, appear to cluster together in Chinese, as compared to English and other familiar European languages, or to Japanese, for example. Space prevents full discussion and illustration of these, but among them are the following: (40)

a. Generalised bare N (denoting kinds) b. A generalised classifier system c. No plural morphology

(41)

a. b. c. d. e.

(42)

a. No agreement, tense, or case morphology b. No wh-movement c. Word order: ‘V2’ counting backwards (no Kaynean VP-movement over adjuncts)21 d. No V-to-T movement

(43)

Radical Pro drop

Action verbs are atelic No simplex accomplishment verbs Resultative compounds or phrases Periphrastic causatives Extensive use of light verbs

Huang proposes that these properties seem to cluster together in one language type to the exclusion of another, because they are manifestations of the same generalisation. He proposes the macroparameter High Analyticity, which states that Chinese lexical items are highly analytic at three levels: the level of lexical categories, functional categories and argument structure. In particular, in a detailed study of the diachronic development of Chinese, Huang shows that the following properties have been lost: (44)

21

a. wh-movement b. VP-movement c. NP movement

This refers to the fact that Chinese canonically shows the order Adjunct V Complement rather than V Complement Adjunct (e.g. John often visits Mary vs. John visits Mary often). Following Kayne (1994) and Cinque (1999), the English order may result from VP-raising into the functional field; Chinese lacks this operation. Hence, in general, the only material following V is the complement (and, given verb serialisation, V usually has exactly one argument). This gives the appearance of ‘reverse V2’.

Introduction: parameters in minimalist theory d. e. f. g.

43

EPP movement Plural morphology Denominal suffix Causativising suffix

These losses have taken place in conjunction with the development of numerous particles, particularly for marking tense, mood, aspect and illocutionary force. It seems tempting, then, to try to maintain that ‘analyticisation’ is loss of movement, associated to some degree with loss of morphology. In particular, it seems that Chinese lacks movement at the lowest structural level, inside the lexical phase. It is plausible to think of the properties in (40) as being related to the lack of N-to-n movement (assuming classifiers to be ns and plural marking to be determined or fed by N-to-n movement). Similarly, at least (41a–c) could be connected to the lack of V-to-v movement, assuming that a complex event structure such as that involved in accomplishments requires some form of incorporation in the vP phase (see Ramchand (2008)). The properties in (42) are more familiar, but perhaps depend on the prior formation of verbs by V-to-v movement (in the case of V-to-T movement) and the prior existence of lexical elements in the n/v position that can be probed by uninterpretable j-features merged outside the first phase, in the case of (42a) for example. Furthermore, there appear to be no EPP-type movement triggers at the higher phase level (see (42b,c)), although, since Chinese is topicprominent, Edge Features must be present on C. We will return to the relation of radical pro-drop to the absence of j-features below (see also Saito (2007); Roberts (Chapter 1)). Putting all of this together with the very well-known fact that in Chinese it is impossible to distinguish nominal and verbal roots by their morphological shape, and that very many basic roots are entirely ambiguous between nominal and verbal interpretation, it is tempting to suggest that head-movement, in particular L-to-l movement (i.e. movement of the lexical root to the local phase-head) is systematically absent in Chinese. More generally, phase-heads appear to lack Agree-related movement triggers (EPP-features). In the non-lexical phases (CP and DP), this has the consequence that probes are largely absent, as their putative goals are too deeply buried in the lexical phase to be accessible (assuming the version of the Phase Impenentrability Condition in Chomsky (2000), rather than the less restrictive version in Chomsky (2001)). So we might, very tentatively, conclude that Huang’s High Analyticity macroparameter results from an aggregate of head-movement parameters acting together, in this case in not triggering movement. In more familiar languages, we have the inherently more marked situation where heads sporadically trigger movement (although fairly systematically at first-phase level; but see Biberauer & Roberts (Chapter 7) for the proposal that (Modern) English lacks V-to-v movement).

Ian Roberts and Anders Holmberg

44

A final, very speculative suggestion comes to mind at this point. If there are macroparameters determining polysynthesis and high analyticity, are there parameters determining other morphological types? In fact, Julien (2002: chapter 3), following a proposal in Kayne (1994), proposed that very many OV agglutinating languages typically showed the following structure, for all (or most) X: (45)

[XP YP [x affix] (YP)]

This is in fact the variant of the general OV parameter, generalised by (38), where the host head contains a bound morpheme. So (38) may fall into two subcases, depending on precisely this, the former giving OV order, the latter agglutinating morphology and OV order. It has been observed many times that rigidly OV languages tend to be agglutinating. Finally, we are led to think that perhaps the very first kind of typology ever proposed, the morphological typology put forward by Schleicher (1862) and Sapir (1921) among others,22 was on the right track after all, but it was really a pre-theoretical observation about syntactic macroparameters. The values of the macroparameters are so salient that one could not fail to notice their effects in the data, but, in the absence of a theory of syntax, it was not possible to discern their true nature. And so the generalisations were mistakenly thought to be morphological.23 Current parametric theory, especially given the distinction between macroparameters and microparameters as construed here, enables us to tentatively begin to do this. Here we have argued first that Baker (2008a) is correct in distinguishing macroparameters from microparameters, but that it is inadvisable to abandon the BCC. Instead, we have suggested that macroparameters are the result of aggregates of microparameters acting in concert, guided by the acquisitionbased markedness constraint in (38). In the next section, we will relate all this to the NSP (and to the somewhat incidental question of whether this is a macro- or microparameter), to GGL’s proposal for ‘parameter schemata’ and to a proposal for learning paths, in the sense of Dresher (1992; 1999), which relate macro- and microparameters.

22

23

The earliest version of this typology was put forward by Schlegel (1817); see MorpurgoDavies (1998: 71–75) for discussion of this and its seventeenth and eighteenth-century antecedents. At this point, it is natural to ask whether there is an Inflectional/Fusional Parameter. We suspect not; this parameter arises where none of Polysynthesis, Analyticity or Agglutination are set to their positive values. Inflection/fusion seems to involve non-uniform behaviour among functional heads, and hence might be seen as a marked system. Certainly, these systems are prone to change, as much of the history of Indo-European attests, and inflectional languages tend to be of mixed type (cf. German, Latin, Sanskrit and other conservative, highly inflecting Indo-European languages).

Introduction: parameters in minimalist theory

3.4

45

Epigenetic parameter-setting

A further critical question concerning the general nature of parameters that Newmeyer (2005: 44) very correctly raises is “whether all parameters are applicable to all languages”. In the Principles-and-Parameters literature, the answer to this question has generally been positive, although it does lead to the questions concerning the status of a complex range of microparameters related to clitics, or agreement, or classifiers, in languages where these properties are lacking, as discussed in Section 2.1. GGL explicitly propose that the answer to this question should be negative. They propose that, instead of innate parameters, UG makes available a small set of parameter schemata, which, in conjunction with the PLD, create the parameters that determine the non-universal aspects of the grammatical system. In this way, parameters are created through interaction with the PLD in a fashion reminiscent of the Piagetian concept of epigenesis.24 They propose the following form for their schemata (2008: 7–8): (46)

a. b. c. d.

Grammaticalisation: is F, a functional feature, grammaticalised? Checking: is F, a grammaticalised feature, checked by X, X a category? Spread: is F, a grammaticalised feature, spread on Y, Y a category? Strength: is F a grammaticalised feature checked by X, strong? (i.e. does it overtly attract X?)

They illustrate the functioning of such schemata in detail, in relation to the feature definiteness, and its effects on the internal syntax of DPs in some of the twenty-four languages for which they have obtained data regarding fortysix parameters. In a similar vein, Roberts & Roussou (2003: 213) propose the following set of options relating to a given formal feature F on the basis of their extensive analysis of grammaticalisation as a diachronic operation affecting the realisation of functional categories: (47)

24

a. is F realised by (external) Merge (i.e. does it correspond to an overt grammatical formative?) b. does F enter an Agree relation? c. if so, does F attract? d. if so, does F attract a head or an XP? e. if (c), does F attract both a head and an XP? f. does F combine realisation by external and internal Merge? g. if so, does F attract a head or an XP?

Boden (2006: 493) characterises epigenesis as follows: “a self-organising dialectic between biological maturation and experience.” A similar suggestion, but restricted to what he calls ‘core parameters’ (approximately equivalent to Baker’s notion of macroparameter), is put forward by Uriagereka (2007: 106ff.). See also Piattelli-Palmarini & Vercelli (to appear).

Ian Roberts and Anders Holmberg

46

(Roberts & Roussou do not assume GGL’s initial question: whether F is present at all, assuming instead that all languages use the same set of formal features; see Roberts & Roussou 2003: 29). What (46) and (47) share is specifying a range of formal operations which can be associated with a given type of substantive feature (a formal feature of a functional head). They differ in detail, and this is certainly not the place to evaluate their relative merits. What we can note is that the sequence of statements involves a steady increase in specificity in each case. In fact, each statement is close to being disjunctively ordered in relation to the previous one, and it would certainly not be difficult to reformulate either (46) or (47) so as to make this more precise. Roberts & Roussou explicitly state that their system reflects a markedness hierarchy; GGL on the other hand make no such claim. Moreover, each set of statements has a kind of ‘branching’ structure, which we can illustrate as follows for (47b–d), replacing (47a) with GGL’s option of the presence of F as a formal feature, and simplifying slightly:25 (48)

F? No STOP

yes Does F Agree? No STOP

yes Does F have an EPP feature? No

yes Does F trigger head-movement? No STOP

yes

(This can also be done for (46).) Each ‘yes’ option entails a further option. The ‘yes’ options that do not dominate anything may entail further options regarding the type of head-movement, or, in the case of XP-movement, piedpiping options (see Richards & Biberauer (2005) and Biberauer & Richards (2006) on the latter option); we leave further specifications aside here. Each more deeply embedded option is more marked than all less deeply embedded ones, since effectively the description of the parameter is the conjunction of all the dominating nodes, and so it increases in length as embedding deepens. ‘STOP’ options on left branches are relatively unmarked options in each case. 25

In the theory of incorporation put forward in Roberts (forthcoming b), a probe which triggers incorporation cannot have an EPP-feature for principled reasons, which justifies the way the options are presented in (48), as opposed to the slightly richer set of options put forward by Roberts & Roussou.

Introduction: parameters in minimalist theory

47

So here we see a parameter schema given as a network of options, each more embedded option representing a more specific, and therefore a more marked, option. Importantly, we can consider networks like (48) to define ‘learning paths’ in the sense of Dresher (1992); again the conservatism of the learner is such that it prefers the path to be as short as possible, and so deeply embedded options are relatively marked owing to the fact that they have longer descriptions. Following GGL, we assume that the schema and the overall pool of possible features are given by UG; the network is created through epigenesis in acquisition, and markedness follows, on one standard construal, from increasing specificity (length of description relevant to F’s role in the grammar, and hence greater computational burden; we will say more about markedness in the next section). Now, parameter schemata of the kind in (46–48) apply to individual formal features. As such, they are classic examples of microparameters (and have many of the advantages of this kind of formulation of parameters discussed in Section 3.2). But we are now in a position to fruitfully combine these, or some of them at least, with markedness statements of the kind in (37), to derive some of the macroparameters discussed in the previous section. In a nutshell, macroparameters quantify over F in networks like (48). To see how this works, consider the EPP option embedded two levels down in (49). The markedness statement in (36) essentially says that the unmarked option for the grammatical system (i.e. not necessarily for F itself) is ‘no F has this value’, and that the next least marked option is ‘all F have this value’, and that the trigger for choosing is the value taken by v. More mixed, and therefore more marked, systems may relate the possession of F to further categorial features, and the options may become progressively more specific (have longer descriptions) and more marked. In other words, we have a crosscutting set of options of the form: (49)

a. Are movement-triggering features absent from all probes? b. If not, are movement-triggering features obligatory on all probes? c. If neither (a) nor (b), are movement-triggering features present on {T, v, . . .}?

The positive value of (49a) gives a rigidly, harmonically head-initial language like Welsh. The positive value of (49b) gives a rigidly, harmonically headfinal language such as Japanese or Turkish. Again, (49c) breaks up into a series of microparameters, with a range of other factors entering here (options regarding pied-piping, and general constraints on disharmonic orders of the kind explored in Biberauer, Holmberg & Roberts (2007)). The existence of this set of cross-cutting options is determined by generalisation of the input. Applying (48) where F is universally quantified, we come very close to deriving the macroparameters discussed in the previous section

48

Ian Roberts and Anders Holmberg

(we have added one further option on the most embedded right branch, for further illustration): F?

(50) No STOP

Yes Does F Agree? No STOP

Yes Does F have an EPP feature?

No Head-initial Does F trigger head-movement? No STOP High analyticity

Yes Head-final Is F realised by external merge?

Yes Polysynthesis

No STOP

Yes Agglutinating

(This may wrongly predict that polysynthetic languages are head-initial, but actually they appear to have free word order, precisely owing to their polysynthetic nature (see Baker 1996: 10ff.)). So we can finally arrive at a picture of the form of parameters as involving generalised quantification over formal features, as follows: (51)

Q(ff ∈ C) [P(f)]

Here Q is a quantifier, f is a formal feature, C is a class of grammatical categories providing the restriction on the quantifier, and P is a set of predicates defining formal operations of the system (‘Agrees’, ‘has an EPP-feature’, ‘attracts a head’, etc.). The longer the characterisation of either C or P, the more deeply embedded in a network/schema the parameter will be, the more marked it will be and the further along the learning path it will be. True macroparameters sit at the top of the learning path, and in fact involve unrestricted universal quantification or its negation, as we saw in (50) (Baker (2008a: 9) also suggests that macroparameters might be a kind of default). This seems to us to be a maximally simple theory of parameters, since ultimately it involves relations between sets of features of categories and predicates defining grammatical operations. Can we relate the NSP to these considerations? Roberts (Chapter 1) speculates along just these lines. He suggests that putting together Saito’s (2007) proposals regarding the nature of radical pro-drop (which were briefly described in Section 1.2.3) with the particular account of the ‘rich agreement’ that facilitates consistent null subjects based on Mu¨ller’s (2005) notion of impoverishment (again, see Roberts (Chapter 1) for details) we arrive at the following generalisations: (52)

a. Radical pro-drop is possible iff j-agreement is not obligatory. b. Consistent null subjects are possible iff there is no impoverishment of T’s j-features.

Introduction: parameters in minimalist theory

49

Where (52a) holds ‘discourse pro’ is possible (i.e. subject pronouns can be merged at LF); where (52b) holds, deletion of subject pronouns is possible prior to LF (again, see Roberts’ chapter for details). We see that the two systems are derivational mirror images of one another, and that this is the direct consequence of the different status of j-features on probes (fully optional vs. obligatorily present and unimpoverished), which in turn is typically reflected in the agreement morphology (totally absent vs. ‘richly’ realised).26 This further suggests a rethinking of the typological generalisations surrounding null arguments: perhaps the fundamental dimension of parametric variation is ‘radical’ vs. ‘consistent’ null-subject (or null-argument) languages, with partial and non-null-subject languages being subcases of the ‘consistent’ type featuring varying degrees of impoverishment of the goal. The basic form of the parameter would then be as in (53): (53)

a.

Are uj-features obligatory on all probes? No Radical pro-drop

Yes b. Are uj-features fully specified on all probes? Yes Polysynthesis

No c. Are uj-features fully specified on some probes? Yes Non-pro-drop

No d. Are the uj-features of {T,v,...}impoverished

As indicated, the ‘No’ value in (53a) gives radical pro-drop. The negative value of (53b) may give rise to a polysynthetic system (or at least to consistent head-marking; Baker’s notion of polysynthesis seems to combine this with consistent head-movement of arguments; see (50)). A positive value for (53c) gives a non-null-subject language like English. (53d) is intended to simply indicate the ways in which the null-subject parameter starts to ‘break up’ into microparameters as individual probes are evaluated in relation to it (cf. Section 1.2.5 (16)). Clearly, a ‘no’ value for T and a ‘Yes’ value for v will give a consistent null-subject language like Italian. In terms of the general schema for parameters in (51), we can state the NSP as follows: 26

The Mainland Scandinavian languages do not show any subject agreement in finite clauses (see Holmberg & Platzack 1995: 10), yet they are not radical pro-drop languages, but instead non-null-subject languages. We maintain that they have a generalised uf-feature in finite T which does not have a morphologically realised valued form, but is nevertheless visible in virtue of obligatory movement of a nominal subject to SpecTP; see Holmberg (Chapter 2).

50 (54)

Ian Roberts and Anders Holmberg ∃ff∈D [S(D, TFin)]

(54) reads ‘For some feature D, D is a sublabel of finite T’, where ‘sublabel’ is understood as in Chomsky (1995b: 268). This captures the force of the informal statement given in (18) above, and shows how the NSP fits with the general format for parameters, and how it is part of the parameter network in (53). Partial null-subject languages, and intricate cases like certain registers of French (see Roberts (Chapter 8)), require still further specification. Again, these represent progressively more marked options, located more deeply in the schema/network and further down the learning path. In these terms, we can immediately observe a connection between the null-subject parameter and the other parameter schemata/networks discussed here. A further advantage of hierarchies of the type sketched above is that they restrict the upper bound of grammars that a given set of parameters can generate. The cardinality of G, the set of grammars, is equivalent to the cardinality of P, the set of parameters, plus 1, to the power of the number of hierarchies: (55)

jGj ¼ (jPj þ 1)n, where n ¼ jHj

Suppose, arbitrarily, that there are 5 hierarchies (we have seen two; there must be one for word order, and it is very easy to imagine that there are at least two more), and suppose that there are 30 parameters. Then jGj is 315, or 28,629,151. This is a large number, but recall that 30 parameters yielded over a billion grammars on our earlier calculation based on unhierarchised microparameters (see Section 3.2). In this section and the previous one, we have clearly made some progress towards resolving the tension between descriptive and explanatory adequacy that we observed in Section 3.1. In particular, the question of the highly specific innate endowment in formal features that the BCC appears to demand has been eliminated, and we have also clarified the relation between microand macroparameters. All parameters ultimately have the extremely simple form in (51), and they form schemata/networks which are related to markedness of the general kind in (48) and (50). The NSP, for example, is a case of the parameter schema in (53), specifically (53c) as it relates to T. We believe that these proposals go a long way towards restoring the explanatory value of parameters (as well as giving wide empirical coverage, if our speculations about macroparameters are on the right track). In the next section, we will suggest that they may show the way beyond explanatory adequacy. 3.5

Why parameters? Comparative syntax beyond explanatory adequacy

In recent work, Chomsky (2004; 2005; 2007) has proposed that an important property of the Minimalist Program is that it can take us beyond explanatory

Introduction: parameters in minimalist theory

51

adequacy (see Chomsky (2007: 19) for a lucid statement of this idea). Accordingly, we attribute the adult state of linguistic knowledge, adult competence, to three factors: (i) the genetic endowment, UG, (ii) experience of the PLD, and (iii) principles not specific to language. The last have become known as ‘third-factor principles’ and have to do in particular with principles of optimal and efficient computation. In this section, we would like to show how the view of parameters arrived at in the preceding sections can begin to take us beyond explanatory adequacy in this domain. The classical view of how principles and parameters interact to produce adult competence was based on the idea that the parametric options were specified as such as part of the genetic endowment, made manifest in the PLD (perhaps with inaccessible properties being triggered by accessible properties in a parametric cluster) and thereby fixed during language acquisition, to give an adult system which was an instantiation of UG with all parametric options fixed. So this view relied entirely in the interaction of factors (i) and (ii) above, characteristic of the classical notion of explanatory adequacy. The view of principles and parameters that follows from the considerations in the previous sections is rather different. UG does not even provide the parameter schemata. In essence, parameters reduce to the quantificational schema in (49), in which UG contributes the elements quantified over (formal features), the restriction (grammatical categories) and the nuclear scope (predicates defining grammatical operations such as Agree, etc.). The quantification relation itself is not given by UG, since we take it that generalised quantification – the ability to compute relations among sets – is an aspect of general human computational abilities not restricted to language. So even the basic schema for parameters results from an interaction of UG elements and general computation. The parameter schemata form networks defined by markedness relations. The markedness notions we have invoked include relative length of description, and generalisation of the input (relevant for statements such as (38), which ultimately define macroparameters). Both rely on a general notion of computational conservatism, which again we can think of as a facet of computational efficiency. Again, then, the schemata arise from third-factor properties. The different points in the schemata all instantiate the schema in (49); they differ in the specificity of the two arguments to the quantifier C, the class of grammatical categories, and P, the predicates defining (conjunctions of) grammatical operations. The more specific either of these arguments, the more marked, and indeed the more ‘micro’, the parameter. Third-factor considerations may contribute to markedness in other ways, too. Mobbs (2008) critically reviews Hawkins’ (2004) three performance efficiency principles we mentioned in Section 2.5, and which were invoked

52

Ian Roberts and Anders Holmberg

by Newmeyer (2005) as an alternative to parametric accounts. We give them here: (56)

a. Minimise Domains (MiD) b. Minimise Forms (MiF) c. Maximise Online Processing (MaOP)

We saw in Section 2.5 how Newmeyer, following Hawkins, tries to invoke MiD in order to account for a well-known Greenbergian implicational universal. Whether or not that account is successful (and there are many reasons to think it is not, see footnote 16), one can wonder as to the precise status of these efficiency principles. Mobbs convincingly argues that Hawkins’ Performance-Grammar Correpondence Hypothesis reflects only a correlation, not causation, and that we might do better to reconsider the efficiency principles as third-factor principles that play a role in defining UG. In these terms, MiD naturally relates to the ‘minimal search’ considerations which underlie locality conditions: the non-intervention condition on Agree and the Phase Impenetrability Condition (PIC). It also relates to the No Tampering Condition (existing relations should not be altered by later operations) and cyclicity generally (Mobbs 2008: 8–9). So a version of one of Hawkins’ efficiency principles, more abstractly construed as a general computational principle informing competence, rather than constituting performance, may lie behind certain fundamental properties of the language faculty. What concerns us more directly here, however, are third-factor constraints related to markedness. Mobbs suggests very plausibly that Hawkins’ Minimise Forms constraint may underlie the markedness preference articulated in Roberts & Roussou (2003) for relatively simple forms. More importantly, he proposes a further efficiency constraint, Generalise Features (GenF), which he states as follows: (57)

Human computation shares features over forms in the same domain (Mobbs 2008: 11).

This, he suggests, may underlie the “generalisation of the input” form of markedness put forward in (38) above. We concur with this suggestion. It seems, then, that Hawkins-style efficiency principles may have a role to play in our theory of universals and typology, but not as performance constraints on processing. Instead, they may reflect deeper, rather general computational principles, which contribute to the third factor determining adult competence. They may contribute to typology to the extent that they inform the markedness principles which determine parameter schemata/ networks. Further, we have suggested that the general form of parameters themselves results from the interaction of UG primitives with the general principles of quantification.

Introduction: parameters in minimalist theory

53

Finally, we can ask the most difficult question of all, but one which the Minimalist Program requires us to ask: why do we have parameters at all? Our general format for parameters in (51), inasmuch as it allows Q to be a negative quantifier, basically states that formal features of functional heads are all in principle optional. UG says nothing more than this, which is about as little as could possibly be said (in particular, this is a more ‘minimal’ statement than either forbidding or requiring the presence of such features). Moreover, the quantificational schema is maximally liberal: it states that the formal features may be in any set-theoretic relation with any predicate defined by the theory of grammar. So parametric variation arises because UG really doesn’t mind about the distribution of formal features in any given grammatical system. But we know that speakers fixate on given grammatical systems during language acquisition, and speech communities recognise given (aggregates of) grammatical systems as languages. Neither of these aspects of parametric variation directly concerns UG, however: ‘fixing’ parameters may be a facet (actually, almost a definition) of learning. So the kind of stable parametric variation we observe in adults arises from the fixation on random values. These values take on cultural and social value – a very different kind of value – as ‘languages’ for the kinds of reasons that have been revealed by sociolinguistics. But, even if UG doesn’t mind, how and why could variation in grammatical systems have emerged? Here, the work of Niyogi & Berwick (1995; 1997) and Niyogi (2006) on modelling the acquisition and change of grammatical systems in populations of learners is revealing. They show that given a learning algorithm A, a probability distribution of linguistic tokens across a population (random PLD), and a restricted class of grammars from which to select (UG), variability will result as long as the time allowed for the selection of hypotheses is restricted. This idea emerges most clearly in the following quotation from Niyogi (2006: 14–15): imagine a world in which there are just two languages, Lh1 and Lh2. Given a completely homogeneous community where all adults speak Lh1, and an infinite number of sentences in the Primary Linguistic Data, the child will always be able to apply a learning algorithm to converge on the language of the adults, and change will never take place . . . Now consider the possibility that the child is not exposed to an infinite number of sentences but only to a finite number N after which it matures and its language crystallizes. Whatever grammatical hypothesis the child has after N sentences, it retains for the rest of its life. Under such a setting, if N is large enough, it might be the case that most children learn Lh1, but a small proportion e end up acquiring Lh2. In one generation, a completely homogeneous community has lost its pure character.

In other words, if we combine the hetereogeneity in any speech community, the random distribution of PLD (poverty of the stimulus) and the limited time

54

Ian Roberts and Anders Holmberg

for learning (i.e. the critical period for language acquisition), change in grammatical systems is inevitable. If change is inevitable in the diachronic dimension, then variation is inevitable in the synchronic dimension. But, again, none of this reflects any aspect of UG except an indifference as to the distribution of formal features, as captured by (51). Here we begin to see the shape of comparative syntax, beyond explanatory adequacy. 3.6

Where are parameters? The locus of parametric variation

We have already discussed the BCC, the idea that parameters are specified in the lexical entries of lexical items, at some length. However, there is also a trend in recent minimalist theory to locate all linguistic variation, including all syntactic variation, in the post-spell-out morphology and phonology components (see, for example, Sigurðsson 2006a; 2006b; Burton-Roberts & Poole 2006a; 2006b; Boeckx, (forthcoming). The following quote is representative: [. . .] it is not implausible to think of Narrow Syntax as completely uniform (meeting LF-demands), and not affected (design-wise) or adapted to cope with or code for variation in the guise of (syntactic) parameters. (Boeckx, forthcoming).

This hypothesis (which Boeckx dubs the the Strong Uniformity Theory, SUT) is still at a programmatic stage, and thus has not been seriously put to the test, yet. It is an interesting hypothesis, though, and very much in the spirit of the Minimalist Program. What we are proposing here can indeed be construed as coming close to the SUT. For example, whether to delete or pronounce a pronoun is clearly a matter of PF, and variation with regard to EPP-features discussed above in Section 3.3. is presumably best construed as taking effect after spell-out (which is made technically possible by phase theory, according to which spell-out applies at several stages in the derivation of a sentence; Chomsky 2000; 2001; 2008; Svenonius 2004). We would maintain, however, that some variation is encoded in Narrow Syntax. In fact, so does Boeckx. He concedes that the following variation is encoded there: (58)

a. Features F1 and F2 may be expressed separately or as a bundle; b. F may or may not exhibit a uF variant; c. A given phase-head may or may not be strong, i.e. uF-bearing, or weak (defective).

(The similarities with GGL’s parameter schema in (46) and Roberts & Roussou’s proposal in (47) are obvious; see also the other proposals mentioned by Boeckx.) But, he says, “all other ‘parametric’ options arise in the post-syntactic morpho-phonological component, such as whether a head H allows its specifier to be filled by overt material, or whether the head or the tail of a chain can/must be pronounced, or whether a given head is affixal and requires its expression in the vicinity of another head, or whether a head

Introduction: parameters in minimalist theory

55

H precedes or follows its complement” (Boeckx, forthcoming). This is not significantly different from what we are proposing. In fact, Boeckx’s parameter schemata (58a,b,c) fall under our more general parameter schema (51) (if P can be ‘is bundled together with F2’, an idea which we have not discussed, but which will be exploited in Holmberg (Chapter 2)).27 Interestingly, Boeckx presents these ideas in the context of a paper where he distances himself from GB-style parametric theory: “[T]he idea that a GB-style Principles-and-Parameters architecture provides the right format for a solution to Plato’s Problem is, I think, seriously mistaken, on both empirical and conceptual grounds” and “if minimalists are right, there cannot be any parametrized principles, and the notion of parametric variation must be rethought” (Boeckx, forthcoming). As should be obvious from the discussion above, this is not exactly how we see it. Instead, we see the theory of linguistic variation which is developed within the minimalist framework as a refinement of the “GB-style Principles-and-Parameters architecture” (see also Biberauer (2008a) for essentially the same proposal). One reason why Boeckx dismisses GB-style parameters, thus concurring with Newmeyer (2004; 2005; 2006) (discussed above in Section 2.3), Culicover (1999) and Jackendoff (2002), is that he thinks of the parameters as being, by definition, principles which come as an addition to the set of universal principles of UG, but with the difference that they specify a range of options for some grammatical property. In this view parameters make UG larger, with more specifications, and thereby also make it more specific to the language faculty, in apparent conflict with the Minimalist Program of reducing UG, as far as possible, to third-factor effects, as discussed in the previous section. It also raises the question how such a rich UG could have evolved (as pointed out by Boeckx, forthcoming). But although the view of parameters as constituting an additional set of specifications to UG is often voiced in textbook presentations of parametric theory, it is not the only view, and it is not inherent to P&P theory. An alternative is that parameters are just those grammatical options which are not specified by UG, as we have tried to specify in the preceding sections. An obvious example is the traditional head–complement parameter. In the case of this parameter the values (head precedes complement or head follows complement) are not given by properties of UG, but instead follow from the ultimately physical fact that words must be linearly ordered, allowing exactly those two options (as Boeckx notes, in fact). However, this does not make it any less of a parameter in the P&P sense, as long as it remains true that there 27

The concept ‘F and G form a feature bundle’ can also be expressed in the restriction, as follows (where ‘S’ is again the predicate ‘is a sublabel of’): (i)

∃f ∈ F∃g ∈ F [S(f,C) & S(g,C)].

56

Ian Roberts and Anders Holmberg

is a finite range of options which are left open by UG in that UG does not make the choice for the learner. Another classical parameter of GB theory which clearly has this character is the wh-parameter of Huang (1982): (59)

Wh-movement takes place before/after S-structure.

Here UG prescribes that wh-movement happens whenever a wh-phrase is selected from the Lexicon, but does not specify when. Given the GB-model, and given the semantic properties of wh-expressions, there are logically two possibilities: before S-structure or after, on the LF-side (in Huang’s (1982) terms “in Syntax or in LF”). In this case the options are given by independently motivated architectural properties of the system. Specifying the options as in (59) is done in the name of explicitness, presumably, but is in fact redundant, as should be obvious to a critical reader. Disagreement between us and Boeckx as regards the historical relation between GB-theory and minimalism is of limited importance, though, in the context of this volume. What is more important is whether we agree on the substantial issues concerning the nature of UG and the explanation of variation. One substantial, empirical claim that Boeckx (forthcoming) makes, which we do not agree with, is that all parameters (he refers to them as ‘nano-parameters’) are independent; hierarchies of parameters and macroparameters do not exist. But the hierarchies we have proposed are determined by third-factor principles, as we have seen, and so there is no cost to UG in proposing these. Unaltered, Boeckx’s view will either be descriptively inadequate (too few parameters to account for the attested variation) or predict astronomical numbers of unattested systems, for the reasons alluded to in Section 3.2. So we see that the opposite view should be taken seriously. An even more substantial point on which we agree with Boeckx is that “the minimalist program offers us a different, more adequate way of exploring how principles and parameters may interact” (Boeckx: note 2). We hope to have demonstrated this in the preceding sections (see also Biberauer 2008a). 4

Conclusion

The above discussions, combined with the chapters to follow and those collected in Biberauer (2008b), attest to the continuing validity of the principles-andparameters approach to UG. Naturally, this approach has changed significantly over the thirty years since it was first suggested (in Rizzi 1978). However, the essential idea can still provide a way to resolve the traditional tension between description and explanation in comparative syntax. Moreover, we have seen that, by attempting to eliminate the same tension at the parametric level (caused by a proliferation of highly specific microparameters), we can

Introduction: parameters in minimalist theory

57

arrive at interesting characterisations of epigenetic parameter-setting, parameter schemata/networks and the relation between macro- and microparameters. A central concept in all of this is markedness, which may be largely determined by third-factor considerations of computational efficiency. As we already said, we are beginning to see how comparative syntax might look, beyond explanatory adequacy.

1

A deletion analysis of null subjects1 Ian Roberts

Introduction One often-noted consequence of the general adoption of the copy theory of movement since its reintroduction in Chomsky (1993) has been the collapse of the typology of empty categories in terms of the values of the features [anaphoric, pronominal] (see, inter alia, Hornstein (1999: 77–78), Manzini & Roussou (2000) for discussion). The non-pronominal empty categories of government-binding theory and related work, i.e. NP-trace and wh-trace, are now generally thought to be copies; their silence is the result of a PF-deletion process affecting non-heads of chains (see in particular Nunes 1995; 2004). The status of the pronominal empty categories is less clear, however. There has been a considerable discussion of the status of the pronominal anaphor PRO, naturally connected to the question of the nature of the various types of ‘control’ dependency (Boeckx & Hornstein 2003; 2004; Culicover & Jackendoff 2001; Davies & Dubinsky 2004; Hornstein 1999; 2003; Landau 2000; 2003; 2004; Manzini & Roussou 2000; Martin 1996; 2001; O’Neil 1997). The question of the status of the ‘pure pronominal’ empty category pro is also uncertain; many authors (including Alexiadou & Anagnostopoulou 1998; Barbosa 1995; 2009; Manzini & Savoia 2005; Nash & Rouveret 1997; Ordo´n˜ez 1997; Platzack 2004; Pollock 1997) have proposed that the pronominal property of the head bearing subject agreement features, observed to characterise consistent null-subject languages since Rizzi (1982), may be enough to derive the Null Subject Parameter and that, as a

1

I’d like to thank the other members of the null-subject project group – Theresa Biberauer, Anders Holmberg, Chris Johns, Michelle Sheehan and David Willis – for their comments on various earlier versions of this work. Preliminary versions of this chapter were presented at the Null Subject Workshop held at York University in November 2003, the Linguistics Association of Great Britain Meeting at Roehampton University in September 2004, the Sounds of Silence Workshop, Tilburg University, October 2004, at the Universities of Geneva, Leipzig and Cambridge, at the Encontro Lingua Falada e Escrita V, Maceio´, Brazil, November 2005, and at the Romance–Japanese Comparative Syntax Workshop held at the University of Siena in May 2007. I’d like to thank the participants at those meetings, especially Claire Blanche-Benveniste, Joa˜o Costa, Eric Haeberli, Ad Neeleman, Liliane Haegeman, Luigi Rizzi and Ur Shlonsky, for their comments and questions. All errors are my own.

58

A deletion analysis of null subjects

59

corollary, the idea that the canonical subject position is occupied by pro is not required, which in turn implies that the Extended Projection Principle either does not need to be satisfied or can be satisfied by something other than a DP occupying the relevant specifier position in these languages. This idea was first put forward by Borer (1986; see also Borer 1989: 70–71), and, following her terminology, we can refer to this type of approach as the ‘I-subject’ approach to null subjects (the idea being that I, as the functional head bearing subject agreement, has the subject role). In this chapter I want to put forward a new suggestion for accounting for the core cases of pro in consistent, agreement-rich, null-subject languages of the Italian type, one which owes much to, but departs from and develops, the ideas in Holmberg (2005). An interesting facet of the analysis is that it also sheds some light on another issue that arises in the context of the copy theory of movement: the question of the conditions under which identical copies are deleted (and indeed the further question of identifying the occurrences of copies). 1

Holmberg (2005)

Holmberg (2005) shows that the widely accepted pro-licensing analysis of null subjects put forward in Rizzi (1986a: 518–23) is incompatible with the feature-valuing system put forward in Chomsky (1995b; 2001). Rizzi’s proposal imposes the following two conditions on pro: (1)

Rizzi (1986a): a. pro must be licensed b. pro must be identified

Let us consider how (1) applies to the case of pro in a null-subject language. Here, pro appears in SpecTP, as shown in (2): (2)

TP proi

T¢ Ti [3PL]

In (2), pro is licensed by T, by assumption a member of the parametrised class of designated licensers in a null-subject language (see Rizzi 1986a). Furthermore, pro’s content is identified by the j-features of T, indicated in (2) as [3PI]. We can think of identification as a process whereby the values of the pro’s features are assigned.

60

Ian Roberts

As Holmberg (2005: 536–37) points out, Rizzi’s account of the identification of pro cannot be maintained in the context of the approach to feature valuing that has emerged in Chomsky’s (1995b; 2000; 2001) recent work. According to this approach, formal features such as j-features may be either interpretable or uninterpretable. Uninterpretable features must be eliminated from the derivation before the LF interface. According to Chomsky (2001), uninterpretable features are unvalued, and part of the process of ‘eliminating’ these features involves assigning them values. Chomsky further assumes that the j-features of T are uninterpretable, and are valued by entering into an Agree relation with the subject DP (I will say more about the technical details of Agree below). Argumental DPs are fully specified for j-features, and as such are fully interpretable and able to value the j-features of T. Concerning Rizzi’s notion that pro is in need of identification, however, this approach runs into difficulties; as Holmberg points out, “[w]ithin this theory of agreement, it is obviously not possible for an inherently unspecified pronoun to be specified by the j-features of I [i.e. T, IGR], as those features are themselves inherently unspecified” (2005: 537). It is worth restating Holmberg’s point in slightly more formal terms, as this brings it out more clearly and will be useful for the discussion of the conditions of deletion in the next section. Let us take formal features to be attribute–value pairs of the general type [Att:val] (see Chomsky 2001: 5). In that case, unvalued features can be seen as being of the form [Att:__]. In the standard case of Agree, the valuing operation consists of copying the values of the valued counterparts of the features into the blank value matrices of the unvalued features. This can be defined as follows: (3)

In a well-formed Agree relation of which a and b are the terms, where a’s feature matrix contains [Atti:__] and b’s contains [Atti: valj], for some feature F ¼ [Atti: (val{..,j,..})], copy valj into __ in a’s feature matrix.

(3) is meant as nothing more than a slightly more formal statement of what I take to be intended in Chomsky (2001).2 According to Chomsky (2001), the originally uninterpretable/unvalued features delete at the end of the relevant phase (which phase is relevant depending on the precise formulation of the

2

Note that this definition does not guarantee the valuing of structural Case features, given Chomsky’s assumption that these features are valued by convention: a DP whose j-features are valued by T is Nominative and a DP whose j-features are valued by v* is Accusative. Valuing of structural Case features must either be the consequence of a separate convention, or it can be subsumed under (3) if we supply T and v* with the features [Case:{Nom, Acc}] and DP with the feature [Case:__]. I will take the former option here. The fact that T and v* do not have Case features may be relevant to the formal characterisation of the nature of weak and strong pronouns, as I will briefly discuss in Section 2 (see footnote 14).

A deletion analysis of null subjects

61

Phase Impenetrability Condition, which need not detain us here; see Chomsky (2001: 13–14)). In terms of (3), both T (by Chomsky’s assumptions regarding the nature of uninterpretable features) and pro (by Rizzi’s account of the licensing of null subjects) will have a feature matrix of the form [Atti:__], hence neither will be able to value the other. This restates Holmberg’s point in terms of the definition given in (3).3 Holmberg observes that there are just two ways of dealing with this situation: one of the two elements, T or pro, must have interpretable, i.e. valued/specified j-features. Whichever one it is will be able to value those of the other one. Accordingly, Holmberg considers the following two hypotheses (see also Section 1.3 of the Introduction): (4)

Hypothesis A: in null-subject languages, the j-features of T are interpretable. SpecTP is therefore either absent or filled by an expletive (depending on whether T’s EPP-feature needs to be satisfied independently of its j-features). Hypothesis B: pro has interpretable features, occupies SpecTP and functions just like an overt pronoun. That pro is silent is thus a PF matter.

These two hypotheses differ empirically in one crucial respect. Hypothesis B implies that no expletive pronoun, overt or null, will be found with a null subject, since pro moves to SpecTP to check T’s EPP-feature. On the other hand, Hypothesis A does not make a clear prediction in this connection: whether an overt expletive is allowed, required or excluded depends on independent assumptions concerning the ability of T’s j-features to satisfy the EPP. Hence, if we can find a language with referential null subjects but at least the possibility of an overt expletive, and if that expletive cannot appear where we have reason to think that there is a referential pro in SpecTP, Hypothesis B is favoured. Holmberg shows that Finnish is just such a language. Finnish has null subjects and an overt expletive pronoun, sita¨ (EXP ¼ expletive): (5)

3

a. Puhun englantia. speak-1SG English ‘I speak English.’

Luigi Rizzi (p.c.) points out that a version of his approach could perhaps be maintained in terms of a system of the kind put forward by Pesetsky & Torrego (2004; 2007), in which uninterpretable and unvalued features are taken to be primitively different elements. In these terms, pro could be regarded as bearing interpretable, but unvalued, features, and the analogue to Rizzi’s (1986a) notion of identification would be valuing pro’s features. This is a technical possibility, but I will leave it aside here, since it entails abandoning what I take to be the conceptually attractive reduction of interpretability to valuing described above.

62

Ian Roberts b. Sita¨ meni nyt hullusti. EXP went now wrong ‘Now things went wrong.’

The expletive sita¨ does not cooccur with referential null subjects: (6)

a. *Sita¨ puhun englantia. EXP speak-1SG English b. Oletteko (*sita¨) ka¨yneet Pariisissa? have-2Pl-Q EXP visited Paris?

Holmberg concludes that Hypothesis B is right: pro occupies SpecTP.4 Since this element is like an overt pronoun in all respects except phonological realisation, Holmberg (2005: 538) concludes that “the null subject is a pronoun that is not pronounced”. Clearly, one way to see this is in terms of deletion: pro is a deleted pronoun. This constitutes a partial return to one of the main ideas in Perlmutter’s (1971) analysis of null subjects, in that the null subject arises through deletion of a subject pronoun. Holmberg goes on to distinguish three varieties of null subject: “a null weak pronoun . . . specified for j-features but lacking D and therefore incapable of (co)referring, without the help of a D-feature in I . . . Another type of null subject is a DP that is deleted under the usual conditions of recoverability. A third type is the classical pro . . . a bare j-featureless noun” (Holmberg 2005: 534). The first type is the ‘canonical’ null subject that we are concerned with here, found in Italian, Spanish, Greek, etc. The second type is exemplified by Finnish and various other languages (Holmberg (pp. 553–54) mentions Brazilian Portuguese, Marathi and Hebrew; see also Holmberg and Sheehan (Chapter 3); Holmberg, Nayudu and Sheehan (2009)). The third type is found in many East Asian languages, including notably Chinese, and is discussed inter alia in Huang (1984; 1989), Tomioka (2003), Neeleman & Szendro˝i (2007; 2008) and Saito (2007). Leaving aside the East Asian type,5 Holmberg distinguishes the first two types of null subject in terms of the features of the licensing T. The first type, that found in ‘consistent’ null-subject languages such as Italian, Spanish, etc., is treated as a weak pronoun in the sense of Cardinaletti & Starke (1999) (this is also proposed in Cardinaletti (2004) and below). More precisely, a “definite null subject is a jP, a deficient pronoun that receives the ability to refer to an individual or group from I containing D”

4

5

Holmberg (2005: 545f.) considers and rejects the possibility that Hypothesis A is correct and that an overt expletive is inserted only to satisfy the EPP. He shows that this is not compatible with the facts of Finnish. I will briefly return to the East Asian kind of null subject in Section 2.5 below, where I look at the nature of ‘rich’ agreement in relation to null subjects.

A deletion analysis of null subjects

63

(Holmberg 2005: 556). The presence of a D-feature on T is what makes a language a null-subject language (this idea appears in different forms in a variety of analyses of null subjects, including Alexiadou & Anagnostopoulou (1998) and Rizzi (1982), and I will maintain a version of it below). In a partial null-subject language such as Finnish, on the other hand, T does not have a D-feature. This has a range of consequences as discussed in Holmberg, Nayudu & Sheehan (2009). Now it is time to introduce the technical ideas which will motivate and restrict the environments of pronoun-deletion which give rise to null subjects.

2

Defective goals, clitics and weak pronouns

2.1

Clitics and incorporation

In Roberts (forthcoming b) I present a general analysis of Romance clitics and cliticisation, adopting and adapting ideas independently developed in Mavrogiorgos (2006). Here I want to summarise that analysis and show how the conditions for pronoun-deletion which can capture the distribution of null subjects can be derived from an extension of that framework. The central idea in Roberts (forthcoming b) is that minimal categories, since they are non-distinct from maximal categories in terms of bare phrase structure, can be phasal (see also Marantz (2001; 2006), where the same idea is implemented in a rather different way). As such, they can attract material to their left edge, and that left edge, unlike all other material inside the minimal category, may be accessible to elements outside the minimal head. This provides a basis for accounting for the puzzling property of clitics: that they act in some respects like affixes, i.e. as parts of the words that host them, and in some respects as syntactically autonomous items. More concretely, I adopt the following definitions of minimal and maximal category, the definition of minimal category departing slightly, but crucially, from the standard conception in Chomsky (1995a; 1995b): (7)

a. The label L of category a is minimal iff a dominates no category b whose label is distinct from a’s. b. The label L of category b is maximal iff there is no immediately dominating category a whose label is non-distinct from b’s.

The definition of minimal category in (7a) allows for head-movement in a highly restricted set of cases, one of which is cliticisation. To see how these definitions work, let us consider the derived structure of head-movement, shown in (8):

64

Ian Roberts

(8)

Y2 X

Y1

By the definition in (7a), Y2 can be minimal, but only if X is minimal and has a label non-distinct from Y. This is the central proposal regarding clitics (and head-movement) in general: clitic-placement can form a derived structure like (8), since clitics are minimal categories (Muysken 1982), and defective in that they do not have a label distinct from their host; this non-distinctness from the host is exactly the content of the notion of defectivity. Because of this, head-movement, adjoining a minimal category to a minimal head, is allowed. This is why clitics can adjoin to heads, and, in fact, why they must adjoin to heads. What exactly does the defectivity of clitics consist in? Cardinaletti & Starke (1999) describe a form of ‘structural deficiency’ in some detail, showing that a general distinction can be made among strong, weak and clitic pronouns. The motivation for the distinction between strong and weak/clitic pronouns goes back to Kayne’s (1975: 82f.) diagnostics for the clitic nature of French complement pronouns: clitic pronouns cannot appear in surface argument positions, ‘peripheral positions’ (including environments where the pronoun is in isolation, owing to ellipsis), be modified or be coordinated. Cardinaletti & Starke (1999: 168–69) also provide evidence that clitics must incorporate with their hosts, primarily from the fact that they cannot appear in initial position in V2 clauses in V2 languages and the fact that, on standard analyses, clitics move with the verb hosting them.6 They propose that structural deficiency amounts to the ‘peeling off’ of layers of functional structure, with the result that clitic pronouns consist only of the inflectional part of the structure of a pronoun. Roberts (forthcoming b) adopts De´chaıˆne & Wiltschko’s (2002: 428–31) terminology in labelling Romance clitics as jPs, rather than DPs. Romance clitics thus differ from the strong complement pronouns of a language like English in being jmin/max, rather than Dmin/max. I will say more about the third element in Cardinaletti & Starke’s typology, the weak pronouns, below.7

6

7

This last idea is challenged in Kayne (1994: 42–46). Poletto & Pollock (2004) and Pollock (2006) develop Kayne’s conclusion and reject earlier approaches to French subject–clitic inversion involving verb-movement to C in favour of remnant TP-movement; these questions are discussed in Roberts (Chapter 8). Implicit in this is a rejection of De´chaıˆne & Wiltschko’s (2002: 421–26) proposal that English 3rd-person pronouns are jPs while 1st- and 2nd-person pronouns are DPs. See Roberts (forthcoming b, Note 28) for discussion.

A deletion analysis of null subjects

65

Since the label of (active, transitive) v* contains j-features, in fact, unvalued versions of the very j-features that make up a direct-object clitic, the clitic’s label is not distinct from v*’s. More precisely, the clitic’s features, essentially j-features, form a proper subset of v*’s features. Thus the clitic can adjoin to v* and form a derived minimal head. The structure that results from cliticisation is as follows: (9)

v*min v*min

[ij] le Root/Vmin voit

v*min [iV, uj]

This account of clitic-incorporation has the following general consequence: (10)

Incorporation can take place only where the features of the incorporee are properly included in those of the incorporation host.

(10) allows for the case where an acategorial root combines with v, as in ‘v-to-V movement’, as in Chomsky (2001: 35). If clitics are jmin/max, it allows for cliticisation of the Romance kind. Where object pronouns are Ds, as in English, cliticisation (to v*) is not possible (Roberts (forthcoming b: Section 3.3) suggests that D-cliticisation is possible, but that the target for this must be C). Thus, (9) is the derived structure of incorporation, and incorporation can take place wherever (10) is met.8

8

One objection to incorporating jmin into v*min as shown in (9) is that the operation violates the Extension Condition (all operations must extend the root (Chomsky 2000: 136–38), and cf. the remarks on head-movement and cyclicity in Chomsky (2001: 38)). However, if XP-movement is triggered by an Edge Feature (EF) in the sense of Chomsky (2008), then movement to a maximal phase vmax or Cmax will always extend the root in virtue of this feature and a separate stipulation is redundant. Movement to a non-phase edge, e.g. SpecTP, does not satisfy the Extension Condition, and is not triggered by EF (being an A-position see Chomsky 2008); raising to SpecCP satisfies the Extension Condition and is triggered by EF. A related point, again mentioned by Chomsky (2001: 38), is that, if we want to ensure that the moved minimal category c-commands its copy in a derived structure like (9), we must in some way complicate the definition of c-command beyond the natural one implied by merge (the transitive closure of sisterhood and containment). However, if cliticisation is triggered by Agree, then it is unclear that any c-command relation is required beyond that between the probe and the goal, postulated independently of movement. If cliticisation is always movement of the goal to the probe, then downwards and sideways cliticisation will be ruled out by the c-command condition on Agree since such movements could only come about where the probe fails to c-command the goal. It is therefore unnecessary and redundant to impose an identical c-command condition on movement. Hence the derived structure of cliticisation in

66

Ian Roberts

What is the trigger for cliticisation? Roberts (forthcoming b) proposes that this is nothing other than Agree, seen as the need to value unvalued features in the most economical way possible. Let us compare the standard Agree relation between v* and a non-clitic direct object with that which obtains under direct-object cliticisation, construed as in (9). The standard case of Agree between v* and the direct object is as follows: (11)

a. Trigger for Agree: v*[Pers:__], [Num:__] b. Outcome of Agree: v*[Pers:a, Num:b]

D[Pers:a, Num:b], [Case:__] D[Pers:a,Num:b], [Case:Acc]

The relations illustrated in (11) hold irrespective of the presence of an EPP-feature on v* which triggers movement of the direct object to its Specifier. The boldfaced features are those copied into the feature matrices of the categories entering into the Agree relation as defined in (3) (see footnote 2 on the ACC feature in (11b)). In cases of clitic-incorporation giving rise to the derived structure in (9) the same Agree relation holds as in (11),9 and, as defined in (3), values v*’s unvalued features. Schematically, then, we have the situation in (12): (12)

a. Trigger for Agree: v*[[Pers:__], [Num:__] b. Outcome of Agree: v*[Pers:a, Num:b]

j[Pers:a, Num:b] (j[Pers:a, Num:b])

Here v’s [Pers:a, Num:b] features are valued as an automatic consequence of Agree of jmin with v*min. An important aspect of the copying of the features of the clitic valuing the features of the probe is that, given the characterisation of incorporation in (10), copying the features of the defective goal exhausts the content of the goal. Therefore the operation is not distinguishable from the copying involved in movement. In the case of incorporation, then, Agree and Move are formally indistinguishable. This means that we can think of the deletion of the copies of the features of the goal in terms of chain reduction, i.e. the deletion of all identical copies in a dependency except the highest one (see Nunes 2004: 22f.). This generally does not apply to Agree, since the content of the goal is not exhausted by Match, and so the goal does not

9

(9) is quite licit, as in this structure, the goal [ij] is incorporated into the probe and the probe c-commands the copy of this feature bundle contained in its sister VP. I follow Cardinaletti & Starke (1999) in assuming that the clitic lacks a Case feature. Thus there may no uninterpretable feature rendering it active in the sense of Chomsky (2001). I assume that goals, especially defective goals, do not have to be active (a possibility entertained in Chomsky (2001: 45 n. 29, 46 n. 38, 48 n. 56)).

A deletion analysis of null subjects

67

constitute an identical copy of the copied feature bundle.10 But, precisely in the case of incorporation, this is what happens. For this reason we see the PF effect of movement, with the j-features realised on the probe and the copy deleted. We can in fact think of the EPP-feature on the probe where the goal is non-defective as an instruction to pied-pipe parts of the goal which are not involved in the Agree relation, giving rise to copying and chain reduction/ copy deletion. But in the case of incorporation, no EPP-feature is required on the probe in order to give the PF effect of movement. This point – that incorporation is movement with no associated EPP (or EF) feature – will be central to what follows, as we shall see below. So, clitic-incorporation is a way for minimal (as well as minimal and maximal) categories to satisfy Agree, which gives the effect of movement. Hence, cliticisation takes place wherever condition (10) is met. It is clear that this instance of Move/Agree is quite distinct from those triggered by or connected with EPP-features. In fact, an important consequence of this analysis is that cliticisation, since it is triggered purely by Agree where (10) holds, is incompatible with an EPP-feature on the probe, since if there is an EPP-feature, the probe will have to Agree with the moved goal, and this goal cannot incorporate into the probe if it is to satisfy the EPP requirement of creating a specifier. We conclude that EPP-features therefore only trigger XP-movement (we might in fact think of them informally as ‘pied-piping’ features).11 Let us state this as a corollary of the above account of cliticisation/ incorporation, as follows: (13)

A probe P can act as an incorporation host only if it lacks an EPP-feature.

In fact, in order to allow a T which attracts the subject to its specifier and the verb to itself, as in standard analyses of French (see Pollock 1989), we need to restate (13) as (130 ):12

10

11

12

Chomsky (2001: 15f.) assumes that uninterpretable features are deleted at the end of the phase after Agree has taken place. But, if we take Agree to function as described in the text, it is unclear how the system can distinguish the originally unvalued features from the originally valued ones once all features are valued by Agree. Perhaps the condition is simply that one set of features has to delete, and, in the case of a defective goal, it is always the goal’s, given the nature of copy deletion and the formal identity between Move and Agree in this context. This is not quite accurate. If heads can move, head-movement to specifier position is definitely a possibility. Such an operation, which has been proposed by Matushansky (2006), Roberts (forthcoming b) and Vicente (2005; 2006) among others, does not involve pied-piping, and yet presumably satisfies the EPP. However, the general conclusion that the presence of an EPP-feature as part of a feature bundle of a given head precludes incorporation as a means of valuing those features holds. The formulation in the text suffices for present purposes, however. Thanks to Eric Haeberli for drawing my attention to this matter.

68

Ian Roberts

(130 )

A probe P can act as in incorporation host for a goal G only if P lacks an EPP-feature capable of attracting G.

(13) has a number of interesting consequences (see Roberts forthcoming b). We will see below that it plays a role in our understanding of null-subject deletion too. Here we can briefly note two consequences of (13). First, if it is correct, following Kayne (1994), to think that surface OV order is derived, either by object-movement to Specv* or by VP-movement to that position (pied-piping the object probed by v*; see Richards & Biberauer 2005), then (13) implies that complement clitics, or at least object clitics, of the Romance kind are not found in OV languages, since v* must have an EPP feature capable of attracting the object in order to trigger movement of the object, or of the category containing the object. As far as I am aware, this prediction is correct; the languages which show adverbial clitics of the Romance type are all quite clearly VO (i.e. the Romance languages themselves, Modern Greek, Macedonian-Bulgarian, and possibly Swahili and other Bantu languages). Also, in the case of the Romance languages and Greek, it is fairly clear that the modern clitic system developed after the change from OV to VO order (see Wanner 1987). More generally, Greenberg’s Universal 25 can be partially derived in these terms. This universal states: “If the pronominal object follows the verb, so does the nominal object” (Greenberg 1963/2007: I, 57). Thus Greenberg observes that three out of four logical possibilities are found, as follows (with “no contrary instances in the sample”, i.e. this is not presented as a statistical universal): (14)

a. b. c. d.

pronoun V & O V pronoun V & V O V pronoun & V O *V pronoun & O V

Assuming that surface OV order is the reflex of an EPP-feature on v*, then, as we have seen, any pronouns and all objects will precede V; this is the situation in (14a). The situation in (14b) is that of Romance and the other languages mentioned above: v* has no EPP-feature but the pronoun appears preverbally as a clitic. (14c) is the case where v* has no EPP and pronouns are not clitics, as in English. Finally, (14d) is impossible, for the reasons we have seen. So, with the proviso that (14b) refers only to clitic objects, our assumptions derive Universal 25. The second prediction concerns verb- and auxiliary-movement. Roberts (forthcoming b: chapter 4) shows that many standard cases of verb-movement to T can be analysed as incorporation of the type in (9). The verb (i.e. the V–v complex) is a defective goal where T has V-features (auxiliaries may always

A deletion analysis of null subjects

69

be defective goals if they lack V-features but have T-features). In this context, it is interesting to contrast verb-movement to T with VP-movement to SpecTP of the kind seen in various VOS and VSO languages, according to recent analyses (Massam & Smallwood (1997), Massam (2000), Rackowski & Travis (2000), and many of the papers in Carnie, Dooley & Harley (2005)). In her study of VOS and VSO in Niuean, for example, Massam (2000) argues that there is an operation fronting a verbal constituent, and that this constituent is fronted to a position within TP. VOS order is derived by VP-fronting, and VSO by object-shift to a VP-external position combined with remnant VP-fronting, as shown in (15):13 (15)

a. [TP [VP V O] T [vP S v.. (VP)]] – VOS b. [TP [VP V (O)] T [vP S v [AbsP O (VP)]]] – VSO

As (15) shows, the landing-site of VP-fronting is taken to be SpecTP; Massam argues that this is motivated by essentially the same property as that which causes the subject to raise to SpecTP in languages like English, French and Mainland Scandinavian: the operations “can be seen as two reflections of a single EPP predication feature” (Massam 2000: 111). This type of analysis, first put forward by Massam & Smallwood (1997), and developed by Rackowski & Travis (2000), has been applied to a number of languages which display both VOS and VSO orders (mainly but not exclusively Polynesian and Mayan languages; Roberts (forthcoming b: chapter 4) argues that VSO order in Celtic is derived by head-movement rather than remnant VP-movement). Now, if VP-movement satisfies the EPP in languages like Niuean, and if V-v must move to T in languages like the Romance languages since where T has V-features V-v is a defective goal, then we predict that VP-movement to SpecTP is found only in languages where V-movement to T is banned, given (13). Further, given the proposal in Biberauer & Roberts (Chapter 7) to the effect that verb-movement to T is connected to richness of tense inflection, we arrive at the prediction that VOS/VSO languages of the Niuean type have impoverished tense morphology. As seen there in more detail, this prediction seems to be borne out. Moreover, the Celtic V-initial languages contrast in the richness of their tense morphology with languages like Niuean (see Biberauer & Roberts (Chapter 7)) for details. This line of argument is developed and documented more fully in Roberts (forthcoming b: chapter 4).

13

AbsP in (15b) stands for Absolutive Phrase, which Massam suggests may correspond to AgrOP in more familiar languages. My summary here glosses over the complication that Niuean is an ergative language and Massam’s treatment of the assignment of ergative and absolutive case.

70

Ian Roberts

Here we have seen the basic mechanics of cliticisation/incorporation, according to Roberts (forthcoming b). The most important point for what follows is the impossibility of incorporation into a probe bearing an EPP-feature, stated in (13). In the light of this, let us return to the consideration of the nature of null subjects.

2.2

Clitics and ‘pro’: similarities and differences

The above account of the derived structure and mechanisms of cliticincorporation yields a notion of defective goal, which derives directly from the general characterisation of the precondition for incorporation in (10): (16)

A goal G is defective iff G’s formal features are a proper subset of those of G’s probe P.

Since clitics are j-elements, they count as defective in relation to v* in this way. Object DPs do not, and neither do the other typical categories of complement (CP, PP, AP, etc.). Defectivity as defined in (16) is a relative rather than an absolute notion; in general a category is defective if it is probed by a category of a similar type. As argued in Roberts (forthcoming b: chapter 4), and briefly summarised above, verbal categories, in particular auxiliaries, can count as defective when probed by categories such as T. As presented so far, a defective goal must be incorporated into its probe: given the nature of incorporation as a feature-copying operation, as described above, it is required as the means of feature valuation for v* in the case of object cliticisation. As already mentioned, a consequence of all this is that head-movement/ cliticisation entails that the probe has no EPP feature and the presence of an EPP-feature entails that the goal is not incorporated. Now, where do null subjects, which I continue to call pro for convenience, fit into this picture? More specifically, can we treat pro as a kind of clitic, and perhaps derive the deletion property from that? This is probably not feasible; both Cardinaletti & Starke (1999) and Holmberg (2005) argue that pro is a weak pronoun. Cardinaletti & Starke (1999: 175–76) give three arguments for their position. First, pro has the semantic properties of a deficient pronoun in that it can be expletive, impersonal, have non-human referents and “cannot occur with ostension to denote a non-prominent discourse referent” (p. 175). These points are illustrated by the following Italian examples: (17)

a. pro/*lui piove molto qui. It rains a-lot here b. pro/*loro mi hanno venduto un libro danneggiato. They me have sold a book damaged ‘I have been sold a damaged book.’

A deletion analysis of null subjects

71

c. pro/*lui e` molto costoso. It is very expensive d. Lui/*pro e` veramente bello. He (over there) is really nice.

Second, pro can only occur in designated specifier positions, like weak pronouns in general. As Cardinaletti & Starke (1999) point out, the special distribution of pro, as analysed by Rizzi (1986a), does not in fact follow from the fact that it is null, pace Rizzi, but from the fact that it is weak. Weak pronouns are required to move as maximal categories to designated specifier positions, although it is unclear why. Third, “[g]iven the choice between a strong pronoun and a pro counterpart, pro is always chosen” (Cardinaletti & Starke 1999: 175), as contrasts like the following show: (18)

Gianni ha telefonato quando pro/*lui e` arrivato a casa. Gianni has called when he is arrived to home ‘Gianni called when he got home.’

So pro is just like other weak pronouns regarding its syntactic and semantic properties, but obviously differs phonologically. In fact, as Cardinaletti (2004: 132) observes, pro shares with weak pronouns the suprasegmental phonological property of being unable to bear word stress. It differs from them only in lacking segmental specification. Cardinaletti’s observation is clearly consistent with Holmberg’s (2005) conclusion regarding pro. The evidence in (16) and (17) clearly shows that pro is not a strong pronoun. Given Cardinaletti & Starke’s (1999) three-way typology of pronominal elements, though, the question arises as to whether pro is a clitic or a weak pronoun. There are several reasons to think it is in fact a weak pronoun. First, pro occupies a specifier position, namely SpecTP. This is clear from Holmberg’s (2005) argument given in the previous section. Furthermore, Cardinaletti (1997: Section 3; 2004: 141) shows that overt weak pronouns such as egli (“he”) cannot appear in dislocated positions and can appear in unambiguously TP-internal positions: (19)

a. Weak pronouns cannot be left-dislocated: Gianni/*egli la nostra causa non l’ha appoggiata. John/ he the our cause not it.has supported ‘John, our cause, he hasn’t supported it.’ b. Weak pronouns can appear in ‘Aux-to-Comp’ contexts (Rizzi 1982): Avendo Gianni/egli/pro telefonato a Maria, . . . Having John/he telephoned to Mary, . . . ‘John/him having called Mary, . . .’

Ian Roberts

72

c. Complementiser deletion: Credevo Gianni/egli/pro avesse telefonato a Maria. I-thought John/he had telephoned to Mary ‘I thought John had called Mary.’

(19a) shows that overt weak pronouns cannot be dislocated, so presumably the same applies to pro. (19b) shows that weak pronouns, both overt and non-overt, can appear in the Aux-to-Comp construction where the gerundive auxiliary has moved to C on standard assumptions (Rizzi 1982: chapter 3). Finally, dislocated elements are not allowed in complementiser-deletion contexts like (19c) (cf. ??Credevo il libro Maria lo avesse dato a Gianni ‘I thought, the book, Mary had given to John’ (Cardinaletti (2004: 141)), but weak-pronoun and other subjects are. All of this, combined with Holmberg’s argument given in the previous section, shows that pro can appear in the preverbal subject position SpecTP.14 Second, Rizzi (1987) showed that only preverbal subjects can license a floated quantifier: (20)

a. Tutti i bambini sono andati via. All the children are gone away b. I bambini sono andati tutti via. The children are gone all away c. Sono andati via tutti i bambini. Are gone away all the children

(21)

a. *Sono andati tutti via i bambini. Are gone all away the children ‘All the children have gone away.’ b. Sono andati tutti via. Are gone all away ‘They have all gone away.’

In (21b), pro must be preverbal, since a post-verbal subject cannot cooccur with a floated quantifier, as the contrast between (20b) and (21a) shows.

14

If egli is a weak pronoun, and as such a Dmin/max, we might ask why it does not delete under the Agree relation with T, since its features appear to be subsumed by those of (null-subject) T (see below for the proposal that null-subject T has a D-feature). I tentatively propose (pace Cardinaletti & Starke 1999: 186f.) that this pronoun, and the similar esso series in Italian, differs from pro in having a Case feature. Owing to its Case feature, egli is not a defective goal in the sense described in the previous subsection, since it has a feature that its probe lacks. Strong pronouns arguably differ from weak pronouns in having the full functional structure of any DP, while, as we have suggested, weak pronouns are Dmin/max and clitics are jmin/max. For an alternative proposal regarding the feature content of egli, which is also compatible with the proposals made here, see Cardinaletti (2004: 149–50). Holmberg (Chapter 5) argues that generic null subjects in Finnish must have Case features; I will return to this point below.

A deletion analysis of null subjects

73

Third, Cardinaletti (1997: 38–39) observes that in the Central Italian dialect spoken around Ancona, 3PL agreement may fail with inverted subjects, but not with preverbal subjects:15 (22)

a. Questo, lo fa sempre i bambini. This, it does(3SG) always the children. b. *Questo, i bambini lo fa sempre. This, the children it does(3SG) always. c. Questo, i bambini lo fanno sempre. This, the children it do(3PL) always ‘The children always do this.’

A 3PL null subject cannot appear with the 3SG verb: (23)

a. *Questo, lo fa sempre. This, it does(3SG) always (impossible with the 3PL interpretation of the subject) b. Questo, lo fanno sempre. This, it do(3PL) always ‘They always do this.’

All of these arguments point to the conclusion that pro must appear in SpecTP. This requirement to occupy a designated specifier position is typical of a weak pronoun, as we have already observed. I conclude then that pro is a Dmin/max not a jmin/max (in this I differ from Holmberg (2005) – see section 1 above; I also differ from Cardinaletti & Starke (1999) in assuming that weak pronouns have no internal structure, or at least that pro doesn’t). This element is a Dmin/max without a morphosyntactic root; I interpret this to mean that there is no internal phase, no nP or NP here. Since, pro is nonetheless a D, it cannot cliticise in the manner of complement clitics in Romance, either to v or to T. 2.3

Pro and the EPP

If pro is a Dmin/max which, in Italian at least, must occupy SpecTP, it presumably moves there to satisfy the EPP. Sun˜er (2002) argues essentially the same for Spanish, as does Ordo´n˜ez (2006); Costa (2004) argues the same for European Portuguese, and Adraga˜o & Costa (2004) argue this for child European Portuguese (Brazilian Portuguese appears to be a partial nullsubject language; see the papers in Kato & Negra˜o (2000), Holmberg, Nayudu & Sheehan (2009), Roberts (forthcoming a) and the references given

15

Manzini & Savoia (2005, I: 338f.) give many examples of this phenomenon, from a range of varieties, some of which have an expletive subject clitic with free inversion and some of which do not.

74

Ian Roberts

there). Given (13), this is also consistent with the conclusion in the previous section that pro is not a clitic. Sheehan (2006; Chapter 6) discusses the nature of the EPP in a range of Romance languages, and concludes that the only case where the EPP does not seem to hold of SpecTP in Romance is in VOS orders in Spanish and Italian. She proposes a variant of Zubizarreta’s (1998) proposal for subject-lowering triggered by prosodic factors to account for these orders. Hence the EPP can be taken to hold for SpecTP here too. A further reason to think that pro can satisfy the EPP comes from the fact that we find null subjects in OV languages. Turkish appears to be an example of a ‘rich agreement’ null-subject language, and is clearly OV (Japanese is also OV, but shows the ‘radical’ type of null subject, while Marathi is OV and shows ‘partial null subjects’ see Holmberg and Sheehan (Chapter 3); Holmberg, Nayudu and Sheehan (2009)). In her sketch of the grammar of Turkish, Kornfilt (2003, IV: 305) says “[b]ecause of the richly differentiated agreement system, subjects of both main and embedded clauses . . . can (and preferably do) remain unexpressed when interpreted as personal pronouns” ¨ ztu¨rk (2001) for a different view). Here is (see also Kornfilt (1997: 132); cf. O a simple example: (24)

Ben eve gel- di- m. pro kitap oku-du- m. pro televizyon seyret-ti-m. I house- Dat come- Past- 1SG book read-Past- 1SG television watch-Past-1SG ¨ ztu¨rk 2001: 241) ‘I came home. I did some reading. I watched TV.’ (O

Assuming that this is a genuine case of an Italian-type ‘consistent’ null subject ¨ ztu¨rk (2001) for a dissenting in Holmberg’s (2005) terminology (and see O view), we see that there is no incompatibility between pro and OV order. Now, if we adopt the analysis of OV (and VP–Aux) order put forward in Biberauer & Roberts (2005), the derived structure of an SOVAux sentence is as follows: (25)

[TP [vP S tV-v [VP tV O]] V-v þ T]

This structure is derived by V-movement to v and onward movement of V-v to T, giving rise to the morphologically complex final verb forms observed in (24) and known to be characteristic of Turkish (see the papers in Taylan 2001), combined with remnant VP-movement to SpecvP and remnant vP-movement to SpecTP. Both VP-movement and vP-movement are triggered by EPP-features (of v and T, respectively). Biberauer & Roberts (2005), following Richards & Biberauer (2005), argue that v and T here probe DP, respectively the object and the subject, and that VP- and vP-movement are the result of a pied-piping option. Biberauer & Roberts (2005) argue this for

A deletion analysis of null subjects

75

Old English and the modern West Germanic languages. The analysis can presumably be extended to OV languages like Turkish. What this means is that T has an EPP-feature in Turkish, one which is associated with j-features on T which probe the subject.16 Turkish merely has the additional extra piedpiping requirement. Most important for our purposes, pro can be the goal; it can be inside a larger pied-piped category which satisfies the EPP. Furthermore, Holmberg’s (2005) argument for a deleted pronoun in nullsubject sentences in Finnish depends on the idea that pro is able to satisfy the EPP (in SpecTP, as Holmberg (pp. 542–43) argues). This can be seen from contrasts like that in (5), repeated here: (5)

a. Puhun englantia. speak-1SG English ‘I speak English.’ b. Sita¨ meni nyt hullusti. EXP went now wrong ‘Now things went wrong.’

Here there is a pro in SpecTP in (5a). So we see that pro can either satisfy the EPP, or that it can be a goal inside a larger pied-piped category which satisfies the EPP. Given (13), this implies that pro cannot be a clitic. 2.4

The trigger for deletion

I will follow the commonly articulated intuition that the core property which null subjects (of the Italian kind) have is that T is ‘pronominal’. In different ways, this idea underlies the analyses in Rizzi (1982), Alexiadou & Anagnostopoulou (1998) and Holmberg (2005), for all that they otherwise differ.17 Following Holmberg (p. 555), I take it that languages of this type have a D-feature in T.18 Also following Holmberg, I take this D-feature to be 16

17

18

At the same time, V-v incorporates with T as a defective goal. There is no incompatibility here between T’s probing the D-feature, leading to pied-piping, and incorporation of V-v triggered by T. Here T’s EPP-feature attracts the category containing D, and cannot attract V-v. Cf. also the following remark by Apollonius Dyscolus on Ancient Greek, also a consistent null-subject language (On Syntax, Book 1, }17; Householder 1981: 25): ‘The nominative [subject] is implicitly present in [finite] verbs, and it is definite (i.e. has definite reference) in the first and second persons, but indefinite in the third because of the unlimited number of possible referents.’ Holmberg (2005: 556) treats this D-feature as valued, and posits an unvalued D-feature on the subject pronoun. I see no strong reason to treat pro as different from other argumental Ds in this respect, and so tentatively treat the D-feature of T as unvalued, with that of pro being valued. This does not greatly affect what follows, however. On the other hand, I concur that null subjects in ‘partial’ null-subject languages may lack an interpretable D-feature, something which explains a number of their peculiarities; see Holmberg, Nayudu & Sheehan (2009) and Holmberg (Chapter 2). In Roberts (forthcoming a) I speculate as to how pro may have lost its D-feature in the recent history of Brazilian Portuguese, where there appears to have been a fairly recent change from consistent to partial null subjects.

76

Ian Roberts

correlated with ‘rich’ agreement (I will consider the nature of rich agreement in more detail in Section 2.5). The postulation of the D-feature on T in null-subject languages means that pro counts as a defective goal in such languages. Its features, j and D, are properly included in T’s. But T also has an EPP-feature, which pro can satisfy, as we have seen. We saw in Section 2.1 that defective goals must cliticise, and that a probe P can act as an incorporation host for a given defective goal only if it lacks an EPP-feature capable of attracting that goal. In fact, we can now see that those conclusions are not exactly correct: a defective goal can satisfy an EPP-feature – this is what we observe with pro. But we can exploit the fact that pro is a defective goal to derive its silent nature, in terms of the following generalisation about defective goals: (26)

Defective goals always delete/never have a PF realisation independently of their probe.

(26) holds for clitics; as we saw in Section 2.1, the copy of the clitic, i.e. the goal itself, deletes and its features are realised on the probe as the incorporated clitic. As we have seen, pro does not incorporate, and in fact cannot, given (13). Nevertheless, it deletes (or fails to be PF realised – see Section 2.5 on this). In Section 2.1, we suggested that the reason for (26) may be connected to the nature of chain reduction in the sense of Nunes (2004). To the extent that pronouns are often defective goals in relation to j- and D-bearing goals, (26) entails the ‘Avoid Pronoun’ principle of Chomsky (1981). It ought to be possible to derive (26) from a suitable generalisation of chain reduction; since we have observed that in the case of incorporation Move and Agree cannot be distinguished, we can think that the occurrence of the defective goal undergoes deletion for the same reason as copies do. This can extend to pro if we see both its first-merged and second-merged occurrence as essentially copies of (the features of) the probe. Clearly, Nunes’ notion of chain reduction needs to be generalised so as to refer to copies of (subsets of) features of the probe. Another, quite unrelated, example of (26) might be null wh-operators. It is natural to see these as deleted wh-phrases (see Chomsky & Lasnik (1977: 446) for a characterisation of ‘deletion in COMP’, which is intended to interact with various ‘surface filters’; we can think of the former as an operation taking place at the end of the Narrow-Syntactic derivation subject to the effects of the latter, which are, or derive from, output conditions at the (PF) interface). It may be that the context for their deletion is exactly analogous to that of null subjects: they delete when attracted, by an EPPfeature, to the specifier of a head which has a superset of their features. In the case of null operators, the relevant head would be a wh-C, and we might think of the deleted element as being a defective goal, perhaps a bare wh-pronoun

A deletion analysis of null subjects

77

rather than a full-fledged wh-DP. Null operators appear in the following contexts (the angled brackets in (27c) are intended to indicate the constituent that has undergone comparative ellipsis): (27)

a. Object relatives (optional in the context of that-deletion): [The man [(who) I saw –]] is John. b. Infinitival relatives (obligatory): John found [a book [(*which) [to read –]]]. c. Comparatives (normatively obligatory): Ruth is stranger [than [(what) [Richard is < – strange >]]]. d. Easy-to-please constructions (obligatory): John is easy [(*wh) [to please –]].

On the other hand, null wh-phrases are impossible in interrogatives and wherever material is pied-piped: (28)

a. *(Who) did you see – ? b. I wonder *(who) you saw – /to talk to –. c. John is the man [to *(whom) [we should talk – /to talk –]].

We can tentatively attribute the fact that interrogative wh-elements can never be deleted to their semantics: an element like interrogative who simultaneously embodies both a wh-quantifier and its restriction, i.e. it has an interpretation like for which x, x a person (Baker 1970; Chomsky 1973). For this reason, it is plausible to think that it is more than just a bare quantifier, whatever its exact structure. On the other hand, in the cases where deletion is either possible or required, the wh-element is arguably not quantificational.19 The impossibility of deletion when the wh-phrase is pied-piped with a preposition, shown in (28c), cannot be attributed to the simple fact of piedpiping alone, since we proposed in the previous section that a null subject can appear inside a pied-piped vP in an OV language. It could be due to the fact that the complement is pied-piped here, rather than the specifier as in the vP case in the previous section,20 or it could be due to the fact that the preposition instantiates a Case feature, not borne by C, and so such elements are not 19

20

It might be that the normatively deleted what in comparatives is quantificational; degree clauses have been standardly interpreted as involving quantification over degrees (see Bhatt & Pancheva (2004: 4–6) and the references given there). However, the morphological realisation of the degree quantifier is -er; the deletable what may correspond to a l-abstractor over degrees, given that the semantics relates the set of degrees to which Bill is tall to the set of degrees to which John is tall (Bhatt & Pancheva 2004: 38–41). In that case, the wh-expression is non-quantificational, and similar to the wh-expressions found in relatives. Deletion is impossible where the relative is a possessive in SpecDP: (i)

The man [(*whose) friends we met – ].

This could be due to the fact that whose is really who combined with the determiner s, and as such a non-constituent (see Chomsky 1995b: 263).

Ian Roberts

78

defective goals and therefore cannot be deleted. I will leave this and other points regarding the nature of null operators open. The central observation in the present context is simply that they constitute a further possible case of a defective goal attracted to the specifier of their probe by an EPP-feature, and as such they are required to delete. In this connection, it is interesting to note that Kayne (1994: 88) mentions that an “avoid relative pronoun if possible” principle is at work in French and Italian relatives, where only the complementiser que/che is allowed in subject and object relatives: (29)

a. la persona che/*cui Bill ha visto b. la personne que/*qui Bill a vue the person that/who Bill has seen

(Italian) (French)

A further prediction stemming from (26) concerns v. If consistent, agreementlicensed null subjects are only found where T has j-features, D-features and an EPP-feature, then we expect exactly the same to hold at the v level. We therefore expect to find consistent, agreement-licensed null objects just where v has j-features, a D-feature and, most important, an EPP-feature. It therefore follows that such null objects will only be found in OV languages, since the presence of an EPP-feature on v will guarantee OV order (perhaps by means of VP-pied-piping; see (25) above). Many of the cases of null objects that have been discussed in the literature appear to be instance of ‘radical’ argument drop; this is the case for Chinese, Japanese and other East Asian languages, arguably also Brazilian Portuguese.21 Whatever is going on here, it is clear that agreement is playing no role in licensing these elements, since there is no object agreement in these languages. The arbitrary null object of Italian, discussed and analysed in Rizzi (1986a), may have more properties in common with ‘partial’ null subjects; again, it is not associated with object agreement and, given its arbitrary interpretation, we conjecture that it is not associated with a D-feature (see footnote 18 and Holmberg and Sheehan (Chapter 3)). There are, however, languages with ‘rich’ object agreement and what seem to be consistent null objects, with properties similar to those of null subjects in languages like Italian and Spanish (notably definite reference and the association with rich agreement). One such case is Pashto, as discussed in Huang (1984). Pashto is a split-ergative language, showing a nominative–accusative

21

These might be cases of topic deletion (in the sense of Huang (1984), Raposo (1986), Modesto (2000a)), which it would be natural to account for in terms of the notion of defective goal. However, if Chomsky (2008) is correct in proposing that topicalisation is triggered by an Edge Feature, and that such cases of movement do not involve Agree, then it is difficult to see how to make such an account work in the terms being assumed here. See below for a further brief comment on ‘radical’ pro-drop, in the light of the proposals in Saito (2007).

A deletion analysis of null subjects

79

agreement pattern in tenses formed from the present verb stem and an ergative–absolutive one in tenses formed from the past verb stem (MacKenzie 2003: 255–56). The agreement marking is consistently rather ‘rich’. In transitive clauses showing the ergative pattern, the verb agrees with the direct object, as shown in (30a), and a null object (with definite reference) is possible, as (30b) shows: (30)

a. ma man¸a wə-xwar-a I apple Prf-eat-3FSG ‘I ate the apple.’ b. ma e wə-xwar-a I Prf-eat-3FSG ‘I ate it (fem.)’

(Huang 1984: 535–36)

We see that the null object can appear exactly where there is agreement with the object, and that object has a definite interpretation: so we interpret Huang’s e as pro. Most importantly in the present context, we can observe from (30a), and other similar examples given by Huang, that Pashto is OV (see also MacKenzie 2003: 256). So this is an example of a consistent, agreement-licensed null object in an OV language. The prediction that emerges from our discussion is that we do not expect to find VO languages without object agreement showing definite, non-discourse-licensed null objects. Unfortunately, it is hard to check this prediction, given the different types of null object that are found, as noted above. In this section we have seen the nature of the trigger for null subjects: the presence of a D-feature on T makes pro a defective goal. Where T also has an EPP-feature, pro cannot cliticise, since this feature is incompatible with incorporation. Nevertheless, since defective goals always delete under feature identity with their probe, pro lacks phonological realisation. In the final part of this section, I turn to the question of the nature of pro’s non-realisation: is it deletion or is it failure to be associated with a phonological matrix?22

22

As we have seen, it is important for our characterisation of pro as a defective goal that it lack a Case feature. If, following Chomsky (2000; 2001), probes lack (structural) Case features, then pro cannot have such a feature and be a defective goal, since its possessing a Case feature will mean that its feature content is not properly included in that of the probe. On the other hand, Holmberg (Chapter 5) argues that the generic null subject pro in Finnish has a Case feature, on the basis of the way in which this element interacts with case-assignment generally in this language. As is well-known, Finnish is rich in morphological case. The basic rule for the casemarking of non-partitive direct objects is that they are marked accusative only if the subject is marked nominative (GEN ¼ genitive): (i)

a. Mina¨ osti-n auton/*auto. I-NOM bought 1SG car-ACC/*car-NOM

‘I bought a car.’

80

Ian Roberts

2.5

Deletion, non-realisation and syntactic impoverishment

Holmberg (2005: 559) concludes that pro may be either a deleted pronoun, or one which fails to have a PF realisation. The central point of his analysis is that, as far as the core computational processes of Narrow Syntax are concerned, pro is just like an overt pronoun; its non-overtness is purely a PF matter. But we should briefly consider whether it is possible, or desirable, to distinguish the two options of deletion and non-realisation. I will try to do this here, and in the process introduce Mu¨ller’s (2005) notion of pre-syntactic impoverishment as a way of accounting for the relation of ‘rich’ verbal agreement inflection to null subjects. I will show how Mu¨ller’s idea can be integrated into the proposals being made here. I will argue that this entails, b. Minun pita¨a¨ ostaa auto/*auton. I-GEN should buy car-NOM/*car-ACC ‘I should buy a car.’ c. Siella¨ ostettiin auto/*auton. There bought-PASS car-NOM/*car-ACC. ‘A car was bought there.’

(Holmberg’s (11a,c,e))

(ia) is a transitive, with a nominative subject and an accusative object; (ib) is a ‘necessive’ construction, where the subject is genitive – here the object shows up as nominative and cannot be accusative. (ic) is a passive, where the single overt argument is nominative rather than accusative. The patterns are the same where we have a generic null subject: (ii)

a. Jos ostaa auton/*auto . . . If buy-3SG car-ACC/car-NOM . . . ‘If you buy a car . . .’ b. Nyt kannattaa vuokrata auto/*auton. Now should-3SG rent car-NOM/*car-ACC ‘It’s good value renting a car now.’

(Holmberg’s (13)) (Holmberg’s (16b))

Holmberg concludes “[b]roadly speaking, the generalization is that the G-pronoun [generic null-subject pronoun – IGR] occurs in any case-environments where overt subjects occur”, and goes on to give examples of the generic null subject in partitive and adessive contexts. However, these observations do not force us to the conclusion that pro has a Case feature, and certainly not to the conclusion that pro has a Case feature which its probe does not have. For partitive, adessive and genitive subjects, it is clear that the probe must be specified somehow as being associated with the relevant feature, and so if pro bears a Case feature then it still counts as a defective goal; this is in fact the line of analysis that Holmberg adopts (assuming therefore that finite T has a nominative feature, for example). But it is not clear that pro must be specified as bearing the relevant feature: overt DPs must be specified with a feature which can be realised as one or another form of case-marking, determined by the nature of their probe, but it is not clear that we have to say this about pro, precisely because it lacks a morphophonological realisation. We could therefore assume that structural Case features (Nominative and Accusative) are not present on probes, are valued in the appropriate way on the relevant, non-defective goals, and absent on pro. Lexical case features such as adessive, partitive and perhaps genitive in Finnish are arguably interpretable features of the probes, directly associated with y-roles. Again, we do not have to say that they are present on pro.

A deletion analysis of null subjects

81

all other things being equal, that null subjects are in fact deleted pronouns rather than feature matrices which fail to have a segmental realisation. Let us begin with impoverishment. In Distributed Morphology, impoverishment is a deletion operation which affects the feature bundles created and manipulated by the syntax, taking place after syntax but before ‘vocabulary insertion’, the post-syntactic operation which pairs phonological and morphosyntactic features (in the functional domain; I will say nothing about the realisation of lexical items here). Impoverishment rules “neutralize differences between syntactic contexts in morphology” (Mu¨ller 2005: 3), thus having the effect of giving the same PF realisation to syntactically (and LF-) distinct bundles of features. In other words, impoverishment rules create what Mu¨ller refers to as “system-defining syncretisms”. This kind of syncretism is distinct from accidental homophony or gaps in a paradigm. System-defining syncretisms hold across a morphological subsystem: two or more distinct feature specifications may have the same realisations where other aspects of the specification varies independently. For example, Mu¨ller (p. 5) gives the following two impoverishment rules for German verbal inflection: (31)

a. [1] ! /[–2,–PL, þpast] __ b. [1] ! /[–2, þPL] __

These rules delete the 1st-person feature ([þ1] is the value of 1st person, [–1] specifies 2nd and 3rd person) in two contexts: non-2nd person singular past tense, and non-2nd plural in all tenses. Since the 1st-person feature distinguishes 1st and 3rd persons, the upshot of (31) is these persons are never distinguished in the singular of past-tense verb forms or in the plural of any verb in any tense. Both of these are correct observations about German verbal inflection. Mu¨ller (2005: 10) proposes the “pro generalisation”, intended to link null subjects and rich agreement in terms of impoverishment: (32)

An argumental pro DP cannot undergo Agree with a functional head a if a has been subjected (perhaps vacuously) to a j-feature neutralizing impoverishment in the Numeration.

(I will return directly to the idea that impoverishment takes place in the Numeration.) We can relate this to the idea adopted in the previous section that the T which licenses null subjects has a D-feature. Let us suppose that this D-feature is really a definiteness feature (this idea is arguably implicit in Holmberg’s (2005) analysis, too; see also Holmberg (Chapter 2)). That is, pro in consistent null-subject languages has a D-feature valued as definite, and, under Agree with T, it values T’s D-feature in this way. This gives us a very concrete way of specifying the fact that null subjects in consistent null-subject

Ian Roberts

82

languages have definite reference (and, as pointed out by Perlmutter (1971), they require a special element, e.g. the subject clitic si in Italian, in order to have indefinite or arbitrary reference; see also the discussion of generic null subjects in partial null-subject languages in Holmberg (2005: 548–50), Holmberg and Sheehan (Chapter 3) and Holmberg (Chapter 5)). Following Russell (1905), I take the notion of definiteness to involve existence and uniqueness.23 In order for existence and uniqueness to be determined, it is arguably natural to require of a definite element that it have a full specification of person and number features. Let us therefore adopt the following postulate, relating definite D to j-feature specifications: (33)

If a category a has D[def], then all a’s j-features are specified.

As we have seen, impoverishment removes certain j-features from a head. So it follows from (33) that where this happens D cannot be specified as definite. In the case of T, on the assumptions we have been developing here, this means that D cannot be valued by pro if any of T’s features have been subject to impoverishment.24 If D is present, then the derivation will crash. Hence a T with impoverished features cannot bear a D-feature. As we saw in the previous section, where T lacks a D-feature pro, being a weak pronoun and therefore a DP, is not a defective goal. And therefore, given (26), pro cannot be null, i.e. cannot undergo deletion or fail to have a PF realisation. We can thus derive Mu¨ller’s pro generalisation from the postulate about the interaction of D[def] with j-features in (33) combined with our conclusions regarding pro as a weak pronoun and the nature of defective goals. This also creates a connection, exactly as postulated by Mu¨ller, between rich agreement and the licensing of consistent null subjects. In these terms, then, consistent null-subject languages have non-impoverished j-features on T, T bears a D-feature, and pro is a defective goal and therefore null. Non-null-subject languages such as English have impoverished j-features on T, and therefore no D-feature and no pro (i.e. if there are weak subject pronouns they are overtly realised). ‘Radical’ pro-drop languages such as Chinese and various other East Asian languages clearly lack j-features on T and hence pattern with English rather

23

May (1985: 8) analyses the as a generalised quantifier as follows: (i)

24

the(X, Y) ¼ 1 iff X ¼ X \ Y ¼ {a} for a ∈ D.

To put it another way, in [TP [DP the k NP] . . . VP] “requires the existence of exactly k individuals” that are in the denotation of VP (Larson & Segal 1995: 320). The features cannot be valued by pro as they are absent, i.e. both the attribute and the value are missing, as indicated in (31). We will see directly that this should be interpreted to mean that they are marked for deletion.

A deletion analysis of null subjects

83

than with consistent null-subject languages in this regard. It seems clear that in these languages other deletion processes are at work, perhaps those identifying null topics in the sense of Huang (1984) (but see footnote 21 above), as well as possibly N0 -deletion as in Tomioka (2003) and/or context-free zero-realisation of morphologically transparent pronouns as proposed in Neeleman & Szendro˝i (2007, 2008).25 As briefly recounted in the Introduction to this volume, Saito (2007) suggests that a single LF-copying mechanism allows for radical pro-drop and argument ellipsis. This mechanism involves copying elements into argument positions at LF from a set of discourse-given entities. As Saito (p. 7) points out, it “is conceivable that pro, being a typical discourse entity, is always included in this set” along with antecedents for the recovery of elided arguments. The precondition for this kind of ‘covert merge’ of arguments is that j-features not be obligatory on probes; if they are, then the absence of a goal when Agree takes place in the pre-LF part of the derivation will mean that these features fail to be valued and the derivation will crash. If, on the other hand, these features may fail to be present, then arguments may fail to be present prior to LF. Presumably, j-features may be absent from probes in systems where they completely fail to have any PF realisation. Thus the relation between radical pro-drop and absence of agreement marking is established (see also Kuroda 1988).26 In our terms, ‘discourse pro’ in these languages would be a bare nominal, presumably lacking j-features and certainly lacking a Case feature. Since it only appears at LF, the only requirement is that it be able to function as an argument; it does not behave as a defective goal since the putative probe does not behave as a probe: there is no Agree relation as the relevant uninterpretable features are missing. How does all this bear on the question of the precise nature of pro’s nonovertness? Mu¨ller (2005: 7–8) gives three conceptual arguments against postsyntactic lexical insertion. One problem is that inflection markers are split into the functional element and the vocabulary item. A second, more serious, problem is that insertion appears to be a very different, and more complex, operation than the elementary operations of current minimalist syntax (primarily Merge and Agree). Third, following Alexiadou & Mu¨ller (2005), Mu¨ller points out that late insertion cannot simultaneously satisfy both the Inclusiveness Condition (which prevents new material from being added during the derivation) and the Legibility Condition (features can be present in a given component of the grammar only if they are interpretable in that

25 26

Neeleman & Szendro˝i’s theory is described in Section 1.2.3 of the Introduction to this volume. This analysis entails, as Saito points out, that the case-markers of languages like Japanese and Korean are not instantiations of structural Case features of DPs to be valued by the j-features of probes. Saito (2007: 17) presents evidence that the Japanese cases are always either ‘contextual’ or inherent.

84

Ian Roberts

component). In this connection, Mu¨ller observes that morphological class features either violate Inclusiveness, by being inserted (late) in the derivation, or Legibility, by being present in the core syntax but irrelevant to it. Mu¨ller concludes that “a morphological component of grammar that meets minimalist requirements should be pre-syntactic rather than post-syntactic” (Mu¨ller 2005: 8), and goes on to suggest that inflectional operations are carried out in the Numeration. In these terms, since syntax and LF need access to information regarding, for example, 1st vs. non-1st person, impoverishment rules applying in the Numeration merely mark certain features for deletion in PF, rather than directly carrying out deletion at this point in the derivation. If Mu¨ller is correct regarding the general impossibility of late insertion, then we cannot treat the non-overtness of pro as a matter of PF non-realisation, where this means post-syntactic insertion of a null segmental matrix. So either pro is present in the Numeration as an empty category, or null subjects are true pronouns which are marked for deletion. Of these, the second seems to be the more attractive alternative in the context of current theoretical assumptions: as mentioned in the Introduction, we are no longer able to embed pro in a general theory of empty categories, while here I have developed an analytical framework in which the conditions for deletion are clear. Moreover, Nunes’ (2004) operation of chain reduction, applied to probe–goal dependencies, provides us with a deletion mechanism. Defective goals will delete since they have no features not also present on the probe; in other words, ultimately the very general principle of recoverability of deletion determines when a goal may delete. Given this, there is no more reason to assume that pro is a theoretical primitive than there is to assume traces. Therefore, I follow Mu¨ller’s proposal regarding impoverishment and treat pro as a pronoun marked in the syntax for deletion. Following the general ‘earliness’ guideline (Pesetsky 1989; Chomsky 2001: 15), we might think that it deletes as soon as it can in the derivation, which, at the latest, is the point of transfer to PF (but see below). It does not seem possible to treat pro as an ‘accidentally silent’ element, i.e. to say that it lacks a segmental matrix since there is no systematic relation between these matrices and the syntactic elements they realise. This would amount to saying that the fact that pro is null in a given context is accidental in the same way that he has an initial /h/; but in that case we would expect a quite random incidence of null arguments, something which we do not find. It is clear that null arguments of various kinds are subject to syntactic, discourse and morphological conditioning; they are not just accidents of PF. We see then that pro is a deleted pronoun. The operation that actually deletes a weak pronoun, giving rise to the silent element pro, is related to that spelling out clitics on their hosts, as we saw in Section 2.1, and, plausibly, to the general operation of copy deletion. Whether this is a PF operation or

A deletion analysis of null subjects

85

internal to core syntax is a further question. One might think that, since copies need to be available at LF, the pro-deletion does not feed LF, and so takes place at spell-out at the earliest. However, a more radical and interesting position suggests itself, which I will now briefly explore. As we have seen, since pro is a defective goal, its features are properly included in those of the T which probes it. Therefore no LF-relevant information is lost by pro-deletion; T’s j- and D-features can recover the relevant properties of the referential, definite pronominal subject. This creates the intriguing possibility that a version of Holmberg’s Hypothesis A, with SpecTP absent (the ‘I-subject’ approach to null subjects), is correct as far as the LF representation of null subjects is concerned. It is interesting to note that many of the arguments for this kind of approach (e.g. in Barbosa 1995; 2009; and Alexiadou & Anagnostopoulou 1998) are interpretative, invoking the topicalised or focalised interpretation of overt subjects as evidence that they are not in the canonical subject position, scope interactions between overt subjects and quantifiers elsewhere in the clause, asymmetries between referential and non-referential quantified subjects, and restrictions on the interpretation of pronouns as bound variables of the kind first observed by Montalbetti (1984) (see in particular Barbosa 2009). All of this could be interpreted, as the authors suggest, as evidence that (expletive/ resumptive) pro is absent, i.e. that SpecTP is absent, at the level where these phenomena are represented: that is, LF. In particular, this idea converges with Alexiadou & Anagnostopoulou’s conclusions regarding expletive pro: this item arguably must delete prior to LF since it is uninterpretable. On the other hand, most of the objections to Hypothesis A, including Holmberg’s (2005) argument from Finnish given in Section 1 and the arguments from Burzio, Rizzi and Cardinaletti in Section 2.2, are distributional: they show that pro is present at some point in the derivation, but are neutral as to pro’s interface status. These arguments are obviously compatible with pro’s absence at PF, and could be compatible with its absence at LF. So it is conceivable that null-subject deletion takes place in core syntax: the deleted pronoun satisfies the EPP and the absence of a subject in SpecTP has the interpretative effects Barbosa and others discuss at LF. One interesting consequence of this line of thought is that consistent nullsubject languages emerge as the inverse of radical pro-drop languages as analysed by Saito (2007). Putting together Saito’s proposals with the particular account of the ‘rich agreement’ that facilitates consistent null subjects based on Mu¨ller’s notion of impoverishment, we arrive at the following generalisations: (34)

a. Radical pro-drop is possible iff j-agreement is not obligatory. b. Consistent null subjects are possible iff there is no impoverishment of T’s j-features.

Ian Roberts

86

Where (34a) holds, covert merge of arguments, including ‘discourse pro’ is possible; where (34b) holds, deletion of subject pronouns is possible prior to LF (and (34b) can clearly be generalised to non-subject arguments in the obvious way). We see that the two systems are derivational mirror images of one another, and that this is the direct consequence of the different status of j-features on probes (fully optional vs. obligatorily present and unimpoverished), which in turn is directly reflected in the agreement morphology (totally absent vs. ‘richly’ realised). This further suggests a rethinking of the typological generalisations surrounding null arguments: perhaps the fundamental dimension of parametric variation is ‘radical’ vs. ‘consistent’ null-subject (or null-argument) languages, with partial and non-null-subject languages being subcases of the ‘consistent’ type featuring varying degrees of impoverishment of the goal in terms of the approach to Mu¨ller’s generalisation described above. The basic form of the parameter would then be as in (35): (35)

a. Are j-features optional or obligatory on probes? b. If obligatory, are j-features impoverished on all probes? c. If j-features can be impoverished, are those of {T, v, . . .} impoverished?27

The ‘optional’ value in (35a) gives radical pro-drop. The obligatory value gives null arguments in general (i.e. null subjects and null objects, licensed by ‘rich’ subject and object agreement, as in Basque). The positive value of (35b) gives a non-null-subject language like English. (35c) is intended to simply indicate the ways in which the null-subject parameter starts to ‘break up’ into microparameters as individual probes are evaluated in relation to it. There is an obvious implicational chain linking (35a–c), and the subcases of (35c) no doubt also break up into further implicational relations. In these terms, we can immediately observe a connection between the Null Subject Parameter as formulated in (35) and the parameter determining head–complement order, discussed in the Introduction. As we suggested there, this may lead to an articulated theory of parameters which is partially driven by third-factor considerations. 3

Conclusion

We began this chapter by observing that the typology of empty categories, including empty pronouns, that was developed in GB theory cannot be

27

In Roberts (Chapter 8), we will see that C’s features must be taken into account in order to understand certain aspects of inversion constructions in French, Franco-Provenc¸al and some Northern Italian dialects.

A deletion analysis of null subjects

87

maintained in the context of the copy theory of movement. Following Holmberg (2005), we have seen that there is evidence for a pronominal empty category, pro, in consistent and partial null-subject languages. Against this background, I developed an account of pro in terms of the general notion of defective goal put forward in the analysis of clitics and head-movement in Roberts (forthcoming b). In these terms, pro is a weak pronoun in the sense of Cardinaletti & Starke (1999), a DP which is required to appear in certain designated positions (SpecTP in the case of subjects), and which undergoes deletion where T has a D-feature, thanks to the general properties of defective goals. T can only have a D-feature if none of its j-features have undergone (pre-syntactic) impoverishment; this establishes the connection between null subjects and ‘rich’ agreement, since non-impoverished j-features can be realised by distinct vocabulary items while impoverished ones cannot (although a certain amount of accidental homophony and null realisation may exist). If impoverishment is pre-syntactic, as Mu¨ller (2005) suggests, then it follows either that pro is marked for deletion in the syntax, or that the deletion operation which gives rise to null subjects takes place in syntax. The latter view has the interesting consequence, yet to be fully explored, that the ‘I-subject’ approach to null subjects (i.e. Holmberg’s (2005) Hypothesis A in (4), originally put forward by Borer (1986) and developed in different ways by Barbosa (1995), Alexiadou & Anagnostopoulou (1998) (for expletive null subjects), Manzini & Roussou (2000) and Manzini & Savoia (2005)), which takes the view that there is no argumental subject DP in null-subject sentences, may be correct for the semantic interface, although not for (all of) the derivation in Narrow Syntax. We saw that this, in turn, creates an interesting connection with Saito’s (2007) theory of radical pro-drop, and suggests a macroparameter along the lines of (35a). Whether or not these speculations turn out to be sustainable, we have arrived here at a general characterisation of the ‘pure pronominal empty category’ of GB theory whose main elements stem from core aspects of current theory, and in particular from the nature of defective goals in the sense of Roberts (forthcoming b). In its reliance on a deletion process, this analysis harks back to the earliest generative analysis of null subjects, that in Perlmutter (1971). An obvious further question is whether PRO could also be analysed in these terms, recalling perhaps Rosenbaum’s (1967) transformation of Equi-NP Deletion, but that remains for future work.

2

Null Subject Parameters Anders Holmberg

1

Introduction: Rizzi’s (1982) Null Subject Parameters

The Null Subject Parameter (NSP) is often talked about in the singular, even though it is widely recognised that null subjects can be derived in more than one way, and that, therefore, more than one parameter is involved in determining whether subject pronouns can be null or not in a given language. In fact, the earliest formulation of the NSP, that in Rizzi (1982: 142), acknowledged two parameters. (1)

a. INFL can be specified [þpronoun]. b. INFL which is [þpronoun] can be referential.

Parameter (1a) is meant to distinguish between null-subject languages (NSLs) and non-null-subject languages (non-NSLs), while (1b) distinguishes between those NSLs which allow all kinds of null subjects, referential and nonreferential, and those which only allow non-referential ones. Other formulations have been proposed over the years (see Huang (2000) for an overview), the most influential one being Rizzi’s own reformulation in Rizzi (1986a). In the present chapter I will argue that the ‘original NSP’ (1) was not so far off the mark. I will argue that a version of (1b) distinguishes what I shall call consistent NSLs from other languages. The crucial notion is not referentiality, though, but definiteness: there is a parameter, a version of (1b), which distinguishes languages that allow definite null subjects (null ‘he/she’), as in (2), from all other languages. (2)

Verra`. come-FUT-3SG ‘He will come.’

(Italian)

Among the other languages there are those which allow indefinite null subjects (null ‘one’) and expletive null subjects (null ‘it’), for example Brazilian Portuguese and Finnish, and those which do not allow null subjects at all, for example Swedish and English. The former make up the class of partial 88

Null Subject Parameters

89

NSLs (on which see in particular Holmberg and Sheehan (Chapter 3)), the latter the class of non-NSLs. I will show that a version of Rizzi’s (1982) parameter (1a) is involved in distinguishing among languages of these two types. However, this is not sufficient to account for all the variation found. There are other parameters affecting the spell-out of subject pronouns that complicate the picture. This will be demonstrated in part by means of a comparison of Icelandic, a partial NSL, and Mainland Scandinavian, a nonNSL. One of the additional parameters is an independent parameter which triggers obligatory pronunciation of SpecTP in some languages, that is the non-NSLs. It is also not the case that consistent NSLs would allow all kinds of null subjects allowed in partial NSLs. There is a type of null subject which is consistently absent from consistent NSLs in active, finite clauses, namely null generic pronouns. This means that the relation between the null-subject parameters is not quite as proposed in Rizzi (1982). One of the theses argued for in this chapter is that there are two ways to derive null subjects: one is by means of incorporation of a subject pronoun in T. In this case the null subject is a deleted copy in a chain headed by T. Definite null subjects can be derived in this way in consistent NSLs only. The other is by deletion of a pronoun in SpecTP, subject to control from a higher clause. This is the only way that definite null subjects can be derived in partial NSLs. The idea that null subjects are derived by incorporation is familiar from the literature (Fassi Fehri 1993; Platzack 2004). In this chapter the mechanism will be in part different from that found in these references. Applying Roberts’ (forthcoming b) theory of pronominal clitics to null subjects, it will be a direct result of Agree, involving no movement.

2

Null subjects not taken up

There are other varieties of null subjects that I will not deal with in this chapter. One is found in replies to yes/no-questions in those many languages where an affirmative reply to a yes/no-question typically consists of just the finite verb of the question. (3) is an example from Finnish, (4) from Marathi. (3)

a. – Luki-ko Tarja sen kirjan? read-Q Tarja that book ‘Did Tarja read that book?’ b. – Luki. read ‘Yes.’

90

Anders Holmberg

(4)

a. – Tara-ni pustak vaache-li? Tara-ERG book read-Q ‘Did Tara read the book?’ b. – ho, vaache. yes read ‘Yes.’

As discussed in Holmberg (2001; 2007) these expressions are derived by movement of the finite verb (more precisely, a remnant verb-headed projection) to the C-domain with ellipsis of the entire IP containing the subject, object and other sentential constituents (see also Holmberg, Nayudu, & Sheehan 2009). One piece of evidence is that the object must be deleted along with the subject, which would be quite unexpected if they are deleted by separate operations, but expected if the subject and object are deleted along with IP. (5)

a. – *Luki sen. read it b. *– ho, tila vaache. yes it read

The parameters in (1) are thus not implicated at all in this construction. Another, related apparent null-subject construction is exemplified in (6), from Swedish, a notorious non-NSL. (6)

– Vad go¨r Johan? what does Johan ‘What’s Johan doing?’ – Sover. sleeps

As discussed in Holmberg (2003a), the subjectless reply is derived by VP-fronting to the CP-domain, with deletion of IP; the syntactic structure of the elliptical reply in (6) is roughly the structure of ‘Sleep (is what Johan is doing)’. Again, the null subject parameters in (1) are not involved. I will also not deal with overtly subjectless, elliptical root clauses found in certain registers in many languages which do not in general allow pro-drop, including English. One particularly well-studied case is that of diary-drop, where a first person subject pronoun is omitted in a narrative style typical of diaries and personal letters (Haegeman 1990; 2000). Another one is expletive drop (as in Can’t be many players like that). These are, I assume, derived by a mechanism different from the one which derives null subjects in consistent and partial NSLs. A characteristic property of null subjects derived by diarydrop or expletive drop, which sets them off from the classical null subjects in NSLs, is that they have to be strictly sentence-initial (*I think that can’t

Null Subject Parameters

91

be many players like that.), suggesting that they are derived by a highly constrained special deletion rule.1 3

Consistent and partial NSLs: the facts

Impressionistically speaking, null subjects are used more in consistent NSLs than in partial NSLs. There are contexts where null subjects are optional in partial NSLs but obligatory in consistent NSLs, and contexts where they are excluded in partial NSLs but allowed in consistent NSLs. The following is an illustration: consider (7), where John is talking about himself, as indicated by the indexing. (7)

John1 said that he1 wanted to buy a car. a. Johan1 sa att *(han1) ville ko¨pa en bil. John said that he wanted buy a car b. Gianni1 dice che (*lui1) vuole comprare una macchina. Gianni says that he wants buy a car c. Ram1 mhanala ki (tyani1) ghar ghetla Ram said that he house bought ‘Ram said that he bought a house.’

(Swedish) (Italian) (Marathi)

In a non-NSL such as English, Swedish, French, etc., the pronoun has to be overt. In a consistent NSL such as Arabic, Greek, Spanish, Turkish, Italian, etc., the pronoun has to be null, assuming for the sake of argument that there is no contrast or topic-shift involved. In partial NSLs, such as Brazilian Portuguese, Finnish or Marathi, the pronoun can optionally be null. Now imagine a context where another person, call him Bill, is being discussed. One of the interlocutors utters (8) as a contribution to the discussion, where the embedded pronoun refers to Bill, as indicated by the index 2. (8)

1

John1 said that he2 bought a house.

The following example is from a cartoon in The Guardian of 5 November 2006. The situation is a mother, a father, and a son with a plate of potato chips in front of him by the breakfast table. Mother utters (i), referring to her son: (i)

Won’t eat his chips. Says they’re made of potatoes. Covered in mud.

Father replies (ii): (ii)

Told you not to let him go on that farm trip.

(ii) is an example of 1st person drop, which, as mentioned, is common. 3rd person drop is less common, but, as evidenced, occurs as well. The sentence Covered in mud exemplifies a different kind of ellipsis common in English, deleting the subject and the auxiliary of a main clause. Note that the subject they of the embedded clause in (i) cannot be deleted, testifying to the strict root character of this type of ellipsis.

92

Anders Holmberg

In a non-null-subject language such as English the pronoun obviously has to be overt. In a consistent null-subject language the pronoun would still typically be null, assuming no contrast or topic-shift. In our partial null-subject languages the pronoun has to be overt in this case. The following is an illustration of this difference, contrasting Finnish, a partial NSL, and Italian, a consistent NSL. (9)

Finnish: a. Juha1ei ole sanonut mita¨a¨n, mutta Pauli2 sanoo etta¨ *1 haluaa ostaa uuden auton. ‘Juha1 hasn’t said anything, but Pauli2 says that he1 wants to buy a new car.’ Italian: b. Gianni1 non ha detto niente, ma Paolo2 ha detto che 1 vuole comprare una macchina nuova. ‘Gianni1 hasn’t said anything, but Paolo2 says that he1/2 wants to buy a new car.’

The problem in (9a) is that the intended antecedent Juha does not c-command the null subject. The only possible antecedent of the null subject is the c-commanding DP Pauli. In Italian c-command is not a requirement, as long as the antecedent is a topic (as will be discussed in Section 5.1 below). This is true only so long as we consider definite null subjects. For indefinite null subjects the situation is, in a sense, reversed: they are common in partial NSLs, but found under more restricted conditions in non-NSLs. In particular, what we do not find in any consistent NSL is a null 3SG indefinite subject, a null ‘one’, in an active finite clause. I illustrate this here by contrasting European Portuguese (EP), a consistent NSL, with Brazilian Portuguese (BP), a partial NSL. (10)

a. E´ assim que faz o doce. is thus that makes the sweet ‘This is how one makes the dessert.’ b. E´ assim que se faz o doce. is thus that SE makes the sweet ‘This is how one makes the dessert.’

(BP)

(EP)

In BP the subject corresponding to the English generic pronoun ‘one’ is null. In EP the overt pronoun se must be present. It is important to make a distinction here between generic inclusive and generic exclusive readings. Characteristic of the generic inclusive reading is that it denotes people in general, including the speaker and the addressee. This is the reading of the English impersonal pronoun one (see Moltmann 2006) and typically also of generic you, the 2SG pronoun used as a generic pronoun, common in many languages. Characteristic of the generic exclusive reading is

Null Subject Parameters

93

that it denotes people in general in some domain which does not include the speaker or addressee. This reading is typically expressed by generic they in English (as in They speak many different languages in India). Consistent NSLs can express generic exclusive reading with a null 3PL subject (null ‘they’), but seem to always resort to other strategies than a null 3SG pronoun to express inclusive generic reading for the subject of a finite clause, typically an overt pronoun, a null 2nd person pronoun (null ‘you’), or some form of passive with a null subject (see Holmberg (Chapter 5)).2 The following is another example of the null generic pronoun in Finnish and two Indo-Aryan languages, Marathi and Assamese (thanks to Modhumita Bora; CLS ¼ classifier); see Holmberg (Chapter 5) for detailed discussion of the generic subject pronoun. (11)

a. Ta¨ssa¨ tuolissa istuu mukavasti. this-IN chair-IN sits comfortably ‘One can sit comfortably in this chair.’ b. Hya khurchi-war aaramani bushushakto. this chair -on comfort-with sit-PRS.3SG ‘One can sit comfortably in this chair.’ c. Ei-khon soki-t aramkori boh-ibo par-i this-CLS chair-LOC comfortably sit-MOD can-3SG ‘One can sit comfortably on this chair.’

(Finnish)

(Marathi)

(Assamese)

I will continue to refer to this pronoun as ‘indefinite’ even though its reading is generic, not a prototypical indefinite reading. Null existential indefinites do not exist in active clauses in any language that I have encountered, and there is some reason to believe that they may not exist at all.3 That is to say, in no language can a sentence such as (2) be interpreted as ‘Someone will come’. This will, in fact, fall out of the theory developed here, as will be discussed in Section 5.2. Of the partial null-subject languages some accept null exclusive as well as inclusive generic pronouns (BP), other just inclusive (Finnish).

2

3

In a new work Abdelkader Fassi Fehri (forthcoming) argues for a qualification (or part-rebuttal) of these generalisations. He points out that consistent NSLs have an indefinite–generic null pronoun, too, but typically in constructions with passive or impersonal voice (of which the Romance se-construction is a special case). So the crucial parameter involves voice, in a way which is not made explicit in the present theory. This is the case in active clauses; in passives such as (i) or impersonal passives such as (ii) the null agent (insofar as it is syntactically represented) is an existential indefinite ‘somebody’. (i) (ii)

My car has been vandalised. Det har stekts bacon ha¨r. EXP has fried-PASS bacon here ‘Someone has fried bacon here.’

(Swedish)

94

Anders Holmberg

If this is correct we have two types of NSLs characterised by the following properties: (12)

Consistent NSLs: Null definite subject pronouns (null ‘he/she’); No null indefinite pronoun (null ‘one’). Partial NSLs: Null definite pronouns only if locally c-commanded by an antecedent; Null indefinite subject pronoun.

This is assuming that the crucial factor in (9a) and (7c) is c-command by the antecedent (see Holmberg & Sheehan (Chapter 3) on the conditions on the antecedence relation). In addition, definite subject pronouns are typically not obligatorily null in partial null-subject languages, while they are, in some contexts, in consistent null-subject languages.4 Both types have null non-thematic pronouns in weather expressions, extraposition sentences and existential/presentational sentences.5 (12) is a first attempt to characterise the two types of system. In fact, things are not quite as clear-cut as indicated here. I will come back to this point after discussing the formal analysis of the differences between the two types of NSLs. 4

The topic antecedent of null subjects

In Holmberg (2005) I proposed that the property which consistent NSLs have, that partial NSLs do not have, is a D(efinite)-feature as part of the f-feature make-up of finite T (see also Roberts (Chapter 1)). This is only a slight reformulation of Rizzi’s (1982) parameter (1b). Two other assumptions were also crucial: (a) Pronouns are either DPs, with the structure [DP D [fP f [NP N]]], or fPs; (b) Null pronouns are fPs. These assumptions, particularly assumption (b), are in line with Cardinaletti & Starke (1999), and also reminiscent of, although actually quite different from, De´chaine & Wiltschko (2002); see also Roberts (Chapter 1). Now when T probes

4 5

One exception found in BP is embedded subject pronouns bound by a robustly non-referential quantifier such as ‘nobody’, which is obligatorily null. It is not the case that overt non-thematic/expletive pronouns are only found in non-NSLs. Even notorious NSLs have expletive pronouns in some constructions (cf. Italian C’e´ del vino sulla tavola ‘There is wine on the table’; see Burzio 1986; Kayne 2008a). Finnish has both an it-type and a there-type expletive. The former is optionally overt, the latter is obligatorily overt in certain constructions; see Holmberg & Nikanne 2002; see also Freeze 1992.

Null Subject Parameters

95

a fP subject, and has its unvalued f-features valued by the subject, the resulting union of the f-features of T and the subject yields a definite pronoun.6 (13) T [uφ, D]

vP φP [3SG]

v'

Now consider a language without D in T. The probe–goal relation between T and a null fP subject does not supply a definiteness value. The result is a D-less, thus indefinite, subject pronoun. If the f-features are 3SG, the interpretation is that of an inclusive generic pronoun, corresponding to one in English. This explains the absence of an indefinite null pronoun in finite, active clauses in consistent NSLs, and why partial NSLs have null subjects with definite interpretation only when they are controlled by a higher definite DP. A problem with the assumption that finite clauses in consistent NSLs have a valued definiteness feature in finite T is that at least some of these languages have indefinite overt subjects which enter an Agree relation with T. The prediction is that this should be impossible, due to feature conflict. An unvalued D-feature in T would allow Agree with indefinite as well as definite subjects, though; the D-feature would inherit the value of the subject it probes. But this would fail to explain the absence of a null indefinite subject pronoun. There is also a fact which is not taken into account in Holmberg (2005), or in the scenario outlined in the previous section: null subjects, particularly 3rd person null subjects, are dependent on an antecedent in consistent NSLs, too. This has been demonstrated by Samek-Lodovici (1996) for a number of NSLs, and, more recently, by Frascarelli (2007) for Italian. As shown by (9), the conditions on the pronoun–antecedent relation are less strict in Italian, representing consistent NSLs, than in Finnish, representing partial NSLs. Nevertheless, there are conditions, the most important one being that the antecedent should be a topic, in fact, according to Frascarelli, a particular type of topic, viz., an Aboutness-shift topic; see Frascarelli & Hinterho¨lzl (2007). Consider the following example, from Samek-Lodovici (1996):

6

Holmberg (2005) proposed that the subject fP has an unvalued D-feature, valued by the feature of T. This may be an unnecessary formal complication. Instead, as will be detailed in the text below, the probe and the goal in a case like this form a chain defined by the union of the valued features of the probe and the goal, which in this case yields a definite pronoun.

96

Anders Holmberg

(14)

a. Questa mattina, la mostra e` stata visitata di Gianni. Pı`u tardi */egli/lui ha visitato l’universita`. this morning the exhibition was visited by Gianni. Later he/he visited the university ‘This morning the exhibition was visited by Gianni. Later he visited the university.’ b. Questa mattina, Gianni ha visitato la mostra. Pı`u tardi  ha visitato l’universita`. this morning Gianni visited the exhibition. Later visited the university ‘This morning Gianni visited the exhibition. Later he visited the university.’

In (14a) a null pronoun is impossible in spite of the (seemingly) unambiguous antecedent in the preceding sentence. Instead an overt pronoun (either the more literary egli or the more colloquial lui) must be used, because the pronoun introduces an Aboutness-shift topic, in Frascarelli’s (2007) terms, which it does because the topic of the preceding sentence is ‘the exhibition’. This insight can provide a solution to the problem faced by the D-in-T hypothesis, as follows. First, following Frascarelli (2007), I assume: (a) that an Aboutness-shift topic (henceforth A-topic) is always syntactically represented in a designated A-topic position in the articulated C-domain, either overtly (for instance in the Italian clitic left-dislocation construction) or covertly; and (b) that the antecedent of a null subject is a null A-topic basegenerated in the C-domain of the clause immediately containing the null subject. This null A-topic is a copy of an A-topic, which may or may not be null, in the locally preceding discourse. This is represented schematically in (15), where the antecedent sentence also has a null topic (derived by covert movement). (15)

[CP [questa mattina Gianni1 ha visitato la mostra]]. [CP <2> [pı`u tardi ha fP2 visitato la mostra]] 1¼2

The referential index of the null subject then ultimately comes from, or is identified with, the index of a spelled out DP in the preceding discourse, via a chain of A-topics. Frascarelli (2007) denies the existence of a topic-chain: “[. . .] this account does not postulate a Topic chain across sentences, but a copying of referential features in different C-domains (through Merge of silent copies), till a new Aboutness-shift Topic is proposed” (fn. 28). This seems like a rhetorical manoeuvre: if the null A-topics are copies, then there is a chain-like relation between them, and this relation is established by an operation across sentences in a discourse. In (15) I have represented this, quite simply, as an index-identifying operation, obviously falling outside sentence grammar but still crucially involved in the interpretation of the sentence,

Null Subject Parameters

97

including the null subject, and no doubt subject to locality and other conditions which I will ignore here. What is important for our purposes is that, as a discourse-grammatical operation, it is not subject to Narrow-Syntactic structural conditions such as c-command. This explains, in part, the facts exemplified by (7) and (9) above. The reason why consistent NSLs cannot have a null ‘one’ is that this pronoun is a bare fP which cannot value [uD], which consequently remains unvalued, causing a crash of the derivation.

5

Incorporation: the role of T

5.1

Consistent null-subject languages

The index-sharing relation between the null A-topic and the null subject in the second line in (15) crucially involves T in the following way: the A-topic values the uD-feature of T, where the valuation consists of uD copying the referential index of the A-topic. I restrict the discussion to 3rd person pronouns, coming back to 1st and 2nd person pronouns below. A defective 3rd person pronoun (labelled ‘fP’, even though it is not strictly speaking a phrase, in terms of Chomsky’s (1995b) Bare Phrase Structure) on its own cannot be definite, as it has no D which could have a referential index. Restricting the discussion to subjects, in a language without uD in T such a pronoun can only be interpreted as impersonal, that is either as generic (inclusive or exclusive) or non-thematic. In a language with uD in T the defective 3rd person pronoun can be, and must be, interpreted as definite if it is merged in the domain of a T whose uD-feature is valued by an A-topic, and from there is incorporated in T, in the manner outlined in the next paragraph. Adopting the theory in Roberts (forthcoming b) (with some modifications, discussed below) I take incorporation of a fP in T to be a direct effect of Agree, in the sense of Chomsky (2001). This works as follows: finite T has a set of unvalued f-features, and therefore probes for a category with matching valued features (step 1 in (16)). The defective subject pronoun has the required valued f-features, and therefore values T’s uf-features, which is to say that the f-feature values of the subject pronoun are copied by T. At the same time T values the subject’s unvalued case feature (step 2 in (16)). I assume that NOM(inative) is encoded as a valued feature of T (not an entirely uncontroversial assumption; see Holmberg (Chapter 5)). (16)

1 [T, D2, uf, NOM] [vP [3SG, uCase] v . . .] ! 2 [T, D2, 3SG, NOM] [vP [3SG, NOM] v . . .] ! 3 [T, D2, 3SG, NOM] [vP [3SG, NOM] v . . .]

98

Anders Holmberg

As a result, T shares all of fP’s feature values. More precisely, T’s feature values are a superset of fP’s values, since T also has uD (valued ‘2’ by the A-topic) and a tense feature. Effectively, the result is the same as if f had moved, by headmovement, incorporating into T, but without the well-known formal problem which classical head-movement has, namely lack of c-command between the links of the head-chain; see e.g. Matushansky (2006). Roberts (forthcoming b) proposes that the probe and the goal in this situation form a chain. As such it is subject to chain reduction. The principal rules of chain reduction are (17a,b):7 (17)

a. Pronounce the highest chain copy. b. Pronounce only one chain copy.

Consequently the subject fP is not pronounced (indicated by the strikethrough under step 3 in (16)). As the chain includes the feature [D], by virtue of T’s D-feature, and since [D] is valued by the A-topic in SpecCP, the result is a definite null subject construction, with the referential index of the A-topic. The chain is pronounced only in the form of an affix on the finite verb or auxiliary (following incorporation of Vþv into T; see Roberts (forthcoming b); Biberauer and Roberts (Chapter 7) for details). As for 1st and 2nd person null subjects, Frascarelli (2007) adopts Sigurðsson’s (2004b) hypothesis that every clause has features representing the speaker and the addressee in the C-domain (in a modern version of Ross’ (1970) performative hypothesis). In this way, the speaker and the addressee are always available as local antecedents. I also adopt this hypothesis. If the subject is a lexical DP or a D-pronoun, T’s uD-feature will be valued by the subject’s D as either definite or indefinite, as the case may be. This will preclude the assignment of a referential index to T by a null A-topic in SpecCP. Instead the subject, if specific-indefinite or definite, will typically be an A-topic. A lexical DP or a D-pronoun cannot be incorporated in T as they have features whose values cannot be copied by T as T lacks the requisite unvalued features. Most obviously, a lexical DP has a root, which is not copied by T under Agree. Consequently, even though T and the lexical subject DP in SpecvP share f-feature values through Agree, they do not form a chain, and consequently the lexical subject, being the highest member of a chain, is spelled out and pronounced (unless it undergoes movement to a higher position). It is less clear what features pronouns such as egli and lui in (14)

7

(17a,b) are default rules; there are various situations where a low chain copy is spelled out, and sometimes more than one chain copy is spelled out. (17a,b) are also plausibly derived from more primitive conditions on spell-out and linearisation; see Nunes 2004; Bobaljik 2002; Landau 2004.

Null Subject Parameters

99

have that prevent their incorporation in T, perhaps particularly egli, which is a deficient pronoun in terms of Cardinaletti & Starke’s (1999) typology of pronouns. In our terms even egli has a valued D-feature, as it can introduce a new A-topic, but since T, by hypothesis, has an unvalued D-feature which gets valued by the subject, T and the subject pronoun will share that feature value, too, in addition to sharing f-feature values. Roberts (Chapter 1: footnote 14) proposes that pronouns which do not incorporate have a case feature blocking incorporation. I do not want to adopt this hypothesis here, though, since the null generic subject pronoun in Finnish has been shown to have case; see Holmberg (Chapter 5), Vainikka (1989) and Roberts (Chapter 1: footnote 22). Since the generic pronoun is incorporated in T (see below, Section 5.2), case must be compatible with incorporation. It will be, if case-assignment is copying of a case-value from the assigner (in this case finite T) to the assignee.8 The result is that the only pronouns that remain null are the ones that are linked to a null A-topic in SpecCP. The generalisations that 3rd person null subjects in consistent null-subject languages are always definite, and always refer to a person or object already introduced as a topic, are thereby explained. A fact which complicates the situation is that there is cross-linguistic variation regarding the syntactic and information-structural role of the discourseantecedent (the DP whose index is ultimately picked up by the null subject), as noted by Samek-Lodovici (1996) and demonstrated by Cole (forthcoming). I return to this complication below. If the subject is incorporated, in (18), for example, how is the EPP checked? (18)

Verra`. ‘He will come.’

(Italian)

A possible hypothesis is that T in consistent NSLs has no EPP-feature. Another is that the EPP can be checked by V-movement to T, as advocated by Barbosa (1995) and Alexiadou & Anagnostopoulou (1998). Yet another one is that the EPP can be satisfied by the incorporation itself, as advocated by Platzack (2004); see, however, Sheehan (Chapter 6) for arguments that at least the Romance consistent NSLs have an EPP condition of the ‘traditional type’ which requires SpecTP to be filled. Maintaining that the subject itself is incorporated in T in (18), and SpecTP therefore is not projected, I propose that the EPP in (18) is checked by the null A-topic which also values [uD] in T. The argument is that a SpecTP-less structure is allowed only in the case of definite null subjects, that is when there is (and must be, if Samek-Lodovici (1996) and Frascarelli (2007) are 8

See Roberts (Chapter 1: footnote 22) for an argument that this conclusion is not warranted.

100

Anders Holmberg

right) an antecedent. The other cases, when SpecTP must be filled to check the EPP, include: (a) when the subject is a null indefinite pronoun; (b) when the subject is a lexical DP or a spelled-out, non-incorporated pronoun; and (c) existential and other thetic sentences, which lack an A-topic. Case (a) will be discussed in the next section, which is about incorporation in partial NSLs. As for case (b), the prediction is that a spelled-out subject, whether lexical or pronominal in the unmarked case, that is when it is not focused, or topicalised, or wh-moved, is preverbal, in a consistent NSL of the SVO type (such as the Romance consistent NSLs). Spelled-out subjects are indeed typically preverbal in these languages. However, there is a longstanding debate whether they are in SpecTP, an A-position, or in an A-bar position in the CP-domain (see Alexiadou & Anagnostopoulou 1998; Barbosa 1995; 2009; Costa 2004). I will not review this debate here, but see Sheehan (Chapter 6). Suffice it to point out that the theory sketched here predicts that if a subject pronoun in such a language is for some reason spelled out even though it is not focused, that pronoun will be preverbal. At least for the Romance NSLs this prediction is correct. Consider again (14), repeated here in a slightly simplified form as (19): (19)

Questa mattina, la mostra e` stata visitata di Gianni. Pı`u tardi lui ha visitato l’universita`. this morning the exhibition was visited by Gianni. Later he visited the university ‘This morning the exhibition was visited by Gianni. Later he visited the university.’

The bolded pronoun is not overt because it is focused or emphatic, but because its antecedent is not an A-topic (in Frascarelli’s (2007) terms).9 As for thetic sentences, which in NSLs may be V-initial with a post-verbal subject, and thus seemingly lack SpecTP, Sheehan (Chapter 6) argues, for the case of Romance languages, that they either have a null locative or a null expletive in SpecTP, checking the EPP. (20)

9

a. Ha telefonato Gianni. has telephoned Gianni ‘Gianni phoned.’ b. Chegou algue´m ao cole´gio. arrived someone to-the school ‘Someone arrived at school.’

See Barbosa (2009) for some cases with an unfocused post-verbal subject pronoun.

(Italian)

(EP)

Null Subject Parameters

101

Pinto (1994; 1997) argued that (14a) has a null locative ‘here’ in SpecTP. It is therefore a case of covert locative inversion, triggered by the EPP. This, Pinto argued, is the case for a large class of sentences with so-called ‘free inversion’ in Italian. As Sheehan (2006; Chapter 6) shows, this is the case in Spanish and Portuguese as well. All other V-initial, seemingly SpecTP-less sentences in the Romance consistent NSLs have an indefinite subject; (14b) is an example. Sheehan argues that they have a null expletive in SpecTP checking the EPP. This analysis is consistent with the theory articulated here. I will come back to the issue of null expletives and locatives in Section 7 below. I conclude that the hypothesis that the EPP in finite sentences with a definite null subject is checked by the null A-topic is consistent with the facts considered so far. More support is offered in the next section. 5.2

Partial null-subject languages

So far we have accounted for consistent null-subject languages. In other languages finite T does not have a uD-feature. In a subset of these languages, viz. the partial null-subject languages, the subject can still be null essentially by the same derivation as in the consistent null-subject languages: T probes for f-feature values. The subject’s f-feature values are copied by T, and the subject has its Case-feature valued in return. In the case where the subject is a bare fP, T will copy all the feature values of the subject. As a result T and fP form a chain, and the subject remains null, by chain reduction. However, in the absence of uD in T, valued by an A-topic, the interpretation of the subject chain cannot be that of a definite pronoun. But it can be indefinite, and that is what we find in partial NSLs. It would be interesting if absence of uD(efinite) on the probe T correlated with absence of definiteness marking on the goal, the nominal argument. The languages which are investigated as exemplars of partial null-subject languages in Holmberg and Sheehan (Chapter 3) are Finnish, Marathi and BP. As it happens, Finnish does not have articles, Marathi, like most or all other Indo-Aryan languages, does not have a definite article, and BP, interestingly, makes more use of bare NP as an argument than EP.10 However, BP still has

10

BP has bare singular NPs functioning as generic subjects, which EP does not. BP also has bare singular NPs with an existential reading, which EP does not. (i)

(ii)

Gato toma leite. cat-SG drink-3SG milk ‘Cats drink milk.’ Crianc¸a gosta de tomate. child like-3SG of tomato-SG ‘Children enjoy having tomatoes.’

102

Anders Holmberg

definite and indefinite articles which are widely used. Furthermore, Icelandic, another partial null-subject language, has a definite article (usually suffixed to the noun), and Hebrew, which is probably another partial null-subject language, also has a definite article. If there is a correlation between definiteness marking on the noun phrase and partial null-subjecthood, it is not a straightforward one, and more research is needed to establish the nature of the correlation, if indeed there is one. When the subject is a DP (lexical or pronominal), it cannot be incorporated, and when the subject is incorporated in the partial null-subject languages it can only be interpreted as indefinite. Yet BP, Finnish and Marathi have null subjects that are interpreted as definite, as shown in Section 1, so there must be an alternative derivation of null subjects. As indicated in Section 1, these null subjects must have a locally c-commanding antecedent; they must be controlled (in some sense; see Holmberg and Sheehan (Chapter 3) for discussion of the control relation). We have also said that a subject which is not incorporated is attracted by the EPP to SpecTP. The prediction is, then, that the definite null subject in partial null-subject languages is in SpecTP and checks the EPP, while the generic null subject is in SpecvP and does not check the EPP. Furthermore, since the indefinite, generic null subject does not have an A-topic antecedent, the EPP must be checked by some other category in these sentences. This prediction can be shown to be right. Consider (21a,b): (21)

a. Jari sanoo etta¨ ta¨ssa¨ istuu mukavasti. Jari says that here sits comfortably ‘Jari says that one can sit comfortably here.’ 6¼ ‘Jari says that he sits comfortably here.’ b. Jari sanoo etta¨  istuu mukavasti ta¨ssa¨ Jari says that sits comfortably here ‘Jari says that he sits comfortably here.’ 6¼ ‘Jari says that one can sit comfortably here’.

(Finnish)

Finnish has an EPP condition which is mostly satisfied by the subject, but can be satisfied by certain other categories, too, including circumstantial adverbials (see Holmberg & Nikanne 2002; Holmberg 2005). In (21a) the 3SG subject has not undergone movement to SpecTP; instead the place adverbial has. The subject is null, by hypothesis due to incorporation by Agree with subsequent chain reduction. In the absence of uD in T, it cannot have definite interpretation, though, but only generic interpretation. In (21b) the subject has moved to SpecTP, satisfying the EPP. In this position the subject pronoun cannot be null by virtue of incorporation in T, but only by virtue of having a local antecedent, hence the interpretation. A similar situation is found in BP, another partial NSL: (22)

a. Joa˜o me contou que na praia vende cachorro quente Joa˜o me told that at.the beach sell-3SG dog hot

Null Subject Parameters

103

‘Joa˜o told me that hot dogs are sold at the beach.’ 6¼ ‘Joa˜o told me that he sells hot dogs at the beach.’ b. Joa˜o me contou que  vende cachorro quente na praia. Joa˜o me told that sells dog hot at.the beach ‘Joa˜o told me that he sells hot dog at the beach.’ 6¼‘Joa˜o told me that hot dogs are sold at the beach.’ BP, based on Rodrigues (2004: 142)

In Marathi the prediction cannot be so easily tested since due to its SOV syntax all arguments and adjuncts precede the finite verb anyway. In the preceding section I proposed that subject pronouns in consistent NSLs are incorporated in T and interpreted as definite by virtue of a D-feature valued by a null A-topic (in the case of 3rd person) or by a speaker/addressee feature (in the case of 1st and 2nd person, following Sigurðsson (2004b)), and that the EPP is then also checked by the A-topic or the speaker/addressee feature. This is consistent with the observations in this section: when a subject pronoun is incorporated in T in a language which has no uD-feature in T which can be valued by an A-topic, then, and only then, the EPP must be satisfied by other means, namely movement of a category to SpecTP. It appears, then, that while definite null subjects in consistent NSLs are incorporated fPs which are interpreted as definite by virtue of a valued D-feature in T, definite null subjects in BP, Finnish, and, I assume, Marathi are DPs which have been second-merged with SpecTP. See Barbosa (2009), who reaches essentially the same conclusion regarding BP (and other partial null-subject languages), and a similar, though not identical conclusion regarding consistent null-subject languages. If all this is right, the relation between the null subject and its antecedent in consistent null-subject languages is indirect: in a clause CL which has a 3rd person null subject NU, the ‘ultimate antecedent’ of NU is a DP which is the A-topic of a clause preceding NU. The intermediate antecedent is a null copy of this A-topic in the C-domain of CL. This A-topic values the uD-feature of T, where valuation means that uD copies the referential index of the A-topic. The immediate antecedent of NU is then the valued uD-feature of T which probes NU. There is thus no c-command relation, or indeed any structural relation, between NU and the DP in a higher sentence or a separate sentence, whose referential index NU will ultimately share. In partial null-subject languages, by hypothesis, inheriting a referential index by this indirect route is impossible due to absence of a uD-feature in T which, once valued, could provide the null subject chain with an index.11 11

Why can a (null) A-topic in SpecCP not control a null subject directly, without involvement of T? Presumably T with its f-features is a ‘defective intervener’ (Chomsky 2000) between the A-topic and the null subject when the subject is in SpecvP. But why can a null A-topic not control a fronted null subject directly? If it did, we would expect to have potentially the same

104

Anders Holmberg

If NU is 1st or 2nd person, the ultimate and also intermediate antecedent is the speaker or addressee ‘speech feature’, which, following Sigurðsson (2004b) are properties of the C-domain in every finite clause. The speech feature values uD of T, T probes the subject fP and the subject fP is incorporated in T.12 What features does the fronted null pronoun have in partial null-subject languages? It seems reasonable to assume that a nominal argument which must enter a structurally determined relation with another argument to be interpretable must be somehow deficient; see Shlonsky (2009), discussing control into finite clauses in Hebrew, for a similar conclusion. I suggest, therefore, that it lacks a D-feature value, and therefore must enter a structurally defined control relation with a valued DP antecedent. That is to say, either the D-pronoun comes with an inherent D-feature value (a referential index), in which case it will be spelled out in SpecTP, being the head of an A-chain, or it has an unvalued D-feature, and must enter a control relation with a valued DP. In that case it remains null by virtue of an extended version of chain reduction. It is an ‘extended version of chain reduction’ since the control relation does not qualify as a chain in the strict sense; see Holmberg & Sheehan (Chapter 3).13 A third possibility, irrelevant here, is that it undergoes A-bar movement to some higher position. 5.3

Summary

In Italian and other NSLs the subject pronoun, if it is a fP, a deficient, D-less pronoun, is incorporated in T as a result of Agree. T is thus the highest member of the subject chain, and as such is pronounced, albeit only as an

12 13

indirect relation between the null subject and the ‘ultimate antecedent’ in partial NSLs as in consistent NSLs. Modesto (2008) in a comparative study of BP and Finnish argues that the controlled null subject in finite clauses is itself in topic position, based on the notion that BP and Finnish are ‘topic-prominent’ languages which always move an argument, usually but not necessarily the subject, to a sentence-initial topic position (see Holmberg & Nikanne (2002) and Holmberg (Chapter 5: 211–13) on Finnish). If so, then this might exclude a separate null A-topic a` la Frascarelli (2007). Consistent NSLs would not have this option because they are not topic prominent. This is an interesting idea. It would, however, add another parameter to an already rather complex theory. It would need to be established whether other partial NSLs which have controlled null subjects are topic prominent in the way BP and Finnish are, and whether they are different from consistent NSLs in this respect. It should be noted that 1st and 2nd person null pronouns are controlled directly in SpecTP in Finnish, by speaker and addressee speech features in the Cþdomain (following Sigurðsson 2004b). See Vainikka & Levy (1999) for a different formal account of how 1st and 2nd person null subjects are recovered (in Finnish), differently from 3rd person null subjects. If control is movement, as argued by Hornstein (1999; 2000) and Boeckx & Hornstein (2003; 2004), then the controlled pronoun is null because of chain reduction in the strict sense. As shown by Holmberg & Sheehan (Chapter 3: Section 4.2) the control relation into finite clauses in the partial NSLs they investigate is not derived by movement, though. See also Landau (2003), Bobaljik & Landau (2007) and Sigurðsson (in press).

Null Subject Parameters

105

affix on the finite verb, while the copy in SpecvP is not pronounced. The chain has a uD-feature provided by T. This feature receives a value, that is a referential index from the null, indexed A-topic in the C-domain. This also has the effect of checking the EPP. The result is a definite null-subject construction. (23)

Ha comprato una macchina nuova. has bought a car new [CP < DP1> [TP haþT[D1,

(Italian)

3SG, EPP] [vP


NOM]>

comprato . . . ]]]

If there is no A-topic in the C-domain, then a subject pronoun has to be a DP with its own index, which will not be incorporated, and therefore will (normally) be attracted by the EPP to SpecTP, where it will be pronounced. In Finnish, BP, and other partial NSLs, the subject cannot be incorporated in T and be interpreted as definite, since T lacks the requisite uD-feature. Instead, a definite subject pronoun must have its own valued D-feature, thereby its own index. Probed by T, it will be forced by the EPP of T to merge again with TP. The second-merged copy will be pronounced, being the highest copy of the chain. (24)

Ha¨n on ostanut uuden auton. he has bought new car

(Finnish)

[TP ha¨n[D1, 3SG, NOM] [T0 onþT[3SG, > ostanut . . . ]]]

EPP] [vP


NOM]

A fP subject can be incorporated, though, as a result of Agree, but in the absence of uD in T, it can only have a generic interpretation (in Finnish only inclusive generic). (25) is the structure of (21a). (25)

[CP etta¨ [TP ta¨ssa¨ [T0 istuuþT3SG, ]]]]

EPP [VP

fP3SG mukavasti

A third alternative is that a pronoun has a D, but a deficient, unvalued D (labelled uDP). This will still prevent incorporation, so the pronoun, if it is a subject, will move to SpecTP, where it can be interpreted if it is controlled by an argument in a higher clause. (26) is the structure of the embedded clause in (21b) prior to control, which will assign the value/index 1 to the unvalued, null subject pronoun. (26)

Jari1 . . . [CP etta¨ [TP uDP [T0 istuuþT3SG, mukavasti ta¨ssa¨]]]]

EPP [VP



In English, Swedish, etc. incorporation is never an option, nor is a controlled uDP, for reasons discussed in the next two sections. In Section 1 the question was raised why there is no null existential indefinite pronoun, a null ‘someone’. We might not expect it in consistent NSLs, which

106

Anders Holmberg

only allow definite null subjects, but why not in partial NSLs?14 We now have an answer: as indicated by the morphology of the English indefinite pronoun, existential indefinites are structurally complex, a quantifier plus a pronoun (someþone). I assume that is the case, universally, also in languages where they are not morphologically complex in the transparent manner of English someone. As such they cannot be incorporated. Universal quantifiers are likewise complex, and, indeed, there are no null universal quantifiers either. Generic ‘one’ is semantically closer to a universal than an existential pronoun, but, as discussed by Moltmann (2006), it is not a universal quantifier. An existential or universal indefinite also cannot be a controlled null subject, quite simply because an indefinite cannot be controlled or bound. Only a deficient definite pronoun (a uD-pronoun in present terms) can be controlled; this seems to be the right result. 6

Non-null-subject languages 1: Scandinavian

In order to unravel the formal properties which make a language a non-NSL it will be useful to compare Icelandic, a partial NSL, with its Mainland Scandinavian (MSc) relatives Swedish, Danish and Norwegian, a group of very closely related non-NSLs. Icelandic is a partial NSL as seen from the fact that it has a null ‘one’ (see Sigurðsson & Egerland, 2009). MSc, here represented by Swedish, has an overt ‘one’. (27)

a. Nu´ ma´ __ fara að dansa. now may go to dance ‘One may begin to dance now.’ b. He´r er __ ekki deyjandi a´ mannsæmandi ha´tt. here is __ not dying in decent manner ‘One cannot die here in a decent manner.’

(Icelandic)

(28)

a. Nu fa˚r man bo¨rja dansa. now may one begin dance ‘One may begin to dance now.’ b. Ha¨r kan man inte do¨ pa˚ ett va¨rdigt sa¨tt. here can one not die in a dignified manner ‘One cannot die here in a dignified manner.’

(Swedish)

That is to say, Icelandic lacks a uD-feature in T. The generic null subject has a somewhat restricted distribution though, and there is also an overt counterpart maður ‘one’ (lit. ‘man’); see Sigurðsson & Egerland (2009) for details.

14

Apparently it does not occur in ‘radical pro-drop’ languages such as Chinese or Korean, either (thanks to Winnie Yiu and Kook-Hee Gill).

Null Subject Parameters (29)

Nu´ ma´ maður fara að dansa. now may one go to dance ‘One may begin to dance now.’

107 (Icelandic)

Unlike the other partial NSLs listed earlier, Icelandic does not allow a controlled null subject in finite clauses. I will come back to this complication in the next section. There are a number of other differences to do with the subject between Icelandic and MSc. In particular, Icelandic has: (1) (2) (3) (4) (30)

oblique subjects (30a); stylistic fronting (30b); null expletives (30c); rich subject–verb agreement. a. Me´r voru gefnar peninga. me were given money(PL) b. [Þeir sem ´ı Oslo´ hafa bu´ið] segja að það se´ fı´nn bær. those that in Oslo have-3PL lived say that it is nice town ‘Those that have lived in Oslo say that it’s a nice town.’ c. Nu´ hafa (*það) komið margir stu´dentar. now have EXP come many students ‘Now many students have arrived.’

The position of the null expletive is due to the V2 condition; this will be discussed below. Subject–verb agreement distinguishes two numbers and three persons, with only some 1/3 syncretism (past tense singular) or 2/3 syncretism (present indicative singular of most verbs). MSc does not have oblique subjects, stylistic fronting or null expletives, and has no subject–verb agreement at all. (31)

a. *Mej blev givet pengar. me was given money b. *de som i Oslo har bott those that in Oslo have lived c. Nu har *(det) kommit ma˚nga studenter. now have EXP come many students ‘Now many students have arrived.’

(Swedish)

In Platzack (1987), Holmberg & Platzack (1991) and Holmberg & Platzack (1995) (hencefort collectively referred to as P&H) these differences between Icelandic and MSc are formally expressed in terms of a parameter closely related to Rizzi’s (1982) Null Subject Parameter (1a). The following is a distillation of the different versions of the theory articulated in P&H. (32)

a. Finite C must govern nominative case (assuming a lexicalist theory of case);

108

Anders Holmberg b. In Icelandic the AGR-component of INFL is a pronoun with inherent nominative case, therefore able to satisfy (a); c. In MSc INFL contains no AGR (hence no subject–verb agreement), so a subject with nominative case has to move to specIP to satisfy (a).

(32a) is assumed to be connected with V2, and thus be parameterised (but the Scandinavian languages are all V2). (32b,c) looks like a version of Rizzi’s (1982) parameter (1a) (INFL can be specified [þpronoun]). If P&H are on the right track, there is a place for (1a) in the parametric theory of UG, if not exactly as Rizzi (1982) saw it at the time. See also Biberauer (Chapter 4) for a partly different approach. Translating P&H’s (32) into our present framework, broadly based on Chomsky (2000; 2001; 2008), let us assume that Icelandic has unvalued person and number [uPn, uNr] in T, while MSc has neither. As the theory requires postulating some unvalued DP or fP-probing feature in T, in order for nominative case to be assigned to the subject (see Chomsky 2000; 2001), I suggest that T in MSc has an unvalued, general noun-probing feature [uN]. Neither language has [uD] in T. This means that Icelandic can, but MSc cannot, incorporate a fP, hence the difference between (27) and (28). The generic/arbitrary pronoun man is a spell-out of 3SG fP. In Icelandic this fP is incorporated optionally (hence (29)). Given the absence of [uD] in T, the incorporation can yield only a generic or expletive null subject but not a definite null subject chain. How does this theory account for the P&H effects? The idea that INFL/T is inherently pronominal in such a way that it can receive nominative case assigned by C is not easily accommodated in the present formal theory. It would mean postulating an unvalued case-feature in T, along with other uf-features, which would satisfy C’s need to discharge nominative case – while in MSc, the subject DP or pronoun would need to move to SpecTP to receive nominative from C. Arguably, a more elegant account is that the crucial difference is the following: in MSc the EPP of T is linked with the probing uf-features in such a way that the only category that can satisfy/check the EPP is the goal of T’s probing, i.e. the DP or pronoun which values T’s uf-feature and is assigned nominative. I will refer to this as a f-dependent EPP. It may be viewed as a special case of ‘bundling’ of features, in the sense of Boeckx (forthcoming); see Roberts and Holmberg (Introduction: Section 3.5). In Icelandic, too, EPP of T will often attract the goal of T’s probing features, but this is a coincidence, as it were, determined by the fact that Agree and the EPP are both subject to locality: uf of T will probe the closest nominal category which is not already case-marked (see Chomsky 2000) and the EPP will attract the category closest to T. Plausibly there is also an economy condition involved which favours minimising grammatical relations, and thus favours picking the

Null Subject Parameters

109

same category as goal for [uf,T] and [EPP,T]. However, if the goal of [uf,T] is not available for movement to SpecTP, because it is too distant, or because it is marked for undergoing wh-movement, or because there is no goal (as in an impersonal passive, for example),15 then the EPP will attract another, more local category, in Icelandic. In that case [EPP,T] is f-independent. Ever since the inception of the EPP in Chomsky (1981) it has commonly been taken for granted that the EPP of T specifically attracts the subject (see Svenonius 2002a). Chomsky (2000) proposed that this was an effect of Agree: the EPP attracts the category probed by T’s uf-features. There are good reasons, though, to take this to be too narrow a view of the EPP, a result of considering too narrow a range of languages; see Holmberg (2000), Holmberg & Nikanne (2002) and Sigurðsson (2004a) (see also Sheehan (Chapter 6) on Spanish, and work by Massam and others on VP-Fronting to SpecTP in various verb-initial Macronesian and Meso-American languages, summarised in Biberauer & Roberts (Chapter 7)). This will account for (30, 31): in (30a) the highest DP (the experiencer) is assigned inherent dative case, so the next DP, the direct object, which does not have a lexical case, is the goal of Ts probing uf-features, hence values T’s f-features and is assigned nominative, but the EPP attracts the closer dative DP, regardless. In the Swedish (31a), on the other hand, the accusativemarked experiencer cannot check the EPP, even though it is the argument closest to T. The only category which can check the EPP is the nominal category probed and assigned nominative case by [uf, T]. (33a,b) are two grammatical alternatives, one making use of the expletive, which is a casemarked and f-feature-bearing category in Swedish (see Holmberg 2002), the other moving the nominative-marked direct object to SpecTP. For reasons which are irrelevant here, the latter construction is better in Norwegian than in Swedish. (33)

a. Det blev gett mej pengar. EXP was given me money ‘I was given money.’ b. Pengene/De ble gitt meg. money-DEF/they was given me ‘The money was given to me.’

(Swedish)

(Norwegian)

In (30b), the subject is a null operator marked for movement to SpecCP in a relative clause. Instead of attracting it to SpecTP, the EPP of T attracts the next category down the tree, in this case the verb complement; see Holmberg

15

See Holmberg (2000) on Icelandic stylistic fronting, where the dissociation of [EPP,T] and [uf,T] is particularly striking.

110

Anders Holmberg

(2000).16 In the Swedish counterpart (31b) this is not an option: the EPP can only ever attract the nominative-marked category to SpecTP. As for the Icelandic null expletives, they occur only in construction with T-to-C, that is only in main clauses, when a non-subject fills SpecCP.17 (34)

a. Nu´ hafa komið margir stu´dentar. now have come many students ‘Now many students have arrived.’ b. Það hafa komið margir stu´dentar. EXP have come many students c. Komið hafa margir stu´dentar. come have many students ‘Many students have come.’

In the absence of a filler of SpecCP the expletive must be overt, as in (34b). (34c) is a case of stylistic fronting in a main clause: a category, here the nonfinite verb, is moved to SpecCP as an alternative to merging an expletive (see Holmberg 2000). In embedded clauses, the expletive must be overt.18 (35)

það er greinilegt að ??(það) hafa ekki komið margir stu´dentar ´ı pro´finu. EXP is obvious that EXP have not come many students in the.exam ‘It is obvious that not many students have come to the exam.’

The most straightforward account of this is that the EPP and V2 are, or can be, collapsed when T is in C. Thus, in (34a) the EPP is checked by the adverb nu´, as permitted since the EPP is independent of uf. In (34b) it is checked by the expletive, and in (34c) by the stylistic-fronted verb, an option allowed by the 16

The analysis in Holmberg (2000) is actually slightly more complicated, proposing that the null operator moves via SpecTP, but without checking the EPP. The analysis sketched in the text is more in line with Rizzi (1982), Rizzi & Shlonsky (2005), and also Holmberg & Hro´arsdo´ttir (2004) and Chomsky (2008). An alternative to (31b) is (i), as predicted if the the verb and the verb complement are sisters and thus are equally close to T. (i)

17

There is one important exception: yes/no-questions. (i)

18

[Þeir sem bu´ið hafa ´ı Oslo´] segja að . . . those that lived have in Oslo say that . . . ‘Those that have lived in Oslo say that . . .’

Hafa komið margir stu´dentar? have come many students ‘Have many students come?’

Plausibly SpecCP is filled in this case, too, with a null question operator. This would be the only case where the EPP and V2 are satisfied by a null category. The two question marks for the null expletive version is the assessment of Halldo´r Sigurðsson (p.c.). I assume that the very marginal acceptability of the expletive-less version is a residue of an earlier grammar. There are embedded constructions where a null expletive (or no expletive) is more acceptable than in (35) (thanks to Halldo´r Sigurðsson (p.c.); see also Ro¨gnvaldsson and Thra´insson 1990).

Null Subject Parameters

111

dissociation of EPP and uf of T. That is to say, there is, in fact, no null expletive in Icelandic. In MSc, the overt expletive is required in constructions corresponding to (34a) because only a nominative-marked, nominal category can check the EPP in MSc. Stylistic fronting is not an option, for the same reason. (36)

a. Nu har *(det) kommit ma˚nga studenter. now has EXP come many students ‘Now many students have come.’ b. *Kommit har ma˚nga studenter. come has many students

We now have two parameters distinguishing the partial NSL Icelandic from the non-NSL MSc: (37)

1. T has unvalued person and number features in Icelandic but not in MSc, and 2. [EPP,T] is f-dependent in MSc, but f-independent in Icelandic.

There is a connection between the two parameters: poor f-feature content in T and f-dependent EPP are both incompatible with incorporation of a subject pronoun in T. In the case of (37.1) this follows since T will be specified for fewer f-features than the probed subject pronoun. In the case of (37.2) it follows if a subject pronoun cannot both be incorporated and move to SpecTP. This, in turn, follows since, when a subject pronoun is incorporated in T as a result of T copying all of pronoun’s f-features, T and the subject form a chain headed by T. The EPP cannot attract the non-head (a ‘trace’) of a chain; in fact no syntactic operation can operate directly on the non-head of a chain (Chomsky 1995b: 304).19 The question is whether they are actually independent parameters. If they are, we expect to find languages that have a T with (a) rich agreement and a f-independent EPP, (b) rich agreement and a f-dependent EPP, (c) poor agreement and a f-independent EPP, (d) poor agreement and a f-dependent EPP, where we take ‘rich’ to mean an agreement system which distinguishes 19

Under slightly different assumptions this would be possible, though. Assume that all operations within a phase are simultaneous, the combined result evaluated before the next phase is introduced (along the lines of Chomsky 2004). In that case Agree between T and a pronoun p and remerge of p with TP would both apply ‘blindly’. In the case of a language with a rich enough system of uf-features in T, the result would be a chain [p [pþT [p]]] (where ‘pþT’ means that T has copied all the features of p). In this chain pþT would have to be spelled out, as it contains features in addition to those of p. It is not unreasonable to assume that, in that case, both of the other copies are deleted/are not spelled out, which yields a null subject construction. This is, in effect, what Roberts is arguing in Chapter 1. One reason why I am not taking this line here or in Chapter 5 is that it does not offer any explanation of why the generic subject pronoun cannot move to SpecTP and be deleted. In other words, it offers no straightforward explanation of the facts in (21, 22).

112

Anders Holmberg

minimally number and person. We know that we have languages of type (a) (Icelandic, Finnish) and languages of type (d) (Mainland Scandinavian, English). As for type (b), Italian and also EP may belong to this type, while Spanish would be type (a): see Sheehan (Chapter 6).20 French is another candidate for type (b): the subject–verb agreement paradigm is highly reduced in the singular, but distinguishes three persons in the plural (see Roberts: Chapter 8), so, if richness of uf-feature content was all that mattered, we might expect incorporation of plural subject pronouns in T, thus null subjects in the plural – which is not found. Instead, French is a non-null-subject language (at the TP-level; Roberts (Chapter 8) argues that a certain common variety of French is, in fact, a null-subject language at the CP-level). This would follow if the EPP in T is strictly f-dependent. As for type (c), it has no representatives, as far as I am aware, among European languages; it would be characterised by lacking or having only poor subject–verb agreement but have PPs and perhaps other non-nominative categories fronted to SpecTP. English has locative inversion, but arguably this is too marked a phenomenon to put English in type (c). I conclude, tentatively, that (37.1,2) are distinct parameters. There may still be a relation between them such that if a language has poor f-feature content in T, there is a preference for having a f-dependent EPP. But the reverse (weak) implication does not hold. This discussion raises the more general question of the role that the agreement system plays for null subjects. This is discussed in some detail in Roberts (Chapters 1 and 8). Roberts adopts Mu¨ller’s (2005) theory of impoverishment, according to which an agreement system with system-defining syncretism will not license pro, predicting (in Roberts’ adaptation) that consistent null-subject languages will not have any system-defining syncretism in the subject–verb agreement paradigm. One problem with this hypothesis is that it does not make a useful distinction between partial null-subject languages such as Icelandic, Marathi and BP on the one hand, and non-null-subject languages such as French and German on the other.21 All of them make number and person distinctions in the subject–verb agreement paradigm (Marathi also gender), and all of them have some amount of syncretism, as the following comparison of the present tense of the verb ‘sing’ illustrates (I give the French 20

As discussed by Masullo (1993), in constructions such as (i) Spanish has an indirect object in SpecTP, satisfying the EPP, while T agrees with the source argument; see Sheehan (Chapter 6: Section 3.4). This puts Spanish in type (a). According to Zubizarreta (1998) Spanish differs from Italian in this respect. (i)

(ii) 21

A Juan le gustan los chocolates. to Juan CL like.3PL the chocolates. ‘Juan likes chocolates.’ ??Le gustan a Juan los chocolates.

German is arguably an expletive null-subject language; see Biberauer (Chapter 4).

Null Subject Parameters

113

paradigm in quasi-phonological spelling; see Roberts (Chapter 8) for a detailed discussion of French).22

1SG 2SG 3SG 1PL 2PL 3PL

French

German

Finnish

Icelandic

Marathi

cha˜t cha˜t cha˜t cha˜to˜ cha˜te´ cha˜t

singe singst singt singen singt singen

laulan laulat laulaa laulamme laulatte laulavat

syng syngur syngur syngjum syngið syngja

gato/gate (M/F) gatos gato/gate (M/F) gato gatat gatat

As shown, Finnish has no syncretism at all (except that many colloquial varieties make no number distinction in the 3rd person), yet it is a partial nullsubject language with only restricted occurrences of 3rd person null subjects; see Holmberg, Nayudu, and Sheehan (2009). I take these facts to indicate that richness of uf-feature content in T plays a role for the distribution of null subjects and also non-nominative subjects, but indirectly, by affecting the choice of value for certain parameters. I return to this question in Section 10. We have now explained the variation between the partial NSL Icelandic and the non-NSL MSc, in terms of an updated version of Rizzi’s (1982) parameter (1a), that is (37.1) and a related, yet arguably independent parameter (37.2). Unfortunately this is not quite the whole story: even the updated version of Rizzi’s parameter (1a) cannot explain why MSc or English or Sindhi23 or any other non-NSL cannot have a null subject controlled by a DP in a higher clause, in finite clauses, the way Finnish, BP, Marathi, Hebrew, Russian and possibly many other partial NSLs can. In other words, why is (38a) not as well

22

23

Note also that the French auxiliary verbs avoir ‘have’ and etre ‘be’, which are extremely common as they are crucial for expressing past tense, have a fuller paradigm than regular main verbs. Sindhi is an Indo-Aryan non-NSL, spoken in India and Pakistan. As shown by (i) it does not have a null ‘one’, and as shown by (ii), it does not allow a null subject in a finite clause, even when there is a potential, local, c-commanding controller for it (data thanks to Pinkey Nayanwani). In both respect it differs from Hindi, Bengali, Marathi, Assamese and probably most other Indo-Aryan languages (DAT ¼ dative). ^

(i)

(ii)

hik/* hini kursi te aaraam-sa vahi sokdo aaheN (Sindhi) one this chair in comfort-P sit can be-PRES ‘One can sit comfortably in this chair.’ John-Khe pato aaheN Ki *(hun-Khe) pensat-me pensan mild-əs John-DAT-M know-M be-PRES that he-DAT sixty-five-at pension get-FUT ‘John knows that he will get a pension at sixty-five.’

114

Anders Holmberg

formed as (38b), featuring control into an infinitival clause in English, or (38c), featuring control into a finite clause, in Finnish?24 (38)

a. *John promised [that  would be quiet]. b. John promised [ to be quiet]. c. Juhani on luvannut [etta¨  pysyy hiljaa]. Juhani has promised that stays quiet ‘Juhani has promised that he will stay quiet.’

(Finnish)

We also have the fact, so far unexplained, that Icelandic, a partial NSL with a null generic subject pronoun, differs from most other partial NSLs in not allowing control into finite embedded clauses. In the next section I will propose a different, ‘brute force’ mechanism to account for why certain languages do not allow a null subject, even when there is a local controller. 7

Non-null-subject languages 2

Imagine a language which has a strict V2 rule, so that the finite verb in main clauses has to be preceded by one and only one spelled-out category, and which is a consistent NSL. The expected effect is that a definite subject pronoun can be null in any context (satisfying conditions on a topic antecedent discussed in Section 5) except when it is the initial constituent of a main clause, since if it is null in that context the V2 rule will be violated. This is what we find in Kashmiri (Wali & Koul 1997: 83f.) and twelfth-century Old French (Roberts 1993b). (39)

Kashmiri (based on Wali & Koul 1997: 83) a. bi chus kita:b para:n I am book reading ‘I’m reading a book.’ b. kita:b chus _ para:n book am (I) reading c. *chus kita:b para:n. am book reading

(40)

Old French (based on Roberts 1993b: 124, 136) a. Apre´s conquist _ Orenge la cite´. afterwards conquered (he) Orange the city ‘Afterwards he conquered the city of Orange.’

24

The fact that (38a) looks like a that-trace construction is presumably irrelevant, as the grammatical Finnish counterpart is not derived by movement (see Holmberg & Sheehan (Chapter 3). As confirmation, consider the fact that (i) is as bad as (38a), even though Fenno-Swedish allows that-trace (see Roberts and Holmberg: Introduction: Section 2.2). (i)

*Johan lovade [att _ skulle ha˚lla tyst]. Johan promised that would keep quiet

(Fenno-Swedish)

Null Subject Parameters

115

b. Je sui le sire a cui _ volez parler. I am the knight to whom (you) want speak ‘I am the knight to whom you want to speak.’

This is a case where pronunciation of a pronoun is clearly forced by a PF-condition. I take V2 to be the effect of the following grammatical property: (41)

V2: declarative C has a feature [P] (a ‘PF EPP’) which triggers pronunciation of a category merged with CP (by internal or external merge).

This is straightforward as an account of the distribution of null subjects in these two languages.25 Once we recognise that this is a plausible account, it has an interesting theoretical consequence: it suggests that non-NSLs differ from NSLs simply in having a P-feature in finite T, forcing pronunciation of the category in finite SpecTP. (42)

T has a [P]-feature.

Other parameter settings conspire to force movement of the subject to SpecTP, as discussed above (the impossibility of incorporation and/or the dependence of EPP on uf in T). The net effect is that a subject pronoun in a finite clause will always be pronounced, whether there is a potential controller or not. See Landau (2006a) for a theory making extensive use of P-features as a mechanism regulating the pronunciation of chain copies. In fact, Landau rejects the idea that there is a general principle to the effect that the highest member of a chain is spelled out, in the unmarked case (see (17a) in Section 5.1 above). Instead, he argues, the choice of which copy to pronounce is always the effect of a P-feature, whose distribution is essentially a languageparticular matter. I will not adopt this version of the P-feature hypothesis here, as that would predict that NSLs would never have a pronounced subject in SpecTP. There must be a principle which determines when a DP is pronounced even in the absence of a P-feature. As for Icelandic, it, too, has P in T, and therefore does not allow control of a null subject in finite clauses. It still allows a generic null subject, because this null subject is not in SpecTP, but in SpecvP, a deleted copy in a chain headed by T. Consistent NSLs do not have P in T. Hence they can have null locatives and null expletives (see Section 5.1) and definite null subject sentences without 25

At least some V2 languages have a restricted form of topic drop, for example Swedish: (i)

(Det) vet jag ingenting om. that know I nothing about ‘I know nothing about that.’

I do not know how to accommodate that with (41); see also Section 2.

116

Anders Holmberg

SpecTP. Partial NSLs other than Icelandic also have no P in T, hence allow (controlled) null subjects in SpecTP.26 The parameter will also single out languages which allow only expletive null subjects (as many Creole languages do, according to Nicolis 2005; 2008) as a subtype of languages which allow no incorporation in T (either because of poor uf-feature content in T or because EPP of T is f-dependent): without P in T these languages will still allow non-thematic null subjects in T. In a recent paper, Belletti, Bennati & Sorace (in press) show that nearnative speakers of Italian with English L1 differ from native Italian speakers in their use of null subjects in subtle ways. They produce and understand null subjects appropriately, but overuse overt subject pronouns in certain contexts, namely contexts such as (7b), repeated here: (7b)

Gianni1 dice che (*lui1) vuole comprare una macchina nuova. ‘Gianni says that he wants to buy a new car.’

In effect, null subjects are always optional in near-native speech where they are obligatory in native Italian (this is my interpretation of Belletti, Bennati and Sorace’s findings; they do not themselves express it exactly in these terms). According to them, “resetting of the null-subject parameter has taken place in the speakers’ L2 Italian grammar”, but they have not learned to fully master the discourse conditions under which null subjects are used. Given the theory detailed here, Belletti, Bennati and Sorace’s findings indicate that the near-natives have set P in T but not uD in T correctly, in their Italian grammar. Their Italian T has neither P nor uD. That is to say, the Italian of the near-native speakers is a partial null-subject language. Definite null subjects are allowed, therefore, but they are not derived by means of incorporation with chain reduction. In embedded clauses they are derived by control (in the extended sense), always optional in partial null-subject languages. As for root clauses, I suspect that definite null subjects are derived in these by (a generalised version of) whatever mechanism derives null subjects in root clauses in spoken English (something I have not touched on here); see footnote 1. 8

A comparison with Holmberg (2005)

In Holmberg (2005) I argued that there are fundamentally two competing hypotheses regarding pro/null subjects in a theory which makes a distinction

26

Matters are complicated by the observation that Finnish has an obligatory overt expletive checking the EPP in some constructions (see Holmberg & Nikanne 2002), unexpected if lack of P in T means a language can employ a null expletive in SpecTP. I will leave this complication for future research.

Null Subject Parameters

117

between interpretable/valued and uninterpretable/unvalued features, following Chomsky (1995b: chapter 4; 2001): Hypothesis A: The subject pronoun is incorporated in finite T, meaning that the inflection on the finite verb encodes interpretable nominal f-features, and absorbs case and f-role. In that case there is no pro, or at most an expletive pro to check the EPP. Hypothesis B: The null subject is a fully specified pronoun which differs from overt pronouns only in that it is not spelled out/ pronounced (see Roberts (Chapter 1) for discussion). What the theory does not allow is a pro which is inherently unspecified and receives values from AGR, as in Taraldsen (1979), Rizzi (1986a) and much other work in the principles & parameters framework. In Holmberg (2005) I showed that null subjects in Finnish are of type B. More precisely I argued that 3rd person null subjects are fPs that receive a definite interpretation by means of a binding or control relation with an antecedent in a higher clause, while 1st and 2nd person null subjects are fully specified DPs which are deleted. In the absence of a controller a 3rd person fP could be interpreted as generic. I then assumed that that was the case in consistent NSLs as well: null subjects are fPs. The crucial difference was that consistent NSLs have a D-feature in T which ensures, and forces, a definite interpretation of the fP even without an antecedent. In Holmberg (Chapter 5) and Holmberg & Sheehan (Chapter 3) it is argued that the generic null subject in partial NSLs is incorporated in T, while the controlled null subject moves to SpecTP. This immediately explains why the generic null subject does not check the EPP while the controlled one does. In the present chapter I am exploring the possibility that Holmberg (2005) was wrong in taking definite null subjects in consistent NSLs to be counterparts of controlled null subjects in partial NSLs. Instead they are counterparts of the indefinite, generic null subject in partial NSLs. Both are fPs incorporated in T, the different interpretation a consequence of the presence/absence of [uD] in T. If this is right there is no ‘pro’, of any kind, in SpecTP in null subject constructions in consistent NSLs or in the generic null subject construction in partial NSLs, but there is one in the controlled null subject construction in partial NSLs. The fact that definite subject pronouns in partial NSLs are always optionally null, regardless of context, is one consequence of this hypothesis (see Holmberg, Nayudu & Sheehan (2009), Holmberg & Sheehan (Chapter 3)). They are not the result of compulsory chain reduction. Things are complicated by the observation that consistent NSLs do not always make use of the incorporation option. Consider again (9), repeated here as (43).

118

Anders Holmberg

(43)

Finnish: a. Juha1 ei ole sanonut mita¨a¨n, mutta Pauli2 sanoo etta¨ *1/haluaa ostaa uuden auton. ‘Juha1 hasn’t said anything, but Pauli says that he1 wants to buy a new car.’ Italian: b. Gianni1 non ha detto niente, ma Paolo2 ha detto che 1 vuole comprare una macchina nuova. ‘Gianni1 hasn’t said anything, but Paolo says that he1/2 wants to buy a new car.’

Here Finnish represents the partial NSL pattern: a null subject must be directly c-commanded by its antecedent, while Italian represents the consistent NSL pattern relying on a null topic strategy (following Frascarelli 2007), which makes direct c-command by an overt antecedent unnecessary. Now consider European Portuguese, uncontroversially a consistent NSL (see (10b)). (44)

O Jose´1 na˜o disse nada, mas o Joa˜o2 disse que *1/2/ele1/2 na˜o quer comprar um carro novo. the Jose not said nothing, but the Joao said that he not wants buy a car new.

Surprisingly, here Portuguese behaves like Finnish, not Italian. Now consider (45): (45)

O carro1 ja foi arranjado, mas a Maria disse que 1//*ele1ainda na˜o anda muito bem. the car already was fixed, but the Maria said that he still not work very well.

The inanimate subject pronoun is, apparently, incorporated obligatorily, and assigned its index not by direct control but via a null topic. This should be compared with Finnish (ADE-adessive): (46)

Auto on korjattu, mutta Maria sanoo ettei *(se) viela¨ka¨a¨n kulje hyvin. car is fixed but Maria says that-not it still go well ‘The car is fixed, but Maria says that it still isn’t running well.’

(47)

Pera¨moottori on hyva¨, jos (se) la¨htee ka¨yntiin ensi yritta¨ma¨lla¨. outboard-motor is good if it gets going first try-ADE ‘An outboard motor is good if it starts up at the first try.’

An inanimate pronoun in Finnish can be null, but only when it is locally c-commanded by an antecedent, the same as animate pronouns. Why this difference between animate-referring (or human-referring) and inanimate-referring pronouns? Part of the answer may be that the overt pronoun ele in (44) is a strong pronoun, in spite of not being in any way emphatic. As discussed by Cardinaletti & Starke (1999), strong pronouns

Null Subject Parameters

119

typically refer only to humans. That the overt pronoun in (44) is a strong pronoun follows if (a) it is a defining property of strong pronouns that they have D (cf. Cardinaletti & Starke 1999), and (b) only D-less pronouns can incorporate, from which it follows that subject pronouns with D have to move and remerge with TP, to satisfy EPP of T. Inanimate-referring pronouns would not have the DP option (in EP; Finnish does not make the same distinction between strong and weak pronouns as EP does), and would thus always be incorporated. This does not explain the source of the c-command requirement on the null pronoun in (44), though. In BP, Finnish and other partial NSLs the null pronouns which are dependent on c-command/control are, by hypothesis, derived by remerger of a [uD]-marked pronoun with TP (movement to SpecTP). Do we want to assume this in the case of the null, controlled pronoun in (44) as well? This is a question I leave open. In as yet unpublished work, Melvyn Cole shows that there is considerable variation also among null-subject languages here classified as consistent, as regards the ‘accessibility’ of an antecedent. For instance, some languages allow the possessor of a subject DP to be the antecedent of a null subject in a subsequent sentence, other languages do not. This variation thus involves rules and parameters of discourse grammar and pragmatics, perhaps adequately described in terms of Ariel’s (1990) theory (as argued by Cole). If I am right, some variation, including that between partial and consistent null-subject languages, involves features and locality principles of sentence grammar, though. 9

Rizzi (1986a) in a minimalist perspective

The formulation of the NSP in Rizzi (1986a) has been very widely adopted throughout the nineties and up until today. (48)

Licensing: pro is Case-marked by X y, where y is parametrised. Identification: pro inherits the f-feature values of X y (if it has f-features; if not, pro gets a default interpretation, typically arb).

In the case of subject pro, X y is finite INFL. (48) differs from the 1982 formulation in that: (a) it generalises the parameter to other null arguments than subjects; (b) it leaves out the features [þreferential] and [þpronoun], replacing them with an unspecified parameterised property ‘y’; (c) it includes arbitrary/generic pro in the picture. The difference between consistent NSLs and semi- or partial NSLs is now ascribed to the features that pro can inherit from X y (as made clear in Rizzi 1986a).

120

Anders Holmberg

As discussed in Holmberg (2005), (48) is incompatible with a theory making a distinction between uninterpretable/unvalued features and interpretable/valued features, following Chomsky (1995b: chapter 4; 2001): pro cannot inherit f-feature values from finite INFL since the f-features of finite INFL are themselves unvalued, in need of valuation. Apart from this obvious flaw (once the Chomskyan feature theory is adopted),27 (b) is a step backwards, compared with the 1982 formulation of the NSP(s). I have endeavoured to show in this chapter that particularly the feature [þreferential], appropriately remodelled as a [uD]-feature, does play a crucial role for null subjects. Including proarb is a step forward, though. The reason for the generalised formulation of the licensing conditions in Rizzi (1986a) is that the focus in that paper is on a form of pro which is not licensed and identified by INFL, namely object pro in Italian, as in: (49)

a. Il bel tempo induce pro a restare. the nice weather induces to stay. b. La buona musica riconcilia pro con se stessi. the good music reconciles with oneself.

In Italian, Rizzi argued, transitive V is an X y, in terms of (48), Case-marking and licensing an object pro, which, as Italian V does not have f-features, is interpreted with arbitrary/generic reference. I have argued that, rather than being dependent on identifying f-features, ‘proarb’, the generic or arbitrary null pronoun, is made possible by the absence of [uD] on the probing head. Assuming that the Italian null object is a counterpart of the generic null subject pronoun found in partial NSLs, and assuming that the head which probes for the object and assigns case to the object is v, then this head should have the following properties: it should have a uf-feature set sufficiently articulated to incorporate an object fP pronoun, but it should not have [P] and it should not have [uD]. Assume that Roberts (forthcoming b; Chapter 1) is right that object clitics are the spell-out of an object pronoun which is incorporated, but incorporated by Agree, without movement. Italian has object clitics, so clearly v in Italian has the uf-features required to incorporate a pronoun. What is unexpected in the present theory is that Italian should allow both a referential 3rd person incorporated pronoun and a generic/arbitrary one. An important component in the theory of null subjects articulated in this chapter, and in Holmberg (Chapter 5) and Holmberg & Sheehan (Chapter 3), is that languages, as it were, have to choose 27

Rizzi’s (1986a) pro is, however, potentially compatible with a feature theory such as the one proposed by Pesetsky & Torrego (2001; 2004), which makes a distinction between (un)valued and (un)interpretable features. In such a theory pro can be inherently interpretable but unvalued, and dependent on the uninterpretable but valued features of T for a value. See Johns (2007) for an implementation of Pesetsky and Torrego’s theory to null arguments.

Null Subject Parameters

121

between either a definite null subject, derived by incorporation, or an indefinite (generic/arbitrary) one. Why should objects be different? This is another question, and a potential challenge to the theory, which I will leave for future research. An idea in D’Alessandro & Roberts (forthcoming b) might be relevant, though. They suggest that a number of asymmetries between subjects and objects, including the non-existence of “pure, consistent null-object” systems without null subjects also being allowed, can be derived from the fact that subjects but not objects are goals in the same phase as their probe. 10

The pro-drop hierarchy

In Roberts and Holmberg (Introduction: Section 3.4) we argue in favour of the hierarchical organisation of parameters. The argument is that the number of parameters that we have to assume is so large that it compromises the explanatory power of parametric theory as (a component of) a theory of language acquisition, unless parameters form hierarchies. The following is a proposal for how the null subject related parameters discussed in this chapter are arranged. (50)

D in T

Yes

No

φ-dependent EPP

Yes Italian

No Spanish

P in T

Yes φ-dependent EPP

Yes French

No Icelandic

No φ-dependent EPP

Yes Kriyol

No 28

Finnish

The hierarchy is represented here in the same yes/no format as in Roberts & Holmberg (Introduction: Section 3.4). I have mentioned a representative language at each terminal node. This tree embodies the hypothesis that f-dependent EPP (Yes/No) is independent of the other parameters, and therefore can distinguish, in principle, between any type of languages which have uf-features, while D in T (Yes/No) and P-in-T (Yes/No) are in a bleeding relation to each other.29 28 29

See Nicolis (2005; 2008). With P in T the most important triggering experience required to postulate D in T as part of the acquisition of a language, namely definite null subjects, will be lacking. A Yes for D in T therefore excludes a Yes for P in T, and vice versa.

122

Anders Holmberg

The first split distinguishes between consistent null-subject languages (Yes) and ‘the rest’ (No). Note that what we have called partial null-subject languages do not make up a node in the tree, under this hypothesis: Icelandic and Finnish are on different branches. They have in common that they both allow incorporation of an indefinite pronoun in T. However, this is an effect of (a) having rich enough uf-feature content but no D in T, and (b) having a f-independent EPP in T. Kriyol in (50) is a representative of a language type which allows only expletive null subjects; absence of P in T makes possible a null SpecTP, but a f-dependent EPP rules out incorporation. The uf-feature content of T is not explicitly included as a parameter in (50). In this way (50) differs from the parameter hierarchy proposed in Roberts & Holmberg (Introduction: 49). The uf-features are involved, though, but indirectly. A positive value for D in T is, plausibly, dependent on rich uf-feature content (as Roberts argues, in Chapter 1). The case of Finnish (see Section 6) shows that this is not a two-way implication, though. Furthermore, as mentioned earlier, f-dependent EPP is probably a typical property, perhaps even a necessary property, of languages with very poor uf-feature content in T. But again, this is not a two-way implication, since a relatively richly specified language like French, not to mention Italian and many other consistent null-subject languages, appears to have f-dependent EPP. The hierarchy (50) should be incorporated in the larger hierarchy proposed in Roberts & Holmberg (Introduction: 49). The following seems a natural way to do it. (51)

uφ

Yes

No Radical pro-drop

uD in all probes

Yes Polysynthesis

No uD in {C, T, v ...}

The first split of the complete pro-drop hierarchy is between languages which do and languages which do not have uf-features in the grammar. For example Chinese, Japanese and Korean do not. This is what allows these languages to have radical pro-drop; see Roberts & Holmberg: Introduction: Section 3.4.30 The next parameter proposed by Roberts and Holmberg (though more 30

It is not the case, apparently, that all languages with radical pro-drop lack uf-features. The Dravidian languages have radical pro-drop (see Jayaseelan 1999), but some of them have subject–verb agreement, even rich agreement (as in Tamil). An alternative, logical given the

Null Subject Parameters

123

speculatively) distinguishes polysynthetic languages (or consistently headmarking languages) from all other languages which have uf-features. I have reformulated it here in terms of the unvalued definiteness feature uD (instead of rich uf-feature content), the idea being that a language with a positive value for this parameter can have a null pronoun with definite interpretation in any argument position. A typical sentence in such a language consists of a predicate and a configuration of null pronouns each assigned an index by a lexical DP merged in peripheral position, or else by a null topic; see Baker (1996). But the index-assignment is via a uD-feature on the relevant probe, according to the theory proposed here.31 The next node in the tree is a set of parameters, specifying which probes, if any, have a uD-feature. The one we have discussed concerns T. The theory articulated here is fairly complex, postulating three parameters (D in T, P in T, f-dependent EPP), with richness of uf-feature content of T as a fourth variable property affecting the distribution of null subjects indirectly, by affecting (but not determining) the choice of value for the three parameters, in order to account for what may be described as a limited range of variation: whether a language has definite null subjects, indefinite null subjects, controlled null subjects (in finite clauses), expletive null subjects or none at all, and in addition whether non-nominative subjects are allowed. One may legitimately wonder whether the theory is any simpler than the facts it is meant to explain. If not, it fails a fundamental criterion of an explanatory theory. As a test, consider the effect of reformulating the parameters in observational rather than theoretical terms. Is anything lost under this reformulation? If not, the theory is on the wrong track (as before, ‘indefinite’ means ‘generic’, and as in (50) I list one representative language of each type).

31

discussion in the text, is that the parameter singling out radical pro-drop languages from the rest concerns specifically definiteness, rather than f-features such as number and person. This is, in fact, what Jayaseelan (1999) proposes, observing that Dravidian and the familiar East Asian radical pro-drop languages lack articles. We could thus replace ‘uf’ at the root of (51) with ‘uD’. However, this can, again, at best be a one-way implication, as there are languages which have no articles, yet do not have radical pro-drop, for example Finnish and many of the Slavic languages. See Roberts & Holmberg (Introduction: Section 1.2.3) for discussion of theories of radical pro-drop. The prediction is that consistently head-marking languages will be like consistent null-subject languages with respect to definite and indefinite null arguments: the latter should either not occur at all, or require some special morphology. I have not seriously researched this question. Mark Baker (p.c.) informs me that Mohawk employs a special form of agreement, the ‘feminine impersonal’, so called because it is homophonous with the feminine-zoic agreement, or uses the 2SG. He also notes that Mapudungun appears to use a passive as a generic/ impersonal construction with transitive verbs when the subject is an indefinite pronoun, but 3SG agreement with intransitive verbs. The latter case could be a real counterexample. Note the observation in the text above about null objects in Italian, that the complementarity between definite and indefinite null pronouns does not appear to hold for (direct) objects.

124

Anders Holmberg

(52)

Definite NS Yes

No Indefinite NS

Italian Yes

No

Controlled NS

Expletive NS

Yes

No

Yes

No

Finnish

Icelandic

Kriyol

French

One fact extensively discussed in this chapter that does not fall out from this tree, but does fall out from the theory behind the tree in (50), is that having definite 3rd person null subjects excludes having indefinite null subjects (in active, finite clauses), and vice versa. In addition, the possibility of nonnominative subjects is not accounted for in (52), but would need to be added as an independent parameter. Another fact which does not fall out from (52), but is accounted for by the theory behind (50), is that an indefinite null subject does not satisfy the EPP, while a definite one does (see also Holmberg (Chapter 5)). Another one is that indefinite null subjects can be generic but not existential or universal. I feel reasonably confident, therefore, that the theory articulated in this chapter has explanatory power, and is not just a technical reformulation of a set of observations.

3

Control into finite clauses in partial null-subject languages Anders Holmberg and Michelle Sheehan

1

Introduction*

This chapter focuses on three partial null-subject languages – Brazilian Portuguese, Finnish and Marathi – which allow null subjects under more restricted conditions than consistent null-subject languages. The conditions include: (a) when the subject is non-thematic; (b) when the subject is a generic pronoun corresponding to English ‘one’ (exemplified by (1a), from Marathi); and (c) when the subject is controlled by an argument in a higher clause (exemplified by (1b), also from Marathi). (1)

a. unahlyat lavkar utthavla jato summer-in early wake go-PRS-3SM ‘In summer one wakes up early.’ b. Ram mhanala ki ghar ghetlə Ram say-PST-3SM that house buy-PST-3SN ‘Ram said that he bought a house.’

Holmberg (Chapters 2 and 5) and Holmberg, Nayudu and Sheehan (2009) argue that property (b) is particularly revealing. It shows that the languages lack a D(efinite)-feature in T: the feature which makes a null subject with definite interpretation possible in consistent null-subject languages. Following the proposal laid out in Holmberg (Chapter 2), we adopt an analysis of the sentential feature composition of partial null-subject languages based on T’s lack of this D-feature coupled with the lack of a ‘phonological EPP’ in these languages, formally a feature [P] in T, as evidenced by property (a).

*

Thanks to the participants in the workshop on partial pro-drop held in Cambridge 2006, especially Aarti Nayudu and Halldor A. Sigurðsson. Thanks are due to Cilene Rodrigues, Marcello Modesto, De´borah Oliveira, Dorothy Bezerra Silva de Brito, Daniel Carvalho, Jose´ da ˆ ngela, Hannu Reime, Urpo Nikanne, Satu Manninen, Itto Takao and Winnie Yiu for help A with the data. Special thanks to Idan Landau, whose comments on an earlier version of this chapter led to considerable revision and, hopefully, improvement. All errors are our own.

125

126

Anders Holmberg and Michelle Sheehan

The focus of this chapter, however, will be to examine the conditions under which null subjects of type (c) are possible in the three languages. A careful examination makes it clear that the conditions under which the languages allow a controlled, externally licensed null subject, vary to some extent, sometimes in puzzling ways. It is our contention, however, that the control relation they display is uniform and distinct from both the obligatory control (OC) and non-obligatory control (NOC) of non-finite and finite clauses (cf. Landau 2004), and from the discourse-based antecedence relation characteristic of consistent null-subject languages, (cf. Samek-Lodovici 1996; Frascarelli 2007). This chapter will be concerned with characterising the differences in the distribution of null subjects in the languages in question and deriving these facts from independent parametric differences between the three languages, largely from the OV status of Marathi vs. the VO status of BP and Finnish. The other main aim of the chapter will be to derive the differences between this control relationship and OC from the type of Agree relation it instantiates. Section 2 will focus on the characterisation of partial null-subject languages, as distinct from consistent null-subject languages and nonnull-subject languages. Section 3 will sketch the analysis of consistent null-subject languages and partial null-subject languages given in Holmberg (Chapter 2), relating it to the typology of pro-drop. In Section 4, we will describe at length the differences between this kind of finite control and the finite OC described by Landau (2004). Section 5 will consider how the availability of null subjects in embedded clauses interacts with the matrix verb type, and Section 6 will consider island sensitivity. It will become apparent that there are many differences between the licensing of null subjects in the three languages, but that many of these differences are derivable from independent parameter settings. Finally, in Section 7 we will return to Landau’s control calculus and consider how the null subjects discussed here are related. Section 8 concludes our discussion, raising some issues for future research. 2

Partial and consistent null-subject languages

2.1

Null subjects that are not licit in partial null-subject languages

As a rough characterisation, null subjects in partial null-subject languages are optional in some contexts where they are obligatory in consistent null-subject languages and excluded in non-null-subject languages, whilst being excluded in other contexts where they are allowed in consistent null-subject languages. The following is an illustration: Consider (2), where John is talking about himself, as indicated by the indexing.

Control into finite clauses in partial null-subject languages (2)

127

John1 said that he1 bought a house.

In a non-null-subject language such as English the pronoun has to be overt. In a consistent null-subject language, such as Arabic, Greek, Spanish, Turkish, etc., the pronoun has to be null, assuming that there is no contrast or shift of topic involved. In partial null-subject languages the pronoun is optionally null, the null-option exemplified in (1b) for Marathi. Now imagine a context where another person, call him Bill, is being discussed. One of the interlocutors utters (3) as a contribution to the discussion, where the embedded pronoun refers to Bill, as indicated by the index 2. (3)

John1 said that he2 bought a house.

In a non-null-subject language such as English the pronoun has to be overt. In a consistent null-subject language the pronoun would still typically be null, assuming no contrast or topic-shift. In a partial null-subject language the pronoun has to be overt in this case.1 We exemplify the difference between the three types with an embedded subject pronoun because a wider range of languages, including at least some partial null-subject languages and non-null-subject languages, allow null subjects in main clauses derived by ‘topic drop’ (see Rizzi 1992; Haegeman 2000; Rodrigues 2002). Even so, the example involves a certain amount of idealisation; for example, there are languages which have some properties of partial null-subject languages, but allow a null subject even in (3), and languages with some properties of partial null-subject languages which do not allow a null subject even in (2), and languages which are otherwise consistent null-subject languages, yet do not allow a null subject in (3).2 Considering just the three languages under investigation in this chapter, there is another complication, which is that Finnish permits 1st and 2nd person null subjects in basically any context (although more commonly in formal and written Finnish). Third person pronominal subjects, though, are subject to constraints which do not apply in consistent null-subject languages, but which closely resemble those which hold for BP and Marathi. We will, from now on, deal with 3rd person subjects only.3 It should also be clarified here that we are discussing spoken Marathi and Brazilian Portuguese. Interestingly, the written 1

2

3

Actually, matters are slightly more complex than this as Brazilian Portuguese does allow null subjects in contexts such as these, through a process of Topic deletion. Crucially, however, it does not allow null subjects with a non c-commanding antecdedent in island configurations. See Sheehan (2006: chapter 4) for a discussion. Bengali and Hindi, two languages closely related to Marathi, appear to belong in the first category, being more permissive than Marathi, BP and Finnish, while Icelandic, being less permissive than Marathi, BP, and Finnish, belongs in the second category, and at least some varieties of European Portuguese belong in the third category. See also Vainikka & Levy 1999; Holmberg 2005.

128

Anders Holmberg and Michelle Sheehan

variety of Marathi is more restrictive in its use of null subjects, whereas the opposite is true in BP and Finnish, where the written form uses more null subjects than the spoken variety.4 For the purposes of this chapter, we will use the term ‘partial null-subject language’ strictly for languages that have properties (a) and (b).5 The following sections will illustrate the cases where the languages in question have null subjects, either optionally or obligatorily, in comparison with consistent null-subject languages and non-null-subject languages. 2.2

Null non-thematic subject

With predicates which do not have a f-marked subject the partial null subject languages generally have no overt subject. One such case is weather predicates.6 (4)

a. Esta´ chovendo. Is raining ‘It’s raining.’ b. Ulkona sataa. outside rains ‘It’s raining outside.’

(BP)

(Finnish)

In Marathi, a different kind of structure is used in these instances, equivalent to ‘rain falls’. 2.3

Null generic subject pronoun

One situation where a null subject is licit in finite clauses in our three partial null-subject languages is when the subject is a generic pronoun corresponding to ‘one’ (INE ¼ inessive). (5)

4

5

6

a. E´ assim que faz o doce. is thus that makes the sweet ‘This is how one makes the dessert.’

(BP)

Duarte (1995) discusses the situation for BP amongst young people, claiming that the null subject, for young people, has become “pra quando a gente escreve” (for when we write), and the overt subject “pra quando a gente fala” (for when we speak) (p. 143). See Huang (2000: 51ff.) on the typology of null subjects. Rizzi (1982; 1986a) discusses what he calls semi-pro-drop languages, which would have only non-thematic null subjects. A further distinction among the semi-pro-drop languages, proposed by Rizzi (1986a) and discussed by Huang (2000), is between languages which allow null subjects with weather verbs (quasiargumental null subjects, for example Icelandic) and those which only allow purely expletive null subjects (German). As it happens, German is not a partial null-subject language in our sense, as it does not have a null generic subject. Icelandic is, as it has a null generic subject, although it does not have controlled null subjects in finite clauses; see Sigurðsson & Egerland (2009). On Icelandic, see Holmberg (Chapter 2). Expletives are not excluded in principle, though. Finnish employs an expletive subject in certain constructions as an alternative way to satisfy the EPP; see Holmberg & Nikanne 2002.

Control into finite clauses in partial null-subject languages

129

b. Nesse hotel na˜o pode entrar na piscina beˆbado in.this hotel neg can enter in.the swimming.pool drunk ‘In this hotel it is not permitted to go in the swimming pool drunk.’7 (BP, Rodrigues 2004: 72) (6)

a. Kesa¨lla¨ hera¨a¨ aikaisin. summer-INE wake-PRS.3SG early ‘In the summer one wakes up early.’ b. Ta¨a¨lla¨ ei saa uida. here not-3SG may-PRS swim ‘One must not swim here.’ c. Nuorten mielipiteita¨ kuuluu arvostaa. youth’s opinions should-PRS-3SG respect ‘One should respect the views of young people.’

(7)

a. unahlyat lavkar utthavla jato (Marathi) summer-in early wake go-PRS-3SM ‘In summer one wakes up early.’ b. asa lokan kade baut dakhavayla paidze nahi like this people towards finger show-INF- DAT should NEG.AUX ‘One should not point at people’. c. mulan-chya vicharan-cha aadar kar-ay-la paidze children’s views respect do-INF- DAT should ‘One should respect the views of young people’.

(Finnish)

Interestingly, in this case consistent null-subject languages such as Spanish and Greek do not allow a plain null subject, but have to resort to some overt strategy. The contrast is seen most clearly when comparing BP, a partial nullsubject language, with European Portuguese (EP), a consistent null-subject language. Compare (5) and (8): (8)

a. E´ assim que se faz o doce. (EP) is thus that SE makes the sweet ‘This is how one makes the dessert.’ b. Nesse hotel na˜o se pode entrar na piscina beˆbado. (EP) In.this hotel neg SE can enter in.the swimming.pool drunk ‘In this hotel it is not permitted to go in the swimming pool drunk.’

In EP, the generic subject reading requires the reflexive clitic se. This clitic is either itself the generic pronoun, or serves to somehow license a null generic

7

The counterpart of (6c) and (7c) is not possible in BP with a null subject. Either a clitic or DP like ‘the people’ is needed in such circumstances: (i)

*(A gente/se) deve respeitar as opino˜es dos jovens. the people/SE must respect the opinions of.the young

This is presumably because, unlike Finnish and Marathi, BP cannot have the object DP satisfy the EPP. We will leave aside a discussion of such differences in this chapter.

130

Anders Holmberg and Michelle Sheehan

pronoun; see Holmberg (Chapter 2). In either case it holds that a plain 3SG null subject with no special morphology is not an option in EP or the other Romance consistent null-subject languages, in this context. It is important to make a distinction in this connection between generic and arbitrary null subject pronouns. By generic pronoun we mean a pronoun best translated into English as either ‘one’ or ‘you’, the semantic defining characteristic being that it denotes people in general including the speaker and the addressee. By arbitrary we mean a pronoun which is best translated into English as they, as in They speak many different languages in India, the semantic defining characteristic being that it denotes people in general (in some domain), but excluding the speaker and the addressee. Consistent nullsubject languages have an arbitrary null subject (null ‘they’) but to express a generic subject pronoun they resort to some overt strategy. Among partial nullsubject languages some have a null arbitrary pronoun as well as a generic one (in active clauses without any special morphology), others do not; for instance BP does, but Finnish does not. Other consistent null-subject languages use other strategies to express inclusive generic meaning, such as using an overt indefinite, or passive voice, or some other dedicated impersonal voice, or generic ‘you’, which may be null but visible on the verb agreement; see Holmberg (Chapter 5). None of them employs a null 3rd person generic subject in a construction with an active, 3SG-marked verb. There is a potential functional explanation for this: in a consistent null-subject language a null 3rd person subject will be interpretable as a definite pronoun. Remove se from (8a), for example, and the sentence can be read as ‘This is how he makes the dessert’. In BP this does not happen, as the language does not have definite null subjects. The reason why consistent null-subject languages resort to overt strategies to express a generic null subject would thus be to avoid ambiguity. This cannot be correct, however, as there are some languages which allow genuine ambiguity of this kind. Consider the case in Cantonese and Japanese:8 (9)

8

a. Ah John waa hai Jinggwok jiu gong Jingman (Cantonese) Prt John say in England need speak English ‘John says that one/he needs to speak English in England.’ b. John-wa kono beddo-de-wa yoku nemu-reru-to iu. (Japanese) John-TOP this bed-in-TOP well sleep-can-COMP say ‘John says that one/he can sleep well in this bed.’

We exclude radical null-subject languages from our discussion here, believing them to lack f-features altogether, and perhaps D. As such, the implicit assumption is that they are immune to the absence/presence of [uD] features on T and the þ/– status of the [PF] sensitive EPP. See Roberts and Holmberg (Introduction) for a discussion.

Control into finite clauses in partial null-subject languages

131

In fact, a degree of ambiguity also arises in partial null-subject languages in certain circumstances. We therefore conclude that the explanation for this difference cannot be purely functional. 2.4

Null subject controlled by an antecedent in a higher clause

Another situation where BP, Finnish and Marathi all allow a null subject is when there is a linguistic antecedent in a higher clause: (10)

a. O Joa˜o1 disse que (ele1) tinha comprado uma casa (BP) the Joa˜o said that he have.PST.3SG bought a house ‘Joa˜o said that he had bought a house.’ b. Os meninos1 ficavam contentes quando (eles1) tinham um dı´a the children were happy when they have-PST.3PL a day de folga of holiday ‘The children were happy when they had the day off.’ c. A Maria1 admite que (ela1) na˜o fala muito bem ingleˆs. (BP) The Maria admits that she not speak.PRS.3SG very well English ‘Mary admits that she doesn’t speak English very well.’

(11)

ostanut omakotitalon. (Finnish) a. Juhani1 kertoi etta¨ (ha¨n1) oli Juhani said that he have-PST.3SG bought house ‘Juhani said that he had bought a house.’ b. Lapset1 olivat mielissa¨a¨n kun (he1) saivat vapaapa¨iva¨n. children were pleased when they get-PST.3PL off-day ‘The children were pleased when they got a day off.’ (ha¨n1) puhu englantia hyvin. c. Marja1 myo¨nta¨a¨ ettei Mary admits that-not-3SG she speak-PRS English well ‘Mary admits that she doesn’t speak English well.’

(12)

ki (tyani1) ghar ghetlə (Marathi) a. Ram1 mhanala Ram say-PST.3SM that he house buy-PST.3SN ‘Ram said that he bought a house’. b. mulan-la1 khushi dzali dzewha (tyan-la1) shalyat-hun radza milali children-DAT happy happen-PST.3SF when they-DAT school-from off get-PST-SF ‘The children were happy when they got off from school’. c. Seema1 kabul karte ki (ti1) chukli Seema agree do-PRS.SF that she mistake-PST.3SF ‘Seema admits that she made a mistake’.

These types of null subjects will constitute the main focus of this chapter.9 9

We exclude from the discussion the null subjects which occur in answer to yes/no-questions in many languages. See Holmberg (2001, 2007) for an ellipsis analysis of these constructions which is independent of the Null Subject Parameter.

132

Anders Holmberg and Michelle Sheehan

Table 3.1. Summary of the null-subject typology Consistent null-subject language Example languages

Spanish, Greek, Turkish10, Arabic, Telugu Non-root, non-controlled null ‘topic’ pronouns Inanimate subject often null pronouns Controlled subject null

Generic null subjects Non-thematic subjects

2.5

overt null

Partial nullsubject language

Non-null-subject language

BP, Finnish, Marathi

English, French11, Sorbian12, Somali13

overt

overt

Overt/null if controlled null/overt (optional) null null

overt overt overt overt

Pronouns with inanimate reference

In BP, Finnish and Marathi, this class of pronouns behaves exactly like those which refer to animate referents: as subjects they are overt unless controlled. In at least some consistent null-subject languages, including the Romance null-subject languages, this is not the case, as pronouns which refer to inanimate objects are always null. This is not always the case, however, as inanimate pronouns can be overt in Arabic, another consistent null-subject language. 2.6

Summary of the null-subject typology

Table 3.1 provides a summary of the properties of the different types of nullsubject languages. Interestingly, the distribution of null subjects works largely on a continuum across the various language classes, with the exception of generic pronouns. We take this to be a crucial property of partial null subject languages. Following Holmberg (Chapter 2), we take it to indicate that T lacks a [uD] (definite) feature, allowing incorporated pronouns to get a generic, rather than a definite interpretation.

10 11 12 13

Cf. Kornfilt 1985. But see Roberts (Chapter 8) for the proposal that French is actually a partial pro-drop language. Cf. Lindseth 1998. Somali subject clitics seem to be obligatory in the absence of a DP subject, optional with referential subjects, and banned with quantificational subjects (Saeed 1996: 165–66). The language also has overt impersonal subject clitics (Saeed 1996: 76; Cabredo Hofherr 2004) and overt expletives (Rebuschi and Tuller 1999: 275–309).

Control into finite clauses in partial null-subject languages

3

133

Null subjects in partial null-subject languages

Holmberg (Chapter 2) provides an account of consistent null-subject languages based on Roberts’ (forthcoming b; Chapter 1) analysis of clitics as incorporated pronouns and Frascarelli’s (2007) analysis of null-subject languages. He proposes that partial null-subject languages differ from consistent null-subject languages in the following way: in consistent null-subject languages, T has an unvalued uD-feature, where D encodes definiteness. This feature gets valued by the subject as part of Agree, the probe–goal relation between T and the subject (cf. Chomsky 2001). In the null subject case, the subject is a D-less jP (made up of valued j-features). This is incorporated in T as a result of Agree (see Holmberg (Chapters 2 and 5) and Roberts (Chapter 1) for details), while the uD-feature gets valued definite by a null topic in SpecCP (following Frascarelli 2007). The result is a ‘null subject’ which is a chain headed by T, valued definite, spelled out as an inflection on the verb or auxiliary. In other languages, finite T does not bear a [uD] feature. In a subset of these languages, the partial null-subject languages, the subject can still be null, essentially by the same derivation as in the consistent null-subject languages: a bare jP subject can be incorporated, as a direct result of Agree. However, in the absence of a uD-feature the resulting chain can only be interpreted as generic or arbitrary, and these are the interpretations we find in partial nullsubject languages. A subject pronoun which is not incorporated in T is attracted by the EPP to remerge with TP. Such a pronoun can be null if it is controlled by a DP in a higher clause. Control is, by hypothesis, only possible if the controlled pronoun is referentially deficient, yet not to the extent that it gets incorporated in T. We propose to capture this in terms of an unvalued D-feature, uD, on the pronoun; see Shlonsky (2009) for a similar analysis. This would also be the case for so-called PRO in non-finite clauses. In finite clauses, this type of null subject is typically found in partial nullsubject languages (though perhaps not exclusively; see Holmberg (Chapter 2)). As discussed in Holmberg (Chapters 2 and 5), the theory predicts that generic/ arbitrary null subjects, being incorporated in T, will not satisfy the EPP, but definite null subjects which are controlled will do so. This prediction can be shown to be right. Consider (13a,b): (13)

a. Joa˜o me contou que na praia vende cachorro quente (BP) Joa˜o me told that at.the beach sell.3SG dog hot ‘Joa˜o told me that hot dogs are sold at the beach.’ 6¼ ‘Joa˜o told me that he sells hot dogs at the beach.’ b. Joa˜o1 me contou que (ele1) vende cachorro quente na praia Joa˜o me told that sell.3SG dog hot at.the beach ‘Joa˜o told me that he sells hot dog at the beach.’ ‘Joa˜o told me that hot dogs are sold at the beach.’ (Rodrigues 2004: 142)

134

Anders Holmberg and Michelle Sheehan

In Finnish as well as BP the EPP is mostly satisfied by the subject, but may be satisfied by other referential categories, including circumstantial adverbials. In (13a) the subject pronoun is incorporated in T, with generic interpretation. In this case another category, here a place adverbial, is attracted by the EPP to SpecTP (i.e. to remerge with TP). In (13b) the subject has moved to SpecTP, satisfying the EPP. In this position it can be null, but only if it is controlled, which it is in this case (see Holmberg (Chapter 5) on Finnish). In Marathi the prediction cannot be so easily tested since due to its SOV syntax all arguments and adjuncts precede the finite verb anyway. It appears, then, that the definite null subjects in BP, Finnish and Marathi are referentially deficient DPs which have been remerged with TP and are interpreted by virtue of control. If Frascarelli (2007) is right, the relation between the null subject and its antecedent in consistent null-subject languages is indirect: the direct antecedent of the null subject is a null Aboutness-topic in SpecCP. This topic is itself interpreted as a copy of a topic in a preceding sentence. It is therefore irrelevant whether this, the ‘ultimate antecedent’ does or does not c-command the null subject. In partial null-subject languages, on the other hand, the indirect antecedence relation, via a null Aboutness-topic, is impossible due to absence of a uD-feature in T which could transmit the topic’s index to the null subject. 4

Control of null subjects in finite clauses

As discussed, BP, Finnish and Marathi all allow null subjects in finite clauses controlled by an antecedent in a higher clause. But unlike in consistent nullsubject languages, the antecedent argument controls the null subject directly (rather than via a null-topic chain). In this respect it is more similar to OC of PRO. However, as we will show here, the type of finite control relation displayed by the three languages under investigation is substantially different from OC. 4.1

How the relationship differs from obligatory control

In the most detailed investigation to date about control into finite clauses, Landau (2004) discusses the case of Hebrew, another partial null-subject language. Hebrew has object control into finite complements of directive verbs, that is verbs of requesting, ordering, proposing, etc.: (14)

Rina himlica le-Gil1 sˇe-1/*2 ya’avod yoter kasˇe. Rina recommended to Gil that will-work.3SG.M more hard ‘Rina recommended to Gil that he should work harder.’

Futhermore Hebrew has subject control into complements of commissive verbs, such as ‘promise’ and ‘declare’, and more marginally into complements of epistemic and declarative verbs (‘believe’, ‘mention’, etc.).

Control into finite clauses in partial null-subject languages

135

(15)

[Sar ha-ocar]1 hic’hir sˇe-1/*2 yorid et ha-misim. minister the-treasury declared that will-lower.3SG.M ACC the-taxes ‘The minister of the treasury declared that he would lower the taxes.’

(16)

et ha-misim. [Sar ha-ocar]1 hosif sˇe-hu1/?1/*2 yorid minister the-treasury added that will-lower.3SG.M ACC the-taxes ‘The minister of the treasury added that he would lower the taxes.’

Landau shows that these complements have in common (with the exception of the marginal epistemic/declaratives) (a) that they have obligatory future tense, and (b) that they correspond to subjunctive complements in other languages. He compares these Hebrew constructions with control into subjunctive complements in Balkan languages, which are notorious for lacking infinitival clauses, and argues that the complement clauses in (14–15) are actually ‘covertly subjunctive’ in Hebrew, too. Furthermore, he argues that what non-finite complement clauses subject to OC and subjunctive clauses have in common is that they have dependent tense. This, he argues, is a necessary condition for OC. He argues that the control relation in (14–15) is indeed OC, and thus that the null subject is effectively PRO, not pro. Interestingly, while it appears that finite control in Hebrew and the other languages Landau discusses can be identified with OC, the kind of finite control seen in Finnish, Marathi and BP differs from OC in many important respects. We will first discuss the apparent similarities between the two types of control, and then the many differences. 4.2

Similarities with obligatory control

Descriptively, both OC and the control relation in BP, Finnish and Marathi appear to be a form of syntactic binding between an antecedent and a null category. The two relations share some properties. In all of the three languages under discussion, the antecedent for the null subject can beþ/– human, as is the case with OC (ERG ¼ ergative):14 (17) 14

This wood1 is dry enough [PRO1 to catch fire].

Holmberg (2005) notes that a null subject is blocked with non-human antecedents in the following context: (i)

a. Jarille tuli selva¨ksi ettei (ha¨n) saa palkintoa. Jari-ALL became clear that-not he gets prize ‘It became clear to Jari that he won’t get a prize. b. Kirjasta tuli selva¨ksi ettei *(se) saa palkintoa. book-ABL became clear that-not it gets prize ‘It became clear from the book that it won’t get a prize.’

The crucial difference between (i) and (18) may be that there is an implicit experiencer (hence human) argument, between the antecedent and the null pronoun in (ib) but not in (18).

136

Anders Holmberg and Michelle Sheehan

(18)

Talo1 kylla¨ paranee jos (se)1 saa uuden katon. house surely improves if it gets new roof ‘The house will surely improve if it has a new roof.’

(19)

na˜o portava nenhum passageiro quando (ele)1 caiu. O avia˜o1 the airplane not carried any passengers when it fell

(20)

dwehwa (te)1 zamini-la aadalə viman-ni1 aag pakadli plane-ERG fire catch-past-3SF when it ground-DAT crash-past-3SN ‘The plane caught fire when (it) crashed to the ground’.

This shows that the relation is not logophoric, in the classical sense discussed in Sells (1987). In the logophoric relation the embedded clause containing the logophoric pronoun is a report of the mental state of the antecedent, or reports an event or state of affairs seen from the point of view of the antecedent (cf. Sells 1987). As all three languages allow inanimate controllers, the relevant relation clearly cannot be logophoricity.

4.3

How this control differs from obligatory and non-obligatory control

In Marathi, Finnish and BP, it is possible to have independent time reference in matrix and embedded clauses where the latter contains a null subject, as illustrated by this example from Marathi: (21)

Seema1 kabul karte aaj ki (ti-chya-ni1) kal chuk dzali Seema agree does today that she-ERG yesterday mistake made ‘Seema admits today that she made a mistake yesterday’.

In this respect, these three languages immediately differ from Hebrew and the other languages discussed by Landau (2004), in which dependent tense is a prerequisite for finite control. In fact, despite apparent similarities, many aspects of the licensing of null subjects in the languages under discussion differ from OC. Landau (2000: 113ff.) shows that a crucial property of OC is the ‘next-clause-up’ restriction. It can be shown that all cases of control across an intervening clause are instances of NOC. Compare (22) and (23). In (22) long-distance control of the embedded PRO is possible across an intervener, whereas in (23) it is not. This shows that (22) is an instance of NOC, not OC: (22)

a. John thinks it is necessary PRO to shave ?himself/oneself. b. John thinks Mary said it was necessary PRO to shave ?himself/oneself.

(23)

*John1 thinks Mary plans PRO1 to shave himself1/oneself.

In previous studies it has been claimed that a similar restriction applies in Finnish (Holmberg 2005), and BP (cf. Modesto 2000a; Rodrigues 2004). In

Control into finite clauses in partial null-subject languages

137

actual fact, we will show that these effects are due only to intervention rather than locality per se.15 None of the three languages allows a control relation across another subject, even if that subject has features incompatible with the null subject (visible on the embedded finite verb or adjective). Marathi is possibly somewhat less strict than BP and Finnish in this regard, though. (24)

a. O Joa˜o1 disse [que os moleques acham [que *(ele)1 e´ esperto]] the Joa˜o said that the kids think that he is smart ‘Joa˜o said that the children think that*he is smart.’ b. Jari sanoo [etta¨ lapset uskovat [etta¨ *(ha¨n) ka¨vi tohtorilla]]]. Jari says that children believe-3PL that he visited-3SG doctor-ADE ‘Jari says that the children believe that he went to see a doctor.’ c. Ram-ni1 mhantlə ki Mary-la watlə ki ?(to)1 doktaran-kade gela Ram-ERG said-3SN that Mary-ACC thought-3SN that he doctor -to went.3SM ‘Ram said that Mary thought that he went to the doctor’.

This can be seen as an instance of defective intervention, of the kind discussed by Chomsky (2000). Implicit (non-PF-realised) arguments also act as interveners so that apparent locality conditions of the following kind can be attributed to this effect: (25)

Marja sanoo [etta¨ on ilmeista¨ [etta¨ *?(ha¨n) saa ensi vuonna ylennyksen]]. Marja says that is obvious that she gets next year promotion ‘Marja says that it’s obvious that she will get promotion next year.’

The predicate ‘obvious’ has an implicit experience argument which may block agreement between the subject in a higher clause and the null subject (see also footnote 1). Crucially though, when the intervening clause contains no argument, all three languages tolerate control by an argument which is more than one clause away. Predicates like ‘true’ and ‘certain’ differ from ‘obvious’ in this respect: (26)

15

a. A Maria1 disse [que e´ verdade [que (ela)1 entornou o copo]] (BP) The Maria said that is true that she knocked.over the glass ‘Maria said it’s true that she knocked over the glass.’ b. Marja1 sanoo etta¨ on varmaa etta¨ (ha¨n1) saa ensi vuonna ylennyksen. (Finnish) Marja says that is certain that she gets next year promotion Marja says that it’s certain that she will get a promotion next year.’

We thank Ian Landau for suggesting this possibility to us.

138

Anders Holmberg and Michelle Sheehan c. Ram-la1 watto [ki he changla dzala [ki (to)1 doktaranRam-DAT thinks-3SM that this good happen-PST that he doctorkade gela]]]. to go-PST-3SM ‘Ram thinks that it was good that he went to the doctor.’

Thus, in some ways it seems that the finite control in BP, Marathi and Finnish is less strict than that seen with PRO. However, in another way the finite control under discussion is actually stricter, as it relies crucially on c-command. A null subject in BP and Marathi, and with some exceptions Finnish, must be c-commanded by its antecedent. The only exception to this condition is that Finnish, but not BP or Marathi, allows control under connectivity by specificational predication, a construction which permits OC into non-finite clauses without c-command (cf. Holmberg, Nayudu & Sheehan 2009; Lyngfelt 2002): (27)

a. John’s plan was [PRO to shave himself and then leave]. b. It’s in Jane’s interest [PRO to be on time]. c. It was a disappointment to John [PRO to be dropped from the team].

Note that a lack of c-command between antecedent and null subjects results in ungrammaticality in BP and Marathi even under specificational predication conditions.: (28)

O plano do Jose´1 realizou-se quando *(ele)1 ganhou o premio. The plan of Jose´ realised-itself when he won the prize ‘Jose´’s plan came true when he won the prize.’

(29)

John-ch vichaar hotə ki *(to) lawkar nighel. J-gen plan was that he early leave-fut ‘John’s plan was that he would leave early.’

(BP)

(Marathi)

In instances of connectivity, though, a non-c-commanding antecedent can bind a null subject in Finnish (CON ¼ conditional, ELA ¼ elative, ILL ¼ illative, ALL ¼ allative):16 (30)

16

a. Jarin1 suunnitelma oli [etta¨ (ha¨n1) ajaisi yhdessa¨ pa¨iva¨ssa¨ Helsingista¨ Ouluun]. Jari’s plan was that he drive-CON one-INE day-INE Helsinki-ELA Oulu-ILL ‘Jari’s plan was that he would drive in one day from Helsinki to Oulu.’ b. Se oli pettymys Tarjalle [ettei (ha¨n) saanut menna¨ mukaan]. it was disappointment Tarja-ALL that-not she could go along ‘It was a disappointment to Tarja that she couldn’t go along.’

The examples with a non-c-commanding antecedent listed in Holmberg (2005) are all predicational constructions, as pointed out to us by Idan Landau (p.c.).

Control into finite clauses in partial null-subject languages

139

In other contexts c-command is a requirement for control into finite clauses in Finnish, too, though. Consider the contrast between (31a), where Jari cannot control the null subject even though pragmatically it is the only possible antecedent, and the roughly synonymous (31b), where it c-commands and licenses the null subject. (31)

a. [Jarin puhe] teki selva¨ksi ettei *(ha¨n) ole syyllinen. Jari’s speech made clear that-not he is guilty ‘Jari’s speech made clear that he isn’t guilty.’ b. Jari teki puhessaan selva¨ksi ettei (ha¨n) ole syyllinen. Jaro made speech-INE-his clear that-not he is guilty ‘Jari made it clear in his speech that he isn’t guilty.’

There are some well-known tests for OC, distinguishing between OC and NOC. OC but not NOC requires sloppy identity in a case like (32a) and a bound reading in a case like (32b); see Landau (1999: 43; 2000); Hornstein (1999). (32)

a. John expects to win, and so does Mary. b. Only John expects to win.

(32b) can only mean ‘John is the only x who expects x to win’ (the bound reading), and cannot mean ‘John is the only x who expects John to win’ (the coreferential reading). Applying these tests to BP and Finnish yields contradictory results. According to Negra˜o (1997; cited by Modesto 2000a; 2000b; 2008), Ferreira (2004) and Rodrigues (2004), (33) can only be interpreted with sloppy identity, and (34) can only have a bound reading. (33)

O Joa˜o1 [acha [que  vai ganhar a corrida]] e a Maria tambe´m the Joa˜o thinks that go win the race and the Maria also (i) ‘Joa˜o thinks he’s going to win the race and Maria thinks she will too.’ [sloppy] (ii) *‘Joa˜o thinks he’s going to win the race and Maria thinks he will too.’ [strict]

(34)

So´ o Jose´ acha que  vai ganhar as eleic¸o˜es. only the Jose´ thinks that goes win the elections (i) ‘Jose´ is the only candidate who expects to win the elections.’ [bound] (ii) *‘Jose´ is the only person who expects Jose´ to win the elections.’ [coref]

In Finnish we do not get this effect. (35)

Marja luulee etta¨ (ha¨n) on ovela, ja niin luulee Jarikin. Marja thinks that she is clever and so thinks Jari-too ‘Marja thinks that she is clever, and so does Jari.’

(36)

Vain Jari uskoo etta¨  voittaa vaalit. only Jari thinks that wins elections ‘Only Jari thinks that he will win the elections.’

140

Anders Holmberg and Michelle Sheehan

(35) allows a strict or a sloppy reading, regardless of whether the pronoun is overt or covert. (36) allows the bound reading but also allows the coreferential reading. Similarly in Marathi, (37) allows a strict or a sloppy reading, regardless of whether the pronoun is overt or covert. (37)

Seema-la wattə ki (ti-ni) hi pustak vaachli aahe ani tasach Ram-la pan wattə Seema-ACC thinks-3SN that she-ERG this book read is and similarly Ram-ACC also thinks-3SN ‘Seema thinks that she has read this book, and so does Ram.’

(38), likewise, allows a bound as well as a coreferential reading, whether the pronoun is overt or covert. (38)

ho, Ram-la wattə ki (to) jinkel yes Ram-ACC thinks-3SN that he win-FUT ‘Yes, Ram thinks that he (himself) will win.’

Thus it can be the answer to either the question (39a), or (39b):17 (39)

a. kontya-hi umiddhwar-la wattə ka ki to jinkel? who-EMPH candidate-ACC thinks-3SN QM that he win-FUT ‘Does any candidate think that he will win?’ b. kontya-hi umiddhwar-la wattə ka ki Ram jinkel? who-EMPH candidate-ACC think-PRES-3SN QM that Ram win-FUT ‘Does any candidate think that Ram will win?’

A well-known difference between OC and NOC is that only NOC allows split antecedents. In BP, split antecedents are acceptable in some limited contexts: (40)

O Ze´ convenceu os meninos que tinham que ir embora The Ze convinced the kids that had that go away ‘Ze convinced the kids that they (all) had to leave.’

In fact, the only reading possible in (40) is that with split antecedents. The reading where the null embedded subject refers only to meninos is ungrammatical for all speakers consulted (see also Modesto 2000a; 2000b; Rodrigues 2004). In Marathi a null subject in a finite embedded clause can have split antecedents. (41)

17

Mary-ni Lucy-la sangitlə ki (te) ekatr jau shaktat Mary-ERG Lucy-ACC said -3SN that they together go happen-PRES-3PL ‘Mary told Lucy that they can travel together.’

The test from ‘only NP’ has been adapted to Marathi this way because for some reason the focusing adverb nusta ‘only’ scoping narrowly over the matrix subject prevents control of a null subject in the embedded clause.

Control into finite clauses in partial null-subject languages

141

Table 3.2. Comparison of control in partial null-subject languages and with English PRO

Binder must be þ human Independent time ref. Next clause up condition Binder must c-command pro Split antecedents allowed? Sloppy reading only with ellipsis?

BP

Finnish

Marathi English OC PRO

N

N

N

N

Y N

Y N

Y N

Y

N Y – otherwise it is NOC N – control under connectivity N

N

Y

Y

N – control under connectivity Y – in restricted Y contexts Y N

Y

In Finnish, split antecedents are unacceptable or at least degraded when compared with a non-split antecedent. (42)

a. Marja kertoi Jarille etteiva¨t ??(he) voi matkustaa yhdessa¨. Marja told Jari that-not-3PL they can travel together b. Marja kertoi Jarille ettei (ha¨n) voi matkustaa ha¨nen kanssaan. Marja told Jari that-not-3SG she can travel him with

We contend that, even in Finnish, the degree of acceptability is far better than that seen in instances of OC, as in (43) and (44), for example. (43)

*Mary told John to leave together.

(44)

*Marja pyysi Jaria [PRO matkustamaan yhdessa¨]. Marja asked Jari travel-INF together

4.4

Summary of similarities and differences

As is clear from Table 3.2, the control relation in finite clauses in partial nullsubject languages is distinct in several crucial ways from that in OC. We will return to the reason for these differences in Section 6. 5

Control with different verb types

Landau (2004) discusses five basic verb types and how they interact with nonfinite and finite control. To his list, in Table 3.3 we add ‘communicative verbs’ which differ from directive verbs in allowing only finite complements, and not triggering subjunctive morphology and dependent tense in their complement clause.

142

Anders Holmberg and Michelle Sheehan

Table 3.3. Landau’s (2004) verb types (amended) Verb type

English examples

OC pattern

commissive communicative epistemic directive

declare, promise, vow, threaten inform, tell, point out, say think, add, state tell, ask, convince, persuade, encourage be glad, regret, know need, want

Always subject control, even across object No non-finite complements No non-finite complements Object control

factive desiderative

Subject control, no matrix object ECM

In the following sections we describe the distribution of null subjects with these verb types in Marathi, Finnish and BP, as compared with the distribution of PRO and finite control in Hebrew. Again, some marked differences emerge. 5.1

Epistemic verbs

All three partial null-subject languages allow subject control into complements of verbs of saying, thinking and perceiving, the groups of verbs which Landau (2004) terms epistemic: (45)

a. O Joa˜o1 disse que (ele1) tinha comprado uma casa. the Joa˜o said that he had bought a house ‘Joa˜o said that he had bought a house.’ b. Marja1 luulee etta¨ (ha¨n1).on hyva¨n na¨ko¨inen. Marja thinks that she is good looking c. Seema1 kabul karte ki (ti-chya-ni1) chuk dzali Seema agree does that she-ERG mistake made ‘Seema admits that she made a mistake.’

Note that epistemic verbs do not typically take non-finite complements and so no comparison with non-finite control can be made: (46)

*John believed/stated [PRO to leave]

In Hebrew and the other languages discussed by Landau (2004) finite control is only marginally possible with these verbs, as they do not select complement clauses with dependent tense. 5.2

Directive verbs

Several directive verbs in Finnish take a finite complement (e.g. ehdottaa ‘propose’, suositella ‘recommend’, sanoa ‘tell (someone to do something)’). In such cases, object control of a null embedded subject is possible (somewhat marginally).

Control into finite clauses in partial null-subject languages (47)

143

Jari1 ehdotti Tarjalle2 [etta¨ (ha¨n*1/2) ostaa uuden telkkarin]. Jari suggested Tarja-ALL that she buys new TV ‘John suggested to Tarja that she should buy a new TV.’

In BP, directive verbs most naturally take a non-finite complement, with a controlled PRO, as in (48b). However, finite subjunctive complements are also possible in formal registers, and in such case a null subject is possible, as in (48a). (48)

a. A Maria sugeriu ao Jose´1 que (ele1) fosse embora. the Maria suggested to.the Joa˜o that he go.IMP./SUBJ away ‘Maria suggested to Joa˜o that he go away.’ b. A Maria sugeriu pro Jose´ PRO ir embora the Maria suggested for.the Jose´ go.INF away ‘Maria suggested for Jose´ to go away.’

In Marathi, too, directive verbs normally take non-finite complements, but can be somewhat marginally construed with a finite subjunctive complement. To the extent that finite subjunctive complements are possible in Marathi, null subjects are banned: (49)

Ram-ni Arun-la shikawlə ki *(tya-ni) mothyaan-cha aader karavə Ram-ERG Arun-ACC taught that he-ERG elders-of respect do-SUBJ ‘Ram taught Arun that he should respect the elders.’

This is a context where Hebrew happily allows object control, provided that the embedded clause is temporally dependent (future relative to the time of the matrix clause) and generally has a subjunctive-like dependent relation to the matrix clause (Landau 2004; see also Gutman 2004). Marathi simply seems to lack the kind of finite control described by Landau, whereas BP and Finnish marginally allow it. In all three languages, control into the complement clauses of epistemic verbs is far better, which is the reverse of the situation in Hebrew and the other languages described by Landau (2004).

5.3

Subject control with commissive verbs

All three languages permit subject control with commissive verbs like declare, promise and vow, which also require subject control with PRO, as illustrated by the following English and BP examples: (50)

I1 promised John2 [PRO1/*2 to be there on time]

(51)

O Joa˜o prometeu a` Maria que iria embora The Joa˜o promised to.the Maria that would.go away ‘Joa˜o promised Maria that he would leave.’

144

Anders Holmberg and Michelle Sheehan

5.4

Subject control with factive verbs

Both BP and Finnish allow subject control into the complement of a factive verb. Marathi, however, does not. We return to this contrast in Section 6.3, claiming that it is a side effect of the fact that Marathi fails to allow control into DPs more generally, assuming factive complements to be covert DPs. 5.5

Verbs of communication

The most interesting contrast between BP and Finnish on the one hand, and Marathi on the other hand, comes from what we term communicative verbs (‘tell (somebody something)’, ‘inform’, ‘remind’). With such verbs Marathi allows subject and object control, somewhat surprisingly. In BP and Finnish this is a marginal possibility, at best. (52)

a. John-ni1 Mary-la kalav-l-ə ki 1/ to1/2 parikshet pas John-ERG Mary-dat informed-PST-3SN that he exam-in pass dza-l-a happen-PST-3SM ‘John informed Mary that he passed the test.’ b. John-ni Mary-la1 kalav-l-ə ki 1/ ti1/2 parikshet pas John-ERG Mary-dat inform-PST-3SN that she exam-in pass dza-l-i happen-PST-3SF ‘John informed Mary that she passed the test.’ c. Lucy-ni1 Mary-la2 kalav-l-ə ki *1/2 / ti1/2/3 parikshet Lucy-ERG Mary- DAT inform-PST-3SN that she exam-in pas dza-l-i pass happen-PST-3SF ‘Lucy informed Mary2 that she2 passed the exam.’

In (52a) the embedded verb is inflected for masculine, consequently only the subject is a possible controller of the null subject. In (b) the inflected verb is feminine, and the only possible controller is the matrix object. In (c) there are two c-commanding DPs with features compatible with those of the null subject. In this case the closer one, that is the object, is the antecedent. This shows that in Marathi, the class of potential interveners is, in this case, relativised to those which are phi-compatible. Neither Finnish nor BP mark gender on verbal agreement morphology, but a similar effect does hold in Finnish, with plural morphology. In Finnish, the subject is the preferred controller in the case where the embedded verb inflection is compatible with both subject and object control: see (53a). However, object control is possible when the inflection on the embedded verb is incompatible with subject control, as in (53b). An alternative reading, in this case, is the split antecedent reading.

Control into finite clauses in partial null-subject languages (53)

145

a. Pekka1 muistutti Juhania2 etta¨ 1/?2 oli luvannut leikata nurmikkoa. Pekka reminded John that had promised mow lawn ‘Pekka reminded John that he had promised to mow the lawn.’ b. Pekka1 muistutti lapsia2 etta¨ *1/2/1þ2 olivat luvanneet leikata Pekka reminded children that had-3PL promised-PL mow nurmikkoa. lawn ‘Pekka reminded the children that they (the children or him and the children) had promised to mow the lawn.’

BP exhibits a slightly different situation: subject control is preferred but split antecedents are possible when the agreement morphology is incompatible with the subject. In most circumstances an in-situ object is not a good antecedent (even if a predicate is phi-matched with the object but not the subject): (54)

a. O Pedro1 convenceu o Joa˜o2 que 1/*2 tinha

que ir embora. (Modesto 2000a) the Pedro convinced the Joa˜o that had.3S that go away ‘Pedro1 convinced Joa˜o that he1 had to leave.’ b. *O Pedro1 convenceu a Maria2 que 2 era esperta the Pedro convinced the Maria that was smart.F ‘Pedro convinced Maria that she was smart.’ c. O Ze´1 convenceu os meninos2 que *1/?2/1þ2 tinham que ir embora the Ze convinced the kids that had.3PL that go away ‘Ze convinced the kids that they had to leave.’

An interesting observation, due to Modesto (2000a; 2008) is that object control is preferred in BP, too, if the object is wh-moved or topicalised. (55)

Quem2 que o Pedro1 convenceu t2 que EC2 tinha que ir embora? Who1 that the Pedro2 convinced that had that go away? ‘Who did Pedro convince that he had to leave?

As discussed by Modesto (2008), Finnish exhibits a similar, though not identical, situation: object control improves if the object is wh-moved or topicalised, but only marginally. (56)

?Keta¨1 Pekka muistutti t1 tetta¨ (ha¨n1) oli luvannut leikata nurmikkoa. who Pekka reminded that he had promised mow lawn ‘Who1 did Pekka remind that he1 had promised to mow the lawn?’

Thus in Marathi, the object is the preferred controller of an embedded null subject with a verb of communication, whereas in BP and Finnish, the subject is the preferred binder, unless the object is fronted, in which case it too becomes a possible binder. An interesting possibility is that this is related to an obvious parametric difference between Marathi on the one hand, and BP and Finnish on

146

Anders Holmberg and Michelle Sheehan

Table 3.4. The possibility of null subjects with different verb types Verb type

pro control pattern in Marathi

pro control pattern in BP and Finnish

PRO control pattern

Commissive (promise) Epistemic (say)

Subject control across object Subject control

Subject control across object Subject control

Factive (regret)

Subject control impossible No finite control

Subject control

Subject control, across object No non-finite complements Subject control

No finite control

ECM

Object control impossible Subject and object control possible

Object control marginally possible Subject control only, unless object is fronted

Object control

Desiderative (want) Directive (recommend) Communicative (inform)

No non-finite complements

the other: OV vs. VO (see Nayudu 2008). Two assumptions are required: (a) clausal complements of this class of verbs are adjuncts (Ferreira 2004), and (b) OV order is derived by object movement out of VP (Zwart 1994; 1997). The result is that, in the unmarked case, the object will c-command the clausal complement (from its derived preverbal position) in the OV language Marathi, but not in the VO languages BP and Finnish. For this reason, the object is the preferred binder in Marathi, but not in BP or Finnish. This is supported by Modesto’s observation concerning object-movement in BP and Finnish: when the object moves out of VP, it becomes the preferred controller in BP and Finnish, too. There is considerable evidence for the idea that the embedded CP is an adjunct, at least in Finnish and Brazilian Portuguese, from the behaviour of epithet subjects, extraction facts, clausal expletive insertion (cf. Rodrigues (2004) for an overview).18 This is a clear example of parametric interaction. The OV status of Marathi means that its object occupies a higher position in the clause and as such has the possibility to bind where objects in VO languages do not. 5.6

Summary of verb type facts

Once again, it is obvious from Table 3.4 that once the OV interaction is controlled for, the control relation in the three languages under discussion forms a class, distinct from the phenomenon of OC.

18

This obviously raises a number of questions. Perhaps the most pressing one is how it is possible for the object to control the null subject in (53b) and (more marginally) (54c)? We will leave this question and other related questions for future research.

Control into finite clauses in partial null-subject languages

6

147

Island sensitivity

The biggest differences between the three languages arise in relation to island effects. While all three languages allow bound null subjects in adjuncts and embedded questions, Marathi fails to allow null subjects in any kind of DP. Finally, only Finnish allows bound null subjects in relative clauses. We include this information here for completeness, though the reason for these differences remain opaque to us at present. 6.1

Adjunct clauses

All three languages (SUBJ¼subjunctive).

allow

control

into

finite

adjunct

clauses

(57)

a. O Joa˜o pode vir, desde que (ele) termine o trabalho. The John may come, since that he finishes.SUBJ the job ‘John can come, if he finishes the job (first).’ b. Eeva saa tulla mukaan jos (ha¨n) lupaa olla hiljaa. Eeva may come along if she promises be quiet c. John khush hota karan (tya-la) pushkar bheti milyala John happy be-past3SM because (he-ACC) very gifts receive-past3PLF ‘John was happy because he received many gifts.’

6.2

Indirect questions

BP, Finnish and Marathi allow control into embedded questions, at least marginally (TRA ¼ translative).19 (58)

a. O Joa˜o perguntou se (ele) podia dormir aqui. The John asked if he could sleep here ‘John asked if he could stay the night.’ b. Jari haluaa tieta¨a¨ saako (ha¨n) ja¨a¨da¨ yo¨ksi. Jari wants know can-Q he stay night-TRA ‘Jari wants to know whether he could stay the night.’ c. John-ni vicharle ki (to) ratri rahu shakto ka? John-ERG ask-past-3SN that (he) night stay happen-PRS-3SM Q ‘John asked whether he could stay the night.’

6.3

Noun complements and factive clauses

BP and Finnish both allow control into the finite clausal complement of a noun selected by a verb, Marathi does not. (60b) exemplifes a very common

19

Control into indirect questions is possibly less marginal in BP than in Finnish and Marathi.

148

Anders Holmberg and Michelle Sheehan

form of verb complementation in Finnish, where the complement is headed by the pronoun se ‘it’ (appropriately inflected) (ABL ¼ ablative). (59)

a. O Joa˜o1 se esquece do fato de que (ele)1 vai ganhar menos no The John SE forgets of.the fact of that will earn less in.the novo emprego. new job ‘John forgets about the fact that he’ll earn less in his new job.’ b. O presidente1 negou os rumores de que (ele)1 tinha recebido The president denied the rumors of that he had received dinheiro de empresa´rios money from businessmen. ‘The president denied the rumours that he had received money from businessmen.’ (example from Modesto (2000a: 99))

(60)

a. Anu ei usko va¨itteeseen etta¨ (ha¨n) olisi maksanut autostaan Anu not believes claim that she has-CON paid car-ABL-her liikaa. too-much ‘Anu doesn’t believe the claim that she would have paid too much for her car.’ b. Jari valitti meille siita¨ etta¨ (ha¨n) joutui maksamaan autostaan Jari complained us-ALL it-ABL that he had-to pay car-ABL-his liikaa. too-much ‘Jari complained to us about the fact that he had to pay too much for his car.’

(61)

John-la [hi goshta ki *(tyan-ni) ti-la dukhavlə] mahiti nahvti John-ACC the fact that he-ERG she-ACC offended was-aware not ‘John wasn’t aware of the fact that he had offended her.’

Since DP objects undergo fronting to pre-VP position, deriving SOV order, in Marathi but not in BP or Finnish (see Nayudu 2008), an initially attractive idea is that this fronting is the reason why control into noun clauses is impossible in Marathi, unlike BP and Finnish. However, another difference is that Marathi also does not allow control into finite factive clauses, while BP and Finnish do. (62)

a. A Maria1 sente muito que (ela1) tenha chegado/chegou tarde. the Maria feels a lot that she has-SUBJ arrived/arrived late ‘Maria regrets that she has arrived late.’ b. Jari1 katuu etta¨ (ha¨n1) tuli takaisin. Jari regrets that he came back c. Ram-la pashchyatap hotə ki *(to) parat aala Ram-ACC regret was that he back came ‘Ram regretted that he came back.’

It seems fairly uncontroversial to assume that factive clauses are underlying nominal clauses, headed by an abstract noun meaning ‘fact’. If so, the correct generalisation is that Marathi fails to allow a controlled null subject in any

Control into finite clauses in partial null-subject languages

149

finite nominal clause. Factive clauses as in (62c) are not fronted, though, so the fronting as such would not be crucial.

6.4

Relative clauses

Of the three languages, only Finnish allows control into relative clauses. (63)

a. Pekka rikkoi maljan jonka (ha¨n) oli saanut lahjaksi sukulaisiltaan. Pekka broke vase which (he) had got gift-TRA relatives-ABL ‘Pekka broke the vase that he had got as a gift from his relatives.’ b. John-ni1 masale tya sauce-madhe misalavale dzo *(tyan-ni)1 John-ERG herbs that sauce-in mixed that he-ERG aadhi-cha kelela. earlier-EMPH prepared ‘John mixed the herbs in the sauce that he had prepared earlier.’

In the case of Marathi this falls under the generalisation that control into nominal complements is not allowed. For BP there must be a different explanation, though, since BP allows control into nominal complements. 6.5

Summary of island facts

The languages in question differ with respect to their degree of island sensitivity (see Table 3.5). Marathi is more sensitive to islands than the other two languages, in particular to all kinds of ‘complex NP’ (or DP) islands. Finnish displays no island sensitivity at all. If both OC and the control relation in partial null-subject languages is an effect of Agree, then the island sensitivity in BP and Marathi must be due to intervention or to some independent fact about their grammar. We leave these matters to future research.

7

The control typology

There are differences among the three languages as regards control of a null subject in embedded finite clauses. Marathi, for some reason, does not allow control into any type of DP. Furthermore, Marathi prefers object control in at least one case where BP and Finnish do not. However, we have argued that at least the second of these differences can be ascribed to the OV nature of Marathi (and we mentioned the possibility that the first one could, as well). The general impression is, then, that we are dealing with variations on a theme: control of a null subject in finite clauses is essentially the same phenomenon, subject to the same constraints, in the three languages, with variations that are (or at least some of them) due to independent parametric differences among the languages.

150

Anders Holmberg and Michelle Sheehan

Table 3.5. Island sensitivity in partial null subjects and with obligatory control

Context

Pro in Marathi

Pro in Brazilian Portuguese

Grammatical in embedded questions Grammatical in adjunct CPs Factive CP complements

Y

Y

Y

Y

Y – subject only N

Y – subject only Y

Y – subject only Y

Y

N

Y

Y

N

N

Y

Grammatical in nominal complements Grammatical in object rel. clauses

Pro in Finnish

PRO in English

Y John regretted having cut himself/*oneself Y J had the desire to hurt himself/ *oneself Y – every man wants something to devote himself to

A comparison with control of PRO in non-finite clauses yields an unambiguous conclusion: control of a null subject in finite clauses in BP, Finnish and Marathi is not OC. The locality conditions are not as strict as for OC, the behaviour with different verb types is different, the possibility of split antecedents is incompatible with OC, as is the strict reading under ellipsis and the coreferential reading in the only-construction (at least in Finnish and Marathi). Since we have proposed that the control relation in BP, Finnish and Marathi is also due to Agree, it remains to be clarified, therefore, how these manifold differences arise. Landau (2004) proposes a control calculus to derive the types of clauses which can license –R pronouns like PRO. He argues that PRO is restricted to environments which lack þT and þAgr after checking between I and C. This has the effect of restricting PRO to non-finite and dependent tense subjunctive clauses. The distribution of uD-pronouns in partial null-subject languages cannot be captured by this control calculus, as they specifically do not occur in the contexts where PRO occurs in other languages. As such, they have the distribution of normal þR pronouns, despite the fact that they need to be controlled. In OC, following Landau, the control relation is essentially Agree between a –R pronoun (PRO) and an antecedent, but crucially this Agree relation is mediated by a head F (v or T) in the higher clause. This means that the

Control into finite clauses in partial null-subject languages

151

relation is very local (strictly ‘next clause up’), as every clause has v or T.20 The type of relation we are proposing, holding between a [uD] null subject in a finite clause and its antecedent in our three partial null-subject languages, is in many ways similar to that proposed by Landau. The uD-feature of the null pronoun, which makes it incapable of independent reference, needs to enter into an Agree relation with a DP, making it similar to Landau’s –R pronouns. Crucially, though, a [uD] pronoun agrees directly with its antecedent (and not via an intermediate head F) and is thus subject to c-command and intervention effects. How does Agree manage to cross the CP phase boundary, though (avoiding the Phase Impenetrability Condition)? The logic of phase theory dictates that the null pronoun moves to the edge of the CP phase, where it will be visible to the DP arguments in the next clause/next phase. However, this still predicts stricter locality than we actually find. Recall that all three languages allowed, at least marginally, the null subject to be controlled by an antecedent two clauses up, provided that there was no intervening DP (see (26a), repeated here as (64)): (64)

A Maria1 disse [que e´ verdade [que (ela)1 entornou o copo]] (BP) The Maria said that is true that she knocked.over the glass ‘Maria said it’s true that she knocked over the glass.’

Successive cyclic movement of the null pronoun is not an attractive solution for at least two reasons. One is that control is possible into finite clauses which are strictly islands for movement: adjunct clauses (all languages), DP-embedded clauses (BP and Finnish) and even relative clauses (Finnish). The other is that a movement account fails to predict the effect of an intervening DP (see (24a), repeated here as (65)): (65)

O Joa˜o1 disse [que os moleques acham [que *(ele)1 e´ esperto]] the Joa˜o said that the kids think that he is smart ‘Joa˜o said that the children think that *he/ is smart.’

Successive cyclic A-bar movement is not hindered by intervening subjects or objects. We are led to acknowledge that spell-out of the uD-marked pronoun at the edge of CP1 in (66) can wait until a DP is merged in CP3, provided that there is no c-commanding DP in CP2. (66)

20

[CP3 . . . DP . . . [CP2 . . . [CP1 [uD] C TP ]]]

In Landau’s (2000; 2004) terms, control is possible across a phase-boundary because the –R feature of PRO is interpretable and PRO is thus ‘visible to Agree from the outside’ (Landau 2004: 843, fn 26).

152

Anders Holmberg and Michelle Sheehan

How to accommodate this fact within a phase-based theory of locality is something we will have to leave for future research. 8

Conclusions

BP, Finnish and Marathi, as partial null-subject languages, allow null subjects: (i) with non thematic predicates, (ii) with a generic interpretation and (iii) in embedded contexts. Type (ii) are the result of incorporation of a nonreferential pronoun into T, and type (iii) are licensed under Agree with a c-commanding DP. These uD-pronouns are not restricted to clauses with dependent tense, and hence their distribution is not regulated by Landau’s control calculus. Rather, their grammaticality is regulated only by the requirement that they be bound by the closest c-commanding antecedent. Marathi for some reason, does not allow these null subjects in DPs, and BP fails to allow null subjects in relative clauses, but apart from this the licensing of these null subjects is not subject to island effects, as is the case with OC, which is also an agreement relation. Differences between OC and the control relation described here stem from the fact that OC is mediated by the heads v and T, whereas the binding of null subjects in partial null-subject languages is not.

4

Semi null-subject languages, expletives and expletive pro reconsidered* Theresa Biberauer

1

Introduction

The focus of this chapter is a (putative) class of null-subject languages often referred to as semi null-subject or semi pro-drop languages (cf. i.a. Platzack 1985/1987; Koster 1987; Gilligan 1987; Grewendorf 1989; Hermon & Yoon 1989; Falk 1993a; Huang 1994). As Huang (2000: 53) notes, there is some inconsistency in the use of the designation semi NSL as it is sometimes equated with the more general notion of ‘restricted pro-drop’, i.e. pro-drop which differs from that found in Italian-style full NSLs in permitting fewer null-subject options, regardless of the type of restriction. Accordingly, Huang (2000) proposes three sub-types of semi NSLs. These are given in slightly adapted form in (1): (1)

*

a. those which only permit non-argumental expletive omission, i.e. omission of ‘true’ expletives, which do not bear a f-role (arguments of weather verbs, etc. are therefore excluded and must be overtly realised) – e.g. Dutch, German; b. those which permit both non-argumental expletive omission and omission of so-called quasi-arguments (i.e. arguments of weather verbs, etc.) – e.g. Icelandic, Yiddish; c. those which permit omission of non-argumental expletives, quasiarguments and, under certain restricted circumstances, referential subjects – e.g. Finnish, Hebrew, Bavarian German, Ukrainian.

This chapter builds on research that has, in various forms, been underway since 2000. Grateful thanks to the following for their input: the audiences at the LAGB Autumn Meetings in Reading (September 2001), Oxford (September 2003) and Roehampton (September 2004), the Joint FGLS/SGL Meeting in London (January 2003), the Workshop on Null Subjects in Cambridge (February 2003), the Cambridge University Linguistic Society (March 2003), the Comparative Germanic Syntax Workshop (June 2004), the Minimalist Inquiries Workshop (June 2004), the London Linguistics Circle (December 2004), the University of California at Santa Cruz Linguistics Colloquium (May 2005), the Workshop on Verb Clusters and Co-ordination in Leiden (November 2005), the Edges in Syntax Conference (May 2006), the Cambridge–Nanzan Syntax Colloquium (May 2007); fellow members of the project; and in Cambridge, particularly Adam Ledgeway, who first got me thinking about the EPP, and Marc Richards, for many hours of discussion. Needless to say, none of these people are responsible for the errors that remain.

153

154

Theresa Biberauer

Our concern in this chapter is exclusively with (1a,b). This is principally for two reasons. First, recent research has shown that (1c) is extremely unlikely to constitute a single (sub-)type.1 Second, (1a,b), but not (1c) featured in the original NSP typology proposed by Rizzi (1986a; cf. also Travis 1984; Platzack 1985/1987) and given in (2): (2)

a. Full NSL: licenses both referential and non-referential null pronominals (here: proREF and proEXP) – Italian, Spanish, Greek. b. Semi NSL Type I: only licenses null non-referential pronominals, i.e. quasi-argumental (here: proEXPþA) and non-argumental (here: proEXPA) expletives – Icelandic, Yiddish. c. Semi NSL Type II: only licenses null non-argumental pronominals (proEXPA), but not referential or quasi-argumental expletives – Dutch, German. d. Non-NSL: does not license null pronominals (pro) at all – e.g. English, French.

As is well known, classification as one of the types in (2) was predicted to depend on inflectional richness, with languages exhibiting systematically morphologically encoded distinctions between different persons and numbers being expected to license all species of pro as in (2a), while those with less systematically distinct encodings permit only one or more of the expletive pro options in (2b,c), and languages (almost) completely lacking morphologically encoded person and number distinctions do not sanction any form of pro, as in (2d). More specifically, Rizzi (1986a) proposed that (a syntactic) person specification on I was crucial for the licensing of proREF, while a lesser specification, (syntactic) number, was essential for the licensing of proEXPþA, languages permitting just proEXPA were, in turn, thought to lack both (syntactic) person and number specifications, featuring only the basic [pronominal]2 specification distinguishing NSLs from non-NSLs (cf. Rizzi 1982).3 This system is schematised in (3): 1

2

3

See Holmberg (2005; Chapter 2), Holmberg & Sheehan (Chapter 3), Roberts & Holmberg (Introduction), Modesto (2008) and Shlonsky (2009) for discussion of the properties of so-called partial NSLs. On Bavarian German, see Fuss (2005). On subject drop in Slavic, see Mu¨ller (2005; 2007) and McShane (2009). For discussion of the danger of viewing all unrealised subjects as manifestations of a single underlying structure/element, see Biberauer (2008a) and Section 4 below. Obviously, [pronominal] here should not be equated with the [þpronominal] specification assumed to be universally common to pronominal elements in the context of GB Binding theory. Instead, this property should be viewed either (a) as an indication of the fact that agreement inflections are independently stored in a manner that they are not in non-NSLs (cf. Alexiadou & Anagnostopoulou (1998) and Bobaljik & Thra´insson (1998) on ‘pronominal Agr’), or (b) as an indication of the fact that I/T in NSLs bears a special nominal specification that is absent in non-NSLs (cf. Rizzi (1982); Holmberg (2005; Chapter 2); Roberts (Chapter 1)). Syntactic is emphasised in each case to highlight the fact that the features intended here are visible to the computational system (i.e. in Narrow Syntax), on a par with I’s [pronominal]

Semi null-subject languages, expletives and expletive pro reconsidered 155 (3)

a. b. c. d.

I specifiation: [þpronominal, þperson, þnumber] ! proREF I specification: [þpronominal, þnumber] ! proEXPþA I specification: [þpronominal] ! proEXPA I specification: [–pronominal] ! no pro

It is clear that we can think of (3) as a small-scale parametric hierarchy of the kind proposed by Baker (2001) and more recently discussed by Gianollo, Guardiano & Longobardi (2008) and Roberts & Holmberg (Introduction) – cf. (4): (4)

I [pronominal]? NO STOP

(i.e. is there ‘rich’ agreement?)

YES [Number] on I?

(i.e. is number distinctively marked?)

(English, French) NO STOP

YES [Person] on I?

(i.e. is person distinctively marked?)

(German, Dutch) NO STOP (Icelandic, Yiddish)

YES (Italian, etc.)

The difficulties associated with the inflectional richness aspect of the system have been much discussed (cf. i.a. Jaeggli & Safir (1989) and Mu¨ller (2005; 2007) and Tamburelli (2006) for more recent discussion) and will not be our primary focus here. Our chief concern will be the difficulties that Rizzi’s expletive-related predictions have encountered. Evidently, the implicational relations noted above say nothing about the possibility of a system having both overt and covert expletives; in reality, however, a great many languages appear to allow for precisely this phenomenon. In the case of the Germanic languages investigated by Rizzi and others, where expletives are systematically realised in clause-initial position, but fail to be realised clause-internally, an explanation appealing to parameter interaction became standard during the GB era: because the languages in question are all verb-second (V2) languages, the positive setting of the V2 parameter rules out the possibility of null expletives in clause-initial position. Leaving aside the complications associated with an specification; they are not just features that are distinctively realised morphologically. As observed in footnote 2, the relevant ‘visibility’ might possibly be ensured by [person] and/or [number] features being bundled together as part of a distinct ‘pronominal agreement’ featurebundle, although this still raises the question of how the computational system would ‘see’ the difference between an NSL system with both [person] and [number] specification and one with just [number] specification. To deal with this problem, Mu¨ller (2005; 2007) proposes a presyntactic Impoverishment operation, the output of which signals to the computational system how ‘rich’ the agreement system in question is. We leave the precise mechanics of Rizzi’s proposal aside here.

156

Theresa Biberauer

account of this type (see Section 2.1 for further discussion), it is not so clear how parameter interaction can account for systems in which overt and null expletives alternate in the same environments. As we will see in Section 3.3, certain varieties of Dutch and Afrikaans and Faroese also seem to exhibit this phenomenon; looking beyond Germanic, we see that it is also attested in various creoles (cf. Nicolis 2005; 2008) and in Finnish (Holmberg 2005). Clearly, this needs to be accounted for. Furthermore, the system represented in (3,4) says nothing about the fact that the various instantiations of proEXPA do not necessarily exhibit uniform behaviour in all the languages in which they occur.4 Thus, for example, it may occur in impersonal passives in Afrikaans, while it may not in presentationals/ existentials. This also requires explanation. What these complications suggest is that understanding of the availability of expletive pro and, by extension, of semi NSLs, requires a theory that extends beyond the agreement properties of I. This chapter aims to highlight some of the considerations that a theory of this type would need to take into account and also reassesses the importance of I’s (syntactically encoded) agreement properties. Empirical concerns aside, there are also theoretical considerations that have arisen within the context of the Minimalist Program that potentially compromise the system illustrated in (3,4). On the one hand, there is, as noted elsewhere in this volume, an influential minimalist analysis of full NSLs, advocated notably by Barbosa (1995) and Alexiadou & Anagnostopoulou (1998; henceforth: A&A), in terms of which V-to-I/T movement in full NSLs suffices to satisfy the requirement that I attract the subject into its local domain (cf. the Extended Projection Principle/EPP of Chomsky 1981). If this is indeed the case, there is no need to postulate projection of SpecTP, at least in cases where all the verb’s arguments are present, and SpecTP would only have been projected to host an EPP-satisfying proEXPA5 This, in turn, implies that proEXP–A may not in fact be a lexical item available to full NSLs, which obviously raises questions about the existence of a lexical item of this type in semi and other non-full NSLs. A second theoretical issue that has come to

4

5

As Rizzi’s semi NSLs exhibit currently still only very poorly understood language-specific variation in respect of the realisation of expletives in raising structures, we leave the behaviour of this expletive aside here (cf. Cardinaletti (1990: 93ff) on ‘external argument es’ and Vikner (1995: chapter 7) for discussion). We also omit discussion of the ‘anticipatory expletive’ in structures like It is possible that he will lose on the grounds that this element may in fact be argumental (cf. Bennis 1986 and Vikner 1995: chapter 7 for discussion). A&A (1998: 531ff) acknowledge that the situation may be more complex in cases traditionally analysed as involving proREF, where the absence of a subject pronominal would necessarily have implications for, among other things, one’s theory of y-role assignment. The same holds for ‘weather’ arguments, i.e. proEXPþA.

Semi null-subject languages, expletives and expletive pro reconsidered 157

the fore during the minimalist era is the question of the universality of the EPP (cf. i.a. Carnie & Guilfoyle 2000; Davies & Dubinsky 2001; Svenonius 2002a; Richards & Biberauer 2005). Whereas projection of a canonical subject position (here: SpecTP) was widely assumed to be a universal feature of natural languages during the GB period, changes in theoretical assumptions about the factors determining movement and specifier projection render this earlier assumption highly suspicious in the context of minimalism (cf. also Epstein & Seely (2006) for detailed discussion). If SpecTP does not in fact have to be projected in all languages, not just because there is an alternative means of satisfying the EPP (cf. full NSLs on the Barbosa/A&A view), but because this requirement may not even hold in some languages, it is clear that questions must once again be asked about the necessity of postulating proEXPA. Doing so is one of the aims of this chapter. To summarise, then, this chapter has two major aims: first, to reconsider the Germanic languages which were initially cited as canonical semi NSLs in order to establish whether their null-subject behaviour does indeed require the postulation of expletive pro (proEXP); and second, to re-evaluate the status of semi NSLs in Rizzi’s original typology. The chapter is structured as follows: Section 2 introduces the subjectrelated phenomena that led to German and Icelandic becoming established as canonical semi NSLs, and also presents some theoretical developments that call for a reconsideration both of this classification and of the very notion ‘semi NSL’. Section 3 revisits the Germanic facts, highlighting several empirical points that call the GB analysis into question and arguing for a genuinely minimalist analysis of the relevant phenomena. Section 4 concludes by briefly considering the implications of the proposed analyses for proExp and the notion ‘semi NSL’ more generally.

2

Background

2.1

Canonical semi NSLs: the case of German and Icelandic

In this section, we will briefly present the data on the basis of which Icelandic and German became established as canonical Type I and Type II semi NSLs respectively. These are of two main types, namely data relating to the absence of overtly realised expletives and data featuring non-expletive subjects that appear not to have undergone raising to SpecTP. Let us first consider the expletive distribution data. Being V2 languages, both Icelandic and German require overtly realised expletives in clause-initial position: in neutral declaratives in which no XP has been focused or topicalised, expletives fill the preverbal slot, which we will assume to be SpecCP

158

Theresa Biberauer

here (cf. Schwartz & Vikner (1996) for detailed discussion of the categorial status of V2 clauses).6 This is illustrated in (5a) for the Icelandic ‘weather’/ ‘ambient’ expletive and in (5b) for a German presentational structure: (5)

a. Það rigndi ´ı gær it rained yesterday ‘It rained yesterday.’ b. Es ist heute ja doch ein Brief gekommen it7 is today MOD.PART.8 a letter come ‘There did after all come a letter today.’

(Icelandic)

(German)

By contrast, neither Icelandic nor German permit overtly realised expletives in the non-neutral counterparts of these structures. This is shown in (6): (6)

a. I´ gær rigndi (*það) yesterday rained it ‘Yesterday it rained’ b. Heute kam (*es) ja doch ein Brief today came it MOD.PART. a letter

Assuming the post-verbal position in structures like (6) to be SpecTP (cf. footnote 6) and also the validity of the EPP as originally formulated in Chomsky (1981; 1982) – every clause must host a subject in SpecTP – structures like those in (6) can be taken as evidence that Icelandic and German license proEXP. Referential subjects located in this position cannot, however, be null, as German (7) shows: (7)

6

7

8

a. Er las das Buch he read the book ‘He read the book.’ b. Das Buch las *(er) the book read he

Because Icelandic (like Yiddish) exhibits V2 in both main and embedded clauses, there is some controversy in the literature regarding the structure of Icelandic V2 clauses (cf. Schwartz & Vikner (1996) for an overview of some of the relevant issues). For reasons which will become clear in Section 3.3 below, we will assume that Icelandic V2 clauses, just like V2 clauses more generally, are CPs (cf. also Richards & Biberauer 2005). Leaving aside the possibility that CP may in fact be an articulated structure (cf. Rizzi 1997), this entails that the post-verbal position in main clauses will be SpecTP, while the post-complementiser position in declarative embedded clauses will be SpecCP (cf. Vikner (1995), Schwartz & Vikner (1996) and Biberauer (2003) for further discussion of CP-recursion in V2 languages). Like Icelandic, Faroese and Yiddish, German employs only a single expletive form in all expletive contexts, namely the neuter pronoun. Literal glosses are given throughout to reflect this fact. Abbreviations employed in this chapter are as follows: MOD.PART. ¼ modal particle; NOM ¼ nominative; DAT ¼ dative; PL ¼ plural; ACC ¼ accusative.

Semi null-subject languages, expletives and expletive pro reconsidered 159

Icelandic and German therefore clearly differ from canonical Romance NSLs not only in respect of the positional considerations determining the availability of null subjects, a property which might be ascribable to the positive setting of a V2 parameter in the former, but not the latter, but also in respect of the type of null subjects that appear to be possible. This latter property cannot straightforwardly be related to a V2 parameter (although cf. Vikner (1995: 57–64) for discussion of earlier approaches to V2 which attempted to do so; cf. also Holmberg & Platzack 1995). Given the muchdiscussed ‘richness’ of Icelandic agreement morphology – it regularly distinguishes five persons just like Romanian, which is an NSL (cf. Biberauer & Roberts (Chapter 7) for references and discussion) – it is also not clear from a morphological perspective why Icelandic at least should not be a full NSL of the canonical Romance type. We return to this issue in section 3.4.3 (cf. also Biberauer & Roberts (Chapter 7) and Holmberg (Chapter 2)). Turning to the empirical facts distinguishing Icelandic and German as, respectively, Type I and Type II semi NSLs, the crucial data are the following: (8)

a. Es schneit heute it snows today ‘It is snowing today.’ b. Heute schneit *(es) today snows it ‘Today it is snowing.’

(German)

As comparison with Icelandic (6a) shows, German differs from Icelandic in consistently requiring an overtly realised ‘weather’ expletive.9 Expletives of this type have been convincingly shown to be quasi-argumental (i.e. nonreferential, but nevertheless y-role-bearing; cf. i.a. Bennis (1986), Cardinaletti (1990) and Vikner (1995: chapter 5) for discussion, and see Bolinger (1977) for insightful discussion of the range of contexts in which these elements surface); therefore it would appear that the difference between Icelandic and German is that the former licenses proEXPþA, while the latter does not. Turning to referential subject distribution, we seem to see further evidence that both languages license proEXPA: on the assumption that (non-argumental) expletives serve to satisfy the EPP wherever subjects fail to raise to SpecTP, the unraised subjects in Icelandic and German structures such as the following

9

Mohr (2005: 178) notes that some speakers of Icelandic do permit það to surface in SpecTP in ‘weather’ contexts. If this usage is consistent, the grammar of the relevant speakers is identical to that of standard German speakers in the relevant respect. We return to the question of optionality in Section 3.2 below.

160

Theresa Biberauer

can be understood as elements in an expletive-associate chain in the manner illustrated below: (9)

a. I´ gær voru proEXPA konunginum gefnir hestar (Icelandic) king-the-DAT given horses-NOM yesterday were-3PL ‘Yesterday horses were given to the king.’ (cf. Sigurðsson 1989) b. ´I eldhu´sinu hefur proEXPA alltaf verið svartur ko¨ttur in kitchen-the has always been black cat ‘In the kitchen, there has always been a black cat.’ (cf. Thra´insson 2007: 319)

(10)

a. . . . daß proEXP–A dem

Mann das

Buch geschenkt wurde (German) that the-DAT man the-NOM book presented became ‘. . . that the book was given to the man (as a present)’ b. . . . weil proEXPA ja doch Linguisten Kammermusik spielen MOD.PART linguists chamber-music play because ‘. . . since there are linguists playing chamber music’ (cf. Diesing 1992: 36)

To summarise, then, Icelandic and German appear to exhibit the same behaviour in respect of their ability to license proEXPA, whereas the ability to license proEXPþA is limited to Icelandic. As noted above, this difference was ascribed to inflectional differences between the two languages, Icelandic (five regular person and number distinctions in verbal paradigms) arguably being inflectionally richer than German (four regular distinctions). The positional restrictions on the licensing of pro in these languages was, in turn, related to their V2 status via a range of proposals in terms of which C’s association with features located on I in other languages, crucially including the Romance NSLs, entailed that C alone was a suitable licenser for pro. In V2 languages, then, pro could only be licensed in SpecTP, with the types of pro a language permitted being determined by its inflectional properties. In the following section, we review certain theoretical considerations which have arisen since the original postulation of the two types of semi NSLs discussed in this section.

2.2

Theoretical developments

As noted in Section 1, at least two developments during the minimalist era would seem to justify a reconsideration of the postulation of proEXP and the class of semi NSLs. The first of these relates to minimalist analyses of canonical and partial NSLs, and the second to the theory of phrasal projection

Semi null-subject languages, expletives and expletive pro reconsidered 161

and EPP-driven movement. Crucial to both developments is Chomsky’s (1995b) recasting of the original EPP, which will therefore be our point of departure here. 2.2.1 Null subjects in minimalism Since Chomsky (1995b), an influential view of the EPP has been that it can be decomposed into two T-related components, namely a nominal feature (D, j, etc.) and an associated movement trigger ([þstrong] in earlier approaches; an EPP-feature or Agree-related movement diacritic in more recent ones – cf. Chomsky 2000 and following). As Roberts & Holmberg note in the Introduction to this volume, the two major minimalist approaches to null subjects analyse these as being the consequence either of: (a) agreement inflection being ‘pronominal’ (i.e. a D-category), with what we might think of as VD-to-T movement thus being able to satisfy both components of T’s EPP-requirements in the relevant languages (cf. Borer’s original (1986) ‘I-subject’ intuition, and Barbosa and A&A’s more recent developments of this idea); or (b) pronouns being deletable under adjacency with T in these languages (cf. Holmberg (2005; Chapter 2), Roberts (Chapters 1, 8), Sheehan (Chapter 6)).10 Under (a)-type analyses, the obligatory projection of a TP-specifier no longer follows, with EPP-satisfaction being achieved via V(v)-to-T movement (henceforth: V-to-T), and specifier projection in principle depending on the presence of a further (possibly ‘free-standing’) movement trigger (cf. Biberauer & Roberts (Chapter 7) and Holmberg (Chapter 2) for discussion of this type of movement diacritic).11 A&A (1998: 513ff) explicitly highlight the fact that a specifierless approach to EPP-satisfaction in NSLs is advantageous in at least one context traditionally assumed to involve proEXPA-type null subjects. Consider (11) in this connection: 10

11

According to Neeleman & Szendro¨i (2007; 2008), null subjects in so-called ‘discourse pro-drop’ or ‘radical pro-drop’ languages like Chinese and Japanese may also be viewed as ordinary pronouns which are deleted at PF. On their view, the difference between canonical NSLs and the more permissive discourse type is simply that the former necessarily entails a context-sensitive deletion operation which references the (nominal) properties of T, whereas the latter applies independently of the properties of T. Operating with a theory that predates the postulation of ‘free-standing’ movement triggers, A&A do not consider the possibility that SpecTP might nevertheless be projected in languages that achieve EPP-satisfaction via V-to-T raising. For them, preverbal subjects therefore necessarily occupy a projection in the CP-domain, i.e. the type of position that is generally assumed to be an A0 -position. See Sheehan (2006; Chapter 6) for discussion of this position.

162

Theresa Biberauer

(11)

a. There remains a problem/*the problem/*every problem b. proEXPA irthe kapios fititis/ o Petros/to kathe pedi arrived some students/the Petros/the every child ‘Some students/Petros/every child arrived.’

(Greek)

As (11b) clearly shows, the putative null expletive – clearly some form of proEXPA (see below) – in NSLs seems to exhibit rather different properties to what appears to be its overt counterpart in non-NSLs. It does not, for example, appear to be sensitive to the Definiteness Effects found with there-type expletives in Germanic. Of itself, this may not be an argument against the postulation of a null expletive in structures of this type as overt expletives lacking sensitivity to Definiteness Effects are attested – consider i.a. Vallader i discussed by Taraldsen (2002), the Arabic clitic –hu cited in A&A (1998), and German es as discussed in i.a. Haider (1993) and Mohr (2005). To summarise, then, one type of minimalist approach to NSLs, the ‘I-subject’ approach outlined in (a) above, raises questions as to whether or not it is necessary to postulate proEXP in canonical NSLs and, if it is, what accounts for its presence (the EPP already having been satisfied via verb-raising) and also what kind of overt expletive(s) it might correspond to. The last of these questions is also crucial in the context of the second major minimalist approach to NSLs, the deletion approach described in (b) above. This approach views all pros and therefore also all proEXPs as the phonologically null counterparts of elements that could potentially be overtly realised. Recall that the deletion approach to null subjects entails SpecTP projection, with the difference between canonical NSLs and noncanonical NSLs being that the feature-composition of T in the former is such that pronouns which have undergone raising to SpecTP will be deleted at PF: in the former, T is assumed to bear nominal features which, in combination with its verbal specification, ensures that a raised weak pronoun (i.e. jP) will always constitute a featural subset of the features found on T, thus justifying pronoun-deletion in accordance with the assumptions more generally made about chains (cf. Roberts (Chapter 1) for detailed discussion). Crucially, Holmberg (2005) and Roberts (Chapter 1) postulate a D(efiniteness)-feature as the NSL-defining property of T, raising the question of how non-definite null subjects will ever arise in these languages. Holmberg (Chapter 2) addresses this issue by proposing that the NSL-defining property of T is in fact an unvalued D-feature, [uD]. This feature would then be expected to be compatible with both definite and indefinite subjects, with the deletion of pronominal versions of the former following from the presence of a null topic in the CP-domain, which is absent in structures featuring indefinite subjects (see Holmberg (Chapter 5) for more detailed discussion, also of generic subjects, which we leave aside

Semi null-subject languages, expletives and expletive pro reconsidered 163

here).12 In the specific case of expletive subjects, it would seem that [uD] on T can be valued either [þ/definite] or [–/indefinite] as T appears to agree with the associate (cf. (11b) above). On the assumption that expletives are always weak pronouns (jPs; cf. also Cardinaletti & Starke 1999: 175 and note 14), with ‘pure’ expletives possibly encoding only [Person] (cf. Chomsky 2000),13 the expectation is, then, that they will be consistently deleted, thereby accounting for the absence of overt expletives in canonical NSLs (but see below).The same analysis can be extended to partial NSLs, assuming expletives in these languages are also jPs. To ensure that expletives are overtly realised in non-NSLs, the proposal then has to be that these expletives differ from their covert counterparts in being DPs – specifically, DPs with D valued [–definite] – which will therefore always encode a superset of the features present on non-NSL T.14 Clearly, then, overt expletives in non-NSLs must differ categorially from their counterparts in canonical and partial NSLs. In semi NSLs, the same is true if C bears j-features (cf. Chomsky (2008) and also Biberauer & Roberts (Chapter 7) on feature-sharing between C and T): in this case, overt expletives must be (indefinite) DPs, while null expletives must be jPs so as to qualify as 12

13

14

Evidently, Holmberg’s proposal departs from those of researchers who view indefiniteness as the consequence of the absence of a D-feature/-projection (cf. i.a. Longobardi 1994; Lyons 1999). See Cowper & Hall (2008) for recent argumentation from the pronominal domain in favour of the former type of approach. Worth noting here in connection with V-to-T approaches to null subjects is that approaches of this type must assume third-person ‘pronominal’ agreement to be underspecified for definiteness/referentiality so as to accommodate non-referential third person subjects, including expletives. To ensure the availability of both referential and non-referential readings, it would then seem that this type of approach, like its deletion counterpart, requires the postulation of a null topic in the relevant referential contexts. If the proposal that structures featuring known/old information involve null topics is correct, we might view the difference between ‘impure’ (it-type) expletives and their referential counterparts in the domain of jP weak pronouns as residing in the presence (referential it) vs. absence (expletive it) of a null topic rather than in any featural difference between the elements themselves (cf. also Ihsane & Puskas (2001) for an in some ways similar idea regarding the locus of specificity – a CP-related property – vis-a`-vis definiteness – a nominal-related property). Alternatively, one could, at least for associate-related expletives, follow Kayne (2008a) in proposing that these elements in fact originate along with their associates as part of ‘big DPs’ (cf. also i.a. Szabolsci 1983; 1994; Cecchetto 1999; 2000; Sabel 2000; Boeckx 2003), with the XP undergoing raising to SpecTP in fact being the overtly realised expletive along with the trace of its evacuated associate. Thus the structure of There is a book would be roughly as given in (i): (i)

[TP [there ti] is [a book]i [there a book]]

On this view, then, we would expect associate-related expletives to be spelled out by virtue of their raising to SpecTP as part of a featurally complex DP. As with remnant movement proposals more generally, something more would need to be said about how PF will be able to correctly identify the copy of the associate that is to be spelled out.

164

Theresa Biberauer

subset-elements relative to T (on the possibility that Icelandic may in fact be ‘Italian-plus-V2’, see Sections 3.3 and 3.3.4 below and also Biberauer & Roberts (Chapter 7)). Leaving aside the desirability of having to postulate a distinction of this type, it is clear that languages permitting both overt and null expletives in the same structural position pose a serious problem to deletion analyses of null subjects. As we will see in Section 3.3, this is the case in various Germanic varieties, i.e. in systems that have previously been characterised as semi NSLs. To summarise, then, both of the influential minimalist analyses of null subjects raise questions about the feasibility of postulating proEXP that did not arise during the GB era. We turn now to a further complicating factor: the nature of SpecTP. 2.2.2 Spec-TP in minimalism As noted at the start of this section, minimalist analyses of the original EPP commonly decompose it into a D-component and a movement diacritic component. The recognition, particularly within the current, Agree-based approach (cf. Chomsky 2001, and following), that syntactic relations between features associated with functional heads and corresponding features on elements c-commanded by those heads can more generally be thought of as relations which may additionally involve a movement-triggering diacritic necessarily forces one to reassess the original EPP. As the operation Agree is independent of the presence or absence of a movement-diacritic – a departure from earlier minimalism where movement was a prerequisite for feature-checking, the precursor of Agree (cf. Chomsky 1995b) – we expect parametric variation in respect of the presence vs. absence of movement diacritics (often rather confusingly called EPP-features; we will use the notation * here). That is, for every functional category (FC), we would in principle expect the options schematised in (12) (F:__ signals uF, i.e. an unvalued/probing feature): (12)

a. X [F: __], i.e. X probes for [F] and Agrees with an appropriate goal b. X [F:__*], i.e. X probes for [F], Agrees with an appropriate goal and then moves the Agreed-with goal

All other things being equal, then, current theory suggests that the original EPP should not be universal as it entails that T will always have the specification T [D: __*], a state of affairs which does not seem obviously explicable in the principled terms demanded by the guiding hypothesis of the Minimalist Program, the Strong Minimalist Thesis (cf. Chomsky 2001, and following). At most, we would expect an English-style EPP to feature in languages which have opted for the parametric setting T [D: __*]. Taking into account the fact that T is likely to be the locus not only of nominal (D-) features, but also of verbal (V-) features, other possible parametric settings for T might include:

Semi null-subject languages, expletives and expletive pro reconsidered 165 (13)

a. T [D: __] [V:__*] b. T [D:__*] [V:__*] c. T [D: __] [V: __]

On the assumption that at least some head-movement phenomena reflect syntactic head-movement (cf. Lechner 2006; Matushansky 2006; Roberts, forthcoming b) and that * may target either the head bearing the probed-for feature or some larger category properly containing this feature (i.e. piedpiping), (13) in fact raises further parametric possibilities. Consider Chomsky (2001: 38) in this connection: (14)

It has always been taken for granted that the strong V-feature [i.e. V* – TB] is satisfied by V-raising to T (French vs English), not VP raising to Spec-T; and the strong NOMINAL-feature [i.e. D* – TB] by raising of the nominal to Spec-T (EPP), not raising of its head to T. But the theoretical apparatus provides no basis for this choice.

Within the Probe–Goal–Agree system, then, systems of the following kind should all in principle be available: (15)

a. b. c. d. e. f. g. h.

T [D: __*][V: __] where DP-raising to SpecTP takes place T [D: __*][V: __] where D-to-T raising takes place T [D: __ ] [V:__*] where V-to-T raising takes place T [D: __ ] [V:__*] where VP-raising to SpecTP takes place T [D:__*] [V:__*] where DP- and V-to-T raising take place T [D:__*] [V:__*] where VD-to-T takes place T [D:__*] [V:__*] where D- and VP-raising take place T [D: __] [V: __] where no raising takes place

(15a) represents the familiar English system. To the best of our knowledge, (15b) is unattested,15 but the Celtic languages, possibly excepting Breton (cf. Jouitteau 2005), may include a T of the kind schematised in (15c), while Massam (2000; 2001; 2005) has argued that Niuean confirms Chomsky’s speculation that VP-to-SpecTP raising languages exist. (15e), in turn, is probably the most generally accepted characterisation of Romance systems, although (15f), where V-to-T movement simultaneously satisfies both movement triggers, arguably captures the behaviour of canonical NSLs which operate in the manner adherents of the ‘I-subject’ view assume (in contrast to deletion theorists, who view the canonical NSLs as (15e)-type systems). Travis (2006) argues that Malagasy is an example of a (15g)-type system. Finally, it is not clear that (15h)-type systems, where nothing ever undergoes movement to T, exist (cf. Baker (2003b) for a checking-based proposal as to 15

On the ‘pronominal’ agreement view of null subjects, D in (15b) could, of course, be the agreement inflection on V, in which case the impossibility of this type of system follows straightforwardly (cf. also Biberauer & Roberts (Chapter 7) for discussion).

166

Theresa Biberauer

why this might be so; cf. also Alexiadou & Anagnostopoulou’s (2001, 2007) so-called Subject in situ Generalisation); on languages in which T may sometimes lack * (i.e. in which T is only optionally associated with a movement trigger), see Section 3.4 below. If the above-mentioned analyses are on the right track, it would seem that the theoretical possibilities opened up by the Probe–Goal–Agree system may in fact be empirically attested, thereby vindicating the notion that the original EPP is but one of a range of options available to language systems. As Davies & Dubinsky (2001) have proposed, languages may in fact differ in terms of their ‘orientation’, with English-style languages and those prioritising nominal-raising being D-oriented, while those, like the V-initial languages and possibly also Slavic, which prioritise V-/predicateraising might be thought of as V-oriented. Additionally, it seems that various ‘size’ options are available to them, with the precise factors underlying the choices that languages make currently only very poorly understood. Logically, one would expect the feature composition of a language’s various FCs to play some role – if v lacks a V-movement trigger, we might, for example, not, on standard assumptions about phase theory (cf. Chomsky 2001, and following), expect subsequent V-movement to T to be possible. The feature composition of the lexical items serving as goals is similarly likely to be relevant – if j-features are spread across nominal categories, with determiners, for example, only encoding a subset of the features probed by a j-complete T, we might expect full DPs rather than just D-heads to undergo raising to T. For our purposes, the crucial point here is, however, that the theoretical possibility that grammars may opt for T-specifications which do not require DP-raising to SpecTP also leads us to expect that languages may not all be equally in need of expletive elements. The observation that overt expletives are in fact cross-linguistically rare has often been made (cf. i.a. Freeze 2001: 944; Svenonius 2002a: 8; Newmeyer 2004: 204). For adherents of the original EPP, this necessarily means that the majority of the world’s languages have null expletives. Given the theoretical ideas outlined above, the possibility, however, arises that at least some languages lacking overt expletives may lack these completely. We will argue below that this is indeed the case. Furthermore, we will show that languages that have been said to feature both overt and null expletives, including the canonical semi NSLs introduced in Section 2.1, are likewise best analysed as lacking null expletives. 3

Reconsidering SpecTP in Germanic

In the light of the theoretical developments outlined in the previous section, this section focuses on the nature of SpecTP in the Germanic languages,

Semi null-subject languages, expletives and expletive pro reconsidered 167

and specifically on the question of whether it is justified to view SpecTP in these languages as a position exclusively reserved for subjects. Obviously, this is the assumption which crucially underlies the postulation of proEXP for Icelandic and German and also for Dutch (cf. i.a. Bennis (1986), Cardinaletti (1990) and Vikner (1995) for discussion of Dutch as a semi NSL). Sections 3.1 and 3.2 consider data which, we argue in Section 3.3, point to the Germanic varieties being D-oriented, but varying in respect of the manner in which this D-orientation expresses itself. Section 3.3 presents an analysis of the Germanic data which draws on and develops that proposed in Richards & Biberauer (2005) and Biberauer & Richards (2006). In terms of this analysis, SpecTP in certain Germanic languages is not uniquely reserved for subjects, raising questions about the need to postulate proEXP for these languages. Section 3.4 specifically reconsiders the canonical semi NSLs, Icelandic and German and also Dutch, with Section 3.4.4 offering some speculations on diachronic considerations in the Germanic domain. 3.1

Empirical facts

This section aims to investigate the plausibility of maintaining the long-held view that SpecTP in the Germanic languages (specifically, English, Icelandic, Faroese, the Mainland Scandinavian languages [Norwegian, Swedish and Danish], Yiddish, German, Dutch and Afrikaans) is specially designated for subjects.16 Given that English has been extensively argued to feature a position of this type and that this language’s behaviour is the reference point for the original EPP, we will take for granted that SpecTP in English is accurately characterised as one systematically requiring DP-raising to it (see Section 3.3 below on expletive insertion).17 To determine the status of the remaining Germanic languages, we will here focus in particular on two properties that have been argued to signal the presence of a canonical subject

16

17

Given the uncertainty surrounding the synchronic description of Faroese (cf. i.a. Thra´insson 2003; Heycock & Sorace 2007), we will only consider it in passing. For similar reasons, we leave the nature of Yiddish SpecTP to future research (cf. Mohr (2005) for recent discussion of the complications surrounding Yiddish subject- and expletive-distribution). We leave aside here the matter of locative inversion, which has sometimes been said to involve a null expletive (cf. Postal 1977, and following). Worth noting, however, is the fact that locative inversion structures are stylistically highly marked, being most natural in narrative contexts. As such, they may in fact be retained relics of an earlier grammar. Leaving aside the specifics of how such structures are synchronically derived, we observe that something similar may be true for the PP-initial structures that appear to license null expletives in Swedish (cf. Falk (1993a: 167ff.) and Rosengren (2002: 157) for illustration): like English locative inversions, these have a distinctive ‘ring’, being reserved for specific registers (Sten Vikner, p.c.).

168

Theresa Biberauer

position, namely subject-raising and expletive insertion (cf. Davies & Dubinsky (2001) for discussion of properties signalling D-orientation more generally). As Biberauer (2004) considers the relevant facts in some detail, we will only summarise the main points here. Of the modern Germanic languages with which we are concerned here, only the Mainland Scandinavian (MSc) languages exhibit English-like SpecTP properties, consistently requiring passive and unaccusative subjects to raise to SpecTP (cf. (16)) and overtly realised expletives to appear where an appropriate subject is either absent (as in impersonal passives like (17a)) or unavailable (as in existential or presentational structures like (17b)): (16)

a. *. . . at proEXP blev spist et æble that became eaten an apple b. . . . at et æble blev spist that an apple became eaten ‘. . . that an apple was eaten’

(17)

a. . . . at *(der) blevet danset that there became danced ‘. . . that there was dancing’ b. . . . at *(der) er kommet en dreng that there is come a boy ‘. . . that there came a boy’

(Danish)

(cf. Vikner 1995: 209, (93d))

(cf. Vikner 1995: 197, (66e))

These properties underlie the consensus that MSc does not license proEXP. As noted in Section 2.1, Icelandic and German do not consistently exhibit the properties one would expect in a language in which DP-raising to SpecTP is exceptionlessly required. Consequently, these languages were diagnosed as semi NSLs licensing proEXP wherever subject-raising apparently fails to take place (cf. (6), (8), (9–10) above). At least one well-known Icelandic phenomenon, however, casts doubt on the validity of the assumption that SpecTP is uniquely subject-related: stylistic fronting (SF). Consider the examples in (18): (18)

a. Það hefur ___verið tekin erfið a´kvo¨rðun there has been taken difficult decision ‘A difficult decision has been taken.’ b. Það hefur tekin verið erfið a´kvo¨rðun there has taken been difficult decision ‘A difficult decision has been taken.’ (cf. Holmberg 2000: 446, (4))

Holmberg’s (2000; 2006) account of these structures entails that SpecTP in (18a) contains the lower copy of clause-initial það, while the same position in (18b) is occupied by the fronted participle (tekin). In the case of subject

Semi null-subject languages, expletives and expletive pro reconsidered 169

wh-interrogatives and relative clauses, Holmberg (2000), like Maling (1980), Jo´nsson (1991) and others before him, actually assumes SpecTP not just to contain the element that has undergone SF, but also, in the case of the interrogative, the lower copy (trace) of the wh-moved subject and, in the case of the relative clause, either a wh-trace or an operator (cf. Holmberg 2000: 471ff. for discussion). Leaving aside how such ‘double occupation’ is possible and why the complementarity effects observed in other cases involving elements that compete for the same structural position do not come into play,18 and also the alternative possibility that the SFed element in fact precludes the need for subject-raising (cf. Holmberg (Chapter 2)), it is clear that SF jeopardises the notion that SpecTP in Icelandic is a position exclusively reserved for subjects. If that is so, a question must arise as to whether this language needs null expletives. We pursue these questions in more detail in the following sections. Although it is generally classified as a semi NSL, Dutch exhibits properties that distinguish it from its West Germanic semi NSL relative, German. The most striking of these is undoubtedly the fact that there are contexts in Dutch where proEXP appears to be in free variation with an overt counterpart, er (“there”). Consider (19) in this connection: (19)

a. . . . dat (er) in dit bordeel een jongen werkt that there in this brothel a boy works ‘. . . that a (non-specific) boy works in this brothel’ (Rosengren 2002: 179, (70)) b. . . . dat (er) gedanst werd19 that there danced becomes ‘. . . that there was dancing’

Significantly, Afrikaans also optionally permits null expletives in impersonal passives like (19b), but not in presentational structures like (19a) (cf. Haeberli 1999: 13; Richards & Biberauer 2005; Biberauer & Richards 2006). This is shown in (20): (20)

18

19

a. . . . dat *(daar) in die dorp ’n internetkafee is that there in the town an internet-cafe´ is ‘. . . that there is an internet cafe´ in town’

Late Insertion at PF (paralleling what has previously been suggested for expletives at LF – cf. the so-called Expletive Replacement Hypothesis, which we discuss in Section 3.4.3) suggests itself as a possibility in the context of a Distributed Morphology (DM) architecture (cf. Halle & Marantz 1993). It is worth noting that not all dialects of Dutch exhibit this optionality (Marjo van Koppen, p.c.).

170

Theresa Biberauer b. . . . dat (daar) gesing word that there sung becomes ‘. . . that there is singing’

That Afrikaans should license a null expletive in any context at all is surprising, given its extreme inflectional poverty20 (cf. Section 1 above), but it is worth noting that impersonal passives in earlier Swedish also permitted null expletives (alongside overt forms) after the last vestiges of rich agreement were lost (cf. Falk (1993a) for detailed discussion and also Richards & Biberauer (2005) and Biberauer & Richards (2006)). We therefore appear to have synchronic and diachronic evidence that the licensing of proEXP is not inflectionally determined. Furthermore, we see that there are systems in which proEXP appears to be able to alternate with its overt counterpart without this alternation resulting in an interpretive effect. Strikingly, consideration of earlier forms of all the Germanic languages reveals that all of them seem to have gone through stages in which proEXP and its overt counterpart were not in complementary distribution (cf. Falk 1993a, 1993b for Swedish; Abraham 1993 for German; Burridge 1983 and van Gelderen 1993 for Dutch; Breivik 1990, Williams 2000 and Biberauer & Roberts 2005, 2006, 2008 for English; and Richards & Biberauer 2005 for overview discussion). Viewed from a GB perspective, this alternation is problematic in view of the Avoid Pronoun Principle (Chomsky 1981) in terms of which overt pronouns should not be possible where null forms (pro or PRO) are available. If proEXP is licensed in a given context (here: SpecTP), we would then expect it to rule out the possibility of overt expletives, counter fact. From the perspective of a minimalist architecture, we would likewise not expect overt and covert counterparts of the same element to alternate freely, unless the lexicons of the systems in question contain duplicate lexical entries, one associated with a phonological form and the other, featurally identical one, lacking phonological specification.21 The problems that arise in the context of a deletion approach

20

21

Barring the verb ‘to be’ and lexical ‘have’, no verb form shows any agreement marking, with all verbs being formally identical to the infinitive (cf. Ponelis 1993; Donaldson 1993). Afrikaans – and also earlier Swedish (see main text) – therefore undermine Gilligan’s (1987) proposal that the semi NSLs consist of two subgroups, the so-called strong agreement semi NSLs which only license proEXP in restricted contexts (German, Icelandic and Dutch were the examples Gilligan cited) and so-called weak agreement semi NSLs which are less restrictive (various creoles and Tagalog are cited as cases in point). Cf. Adger (2006a) for discussion of free variation involving two fully identical lexical items and Kroch (1994) for the proposal that such doublets cannot be stable. For discussion of the very specific circumstances under which PF might license both the spell-out and the deletion of a given feature bundle, see Neeleman & Szendro¨i (2007; 2008).

Semi null-subject languages, expletives and expletive pro reconsidered 171

to null subjects of the kind discussed in Section 2.2.1 above were already mentioned there. What we see, then, is that while the postulation of proEXP may enable us to maintain that the original EPP is always satisfied in languages like German, Dutch and Afrikaans, its postulation comes at a cost since it raises various non-trivial and currently unanswered technical questions. Furthermore, we have also seen that there are Icelandic structures which seem to undermine the assumption that SpecTP in that language is specifically reserved for subjects. In the following section, we consider word-order variation data from Afrikaans which similarly undermines this assumption for a superficially much more ‘well behaved’ Germanic language.

3.2

Afrikaans word-order variation as a window on the nature of SpecTP

Biberauer (2003) observes that modern spoken Afrikaans (MSA) permits specific ‘verb-early’ orders to alternate with the prescriptively correct verbfinal orders in embedded declaratives (dat-clauses), but not others. Consider (21–22) in this connection: (21)

a. Ek weet dat sy dikwels Chopin gespeel het I know that she often Chopin played has ‘I know that she has often played Chopin.’ b. Ek weet dat sy het dikwels Chopin gespeel I know that she has often Chopin played ‘I know that she has often played Chopin.’

(22)

a. Ek weet dat sy dikwels Chopin speel I know that she often Chopin play ‘I know that she often plays Chopin.’ b. %Ek weet dat sy speel dikwels Chopin I know that she play often Chopin

Whereas the superficially V2 structure in (21b) is readily produced by native speakers and also judged by them to be interpretively identical to the prescriptively correct structure in (21a), the same is not true for the corresponding structures in (22): here (22b) is consistently assigned a matrix-clause interpretation, with speakers viewing the structure as one which one might use if one has forgotten that one was uttering a subordinate clause or if one wanted to assign matrix force to that clause, despite the presence of the subordinator (cf. Holmberg & Platzack (1995: 179ff.) on so-called embedded root phenomena and also Biberauer (2003; 2009) for more detailed discussion of the Afrikaans facts). The crucial point for our

172

Theresa Biberauer

purposes is that ‘verb-early’ structures are generally possible as an interpretively identical alternative to verb-final structures wherever (i) the subject follows dat and (ii) an auxiliary occupies the second position; where a lexical verb surfaces in (non-trivial) clause-second position (i.e. where the clause consists of more than just a subject and an object), the structure cannot be interpreted in this way.22 Various considerations point to many Afrikaans auxiliaries being T-elements, just like their English counterparts (cf. i.a. Roberts 1985; 1993b; Lightfoot 1991; Biberauer & Roberts (Chapter 7)) and unlike their West Germanic relatives (cf. the thematically-sensitive auxiliary selection that characterises German and Dutch, which is absent in Afrikaans). In view of this fact and taking into account the so-called Fox–Reinhart intuition on optionality in terms of which “optional operations can apply only if they have an effect on outcome” (cf. Chomsky 2001: 34), Biberauer (2003) proposes that the MSA variation can be understood as diagrammed below (VP and vP are represented as head-final structures for expository convenience; see Biberauer (2003) and Biberauer & Roberts (2005) for discussion of a consistently Kaynian analysis of SOV Germanic): (23)

a. Ek weet dat sy dikwels Chopin gespeel het

(cf. (21a))

CP C dat

TP vP

DP sy

T¢ v¢

Adv dikwels

het v¢

VP DP Chopin

22

tvP

v V gespeel

There are exceptions to this generalisation as lexical verb-second structures followed by a ‘leaked’ constituent may also be interpreted as embedded-clause alternatives to the standardly sanctioned verb-final structure. In these cases, however, it can be shown that the lexical verb occupies a low position despite its ‘earliness’ (see Biberauer (2009) for more detailed discussion). By contrast, the auxiliary-second structures would all seem to involve verbs located in T (see main text).

Semi null-subject languages, expletives and expletive pro reconsidered 173 b. Ek weet dat sy het dikwels Chopin gespeel

(cf. (21b))

CP C dat

TP DP sy

T¢ het

vP v¢

tDP Adv dikwels

v¢ VP

DP Chopin

v V gespeel

As the trees show, the proposal is that MSA’s superficially V2 structures are in fact structures in which the second-position verb is located in T. As lexical verbs do not undergo V-to-T movement in Afrikaans (cf. Biberauer 2003; 2009; Vikner 2001; 2005), the impossibility of ‘verb-early’ structures featuring second-position lexical verbs follows: for a lexical verb to surface before the VP adverb and object, as in (22b), it would have to undergo V-to-T raising, an operation which embedded T cannot trigger.23 By contrast, auxiliaries are either merged in T – plausible for het (“have”) and possibly also for wees (“be”) and some modals – or move there – likely to be the case for word (“become”) and some modals – with the result that we would expect these elements to surface second wherever the derivation involves English-style subject DP-raising to SpecTP. Standardly, Afrikaans auxiliaries should, however, be final (non-modal auxiliaries) or pre-final (modal auxiliaries, which consistently trigger Verb Raising – cf. Evers 1975). Given the common nature of the ‘verb-early’/verbfinal alternation illustrated in (21), the fact that it is limited to structures featuring second-position auxiliaries, and the fact that there is no interpretive difference between the structures concerned, Biberauer (2003) proposes that SpecTP in Afrikaans is not in fact a specially designated subject position, but that instead, it is a position which hosts the entire vP containing the subject. More specifically, the proposal is that Afrikaans T is endowed with the same 23

Cf. Biberauer & Roberts (Chapter 7) for the proposal that T consistently fails to trigger V-to-T movement in V2 languages other than Icelandic and Yiddish.

174

Theresa Biberauer

featural specification as English T – i.e. [D: __*] [V: __] as in (15a) above – but that it differs from English in that T probes the subject DP in Spec-vP, whereafter * then moves not just the goal (the subject DP), but the entire vP, which properly contains the subject. In other words, the difference between English and Afrikaans is that the latter standardly employs a pied-piping strategy in order to satisfy the * diacritic associated with T’s D-probe, whereas English moves just the Agreed-with category. On the assumption that the computational system is only concerned with Agree-mediated movement resulting in the movement of the Agreed-with category (the goal), we can understand the MSA alternation as following from the more general pied-piping vs. stranding alternations that natural language systems appear to permit under the appropriate circumstances.24 Crucially, then, structures such as those illustrated in (21) are outputs of identical numerations, with the observed word-order differences following from different choices as to the ‘size’ of the moved category, choices which are immaterial to the computational system which, in this case, cares only that the subject DP should ultimately reach SpecTP.25 In the present context, the primary significance of this proposal is that it entails an analysis in terms of which SpecTP in Afrikaans is not uniquely reserved for subjects, despite the fact that T probes the subject. This obviously raises the possibility that other members of the Germanic family, in particular Afrikaans’s West Germanic cousins, may similarly lack an English-style SpecTP, a possibility to which we now turn.

3.3

Germanic EPP-satisfaction: a typology

Taking into account the Probe–Goal–Agree system’s indeterminacy in respect of the precise identity of the elements that undergo Agree-driven movement (cf. (14) above) and building on the insights from Afrikaans, Richards & Biberauer (2005) and Biberauer & Richards (2006) propose that the Germanic

24 25

The fact that this alternation is absent in standard Afrikaans is presumably the result of prescriptive pressure proscribing a structure that the grammar does in fact make available. Given the interpretive identity of the structures and the fact that all native speakers appear to employ both variants, we do not consider an alternative proposal that might suggest itself – namely, one involving two ‘competing grammars’ (cf. Kroch 1989). Grammars of this type would seem to us to be best motivated in cases where the variants concerned are associated with specific sociolinguistic values which will signal to speakers that there are in fact two distinct grammars to acquire (cf. Roberts (forthcoming b: 330ff.) for discussion of this ‘syntactic diglossia’ – also mooted in Kroch (2000) – and the proposal that the Brazilian Portuguese situation may be amenable to an analysis along these lines). This is very clearly not the case in Afrikaans, where there has never been a variety featuring only English-style subject-raising (i.e. (21b)-type structures).

Semi null-subject languages, expletives and expletive pro reconsidered 175

languages in fact exhibit a range of T-related EPP-satisfaction strategies. In particular, they propose the typology in (24): (24) Language

Source of f-features (goal)

EPP movement

(i) English, MSc (ii) Greek, Italian (pro-drop) (iii) German, Icelandic (iv) Afrikaans, Dutch

D(P) in Spec-vP f-features on V-morphology f-features on V-morphology D(P) in Spec-vP

DP-to-SpecTP v-to-T vP-to-SpecTP vP-to-SpecTP

In terms of this typology, languages may differ as to the source of the goal and the size of the goal-containing category that is moved to satisfy T’s EPP-feature. Following A&A (1998), it is assumed that the nominal category that agrees with T (the goal) may be associated with one of two categories: either (a) the j-features of the DP argument in spec-vP, or (b) the j-features of the agreement morpheme on the verbal head in richagreement languages (“argumental morphology”/“pronominal agreement” – cf. also Bobaljik & Thra´insson 1998). Without adding to earlier assumptions, it is then clear that we would expect there to be two possible sources for T’s goal. If Agree-driven movement were only able to target goals, we would then only expect the EPP-satisfaction mechanisms in (24i) and (24ii) above, i.e. the familiar English and null-subject mechanisms. Given the incontestable existence of the currently still rather mysterious operation pied-pipe (cf. Chomsky 2000, and following), it, however, follows that there should also be options regarding the size of the moved category, with the Move-Goal-Only options in (24i–ii) existing alongside two Move-Goaland-pied-pipe options, namely those in (24iii–iv). In a sense, then, the proposal is that German and Icelandic are ‘Italian/Greek-plus-pied-pipe’ (i.e. head-pied-piping languages), their inflectional richness rendering the agreement morphology on V/v an appropriate goal, whereas inflectionally poorer Afrikaans and Dutch are ‘English-plus-pied-pipe’ (i.e. spec-piedpiping languages), with the subject DP serving as the goal. As Richards & Biberauer (2005) note, the assumption that rich inflectional morphology is a prerequisite for head-pied-piping status makes a strong diachronic prediction: wherever a language’s inflectional morphology has undergone impoverishment to the point where acquirers will not postulate that it should be stored in the form of distinct lexical entries, they also will not be able to postulate that the language they are acquiring is a head-piedpiping language. Given the superficial resemblances between head- and spec-pied-pipers (compare German vs. Dutch or Afrikaans in the OV

176

Theresa Biberauer

domain26), the prediction is that formerly head-pied-piping languages will be reanalysed as spec-pied-pipers once their inflectional morphology has become too impoverished to support a ‘pronominal agreement’ analysis. In the present context, Richards & Biberauer’s proposal has the following entailments and consequences. First, it is assumed that the Germanic and Romance languages are all D-oriented in the sense that T is specified [D: __*]. This seems justified, given that all the languages in question exhibit raising-to-subject phenomena (although they do not all do so consistently – see Section 3.4 below; and they also all seem to meet Davies & Dubinsky’s more general criteria for D-prominence and also those proposed by Li & Thompson (1976) to hold of ‘subject-prominent’ languages). Second, it leads us to expect that raising-to-subject operations, moving a subject to SpecTP, may not be equally necessary in all of these languages. We have already discussed this point in relation to the V-to-T raising approach to null subjects (cf. Section 2.2.1 above): if this general approach is on the right track, V-to-T raising will fulfil the role of non-focused pronoun subject-raising, with the consequence that subjects of this type will not feature as part of the Numeration and will therefore also not be available to undergo raising to SpecTP. Similarly, we would expect the subject-raising patterns in the pied-piping counterpart of V-to-T NSLs to be similar. Insofar as this relates to referential null subjects, this prediction clearly does not hold for Icelandic and German, a point to which we return in Section 3.4.3 below. It is, however, worth noting that the proposal that these languages employ vP-fronting to satisfy T’s EPPrequirements offers a way of understanding why ‘low subject’ structures like those in (9–10) are possible (cf. also Mohr 2005). As (25) illustrates, fronting of a vP in which the Recipient and Theme arguments have been merged in their unmarked order (cf. Grewendorf 1989; Haider 1993) derives the desired order:27

26

27

Richards & Biberauer (2005) and Biberauer & Richards (2006) speculate that Icelandic and Faroese represent another head- vs. spec-pied-piping pair, this time in the VO domain. Cf. Section 3.4 for further discussion of Icelandic. Biberauer & Roberts (2005; 2006; 2008; 2009) further argue that Old and Middle English were spec-pied-piping varieties, which may well have developed from an inflectionally richer head-pied-piping ancestor. Worth noting here is that the neutral interpretation and also the neutral intonation assigned to structures like (25) precludes the feasibility of an analysis in terms of which the indirect object has scrambled over the raised direct object – cf. (i): (i)

[CP dass [TP [DP dem Mann]i [DP ein Buch]k [vP [VP ti tk geschenkt] wurde]]]

Structures of this type – variously referred to as focus scrambling or A0 -scrambling (Neeleman 1994; Neeleman & van de Koot 2008), I-topicalisation (Jacobs 1997) T-scrambling (Haider & Rosengren 1998) and S-scrambling (Hinterho¨lzl 2005) structures – require a very special intonational pattern, the so-called hat contour (fall–rise–fall), which is very evidently absent in the cases under discussion here.

Semi null-subject languages, expletives and expletive pro reconsidered 177 (25)

. . . dass dem Mann ein Buch geschenkt wurde . . . that the-DAT man a-NOM book presented became

CP TP

C dass vP VP

T¢ vdef wurde

T

tvP

DP V¢ dem Mann DP V ein Buch geschenkt

Significantly, vP-fronting does not require us to postulate proEXP in this case; we can simply assume that the passive participle has ‘absorbed’ v’s external argument (cf. Baker, Johnson & Roberts 1989), with the result that thematic Spec-vP fails to be projected. As EPP-satisfaction in German is achieved via vP-fronting, there is then no need to postulate expletive insertion into SpecTP. Since German T probes vP for VD (i.e. the agreement morphology on the verb), the same is true for the unraised subjectcontaining structure in (10b): vP-raising will ensure that the VD goal undergoes movement, thereby satisfying T’s EPP-requirements. In terms of the analysis proposed here, the structure for (10b) can therefore be schematised as follows (for expository purposes, we once again gloss over the internal structure of vP): (26)

[CP . . .weil [TP [vP ja doch Linguisten Kammermusik spielen]]] MOD.PART linguists chamber-music play because

Modulo the fact that Icelandic is a symmetric V2 language (i.e. V is located in C in both main and selected embedded declarative clauses – cf. footnote 6), the same proposal can be extended to the Icelandic examples in (9). Consider the derivation of (9a) by way of illustration (strikethrough here and elsewhere indicates unspelled-out lower copies): (27)

´I gær voru konunginum gefnir hestar (Icelandic) yesterday were-3PL king-the-DAT given horses-NOM

178

Theresa Biberauer

CP C¢

Spec Í gær C voru

TP Spec vP

T¢ T

vP v

VP Spec konunginum V gefnir

V¢ O hestar

As (27) shows, vP-fronting once again precludes the need to postulate the merger of proEXP in SpecTP. The same is true for (9b), schematised below (clause structure simplified as before): (28)

[CP I´ eldhu´sinu hefur [TP [vP alltaf verið svartur ko¨ttur]T]] in kitchen-the has always been black cat

A point that deserves some discussion is V-to-T movement. Recall that Icelandic is ascribed the status of a head-pied-piping language in the typology in (24). In the examples in (9), T will therefore be probing the auxiliaries voru and hefur. If V-to-T movement does indeed take place, as illustrated in (27), the question is why vP-movement would ever take place, i.e. how can a V-to-T movement language be a head-pied-piping vP-raiser? One possibility that suggests itself given the phase-based architecture in which the Probe–Goal– Agree theory is embedded is that V-to-T raising and vP-raising in fact take place simultaneously. Recall that Chomsky (2008) suggests that syntactic operations within a given phasal domain all take place simultaneously upon completion of the phase in question. If this is indeed correct, we would expect T and C to probe simultaneously, with the result that T can probe v for D- and V-features, thereafter raising vP so as to simultaneously satisfy its featural requirements (assumed to be [D:__*], [V:__*]) while C probes v to value its verbal probe and raises it to satisfy its movement diacritic. In terms of this proposal, then, there is no V-to-T movement (cf. also Biberauer & Roberts, Chapter 7). A non-trivial question that arises on this view, however, is how PF will be able to establish which copy of v to spell-out. Given this problem, an alternative in terms of which movement operations proceed in sequence as each FC is merged (in accordance with something like Pesetsky’s (1989)

Semi null-subject languages, expletives and expletive pro reconsidered 179

Earliness Principle) and where the finite verb can subsequently be extracted from SpecTP may be preferable. (This would seem to violate the ban on Freezing – cf. Mu¨ller (1998), but see Abels’ (2008) Universal Constraint on the Order of Operations (UCOOL), which tries to pin down the circumstances under which moved XPs will not be frozen.) Finally, it may be that Icelandic is not in fact a head-pied-piping language, but that it is actually a head-raiser like the Romance NSLs (the Icelandic¼‘Italian-plus-V2’ view; cf. Biberauer & Richards (2006: note 14) for speculation on this point). Given the problems we have identified here and also the fact that the proposal as it stands does not offer any obvious way of dealing with SF, we will return to the question of Icelandic’s status in the typology of T-related EPP-satisfaction in Section 3.4.3 below. Leaving aside the problems we have just discussed, what we have seen so far is that the proposal that German and Icelandic may in fact be vP-fronting languages allows us to do away with the idea that these languages merge a null expletive in SpecTP. As it stands, we do not yet have an account of the fact that German weather expletives (i.e. proEXPþA) may surface clauseinternally (cf. (8b) above) while the same is not possible in Icelandic (cf. (6a)). This undoubtedly results from the fact that Icelandic does not, in general, lexicalise impersonal subjects (cf. Sigurðsson 1989; Sigurðsson & Egerland 2009; Holmberg, Chapter 2), whereas German does (cf., for example, the use of man in generic contexts). As Mohr (2005: 178ff.) convincingly argues, það cannot be viewed as a genuinely quasi-argumental element, even when it surfaces clause-initially. Among other things, it differs from genuine quasi-arguments in not being able to control PRO (cf. Chomsky 1981: 323ff.; Bennis 1986: 99ff.; Vikner 1995: 228); as (29b) shows, Icelandic requires a finite embedded clause to express the meaning realised by a weather predicate-containing control clause in German or English: (29)

a. Es regnete ohne PRO zu schneien (German) it rained without to snow-INF ‘It rained without snowing.’ b. Það rigndi a´n þess að það hafi snjo´að (Icelandic) it rained without it-GEN that it had snowed ‘It rained without snowing.’ (Mohr 2005: 179, (10a))

On the assumption that quasi-arguments resemble other arguments in being merged within the thematic domain (vP), we can account for það’s inability to control PRO by saying that Icelandic lacks a vP-expletive, with það consistently behaving like an element which is exclusively associated with the CP-domain, regardless of whether it surfaces in ‘weather’, presentational, existential or SF constructions (cf. i.a. Platzack (1985), Sigurðsson (1989),

180

Theresa Biberauer

Holmberg & Platzack (1995), Svenonius (2002a), Mohr (2005), Richards & Biberauer (2005) and Sells (2005), who have previously argued that það is a CP-expletive without, however, addressing the question of the implications of this analysis for the original EPP). By contrast, the formally identical expletives surfacing in German ‘weather’, presentational and existential constructions do not all appear to be the same element. (29a) indicates that German has a genuine quasi-argumental expletive, which we would therefore expect to be merged in Spec-vP. If this is the case, we would in turn expect this element to surface clause-internally wherever it undergoes vP-fronting to SpecTP, but does not subsequently undergo fronting to SpecCP, the correct expectation (cf. (8b) above). On the analysis presented here, Icelandic and German therefore differ as a consequence of a difference in their lexical inventories: while there is only one expletive in Icelandic, a CP-related ‘topic expletive’, German has two distinct es-expletives – the quasi-argument which is merged in Spec-vP and the ‘topic expletive’ which is merged in SpecCP. Crucially, neither language has an expletive that is merged in SpecTP (henceforth: TP-expletive) as the manner in which T’s EPP-requirements are met precludes the need for a lexical item of this type. Afrikaans and Dutch, likewise, lack TP-expletives. Like German, their lexical inventories include a weather expletive (realised as het and dit respectively) which is distinct from the expletives found in presentational and impersonal passive structures (realised as er and daar respectively) and which is obligatorily merged in Spec-vP wherever the Numeration contains a weather predicate. Unlike German, whose inflectional wealth (cf. footnote 3) allows it to be a head-pied-piper (i.e. a language in which T probes D-onV/VD), both of these inflectionally poorer languages rely on the presence of a subject-element in vP which can serve as the goal for T’s D-oriented probe: an expletive which can be merged within the vP-domain as a ‘last resort’ device for T’s D-probe.28 As we will see here, however, Dutch and Afrikaans are not equally dependent on this ‘last resort’ device. Let us first consider the latter. As noted in Section 3.1 above, daar is obligatory in presentationals and existentials, but optional in impersonal passives (cf. (20)). Given the fact that Afrikaans verbs consistently lack agreement marking of any kind, the 28

As argued in Richards (2004) and Richards & Biberauer (2005), the simplest implementation of the Probe–Goal–Agree system arguably requires English-type expletives to be merged in Spec-vP rather than in SpecTP as was initially assumed. Unless probes may probe their specifiers (as Rezac (2003) and Mu¨ller (2004b), for example, assume), it could therefore be the case that languages universally lack TP-expletives, with expletives which are spelled out in SpecTP originating within the vP-domain. For other arguments that expletives are merged in the vP-domain, see i.a. Hoekstra & Mulder (1990), Moro (1997), den Dikken (1995), Basilico (1997), Hale & Keyser (2000), Sabel (2000), Hazout (2004) and Deal (2008).

Semi null-subject languages, expletives and expletive pro reconsidered 181

obligatoriness of daar comes as no surprise if the inflectional basis of the typology in (24) is on the right track: the expectation would be that languages of this type would be rigidly spec-pied-piping, with the goal for T’s D-probe consistently being filled either by an appropriate (non-rhematic) subject DP or by an expletive (cf. also the oft-observed fact that pied-piping structures very commonly require pre-pied-piping movement of the pied-piper to the specifier of the moved category; cf. i.a. Richards (2001) and Horvath (2000; 2006)).29 As shown in the diagrams in (30), which represent simplified derivations at the point at which T is merged, this is precisely what we see in Afrikaans: (30)

a. . . . dat daar in die dorp ‘n internetkafee is (¼ (20a)) [TP . . . T [vP daar [PP in die dorp] [DP ’n internetkafee] is . . .]

^

b. . . . dat die boek vir die man gegee is that the book for the man given is ‘. . . that the book was given to the man’

(contrast German (25))

[TP . . . T [vP [DP die boek] [VP [PP vir die man] [DP die boek] gegee] is]]

^

In (30a), daar-insertion permits T to locate its goal in outermost Spec-vP and in (30b), raising-to-subject of the direct object, die boek, facilitates the same outcome. Low subject structures like the German counterpart of (30b) are systematically barred in Afrikaans, just as we would expect. Against this background, the fact that the language permits impersonal passives lacking daar (cf. (20b) above) is surprising: while daar-containing impersonal passives operate in exactly the manner schematised in (30) above, it would seem that the daar-less counterparts of these structures would require the postulation of proEXP–A in order to comply with Afrikaans’s spec-piedpiping requirement. Biberauer & Richards (2006), however, suggest an alternative solution: once again drawing on the idea that passive participles ‘absorb’ the external arguments in structures in which they occur, thereby rendering themselves suitable nominal goals for T, they propose that Afrikaans’s status as a spec-pied-piping language makes it ‘backwards compatible’ with head-pied-piping in the absence of a suitable goal in Spec-vP.30 The idea here is that, since the grammar is one which is [þpied-piping] – i.e. one which, unlike its [–pied-piping] counterpart, does not explictly specify 29 30

Horvath (2000; 2006) in fact observes that pied-piping is generally only triggered either by heads or specifiers, which is in line with the proposed typology in (24). ‘Backwards’ here reflects diachronic/chronological considerations since there appear to be a range of cases in which the loss of sufficiently rich inflectional morphology has led to a change from a head-pied-piping to a spec-pied-piping system (cf. discussion of the strong diachronic prediction mentioned in the main text).

182

Theresa Biberauer

the exact category targeted by T’s *-feature – T can in fact probe the participle rather than a subject(-related) element located in Spec-vP since the participle would always have been part of the vP that undergoes *-driven movement to SpecTP. In terms of this analysis, then, the daar-less impersonal passives found in Afrikaans do not require the postulation of proEXP–A since both the daar- and the daar-less variants of this structure involve vP-fronting to SpecTP. This is illustrated in (31) (D indicates the goal probed by T): (31)

a. . . . dat daar gesing word that there sung become ‘. . . that there was singing’ [CP dat [TP [vP daarD gesing word] T [vP daarD gesing word]]] b. . . . dat gesing word that sung become ‘. . . that there was singing’ [CP dat [TP [vP gesingD word] T [vP gesingD word]]]

Suggestive evidence that the above approach is on the right track comes from Dutch.31 As is well known, this language permits word-order variation in the context of verbal clusters involving a participle and temporal auxiliary – cf. (32) below: (32)

a. . . . dat ze een goed werkstuk afgeleverd heeft that she a good assignment submitted has ‘. . . that she has submitted a good assignment’ (‘green order’ ¼ V-AUX) b. . . . dat ze een goed werkstuk heeft afgeleverd that she a good assignment has submitted ‘. . . that she submitted a good assignment’

(‘red order’ ¼ AUX-V) Strikingly, both orders are also possible in impersonal passives where expletive er is present, whereas only the participle-initial order is available in the absence of er: (33)

a. . . . dat (er) gedanst wordt that there danced becomes ‘. . . that there is dancing’ b. . . . dat *(er) wordt gedanst that there becomes danced ‘. . . that there is dancing’

(V-AUX)

(AUX-V)

On the analysis proposed in Richards & Biberauer (2005), the above patterns can be explained as follows: er is optional in (33a) as Dutch is ‘backwards 31

Thanks to Hans Bennis for having originally drawn these data to our attention.

Semi null-subject languages, expletives and expletive pro reconsidered 183

compatible’ with head-pied-piping, i.e. the participle can serve as the Agreedwith goal wherever er is absent; in (33b), by contrast, the participle necessarily constitutes T’s goal in er’s absence, but fails to undergo raising, which is not possible if, as is standardly assumed, Agree-mediated movement always requires movement of the Agreed-with goal. If we consider a proEXP–A-based alternative, it is not clear how the facts can be accounted for: if proEXP–A is available wherever er fails to surface, (33b) should be just as grammatical as (33a). This is shown in (34): (34)

a. . . . dat proEXP–A gedanst wordt b. . . . dat proEXP–A wordt gedanst

Here, then, we seem to have a clear case where a proEXP–A-based analysis fails to account for empirical facts that are amenable to explanation via a proless alternative.32 Also worth noting in this connection is that the more general prediction made by the proposed pied-piping-based theory – in terms of which there is some scope for expletive- (and, more generally, subject-) related optionality in the context of languages which have undergone/are in the process of undergoing inflectional impoverishment – seems to hold out diachronically. Thus Falk (1993b: 166) makes the observation that “det with a purely grammatical function, i.e. it fills the otherwise illicit, empty subject position [SpecTP – TB]” surfaced for the first time in early modern Swedish (1600 onwards) after the loss of agreement, which she correlates with the loss of V-to-I movement. The crucial point for our purposes is that structures in which SpecTP does not appear to be filled continued to be possible after this period, with expletives only gradually becoming obligatory in all their modern contexts (cf. Falk 1993b: 161, Table 1; cf. also Platzack 1985/1987). In terms of the proposals outlined here, this optionality is to be expected for as long as SpecTP in earlier Swedish remained a position not specifically reserved for subject(-related) elements, i.e. one not exclusively requiring subject/spec-raising. As we will see in the following section, there are strong indications that the modern Swedish, English-style canonical subject position did not come into being in Swedish until quite some time after the loss of V-to-I raising. In terms of the standard analysis of semi NSLs, the non-correlation between the loss of

32

Taking into account the findings of Barbiers (2005), we leave aside the possibility that the (b) orders in (32) and (33) are simply the consequence of a PF operation ‘flipping’ the verb and the auxiliary (cf. Wurmbrand (2004) for a recent suggestion along these lines and references cited therein). If this were the case, neither the account proposed here nor the pro-based alternative would of course offer any insight as to the grammaticality differences at issue. Similarly, we leave aside the possibility that expletive insertion may in fact be a PF operation of the sort assumed in Holmberg (2000) and Bobaljik (2002), i.e. an operation inserting material whose features did not play any role in Narrow Syntax.

184

Theresa Biberauer

sufficiently rich agreement and the loss of null expletives is surprising; in terms of the proposals made here, the dissociation is not unexpected. Biberauer & Roberts (2005; 2006; 2008; 2009) show that the same is true for the history of English, where unrealised expletives continued to be possible clause-internally during the Middle English period despite the inflectional impoverishment the language had undergone by this stage (cf. i.a. Allen (1995) and Haeberli (1999/2002) for discussion).33 Similarly, synchronic evidence from Faroese (cf. Thra´insson 2003) highlights the non-definitive nature of the correlation between inflectional richness and null expletives: although modern-day Faroese34 is undeniably inflectionally poorer than Icelandic, whose syntax is generally viewed as the template for earlier Faroese, with V-movement seeming to be at best optional (cf. also Heycock & Sorace 2007), null expletives are undoubtedly still licensed (specifically, in impersonal passives and presentational/existential structures – cf. Thra´insson, Petersen, Jacobsen & Hansen 2004). We will return to these matters in more detail in the following section. Here we conclude by summarising the main predictions of the proposals made in this section: (a) Languages which employ D-oriented vP-raising to satisfy T’s EPPrequirements will not require proEXP–A as vP-raising precludes the need for expletive insertion in SpecTP. The same should clearly be true for V-oriented vP/XP-raising, a point to which we return in Section 4; (b) Languages which employ vP-raising may be either inflectionally rich or inflectionally impoverished, with inflectional considerations determining the source of T’s D-goal: either the inflected verb (head-pied-piping) or a subject/expletive D(P) (spec-piped-piping). Spec-pied-piping languages may exhibit optionality in respect of the presence vs. absence of nonargumental expletives owing to their ‘backwards compatibility’ with a grammar in which T can locate an alternative D-bearing goal. Several empirical issues, however, remain unresolved and we now turn to these.

33

34

Cf. also Biberauer & Roberts (Chapter 7) for discussion of the status of V-to-T movement in the history of English. As they argue, Old and Middle English are likely to have resembled modern German and the West Germanic languages in lacking V-to-T movement (cf.Vikner (2001; 2005) for convincing argumentation). If this is correct, there certainly cannot be a direct correlation between V-to-T movement and the availability of proEXP–A, as is generally argued in the context of parametric accounts inspired by (3, 4). As noted in Section 3.1, much remains to be learned about the syntax of modern Faroese and its dialects. Worth noting, however, is that recent research (cf. Thra´insson (2003) and references cited therein) suggests that it is too simplistic to distinguish a conservative, Icelandic-like variety from a less conservative, more MSc-like variety, as suggested in Jonas (1996). In terms of this distinction, the correlations between inflectional richness, V-movement, and subject and expletive distribution seem to fall out precisely as (3, 4) would lead one to expect, but the reality seems significantly more complicated.

Semi null-subject languages, expletives and expletive pro reconsidered 185

3.4

A closer look at Dutch, German and Icelandic

3.4.1 Dutch So far, we have an account of expletive optionality in Dutch and Afrikaans impersonal passives. What is not yet clear is why expletives should also be optional in Dutch, but not Afrikaans, presentational/existential structures (cf. (19a) vs. (20a/30a) above). Observing that head-pied-piping systems like German do not require an overtly realised expletive in these contexts, we might be tempted to extend the ‘backwards compatibility’ proposal discussed in Section 3.3 to this structure. This would, however, miss an important difference between Dutch and Afrikaans that is evident in domains other than presentationals/existentials. Recall that Afrikaans does not permit ‘low subjects’ of the kind found in German (cf. (30b)); Dutch, by contrast, does:35 (35)

a. . . . dat het meisjeDAT de ergste rampenNOM overkwam (Dutch) that the girl the most-terrible disasters happened ‘. . . that the most terrible disasters happened to the girl’ (cf. Rosengren 2002: 182, (76a)) b. . . . dat die meisie die verskriklikste rampe oorgekom het (Afrikaans) that the girl the most-terrible disasters happened have ‘. . . that the most terrible disasters happened to the girl/her’

That die meisie in (35b) is indeed the nominative subject rather than a dative parallel to Dutch (35a) is clearly shown by the fact that pronominal substitution requires the nominative sy (“she”) rather than oblique haar (“her”) – thus: . . . dat sy/*haar die verskriklikste rampe oorgekom het. Strikingly, Dutch exhibits a further similarity to German in respect of its subject distribution, also exhibiting both subject- and object-related ‘Diesing Effects’ (cf. Diesing 1992; Diesing & Jelinek 1995; Broekhuis 2000; de Hoop 1996; Neeleman & van de Koot 2008):36 (36)

35

36

a. . . .dat toch wel veel mensen in grote groepen communiceren that MOD.PART. many people in big groups communicate ‘. . . that there are after all many people who communicate in big groups’

Like German (25), structures like Dutch (35) cannot be analysed as scrambling structures in which the dative argument has scrambled over a raised subject as the interpretive and intonational properties of ‘low subject’ structures of this type do not correspond to those typically associated with the relevant type of scrambling. The precise nature of ‘Diesing effects’ has been much disputed in the literature (cf. i.a. Haider & Rosengren 1998; 2003; Frey & Pittner 1998; Frey 2001; Rosengren 2002; Neeleman & van de Koot 2008). What seems fairly clear is that the full range of interpretive effects entails the postulation of a clause-medial scrambling domain (the Mittelfeld of traditional descriptions), which we abstract away from for the purposes of this chapter (but cf. Haider (2005) for overview discussion). Since it is generally accepted that clause-medial scrambling should be distinguished from focus scrambling (cf. footnote 27), with the latter, but not the former, targeting positions above the canonical subject position, SpecTP, this seems justified: the interpretively significant ‘Diesing-type’ subject-related scrambling that we are concerned with

186

Theresa Biberauer b. . . . dat veel mensen toch wel in grote groepen communiceren that many people MOD.PART. in big groups communicate . . .‘that many people after all communicate in big groups’

In Afrikaans, by contrast, weak/new information ‘low subject’ readings can only be obtained in the presence of an expletive (object behaviour mirrors that in German and Dutch): (37)

. . . dat *(daar) tog baie mense in groot groepe kommunikeer that there MOD.PART. many people in big groups communicate ‘. . . that there are after all many people who communicate in big groups’

An expletive (er) is also optionally possible in Dutch structures like (36a), which might lead one to conclude that Dutch is after all less similar to German than we have begun to suggest. In this connection, it is, however, important to note two things. The first of these is that the presence of er does not simply seem to guarantee a ‘low subject’ reading for the unraised subject in the way English there or Afrikaans daar does (cf. Bobaljik 2002); it also seems to add a currently not very well understood situative ‘here and now/ there and then’ interpretation (cf. Mohr (2005: 143ff.) for detailed discussion and references), one that can, in the absence of er, also be conveyed by the presence of appropriate situative (e.g. locative) adverbials. Er, then, seems to be ‘more than’ just an interpretively vacuous element (though see Section 3.4.3 on the interpretive significance of English-style expletives). Second, spoken German features an element which mirrors the behaviour of Dutch er – da (cf. Koeneman & Neeleman 2001: 228ff.; Rosengren 2002: 175; Mohr 2005: 146ff.; Hartmann 2008: chapter 4): (38)

Es hat da jemand einen Apfel gegessen It has there someone an-ACC apple eaten ‘Someone (in that context) ate an apple.’ (cf. Koeneman & Neeleman 2001: 229)

As (38) shows, da may cooccur with es, clearly signalling that it is either a vP-expletive cooccurring with German’s CP-expletive, non-(quasi)argumental es (cf. Richards & Biberauer 2005) or that it is in fact not an expletive element, takes place in a domain that appears to be ‘closed off’ by SpecTP. Furthermore, the interpretations associated with subjects at the ‘top’ and ‘bottom’ of the potential subject field (cf. Neeleman & van de Koot 2008) seem uncontested – the former must be strong and/or informationally old and the latter must be weak and/or informationally new (cf. Frey 2001). Needless to say, more detailed work is required to establish the precise structure of the Mittelfeld and how and if it maps onto the cartography of (T-related) subject positions that has been suggested i.a. by Kiss (1996), Bobaljik & Jonas (1996) and Cardinaletti (2004); for the proposal that particular interpretations are not necessarily associated with particular cartographic positions, but fall out relationally as a result of differently ordered derivational steps, see Neeleman & Weerman (1999), Zwart (2007) and Neeleman & van de Koot (2008).

Semi null-subject languages, expletives and expletive pro reconsidered 187

but instead something different, possibly a discourse-related element belonging to the same general class as the modal particles (cf. Koeneman & Neeleman 2001; Rosengren 2002: 177; Mohr 2005; Hartmann 2008). Dutch, too, permits peripheral expletives to occur with clause-internal er: (39)

Er heeft (er) iemand een appel gegeten there has there someone an apple eaten ‘Someone (in that context) has eaten an apple.’ (Koeneman & Neeleman 2001: 230)

If clause-internal er were in fact a vP-expletive on a par with Afrikaans daar, one would not expect this element to be spelled out both clause-internally and peripherally within the same presentational sentence ((40) reflects the vP-fronting analysis proposed in Richards & Biberauer (2005)):37 (40)

[CP Er heeft [TP [vP er iemand [vP [VP een appel gegeten] heeft] heeft]]]

As (40) shows, standard views on chain reduction would lead us to expect the lower copy of a fronted vP-expletive to be deleted. Crucially, the Afrikaans counterpart of (39) is impossible, with post-verbal daar in a daar-initial structure necessarily receiving a locative adverbial interpretation.38 As (41) shows, the Afrikaans facts follow straightforwardly on the basis of the assumptions made about this language so far: (41)

a. Daar het (*daar)/DAAR iemand ’n appel gee¨et There have there someone an apple eaten ‘Someone over there ate an apple.’ b. [CP Daar het [TP [vP daar iemand [VP ’n appel gee¨et]] het]] c. [CP Daar het [TP [vP daar DAAR iemand [VP een appel gee¨et]] het]]39

Given the above, it would seem that Dutch may in fact be more like German than is generally thought and that this language may lack a non-argumental vP-expletive just as German does. If this is correct, er-optionality may simply be a consequence of the optionality of the non-expletive class of adverbial elements to which it belongs, which, in turn, implies that its absence cannot, as has standardly been done, be viewed as evidence in favour of the availability of proEXPA in Dutch. Evidently, much more needs to be learned about the precise nature of da and er (which is famously homophonous – cf. Bennis (1986) and Neeleman & van de Koot (2006) for discussion of Dutch’s

37

38 39

The fact that (39) is interpreted as a presentational on a par with other structures introduced by clause-initial er suggests that it is correct to analyse the initial element as a CP-expletive, parallel to German es and Icelandic það. Like Dutch er and English there, Afrikaans daar is homophonous with the locative adverbial. We ignore the possibility that locative daar may in fact have been first-merged lower in vP than indicated here as it does not affect our argument.

188

Theresa Biberauer

numerous ers; cf. Hartmann (2008: chapter 4) on partially similar problems afflicting syntacticians’ understanding of German da). What seems clear, however, is that these elements are different from Afrikaans (non-adverbial) daar and that our findings regarding the inventory of expletive elements in Germanic accordingly needs to be revised as follows: (42)

Inventory of expletive elements in Germanic Language

CP-Expl

vP-Expl

Icelandic German Dutch Afrikaans

það es er –

English

–

Mainland Scandinavian

–

– Quasi-argumental es quasi-argumental het quasi-argumental dit; non-argumental daar quasi-argumental it; non-argumental there quasi-argumental det; non-argumental der

Based on the discussion in this section, then, we might conclude that Dutch’s similarity to German in the relevant respects entails that it should be analysed as a head-pied-piping vP-raising language rather than as a spec-pied-piper, as suggested by Richards & Biberauer (2005). The following section reconsiders the feasibility of postulating a head-pied-piping analysis for German. 3.4.2 German As noted above, German exhibits ‘Diesing effects’ in both subject and object contexts, with ‘higher’ subjects necessarily receiving an old information/strong interpretation that is not assignable to ‘low’ subjects. In the object domain, this interpretive alternation has been analysed by appealing to the presence vs absence of optional EPP-features (cf. Biberauer & Richards (2006) for discussion and references). Consider (43): (43)

a. Er hat oft ein Buch gelesen he has often a book read ‘He often read a (non-specific) book.’ (weak reading; cf. Diesing 1992) b. Er hat ein Buch oft gelesen he has a book often read ‘There’s a book that he often read.’ (strong reading; cf. Diesing 1992)

In accordance with the Fox–Reinhart intuition on optionality, in terms of which optional features, operations, etc. must deliver some extra effect that would have been absent in the absence of these features, we can legitimately view (43a) and (43b) as the outputs of two different numerations: specifically, that which

Semi null-subject languages, expletives and expletive pro reconsidered 189

delivers (43b) can be said to feature a movement trigger which is absent in (43a). Following Chomsky (2001: 34ff.), Biberauer & Richards propose that the ‘Diesing Effects’ in (43) are the consequence of v’s D-probe bearing an optional movement trigger (i.e. being [D*]) in (43b), but not in (43a). Extending this proposal to German – and Dutch – subject effects, we might most naturally expect the D-probe on T, likewise, only optionally to be associated with *. Rather than being a head-pied-piping language, then, German and Dutch would be languages in which SpecTP is only optionally occupied by the subject (cf. also Wurmbrand (2006) and Kratzer & Selkirk (2007), who arrive at the same conclusion on the basis of, respectively, LF and PF considerations). Given the lack of evidence for a clause-internal (nonargumental) expletive in these languages (cf. Section 3.4.1 above), the fact that there is no position that always needs to be occupied by a subject-element and the fact that both German and Dutch have at least some auxiliaries which are most plausibly merged within the thematic domain (both languages exhibiting auxiliary selection), this would seem to be the simplest hypothesis for German- and Dutch-acquiring children. Consider (44) below: (44)

a. [CP weil [TP [vP ja doch Linguisten [VP Kammermusik gespielt] haben]] tvP] b. [CP weil [TP [vP ja doch Linguisten [VP Kammermusik gespielt] haben]]]

As (44) clearly shows, the input motivation for postulating vP-raising in languages with v-auxiliaries is not the same as that for languages like Afrikaans in which, lack of auxiliary selection aside, alternations such as those discussed in Section 3.2 signal the plausibility of postulating a system in which vP-raising alternates with English-style DP-raising. This is shown in (45): (45)

a. [CP dat [TP [vP sy dikwels [VP Chopin gespeel] v] het] tvP] b. [CP dat [TP [DP sy] het] [vP tsy dikwels [VP Chopin gespeel] v]]

Given the above, we therefore conclude that Afrikaans on the one hand and German and Dutch on the other are, in fact, quite different systems where their T-related EPP-satisfaction mechanisms are concerned. Having questioned the status of German, we now reconsider Icelandic. 3.4.3 Icelandic At least three empirical facts point to the possibility that Icelandic may in fact resemble German and Dutch in having an optional EPPrequirement on T: the interpretive effects associated with different subject positions in this language (cf. i.a. Bobaljik & Jonas (1996), Bobaljik & Thra´insson (1998), Sigurðsson (2000), Vangsnes (2002), Rosengren (2002), and see Thra´insson (2007) for overview discussion); the option of not realising generic/impersonal subjects (cf. Holmberg (Chapter 2)); and SF. Let us consider these in turn.

190

Theresa Biberauer

The following example (adapted from Jonas 1996: 63) illustrates the fact that occupation of the Icelandic subject-position is, among other things, closely connected to information-structural considerations: (46)

Við settum færeyskar bækur fyrir ma´lstofuna ´ı lestrarsalinn . . . we put Faroese books for seminar-the in reading-room-the ‘We put Faroese books for the seminar in the reading room . . .’ (i) . . . en það hafa margar af þessum bo´kum verið lesnar a´ður . . . but there have many of these books been read before ‘. . . but many of these books have been read before’ (ii) %. . . en það hafa ___verið lesnar margar af þessum bo´kum a´ður but there have been read many of these books before (iii) %. . . en það ho¨fðu ´ıslenskar bækur verið lesnar ´ı fyrra but there had Icelandic books been read last year (iv) . . . en það ho¨fðu ___ verið lesnar ´ıslenskar bækur ´ı fyrra . . . but there had been read Icelandic books last year ‘. . . but Icelandic books had been read last year’

As the continuations indicate, Icelandic subjects can occupy ‘lower’ and ‘higher’ positions, with old information margar af þessum bo´kum, referring back to the Faroese books previously mentioned in the scene-setting phrase, having to occupy the higher immediately post-verbal position (i.e. SpecTP) – hence the infelicity of (46ii) in this context – and new information ı´slenskar bækur necessarily occupying the lower position, as shown in (46iv). As the well-formedness of structures like (46iv) shows, Icelandic SpecTP need not be (overtly) filled. Taken in isolation, one might be tempted to propose firstmerge of það in SpecTP in structures of this type (assuming TP-expletives to exist – cf. footnote 28), but this proposal seems implausible if one further considers the fact that SpecTP may likewise lack overt content in the context of impersonal and yes/no interrogative structures. Consider (47) in this connection: (47)

a. Nu´ ma´ (maður) fara að dansa now may one go to dance ‘One may begin to dance now.’ b. Hafa (*það) komið margir stu´dentar? have there come many students ‘Have many students come?’

(47b) is particularly telling since this structure clearly features a ‘low’ subject (margir stu´dentar may, in fact, be located in its first-merge position as the complement of unaccusative komið), and its unmarked declarative counterpart (Það hafa komið margir stu´dentar) obligatorily requires the expletive that is barred in (47b). Given these facts, then, it seems justified to propose that Icelandic SpecTP resembles its German and Dutch counterparts in not

Semi null-subject languages, expletives and expletive pro reconsidered 191

necessarily having to be projected (cf. also Wurmbrand (2006) for further arguments in favour of this conclusion). There does, however, appear to be an important difference between Icelandic on the one hand and German and Dutch on the other. While the nonobligatory projection of SpecTP can account for the interpretive effects illustrated in (46) and can also, in view of Icelandic’s more general tendency not to lexicalise arguments of very general or indeterminate reference,40 account for the non-obligatory presence of overtly realised impersonal subjects, it cannot account for its SF structures. Consider (18) above and also the subject wh-interrogative in (48): (48)

a. Hver heldur þu´ að stolið hafi hjo´linu? who think you that stolen has bike-the ‘Who do you think has stolen the bike?’ b. Hver heldur þu´ að ___ hafi stolið hjo´linu?

SFed elements are generally thought to involve movement to SpecTP (cf. Holmberg 2000; 2006; pace Jo´nsson 1991; and Hrafnbjargarson 2004a; 2004b), in which case SpecTP in (48a) must be filled (at least) by stolið, while the corresponding position in (48b) is either unoccupied or occupied by the lower copy of the subject or a subject operator. Regardless of SpecTP’s contents in (48b), however, (48a) – and, likewise, (18b) – constitutes a problem for the optional subject-raising analysis of Icelandic: [D*] on T would be expected to target only Agreed-with subjects, and not the range of elements that can potentially undergo SF (cf. Maling 1980; Holmberg 2006). We therefore propose that Icelandic T in fact differs from German and Dutch T in bearing an optional free-standing (i.e. non-Agree-related) movement diacritic (cf. also Holmberg (Chapter 2)). Like free-standing movement triggers more generally (cf. Chomsky (2008) on ‘Edge Features’), this * on T ‘blindly’ targets the highest specifier of T’s complement (vP), thus raising the subject wherever this occupies this highest Spec-vP – with the result that the subject in question will be interpreted as old information (cf. the discussion of (46) above) – and raising a higher-merged adverb (e.g. sentential adverbs)41 or a fronted element originating lower within the vP-domain

40 41

Cf. the discussion of the ‘weather’ expletive in Section 3.3, and see also Bolinger (1977). This proposal clearly requires one to assume that sentential adverbs, including those which are often thought to be merged within the TP- and CP-domains (cf. Alexiadou 1997; Cinque 1999), are merged at the edge of vP. While this may seem undesirable, it is worth noting that analyses of SF generally would seem to have to make this assumption as the proposal is consistently that adverbials occupying SpecTP in SF structures – or adjoined to T, in the case of T-adjunction proposals – have undergone raising from the lower position that they occupy in the corresponding non-SF structures. Thus (ignoring the first-merge position of the subject and the question whether relative clauses are raising or operator structures):

192

Theresa Biberauer

wherever one of these occupies outermost Spec-vP. In the absence of *, no raising takes place, with the result that SpecTP remains unprojected.42 The proposal that Icelandic T bears an optional free-standing * has various positive consequences. First, it enables us to understand why this language has no need for TP-expletives (strictly speaking, vP-expletives – cf. Section 3.4.1 above). Let us firstly consider the role of expletives that are ultimately spelled out in what would otherwise be the canonical subject position (thus not CP-expletives of the type found in Icelandic, German and Dutch). As Bobaljik (2002) shows, there is a very important difference between structures like (49a) and (49b): (49)

a. Someone must be (someone) in the garden must >> someone someone >> must b. There must be someone in the garden must >> someone *someone >> must

As (49a) shows, the expletive-less construction is ambiguous, with ‘A-reconstruction’ being possible: in Bobaljik’s terms, either the higher or the lower copy of someone may be privileged by LF for interpretation in this case. In (49b), by contrast, this choice is unavailable: only the lower copy may be interpreted. Because of the expletive facts discussed in earlier sections, we reject Bobaljik’s specific interpretation of these ‘right corner effects’ as indicative of PF expletive insertion, our focus, instead, being the interpretive effects at issue. The view that expletives are semantically vacuous is widespread (cf. Chomsky (1986) on the Expletive Replacement Hypothesis), but this view is undermined by the systematic interpretive differences between (49a)- and (49b)-type structures: wherever an expletive is present, only a single interpretive possibility arises, with DPs necessarily taking narrow

(i) (ii)

42

sa´ [CP sem [TP ___ hefur [vP sennilega skrifað þessa bo´k]]] he that has probably written this book sa´ [CP sem [TP sennilega hefur [vP sennilega skrifað þessa bo´k]]]

Mu¨ller’s (2004a) vP-fronting analysis of V2 similarly requires him to assume that all adverbials are merged vP-internally, as does Biberauer’s (2003) vP-fronting analysis of Afrikaans. This proposal raises a question about the lack of obligatory interpretive effects associated with SF: since SF entails the presence of the optional * on T, whereas its absence correspondingly signals the absence of *, the Fox–Reinhart approach to optionality would lead us to expect an obligatory interpretive effect. The analysis proposed here and also in Holmberg (Chapter 2) therefore only readily accounts for cases of SF that involve some ‘extra’ meaning (cf. Hrafnbjargarson (2004a; 2004b) and Mathieu (2006) for discussion). We leave this matter aside for future research.

Semi null-subject languages, expletives and expletive pro reconsidered 193

scope wherever scope questions arise.43 Overt expletives located in SpecTP, then, are not semantically vacuous as their presence in a structure precludes an otherwise available interpretive reconstruction possibility. To the extent that deleted expletive pronouns (i.e. proEXP–A) in canonical NSLs consistently ensure the same interpretive effects, they are, then, equally well motivated. This is precisely the argument made in Sheehan (2006: chapter 5) for a range of Romance NSLs, all of which can be shown to have obligatory D* on T.44 If we consider languages which lack this obligatory D*, however, it is not at all clear that the postulation of proEXP–A is motivated. Unlike in obligatory D* languages, there is no imperative to raise subjects to SpecTP in these languages, with the consequence that the interpretive effects that are facilitated by expletives can be achieved simply by the presence vs absence of * on T, whether this is associated with D, as in German and Dutch, or not, as in Icelandic. A second positive consequence of the proposal that Icelandic T bears an optional free-standing * is that it allows us to understand why this language lacks that-trace effects (cf. Lohndal (2009) for recent discussion and references): wherever T lacks *, subject-raising is predicted not to occur, with the consequence that the subject in question can be extracted from a lower position (cf. Rizzi 1982; 1986a; and also Rizzi & Shlonsky 2006).45 Furthermore, wherever * is present and SF satisfies this movement imperative, we also expect that-trace effects to be absent. This is correct, as (48a) illustrates. A further advantage of the optional free-standing * is that it opens up the possibility of understanding why Icelandic is in fact not a canonical NSL. On the assumption that this language is indeed ‘Italian-plus-V2’ (cf. Sections 2.2.1 and 3.3 above), we would expect T’s feature specification to be as follows: [D*], [V*] (*). V-to-T raising, then, simultaneously satisfies both T’s V- and D-related movement requirements, just as, on the ‘I-subject’ approach, it does in canonical NSLs (cf. Biberauer & Roberts (Chapter 7) 43

As has frequently been observed, the non-equivalence of sentences such as the following (cf. Milsark 1974) also strongly undermines the Expletive Replacement Hypothesis: (i) (ii)

44 45

There were not many people in the room. Many people were not in the room.

The lack-of-interface-properties argument that is often used as an argument against the postulation of proEXPA can therefore be cast aside as irrelevant. The same is, of course, predicted to be possible in German and Dutch wherever T’s D-probe lacks its optional *. The empirical facts for these languages are not entirely clear at this stage (although cf. den Dikken (2007) for a recent investigation of the Dutch facts, Haegeman (1992) on West Flemish, and Bayer (1984) and Mayr (2009) for discussion of Bavarian German), but it does seem that varieties which permit extraction from finite complements more generally also permit (superficial) that-trace violations under circumstances that are sensitive to discourse-linking considerations. We leave this matter to future research.

194

Theresa Biberauer

for discussion). Icelandic T, however, optionally bears an extra movement diacritic which will target the subject wherever no other element is present at the vP-edge, with information-structural consequences for the interpretation of raised subjects. On this view, the difference between Icelandic and the canonical NSLs is therefore that subject-raising in the former is the consequence of an optional free-standing * on T, whereas in the latter, this movement is the consequence of an optional free-standing * on a head within the C-domain (cf. A&A on the A0 -nature of initial subjects in canonical NSLs and see Sheehan (Chapter 6) for discussion). Regardless of which analysis of canonical NSLs turns out to be correct, Icelandic will, on the account proposed here, thus differ from these languages in having a T associated with an optional free-standing *. If we now make the not unreasonable assumption that the difference between an Agree-related and free-standing * is legible to PF46 and, furthermore, that free-standing * on T differs from free-standing * on C in categorically disallowing post-spell-out deletion, we have an account of why Icelandic pronominal subjects must always be overtly realised in Icelandic, but not in the canonical NSLs: pronouns moved by Agree-mediated * may, under appropriate circumstances, be deleted (see Roberts, Chapter 1, Sheehan, Chapter 6 and Holmberg, Chapter 2 for discussion);47 those moved by freestanding *, by contrast, may not. An ‘architectural’ consideration comes to mind as a possible reason as to why free-standing * on T may behave differently to a corresponding * on C: if one adopts the stricter version of the PIC originally proposed in Chomsky (2001) (cf. i.a. Mu¨ller 2004c; Svenonius 2004; and Richards 2004, 2007 for discussion), there is a question about how material located in root C and its edge comes to be spelled out. Various researchers, including Fitzpatrick (2006), Rizzi (2006a), Haegeman (2006) and Sigurðsson & Maling (2008) have proposed that material in this ‘final edge’ may, but need not, be subject to deletion operations that cannot apply to non-final edges – in Fitzpatrick’s terms, material in final edges may, uniquely, remain ‘uninterpreted’ by PF. As this option will not be available to material moved to SpecTP, we predict that such material will always, in the absence of a Romance-style, context-sensitive deletion possibility, be

46

47

This difference may in fact also relate to the parametric difference proposed in Baker (2008a; 2008b). In terms of his Direction of Agreement Parameter, the various functional heads in a language either do or do not Agree with the XPs located in their specifiers. We leave this matter to future research. In our terms, German is a language in which optional Agree-mediated movement to SpecTP takes place. Null subjects will, however, be ruled out in this language either because T does not host the nominal features required to render pronouns a subset of the features on T (the deletion analysis) or because V fails to raise to T (the ‘I-subject’ approach; cf. Section 2.2.2 above on the non-existence of (15b)-type grammars).

Semi null-subject languages, expletives and expletive pro reconsidered 195

spelled out. The fact that Icelandic is not a canonical NSL may therefore be the consequence of a more widely relevant distinction between phase-internal and phase-edge specifiers and the circumstances under which their contents may remain unrealised. 3.4.4 Summary and some diachronic speculations In the preceding sections, we have argued that the Germanic languages cannot all be viewed as systems in which SpecTP is a specially dedicated subject position; they in fact vary in respect of their degree of subject-prominence. Specifically, the proposal is that only English, MSc and, on the deletion analysis, canonical NSLs have a canonical subject position in the sense of the original EPP, i.e. a position which must always be occupied either by a subject or a subject-related expletive. Afrikaans consistently requires SpecTP to be filled, but by a fronted vP containing a subject-element in its highest specifier, with English-style subject DP-raising being an alternative option in the modern spoken language.48 Icelandic, German and Dutch only optionally require subject-raising to SpecTP, with this raising being Agree-based in the latter two languages, but not the former. We have also argued that the different status of SpecTP in the various Germanic languages correlates in important ways with their expletive inventory, with the languages only requiring optional SpecTP projection systematically lacking the vP-expletives (i.e. expletives capable of standing in for subjects in order to fill SpecTP) found in English-type languages. Given that these languages lack overt vP-expletives and demonstrably have an alternative strategy to achieve what English-style languages achieve via vP-expletive insertion – optional subject-fronting – we concluded that none of the optional SpecTP projection languages license proEXP–A. Furthermore, the absence of a specially designated subject position leaves optional SpecTP projection languages with the option of not obligatorily lexicalising referentially indeterminate subjects, an option which Icelandic, which lacks an overtly realised weather quasi-argument, employs to a greater extent than German and Dutch, which, depending on one’s analysis of es and er in impersonal passives, may 48

This is true for embedded clauses, which do not permit scrambling over the subject, i.e. focusscrambling of the sort discussed in footnote 27 is not possible in Afrikaans embedded clauses. Scrambling to the edge of the vP-phase – the sort that is often thought of as being triggered by an extra Edge Feature on v – is, of course, available in matrix contexts. As scrambled XPs of this type are, however, ultimately spelled out clause-initially (in SpecCP), vP-raising will still result in the appearance of the raised vP in SpecTP having a subject(-related) element in its highest specifier. This is schematised below: (i)

[CP XP Vf [TP [vP XP subj . . . v] Vf] [tvP]]

Obviously, various technical questions remain – e.g. whether C is able to extract the clauseinitial XP from SpecTP or whether C and T probe simultaneously (cf. Chomsky 2008), how appropriate copies are spelled out and deleted, etc. These are left for future research.

196

Theresa Biberauer

permit non-lexicalisation of this type of subject. In the quasi-argumental expletive domain too, then, Icelandic, German and Dutch do not require the postulation of a null subject (proEXPþA). This leads us to the conclusion that none of the canonical semi NSLs in fact licenses any species of proEXP. This, in turn, entails that these languages cannot belong to the class of semi NSLs as originally defined. More generally, what we have seen in the Germanic context here suggests that semi NSL-type behaviour depends more on the status of SpecTP than, as has traditionally been assumed, on inflectional considerations. Inflectional considerations can, of course, be relevant in that insufficiently rich languages cannot have ‘pronominal agreement’, which further entails that they will not be able to satisfy D* on T via V-to-T movement and will, therefore, if D* is obligatory, require expletives. Like its predecessors, then, the present proposal predicts loss of inflectional richness to bring about changes in V-raising and subject-related behaviour: where a D* was previously satisfied via V-to-T raising, this will no longer be possible when agreement morphology can no longer be analysed as ‘pronominal’ by acquirers. Unlike its predecessors, however, our proposal does not entail that the loss of ‘pronominal agreement’ will trigger the abrupt introduction of English-style subject phenomena. Instead, the possibility exists that a language which formerly satisfied D* via V-raising and which permitted optional XP-fronting to SpecTP (triggered by free-standing *) may in the first instance be reanalysed as a language lacking V-to-T movement, with optional XP-raising continuing as before. Since subjects are likely to be the most commonly raised XPs, particularly if subject-raising vs. nonraising is associated with information-structural considerations of the type that seem to be operative in Icelandic, free-standing * on T may subsequently be reanalysed as D* on T, at which point English-style subject behaviour will become obligatory. This gradual change from D* satisfied via V-raising to D* satisfied via DP-fronting is particularly likely to occur in V2 languages, where an additional V-movement operation (V-to-C) obscures the extent to which V-movement occurs independently of V2. This, we submit, is in fact precisely why the MSc languages, which at one stage in their history exhibited Icelandic-style generalised V2 (cf. Holmberg & Platzack 1995), underwent the changes they did in the way they did (cf. Falk (1993a; 1993b) for detailed consideration of the verb- and subject-related changes that Swedish underwent during the course of its history. The ‘lag factor’ between the loss of agreement and final loss of subject-position-related phenomena like missing expletives, SF and consistent subject-raising is precisely what is predicted here.). Similarly, the ‘lags’ often noted in the rise of an English-style canonical subject position in the history of French may be the consequence of this language – which has also often been said to have initially been V2 (cf. i.a. Adams 1987a; Roberts 1993b; Vance 1997) – having initially had an

Semi null-subject languages, expletives and expletive pro reconsidered 197

Icelandic-style optional * on T, with the modern-day D* only having developed subsequently.49 Clearly, much further research is required to establish the validity of these proposals; our main objective here has simply been to sketch out the diachronic possibilities opened up by the acknowledgement that even very closely related Indo-European languages may not all have the same D-related T-specifications. 4

Conclusion

This chapter set out to re-evaluate the feasibility of postulating proEXP and a class of NSLs permitting only this null subject, the so-called semi NSLs. Based on empirical and theoretical arguments, we have concluded that expletives – both quasi-argumental and non-argumental – may be ‘missing’ for a range of reasons, that it is thus not justified to postulate proEXP wherever an expletive fails to surface and, consequently, that it is not possible to identify a coherent class of semi NSLs, at least not in the sense of Rizzi (1986a). Foremost among the considerations determining the presence vs. absence of overtly realised English-style expletives is the nature of SpecTP, which we have seen, does not universally appear to be uniquely reserved for subjects and subject-related expletives; instead, as current Probe–Goal–Agree theory and the so-called Borer-Chomsky conjecture (cf. Baker 2008a and the Introduction to this volume) would lead us to expect, T’s association with movement diacritics appears to be subject to parametric variation. We have shown that this is even true for the Germanic language family, whose members do not all share an English-style subject position. As far as expletives are concerned, we have shown that only languages with an English-style SpecTP systematically require overtly realised expletives; languages in which SpecTP is either associated with an optional *, or in which T lacks a D-oriented * entirely, or in which T’s V-features are associated with * will therefore all be predicted to lack the systematic expletive behaviour observed in English on the basis of which proEXP was initially postulated. This is so because languages with an optional subject-raising trigger may use this mechanism to achieve what English-style languages achieve by means of expletive insertion (i.e. non-raising of the subject, with concomitant interpretive effects); they may, however, have ‘imposters’ which superficially resemble English-style expletives, as our discussion of Dutch er and German da has

49

If Biberauer & Roberts (Chapter 7) are correct, V-to-T raising in French and Romance more generally is not contingent on ‘pronominal agreement’, but instead on the richness of the tense system in these languages. Since all the languages in question have remained ‘tense-rich’ in the sense proposed by Biberauer & Roberts, the fact that they have retained V-to-T emerges as unsurprising.

198

Theresa Biberauer

shown (cf. also i.a. Sheehan (2006), Nicolis (2008) and Carilho (2008) and therein on expletive-like elements in Romance NSLs). Further, the existence of topic-related CP-expletives should also be borne in mind: as the discussion has shown, languages may have these without having English-style expletives, and, indeed, the expectation is that expletive elements surfacing in so-called topic-prominent languages that have an obligatory topic position (e.g. SpecCP or SpecTP where C or T bears an obligatory free-standing *) will be of this exclusively peripheral type. In general, though, we might expect topic-prominent languages to lack expletives completely as the presence of an optional * on C or T will deliver the same effect (cf. Li & Thompson (1976), who register a lack of expletives as one of the defining properties of topic-prominent languages). In turn, we would expect English-style expletives to be entirely superfluous in languages which completely lack * on T or with V*. This seems to be correct for V-oriented languages like the Slavic languages (cf. Franks 1995), Irish (cf. McCloskey 2001),50 and also for V-initial languages like Niuean (cf. Massam & Smallwood 1997), Malagasy (cf. Paul 2000), Chamorro (Chung 2005), Tagalog and Seediq (Edith Aldridge, p.c.), all of which have been argued to involve VP/vP/TP-movement either to SpecTP or to SpecCP (cf. Aldridge 2006 for discussion), since all of these languages lack overt expletives.51 Breton deserves special mention here: Jouitteau (2005) analyses this language as one featuring D-driven vP-raising to SpecTP – in Richards & Biberauer’s (2005) terms, Breton is a head-pied-piping language as T’s goal is a vD. Unsurprisingly, then, this language lacks English-style overt expletives, although Jouitteau (2005) highlights the existence of a preverbal expletive,

50

Welsh may be different as it appears to permit optional expletive realisation that in many ways seems to echo the situation in Dutch (David Willis, p.c.; as the gloss indicates, optional ’na is homophonous with the locative adverbial): (i)

51

a. Mae ’na wastad rhywun yno is there always someone there ‘There is always someone there.’ b. Mae wastad rhywun yno is always someone there ‘There’s always someone there.’ c. Mae rhywun wastad yno is someone always there ‘Someone is always there.’

The precise nature of Welsh ’na and of the language’s T-specification awaits closer investigation. The same is true of T in Arabic varieties which permit overt expletives and which additionally exhibit so-called anti-agreement effects (cf. Ouhalla 1991). Aldridge (2006: 6), in fact, registers optionality in respect of subject-raising in Tagalog, a topic-prominent language. In our terms, the absence of null expletives is therefore expected on at least two counts.

Semi null-subject languages, expletives and expletive pro reconsidered 199

bez (a shortening of infinitival bezan~ – ‘to be’), whose sole function is to prevent the inflected verb from raising to initial position. Evidently, then, much more needs to be learned about the precise range of expletives and expletivelike elements attested in the world’s languages; what seems clear, though, is that the type(s) of overtly realised expletives found in a language will reflect ‘core’ aspects of that language’s grammatical orientation (D-prominence vs V-prominence, subject-prominence vs. topic-prominence, etc., with these notions clearly not being primitive as overlaps are very evidently possible – cf. Tagalog, which is both V-oriented and topic-prominent). Given the above, we therefore expect many languages to lack overtly realised expletives, exactly as the empirical record suggests. Further, since ‘missing’ expletives can be the consequence of quite different parametric settings – just as ‘missing’ referential subjects can be the reflex of a range of syntactic and post-syntactic factors (cf. Holmberg, Chapter 2 and Biberauer 2008a) – it is clear that postulating proEXP wherever a structure lacks an English-style expletive is unjustified. The proposals discussed here lead us to expect that proEXP may in fact only be licensed in NSLs which can be demonstrated to have an English-style EPP-requirement co-existing with the kind of nominal T-specification that sanctions pronoun deletion; for other systems, English-style expletives are expected to be extraneous. Without proEXP and acknowledging that a range of grammatical properties may underlie the absence of overt expletives, it is clear that the notion ‘semi NSL’ and, with it, the null-subject typology in (3,4) cannot be upheld. This, however, does not entail that a parametric approach to null-subject phenomena is likewise unsustainable (cf. Newmeyer 2004; 2005; 2006). On the contrary, the discussion in this chapter has shown that it is precisely the featural specifications of T and those of the substantive lexical items and FCs with which it interacts that determines whether a given system will or will not exhibit what appear to be expletive null-subject phenomena and what other subject- and non-subject-related properties this will correlate with. The original insight that parameter interaction underlies the availability of different types of null-subject systems therefore remains unchanged; what has changed is our understanding of the nature of parameters and, consequently, of the types of grammars that may each produce what, superficially, looks like ‘the same’ phenomenon.

5

The null generic subject pronoun in Finnish: a case of incorporation in T Anders Holmberg

Introduction1

1

Finnish does not have an overt generic pronoun corresponding to English one, French on, German man or Italian si. The Finnish counterparts to English, German, etc. constructions with a generic pronominal subject have no overt expression of a subject. (1)

a. Ta¨ssa¨ istuu mukavasti. here sits comfortably ‘One can sit comfortably here.’ b. Kesa¨lla¨ hera¨a¨ aikaisin. summer-ADE wakes.up early ‘You wake up early in the summer.’ c. Sita¨ ei kannata aina valittaa. EXP not should always complain ‘It’s no use always complaining.’

A possible analysis of (1a,b,c) is that they have a null generic pronoun subject. Indeed, scholars who have investigated or commented on the construction have assumed that it contains a syntactically active generic subject pronoun, and have presented arguments to that effect (see Hakulinen & Karttunen 1973; Laitinen 1995; 2006; Vainikka 1989; Vainikka & Levy 1999; Holmberg 2005). In the present chapter I will try to determine in as much detail as possible some of the formal syntactic properties of the Finnish generic pronoun, or G-pronoun, as I will call it. I will name the construction which, by hypothesis, contains this category the generic subject construction, or GSC for short. The following questions will be adressed:

1

Thanks to the following people for data and comments on earlier versions of the chapter: Seth Cable, Abdelkader Fassi Fehri, Mayumi Hosono, Dimitra Kolliakou, Satu Manninen, Marc Modesto, Aarti Nayudu, Hannu Reime, Norvin Richards, Eugenia Romanova, Ur Shlonsky, ´ . Sigurðsson, K.V. Subbarao, Helena Sulkala, Kriszta Szendroi, Anne Vainikka. Halldor A The following less common abbreviations are used: ABL: ablative; ADE: adessive; ALL: allative; CON: conditional; ELA: elative; INE: inessive; PAR: partitive; PRC: participle; PX: possessive suffix; TRA: translative.

200

The null generic subject pronoun in Finnish

201

(a) Where does the Finnish GSC fit in the typology of impersonal constructions? (b) What are the arguments that the GSC has a subject? (c) What features does this subject have? Specifically: (d) Does it trigger agreement? (e) Does it receive case? (f) Why does it not satisfy the EPP? It will be shown that the pronoun receives case, triggers agreement, binds anaphors and generally does everything overt subjects do, except that, unlike Finnish definite null subjects, it is unable to satisfy the (Finnish variety of the) EPP. The chapter will focus on the explanation of this, in a sense surprising, combination of properties, and consider the theoretical implications of it for a theory of null arguments. An analysis is considered according to which the Finnish GSC does not have a subject other than the 3SG agreement on the finite verb or auxiliary, which in Finnish would have the properties of a generic pronoun. This analysis is rejected. Instead, building on Roberts’ (forthcoming b; Chapter 1) theory of incorporation, it is argued that the reason why the G-pronoun is obligatorily null in Finnish, and fails to satisfy the EPP, is that it is a non-head member of an argument chain headed by finite T, and as such is not spelled out and is inaccessible to the EPP. In the final section it will be shown that the Finnish GSC is characteristic of partial null-subject languages, a type of languages which includes Brazilian Portuguese, Hebrew, Marathi and Icelandic, but is different from the GSC of consistent null-subject languages such as Arabic, European Portuguese, Greek, Italian, etc. 2

The Finnish generic pronoun and the typology of impersonal pronouns

In a recent paper on the typology of impersonal pronouns Egerland (2003) distinguishes three distinct readings of the Germanic and Romance impersonal pronouns man (Swedish), on (French) and si (Italian): He dubs them the generic, the arbitrary and the specific reading.2 These are exemplified in (2a,b,c), respectively, with French examples (from Egerland 2003). (2)

2

a. On doit travailler jusqu’a l’age de 65 ans. ‘ON must work until the age of 65 years.’

(generic)

This terminology differs from that in the seminal work by Cinque (1988). In that paper Egerland’s generic and arbitrary readings are referred to as two readings of ‘arbitrary pro’ (proarb): the quasi-universal reading and the quasi-existential reading, respectively.

202

Anders Holmberg b. On a travaille´ pendant deux mois pour re´soudre le proble`me. (arbitrary) ‘ON has worked for two months to resolve the problem.’ c. Hier soir on a e´te´ conge´die´. (specific) ‘Yesterday evening ON was fired.’

In (2a) the reading of on is ‘people in general’. In (2b) the intended reading is roughly ‘some people’, and in (2c) it is ‘we’. In English the three readings are encoded by distinct pronouns: the generic reading by one or you, the arbitrary reading by they and the specific reading by we. In Finnish, the three readings are distributed over two impersonal constructions: the GSC and the passive:3 The GSC only has the generic reading, while the passive has the arbitrary or the specific reading. (3)

a. Ta¨a¨lla¨ saa tyo¨ta¨ jos puhuu saksaa. here gets work if speaks German ‘You get a job here if you speak German.’ b. Ta¨a¨lla¨ puhutaan saksaa. here speak-PASS German ‘German is spoken here/They speak German here.’ c. (Me) puhuttiin saksaa. we spoke-PASS German ‘We spoke German.’

(generic)

(arbitrary)

(specific)

Egerland shows that the arbitrary and specific readings of man, on, etc. are closely related, having essentially the same syntactic properties, distinct from the generic reading (see also Cabredo-Hofherr 2004). In Finnish this is reflected in the form of the verb: active 3SG for the generic pronoun, passive non-agreeing for the arbitrary and specific reading. In Finnish the affinity between the arbitrary and specific reading is particularly striking, as it has led to the passive becoming the normal form for expressing 1PL active meaning in colloquial Finnish, with an optionally overt 1PL pronoun subject, as exemplified by (3c) (see Helasvuo 2006).4 Egerland’s generic–arbitrary distinction does not, however, draw the line between the reading of the GSC and the reading of the passive construction in Finnish in quite the right place. The passive can have a generic, quasiuniversal reading ‘people in general’, as in (4a,b):

3

4

The proper classification of the Finnish passive is a controversial issue; see Shore 1988; Blevins 2003; Manninen & Nelson 2004; Helasvuo 2006. The issue is whether it is a passive or a form of active impersonal construction. For the present purposes the important property of the Finnish passive is that the understood subject, when not specific (meaning ‘we’), has arbitrary reference in Egerland’s sense, or, as discussed in the text below, the meaning of generic ‘they’. See Section 11 for some discussion of the passive. See Cinque (1988) for a suggested explanation for the (universal) tendency of the quasiexistential reading of proarb to be interpreted as 1PL. As shown by Lo¨flund (1998), the use of the passive form with specific, active meaning is, in fact, not restricted to the 1PL in Finnish.

The null generic subject pronoun in Finnish (4)

203

a. Keskiajalla ei syo¨ty sokeria. Middle-ages-ADE not ate-PASS sugar ‘They/People didn’t eat sugar in the Middle Ages.’ b. Ita¨vallassa puhutaan saksaa. Austria-INE speak-PASS German ‘They/People speak German in Austria.’

The difference between the GSC and the passive when used with generic meaning, is that the GSC includes the speaker and the addressee in in its reference (henceforth called the inclusive generic reading), while the passive is typically used to express a generic reading when the speaker and the addressee are not included (the exclusive generic reading). The difference can be exemplified by (5): (5)

a. Ranskassa syo¨ hyvin. France-INE eat-3SG well ‘You eat well in France.’ b. Ranskassa syo¨da¨a¨n hyvin. France-INE eat-PASS well ‘They eat well in France.’

The former implies that the speaker and the addressee also eat well in France, if, for instance, they visit the place. The latter does not have that implication; it is a statement about people in France, with no implication that it would ever be relevant for the speaker and the addressee to eat in France. The Finnish G-pronoun is therefore a close synonym of impersonal one in English, which is also known to have inclusive reference (see Moltmann 2006).5’6 Moltmann, in her semantic analysis of generic one, argues that it 5

Generic you is also typically inclusive (in English, and presumably in other languages making use of this form of generic pronoun). The use of 2SG with generic meaning is also not uncommon in colloquial Finnish (see Laitinen 2006), and the use of 3PL with arbitrary meaning occurs as well. (i)

(ii)

6

Sa¨ saat to¨ita¨ jos sa¨ puhut saksaa. you.SG get-2SG work if you.SG speak-2SG German ‘You get a job if you speak German.’ Ne puhuu Ita¨vallassa saksaa. they speak Austria-INE German ‘They speak German in Austria.’

The GSC can be used to refer specifically to the speaker, as in (i), from Laitinen (2006: 212). (i)

Aamulla sai kalaa. morning-ADE get-PAST.3SG fish ‘In the morning I/we got fish.’

Laitinen claims that the null pronoun in (ii) could refer to the addressee only, presumably when uttered by a doctor or nurse to a patient.

204

Anders Holmberg

is an indefinite expression introducing a variable in LF, bound by a generic null operator situated in SpecCP (with the effect that the generic subject always has scope over other quantifiers; Moltmann 2006: 263–64). The operator, then, has a feature [generic], and the pronoun has a corresponding uninterpretable feature [generic] (along the lines of Chomsky’s (1995b; 2000; 2001) feature theory), which means that it must be bound by the generic operator. Moltmann further argues that the variable in question also has an interpretable component responsible for the association with the speaker (a ‘first person feature’ of a sort). The precise feature composition of the pronoun will become relevant in Section 8, and I will return to it there. 3

There is a syntactically projected subject in the GSC

The most obvious argument against postulating a syntactically projected subject in the Finnish GSC is that it has no visible (i.e. audible) subject, and cannot have one; Finnish doesn’t have any overt generic pronoun. Furthermore, the EPP is not checked by any covert subject in the GSC, but has to be checked either by an expletive or by moving some other category, such as an object or a locative or temporal adverbial to the preverbal ‘EPP-position’ (see Holmberg & Nikanne (2002) and Section 6 below on the EPP in Finnish) (EXP ¼ expletive). (6)

a. *Istuu mukavasti ta¨ssa¨. sits comfortably here b. Ta¨ssa¨ istuu mukavasti. here sits comfortably c. Sita¨ istuu mukavasti ta¨ssa¨. EXP sits comfortably here ‘One can sit comfortably here.’

The most straightforward explanation of this is that the construction has no syntactically represented subject. Nevertheless, scholars who have investigated the GSC (Hakulinen & Karttunen 1973; Laitinen 1995; 2006; Vainikka 1989; Vainikka & Levy 1999) have argued that it does contain a covert but syntactically active subject. One piece of evidence that the GSC contains a syntactically active subject is that it may contain a subject-oriented anaphor (PX ¼ possessive suffix) (see Vainikka 1989: 232f.; Laitinen 1995). (ii)

Tuolla tavalla ei parane. that/ADE way-ADE not-3SG recover ‘You won’t recover that way.’

Alternatively the sentence meaning is the usual inclusive generic one, and the specific reference to the addressee is an implicature.

The null generic subject pronoun in Finnish (7)

205

a. Shelliasemalla voi pesta¨ auto-nsa. Shell-station-ADE can-3SG wash car-PX ‘You can wash your car at the Shell station.’ b. Sita¨ ei kuulu ottaa itsea¨a¨n liian vakavasti.7 EXP not-3SG should take SELFPX too seriously ‘One shouldn’t take oneself too seriously.’

The Finnish possessive reflexive suffix is an anaphor falling under Principle A of the Binding Theory (see Kanerva 1987; Vainikka 1989; Trosterud 1993). In (7b) it combines with the reflexive itse ‘self’, also an anaphor. The antecedent in (7a,b) is, presumably, the null G-pronoun subject.8 Another piece of evidence, put forth in Laitinen (1995), of a generic subject in the GSC is that it may contain an agent-oriented adverbial. (8)

Sinne ei muuta vapaaehtoisesti. there not move voluntarily ‘One doesn’t move there voluntarily.’

Furthermore, the GSC may contain an infinitival clause with a PRO subject evidently controlled by a generic subject in the matrix clause. (9)

a. Sita¨ voi pa¨a¨tta¨a¨ [PRO muuttaa Lappiin]. EXP can decide move Lapland ‘You can decide to move to Lapland.’ b. Ta¨nne tulee mielella¨a¨n [PRO ostamaan keramiikkaa]. here comes with-pleasure buy-INF pottery ‘It’s nice to come here to buy pottery.’

Whether we take these as valid arguments for a syntactically projected subject depends on our prior assumptions, though. For example, if an object anaphor can be bound by a verb which includes a subject argument in its argument structure, as in Williams’ (1994) f-theory, then obviously (7a,b) do not indicate a syntactically projected subject. This also applies in the case of control, as in (9). In Section 4, and again in Section 9, I will discuss an argument for a syntactically projected subject in the Finnish GSC which is not dependent in this way on prior assumptions about the mapping of argument structure onto syntactic structure.

7 8

The negation is Finnish is an auxiliary which inflects for subject agreement. The passive does not license a subject-oriented anaphor. (i)

Shelliasemalla pesta¨a¨n auto (*-nsa). Shell-station-ADE wash-PASS car (-PX) ‘A car is being washed at the Shell-station.’

This will be discussed below in Section 9.

206

Anders Holmberg

4

Does the G-pronoun trigger agreement?

Which f-features, if any, does the abstract G-pronoun have? Insofar as the G-pronoun triggers agreement, it should be possible to determine this from the agreeing predicate. The finite verb or auxiliary is inflected for 3SG in the GSC. This could be either because T agrees with a 3SG G-pronoun or it could be the default feature values of T, as seen in weather expressions, for example. (10)

Nyt sataa. now rain-3SG

Considering that the reference of the G-pronoun is typically semantically plural, often paraphrasable as ‘people in general’, there is an apparent mismatch between the inflection and the reference which indicates that the 3SG inflection is by default. As discussed by Vainikka (1989), in Finnish it is actually possible to determine with near-certainty which is the right analysis (see also Laitinen 2006). This is because there is a correlation between subject agreement and object case in Finnish. There are two non-partitive structural object cases in Finnish, morphologically distinct on singular lexical NPs: nominative and accusative. The nominative is the stem form of the noun, the accusative is marked by a suffix -n.9 The choice between them depends on the presence in the sentence of a nominative subject triggering agreement on T. If there is one, the object is marked accusative, if not the object is nominative (see Timberlake 1975; Maling 1993; Reime 1993; Kiparsky 2001). (11)

9

a. Mina¨ osti-n auton/*auto. I-NOM bought-1SG car-ACC/car-NOM ‘I bought a car.’ b. Mina¨ aion ostaa auton/*auto. I-NOM intend-1SG buy car-ACC/car-NOM ‘I intend to buy a car.’ c. Minun pita¨a¨ ostaa auto/*auton. I-GEN should-3SG buy car-NOM/car-ACC ‘I should buy a car.’ d. Osta auto/*auton! buy-IMP car-NOM/car-ACC ‘Buy a car!’

In traditional grammar the nominative object form is sometimes referred to as ‘the second accusative’; see also Reime 1993. It differs from the nominative subject case in that it does not trigger agreement. An(other) argument in favour of taking the bare NP object to be a form of accusative is that [þhuman] pronominal objects do not exhibit the nominative–accusative variation, but have one distinct accusative object form. An additional complication is that the form marked -n is homonymous with the genitive; see Vainikka (1993) for discussion.

The null generic subject pronoun in Finnish

207

e. Siella¨ ostettiin auto/*auton. there bought-PASS car-NOM/car-ACC ‘A car was bought there.’

In (11a) the verb agrees with the nominative subject, and the object is accusative. (11b) is a biclausal construction where the embedded verb that takes the object is infinitive, but the matrix verb agrees with a nominative subject, which is sufficient for triggering accusative on the object. In (11c) the matrix verb assigns genitive, a lexical case, to its subject. The genitive does not trigger agreement on T, which therefore has the default 3SG form. Consequently, the object of the embedded infinitive gets nominative. In (11d) the verb is imperative, a subjectless, non-finite, non-agreeing verb form, so the object gets nominative. In (11e) the verb is in the passive form, which is also a non-agreeing form, so again the object gets nominative.10 (12) shows that it is not the presence of an overt nominative subject which is the crucial factor triggering assignment of accusative on the object, but the presence of subject agreement. (12)

Me ostetaan auto/*auton. we buy-PASS car-NOM/car-ACC ‘We’re buying a car.’

(12) exemplifies the use of the passive in construction with a 1PL subject to encode 1PL active meaning. The presence of the overt subject pronoun, which is nominative but does not trigger agreement on the passive verb form, does not affect the form of the object. Now compare with the GSC: (13)

Jos ostaa auton/*auto, . . . if buy-3SG car-ACC/car-NOM ‘If you buy a car . . .’

Compare especially (13) and (11c). Both have a 3SG verb, but while the object is nominative in (11c), due to the absence of subject agreement, it is accusative in (13). This shows quite clearly that the 3SG form in (13) is not assigned by

10

The imperative provides a challenge for the theory of Finnish object case expounded in the text as it actually agrees with the subject (whether overt or covert). (i) (ii)

Osta (sina¨) auto. buy-IMP you.SG car-NOM Ostakaa (te) auto. buy-IMP-PL you.PL car-NOM

The relation between subject agreement and object case seems to be sensitive specifically to finite verb agreement, not to the morphologically distinct agreement found in the imperative.

208

Anders Holmberg

default, but is assigned by a subject. That is to say, there is a subject, a G-pronoun, in the GSC capable of triggering subject agreement. The form of the agreeing verb is 3SG, indicating that the G-pronoun has 3SG value. Note that VP-internal predicates open to subject agreement are also singular in the GSC, as pointed out by Laitinen (2006) (ESS ¼ essive).11 (14)

Cocktail-kutsuihin ei kuulu menna¨ na¨lka¨is-e-na¨/ *na¨lka¨is-i-na¨ Cocktail-party-ILL not should go-INF hungry-SG-ESS/ hungry-PL-ESS ‘One shouldn’t go hungry to a cocktail party.’

How come a 3SG pronoun is used to refer to what is a plural entity (‘people in general including you and me’)? The question obviously applies equally to overt generic pronouns such as English one, French on and German man. Presumably part of the answer is that 3SG is the least marked form and thus is open to the widest range of reference of all the forms inflected for f-features (although, in this light, it is unexpected that the 2SG form is also very widely used as a G-pronoun).12 The Finnish G-pronoun is a covert ‘one’, to put it simply. An important question is whether the G-pronoun is specified [human] (as claimed for the null arbitrary object pronoun in Italian in Rizzi (1986a), and for the Icelandic G-pronoun in Sigurðsson and Egerland (2009)). In fact, it follows from the inclusive property of the null G-pronoun that it has human reference. This suggests that postulating the formal feature [human] is redundant. On the other hand it does not follow from inclusiveness that the G-pronoun could not include inanimate objects along with the speaker in its reference, which, arguably, it cannot. It seems that (15) cannot include, for example, wardrobes in its reference along with people including the speaker. (15)

Ta¨ssa¨ ei voi seista¨. here not can stand ‘One can’t stand here.’

The alternative is that the reference to humans including the speaker follows from other properties of the construction (this seems to be what Laitinen (2006) is arguing). We return to the question of the feature [human] in Section 8. 11

Laitinen’s example sentence (i) is not, in fact, a good example of a G-pronoun triggering singular agreement. The null subject in the embedded clause, which the predicate pitka¨ ‘tall’ agrees with, is a null pronoun controlled by an implicit arbitrary argument, roughly ‘to anyone’, in the main clause. (i)

12

Joskus on eduksi etta¨ on pitka¨. sometimes is advantage that is tall-SG ‘Sometimes it’s an advantage to be tall.’

See Laitinen (2006) for discussion of these issues.

(Laitinen 2006: 211)

The null generic subject pronoun in Finnish

5

209

Does the G-pronoun have case?

For overt DPs, only those with nominative case can trigger agreement in Finnish. So it is entirely reasonable to take the G-pronoun in, for example, (15) to have nominative case. The G-pronoun may also occur in environments where other cases than nominative are assigned. Consider (16a,b): (16)

a. Sinun kannattaa vuokrata auto/*auton. you-GEN should-3SG rent car-NOM/car-ACC ‘You would do well to rent a car.’ b. Nyt kannattaa vuokrata auto/*auton. now should-3SG rent car-NOM/car-ACC ‘It’s good value renting a car now.’

The verb kannattaa ‘should, do well to’ along with a host of other necessive predicates (ta¨ytyy ‘must’, pita¨a¨ ‘should’, on pakko literally ‘is obligation’, i.e. ‘must’ etc.) takes a subject with genitive case (see Laitinen & Vilkuna 1993). Consequently no agreement is triggered, the verb has default 3SG form, and the object, if the main verb is transitive, gets nominative case. In (16b) the subject is the putative covert G-pronoun. The fact that the object in this case, too, has nominative case then strongly indicates that the G-pronoun gets genitive case, hence fails to agree with the verb, causing nominative case to be assigned to the object. If a predicate which assigns a lexical-semantic case, as the necessive predicates do in Finnish, has to assign its case, then (16b) provides additional evidence that the generic subject is structurally represented, as there is no other recipient of the genitive case (I return to this point in Section 7). Broadly speaking the generalisation is that the G-pronoun occurs in any case-environments where overt subjects occur. (17) is an example of a partitive subject, (18) an example of an adessive subject (PAR ¼ partitive).13 13

There are some subject contexts where the G-pronoun is excluded. For instance, in contrast with (16b), (ii) is preferably read as ‘if you are a dog’, not as ‘if you have a dog’. (i)

(ii)

Minulla on koira. I-ADE; is dog ‘I have a dog.’ Jos on koira, . . . if is-3SG dog

Conceivably the reason is that there are two competing derivations in this case, where the one leading to ‘if you are a dog’ wins, perhaps on account of not requiring oblique case on the subject. The following is another case: (iii)

Kallesta tulee kapteeni. Kalle-ELA comes captain ‘Kalle will become a captain.’

210

Anders Holmberg

(17)

a. Kallea a¨rsytta¨a¨ kun telkkarista tulee urheilua. Kalle-PAR irritate-3SG when TV-ELA comes sports ‘It irritates Kalle when they show sports on TV.’ b. Jos a¨rsytta¨a¨ kun telkkarista tulee urheilua, voi sammuttaa telkkarin. if irritate-3SG when TV-ELA come sports can turn-off TV ‘If it irritates you when they show sports on TV, you can turn it off.’

(18)

a. Minulla on pa¨a¨nsa¨rky. I-ADE is-3SG headache ‘I’ve a headache.’ b. Jos on pa¨a¨nsa¨rky se voi johtua nesteen puutteesta. if is-3SG headache it can derive.from dehydration-ELA ‘If you have a headache, it may be due to dehydration.’

We conclude that the G-pronoun has case, like overt subjects. 6

Why does the G-pronoun not satisfy the EPP?

One clear difference between the covert Finnish G-pronoun and other arguments in Finnish, overt or covert, is that it does not satisfy the EPP. For that reason the preverbal position of a GSC must generally be filled by a nonsubject category, often a locative phrase, or the expletive sita¨. The examples in (6) are repeated here as (19a,b,c). (19)

a. *Istuu mukavasti ta¨ssa¨. sits comfortably here b. Ta¨ssa¨ istuu mukavasti. here sits comfortably c. Sita¨ istuu mukavasti ta¨ssa¨. EXP sits comfortably here ‘One can sit comfortably here.’

The fact that the G-pronoun is phonetically null is not in itself a sufficient reason for this. Finnish is a partial pro-drop language, where 1st and 2nd (iv)

Jos tulee kapteeni, . . . if comes captain

Here (iv) cannot be read as ‘if one becomes a captain’, but only as ‘if a captain comes’. This may, again, be due to the existence of a competing derivation, leading to the same interpretation as (iii), namely: (v)

(vi)

Kalle tulee kapteeniksi. Kalle comes captain-TRA ‘Kalle will become a captain.’ Jos tulee kapteeniksi, . . . if comes captain-TRA ‘If one becomes a captain, . . .’

If so, this has potentially interesting consequences for the role of ‘transderivational constraints’ or competing derivations in the grammar.

The null generic subject pronoun in Finnish

211

person subject pronouns can always be null (more commonly in formal registers) and 3rd person subject pronouns can be null in embedded clauses if they have a close enough antecedent (see Holmberg and Sheehan, Chapter 3; Holmberg 2005; Chapter 2). As discussed particularly in Holmberg (2005), these definite null pronouns do satisfy the EPP, and are consequently incompatible with the expletive (as first noted by Hakulinen (1975)).14 (20)

a. (Mina¨) istu-n mukavasti ta¨ssa¨. I sit-1SG comfortably here ‘I’m sitting comfortably here.’ b. *Sita¨ istun mukavasti ta¨ssa¨. EXP sit-1SG comfortably here

According to Holmberg & Nikanne (2002) the EPP in Finnish requires a topic in the sentence-initial position here referred to as SpecTP. A topic is an expression denoting an individual or a group already established in the discourse, about which the predicate says something new (see Kiss 1995). Often the subject is a topic, in which case it occupies that position, but if the subject is focused, or if the subject is missing altogether, some other referential, topic-worthy category may take over filling SpecTP, for instance the object, as in (21), or a locative adverbial as in (19b). Failing that, the expletive sita¨ is inserted. (21)

a. Aila keitti kahvin. Aila-NOM made coffee-ACC ‘Aila made the coffee.’ b. Kahvin keitti Aila. coffee-ACC made Aila-NOM ‘The coffee was made by Aila/The one who made the coffee was Aila.’

Holmberg (2005) proposes the following formulation of the Finnish EPP:15

14

A 1st or 2nd person pronoun can cooccur with the expletive if the pronoun is not in specIP, but in a focus position, either a low focus position as in (i) or a high one (SpecCP), as in (ii); see Holmberg 2005 and Holmberg & Nikanne 2002 (PCL ¼ particle). (i)

(ii)

15

Sita¨ istun mina¨kin mukavasti ta¨ssa¨. EXP sit-1SG me-too comfortably here ‘I, too, can sit comfortably here.’ Mina¨ha¨n sita¨ istun aina ta¨ssa¨ tuolissa. I-PCL EXP sit always this chair-INE ‘Me, I always sit in this chair.’

I will follow the convention of assuming just one sentential functional head between v and C, labelled T. In fact there are several good reasons to assume a series of heads in this domain: see Holmberg, Nikanne, Oraviita, Reime & Trosterud 1993, Holmberg 2003b. The heads undergo movement/incorporation, though, so their features are (typically) combined into

212

Anders Holmberg

(22)

If the sentence contains one or more categories which can check the EPP, then one of them must remerge with TP /i.e. move to SpecTP/, or an expletive be merged with IP.16

The reason for the conditional clause is that the EPP is suspended if the sentence does not contain a referential category capable of serving as topic (see Holmberg & Nikanne (2002, for discussion). In (23), for example, the word kiire is not an argument but a (nominal) part of a complex predicate. In this case the sentence can be verb-initial (although merging the expletive sita¨ is also possible). (23)

16

Tuli kiire. came haste ‘There was a rush.’

one chain. As far as I have been able to determine, the simplification adopted here is not crucial for the argument, but greatly simplifies the presentation. Expressions such as (13), repeated and slightly modified in (i), look like exceptions to the EPP. (i)

Jos/kun ostaa auton Saksassa, . . . if/when buy-3SG car-ACC Germany-INE ‘If/when you buy a car in Germany, . . .’

This is quite general in conditional and temporal clauses. Typically these clauses are embedded in a main clause with a generic subject (but are optionally topicalised). (ii)

Sita¨ on oltava hyvin tarkka, jos ostaa auton Saksassa. EXP is be-PRC very careful if buys car Germany-INE ‘You have to be very careful if you buy a car in Germany.’

This would then be a case of an embedded null subject controlled by a matrix subject. As discussed by Holmberg (2005) and Holmberg, Nayudu & Sheehan (2009), in this case the null subject must be in SpecTP. But the phenomenon is more general: In (ii) the embedded clause subject cannot be controlled. (iii)

Tulli ei tuota ongelmia jos ostaa auton Saksassa. customs not cause problems if buys car Germany-INE ‘Customs don’t cause problems if you buy a car in Germany.’

I have no explanation for why conditional and temporal clauses look like exceptions to the EPP in Finnish. It should be pointed out that verb or auxiliary-initial GSCs occur in main clauses, too, as in (iv) and (v), from Laitinen (2006): (iv)

(v)

Ta¨ytyy harjata hampaat. must brush teeth ‘One/I/you must brush one’s/my/your teeth.’ Ei saa tulla sisa¨a¨n kenga¨t jalassa. not may come in shoes foot-INE ‘You may not come in with shoes on.’

However, minimal pairs (or triplets) like that in (6)/(19) can easily be multiplied, and clearly represent a productive phenomenon in Finnish, so the importance of the EPP of T in Finnish is hardly in doubt.

The null generic subject pronoun in Finnish

213

Conceivably, then, the reason why the G-pronoun cannot check the EPP is that it is not referential in the right way; it cannot be a topic. It is, however, not true that the EPP can only be checked by referential categories, capable of serving a topic function. (24)

a. Kuka tahansa voi tulla ta¨nne. who ever can come here b. Ta¨nne voi tulla kuka tahansa. here can come who ever c. *Voi tulla kuka tahansa (ta¨nne). can come who ever here

Kuka tahansa ‘whoever’ is an indeterminate quantifier, and as such not referential and capable of functioning as topic. In fact, if the EPP strictly called for a topic, we wouldn’t expect an expletive to be able to check the EPP. A closer approximation to the truth is that the subject may check the EPP even if it is not referential, but, for example, a quantified NP, while nonsubjects have to be referential and interpretable as topics, to check the EPP.17 (25)

a. Aila puhuu kenelle tahansa. Aila speaks who-ALL ever ‘Aila speaks to anyone.’ b. *Kenelle tahansa puhuu Aila. who-ALL ever speaks Aila

So the question then remains, why can’t a G-pronoun satisfy the EPP? An obvious difference between the null definite pronoun in (20a) and the null generic pronoun in, for example, (19b), is that the former is optionally null, while the latter is obligatorily null: there is no counterpart to (19b) with an overt G-pronoun. The Finnish Lexicon does not contain an overt G-pronoun. Is it possible that the EPP just does not see such a radically null category? Apart from raising problems with regard to the proper formulation of the EPP, this raises the question of why the G-pronoun does not have an overt form, in Finnish. Is it just an accidental gap? There are good reasons to think that it is not; as will be discussed in Section 9, it seems to be characteristic of partial null-subject languages that they have a null G-pronoun. Consequently, what we want is a theory which explains why the G-pronoun has no overt form, which will also explain why it cannot satisfy the EPP. In the following I will first sketch a theory which appears initially appealing, as it provides a straightforward explanation of the failure of the G-pronoun to satisfy the EPP. That theory will quickly be discarded, though, as being descriptively inadequate. Instead I will articulate a theory which 17

The problem in (25b) may be that the indeterminate pronoun object is necessarily information focus, which the subject left in situ is as well. If a sentence cannot have two disjoint information foci this is sufficient to rule out (25b).

214

Anders Holmberg

is descriptively adequate and provides a straightforward explanation why the G-pronoun is obligatorily null and fails to satisfy the EPP. 7

An inadequate theory

The following is a hypothesis based on a theory where the finite subject agreement marking in null-subject languages is an interpretable category, basically a pronoun which is morphologically realised as an affix. In a radical version of this theory finite sentences with null subjects actually have no other subject than the subject agreement morpheme; there would be no subject pro. Within such a theory the G-pronoun would not satisfy the EPP because it is a head, an AgrS, which in the absence of a lexical DP subject is able to itself carry the the subject y-role and absorb the subject case, although it is morphologically realised as a 3SG affix on the finite verb or auxiliary. Recall that Finnish has 1st and 2nd person null subjects but 3rd person definite null subjects only in restricted circumstances. (26)

(Mina¨) hera¨sin. I awoke-1SG (Sina¨) hera¨sit. you awoke *(Ha¨n) hera¨si. he/she awoke

Within a theory where the person inflection is an interpretable pronominal category AgrS, this would follow if 3rd person AgrS (singular and plural) in Finnish is an interpretable nominal category capable of carrying the subject y-role, but unspecified for definiteness. A definite subject reading, then, requires merge of a definite pronoun. In the absence of a definite pronoun, 3SG AgrS is still interpretable and able to carry the subject y-role, but can only be interpreted as generic. In other null-subject languages, 3rd person AgrS is, or at least can be, inherently definite, hence a definite reading does not require merging a definite pronoun. This theory is initially appealing because it accounts for the null subject facts in (26) and (potentially) explains why the G-pronoun does not satisfy the EPP. Assume that the EPP can, perhaps universally, be satisfied by a definite, pronominal AgrS (as argued by Barbosa (1995) and Alexiadou & Anagnostopoulou (1998)) or, at least in some languages including Finnish, by a category, which need not be the subject, merged as a specifier of T-AgrS. Then it follows that the EPP is not satisfied in the GSC unless a phrasal category is moved to specIP/AgrSP, or an expletive pronoun is merged there. The G-pronoun qua head cannot satisfy the EPP, while there is no obvious reason why it could not bind an anaphor, control PRO, or support an agent-oriented adverbial, all properties which the G-pronoun was shown to have, in Finnish.

The null generic subject pronoun in Finnish

215

This would provide a straightforward explanation of the generalisation, observed by Holmberg (2005), that languages which have 3rd person definite null subjects do not have a generic null subject, and vice versa, languages which have a generic null subject do not have a 3rd person definite null subject (I will come back to this correlation in Section 9).18 However, the contrast between (13) and (16b), already discussed in Sections 4 and 5, and repeated here as (27) and (28), shows that the hypothesis that 3SG AgrS is itself the generic pronoun cannot be maintained. (27)

Jos ostaa auton/*auto, . . . if buy-3SG car-ACC/car-NOM ‘If you buy a car, . . .’

(28)

Nyt kannattaa vuokrata auto/*auton. now should-3SG rent car-NOM/car-ACC ‘It’s good value renting a car now.’

As discussed, the contrast as regards object case was due to presence or absence of subject agreement: in terms of the theory expounded in this chapter, in (27) the verb agrees with the nominative covert G-pronoun, just as it would with an overt, definite subject. In (28) the verb does not agree, because the covert G-pronoun has genitive case, just as an overt definite subject would. The 3SG form in (28) is the default finite form; the verb kannattaa ‘should, be worth it’, like other necessive verbs, does not occur in any other finite form than 3SG.19 18

Under this theory the binder of the possessive reflexive in a GSC would be the 3SG AgrS. (i)

Siella¨ voi pesta¨ auto-nsa. there can-3SG wash car-PX3 ‘You can wash your car there.’

The theory can then explain why the Finnish passive construction does not license a possessive reflexive: the passive finite verb does not show any agreement, i.e. it does not have AgrS (see Holmberg, Nikanne, Oravita, Reime & Trosterud 1993). (ii)

19

Siella¨ pesta¨a¨n auto (*-nsa). there wash-PASS car (PX3) ‘A car is being washed there.’

Within this theory, the Finnish passive would be a radically subjectless construction. See Section 11. They may have past tense (kannatti) and conditional mood (kannattaisi). The necessive verbs ta¨ytyy ‘must’ and pita¨a¨ ‘should’ do not occur in non-finite form at all, while kannattaa may occur in infinitival form, as in (i). (i)

Nyt voisi kannattaa ostaa auto. now could be-worth buy car ‘It might be worth it now to buy a car.’

Kannattaa also occurs with agreement and a nominative subject with the meaning ‘support’. This, I assume, is a different verb.

216

Anders Holmberg

In a theory where AgrS itself is the G-pronoun, the existence of (28) is unexpected. Within this theory it is presumably the presence vs. absence of AgrS (of the interpretable kind, rather than just the default form of a verb) which determines the form of the object: if the sentence contains (interpretable) AgrS, the subject gets nominative and the object accusative case, if not, the object gets nominative case. In (28) there is no (interpretable) AgrS, shown by the nominative form of the object. But if there is no (interpretable) AgrS, then (28) has no category which could carry the y-role. We expect the sentence to violate the y-criterion. Seen from a different angle, if (28) were sharply ill-formed, this would constitute strong evidence in favour of the theory where there is no other G-pronoun in Finnish than AgrS itself, as under that theory (28) ought to violate the y-criterion. Correspondingly, since (28) is perfectly well-formed (with the right choice of case), this must be considered evidence against that same theory. 8

An adequate theory

Consider, however, the following alternative. First, we assume a theory of agreement and structural case along the lines of Chomsky (2001). According to this theory, finite T has a complex of unvalued f-features consisting of person, number, and in some languages (but not Finnish) gender. Once merged, T becomes a probe searching its c-command domain for a goal, that is a category which can value its f-features. The first category it encounters which (a) has a set of interpretable, that is inherently valued, features matching the unvalued features of T, and (b) has at least one unvalued feature itself which matches an inherently valued feature of T, will be the goal of the probe T. In a transitive construction the goal will be a DP in SpecvP, in an intransitive one, a DP in VP. This DP, unless it is already assigned a Case value (by an adposition, for example), assigns values to T’s f-features, and receives nominative Case value from T. If T also has an EPP-feature, the DP probed by T will move to SpecTP, more precisely, a copy of the DP will merge with TP, thus checking and thereby deleting the EPP-feature. That is to say, finite T has an inherent nominative (NOM) feature which values the [uCase]-feature of the nominal category probed by T. A more sophisticated version of this theory is proposed by Pesetsky & Torrego (2001; 2007), who argue that so-called nominative case is an unvalued tense-feature on noun phrases, assigned a value by the valued tense-feature of finite T. I will here assume the less sophisticated version, mainly for reasons of presentation. The G-pronoun is merged like a regular pronoun with vP in a transitive sentence, with V in unaccusative sentences, receiving the (subject) f-role exactly like other pronouns or referential expressions. It consists of nominal features, meaning that it has number (singular), person (3rd) and an unvalued Case-feature (which needs to be assigned a value). It is, however, deficient in

The null generic subject pronoun in Finnish

217

that it lacks a D-feature (a property it shares with overt generic pronouns such as English one, German generic man, etc.). As a pronoun, it also lacks a root: it is a bare f-feature complex. This means that when T probes this pronoun, and has its f-features valued by it, while at the same time valuing the Casefeature of the pronoun, T and the pronoun end up having the same feature values. More precisely, since T may have additional features (tense, and also perhaps mood), the subject pronoun’s feature values are a subset of those of T. (29)

T[uNr, uPn, NOM] . . . [vP [SG, 3, uCase] [v . . .]] ! T[SG, 3, NOM] . . . [vP [SG, 3, NOM] [v . . .]]

Now assume, following Roberts (forthcoming b; Chapter 1), that this means that the grammar takes them to be copies, forming an argument chain, effectively as if the subject pronoun had moved by head-movement to T, except there has only been Agree (mutual feature valuing), no movement. According to Roberts (forthcoming b), this is the derivation of what is commonly termed incorporation (of a head into a head): a probe–goal relation is copying the feature values of the goal onto the matching but unvalued features of the probe, and vice versa. In the special case where the goal’s features are a subset of the features of the probe, all the goal’s feature values will be represented in the probe as well, and the goal will thus be a copy of the probe. As such, the probe and the goal form a chain. It holds universally for chains that normally only one link of the chain is pronounced (unless there are special reasons to pronounce more than one, say, as a resumptive pronoun). Furthermore, generally the link which is pronounced is the highest one (unless there are special reasons not to pronounce the highest link) (see Bobaljik 2002; Landau 2006a). Thus we derive the effect of head-movement/incorporation as the result of feature copying under Agree. The probe, with features valued by the goal, will be spelled out (i.e. will be assigned phonetic form), the goal will not. If the goal has features not copied by the probe, then the goal will not be a copy of the probe. Then the goal will be the head of a chain, and as such will be spelled out (unless it undergoes other movement, i.e. a second merge in an even higher position). In the case of (29), the G-pronoun is a copy of T, and they thus form a chain, in fact, an argument chain (A-chain). Consequently, T is spelled out, as an affix on the finite verb or auxiliary, while its copy, the subject G-pronoun in SpecvP is not spelled out (or is spelled out as null).20 This means that there is a principled reason why the G-pronoun is null in Finnish; it is not just

20

If Narrow Syntax operates not with full lexical items but with syntactic feature complexes, along the lines of Distributed Morphology (Halle and Marantz 1993), then ‘spell-out’ means ‘vocabulary insertion’. Normally only one copy, the highest copy, in a chain is spelled out, the other copies remain null. Later (in Section 9), a crucial distinction is made between ‘spell-out’ and ‘pronunciation’: a chain can be spelled out and not be pronounced.

218

Anders Holmberg

an accidental lexical gap. It is null because it is a D-less pronoun probed by finite T. Pronouns probed by T which have a D-feature, that is definite pronouns, are not copies of T because they have a feature which T does not have, namely D. Instead, they are the highest member of their argument chain, and as such are spelled out, unless they undergo movement to the C-domain (i.e. merge with CP), as in the case of subject wh-pronouns, or if they have a local enough antecedent in a higher clause, which licenses deletion of the pronoun (see Holmberg, Chapter 2, and Holmberg and Sheehan, Chapter 3; I return to this case below). Lexical subjects, definite or indefinite, are never copies of T as they have a root which is not copied by T. I return below to the question of why in some languages all subjects, including G-pronouns, are pronounced. What about the EPP? The EPP is a feature of T which requires merge of a category with TP. As discussed, in Finnish this category is usually the subject, but if the subject is, for some reason, not available, another category may satisfy the EPP. Formally, by merging with TP the category in question checks and thereby deletes the uninterpretable EPP-feature. One situation when the subject is not available for merge with TP is when it is a chain headed by T, for obvious reasons: grammatical operations applying to a chain can only apply to the head of the chain, for reasons of locality, so the non-head copy in SpecvP is not accessible, and the head of the chain, T, cannot merge with itself.21 Holmberg (2005) showed that the EPP in Finnish cannot be checked by a D-feature, or other nominal features, in T, as has been claimed by Barbosa (1995) and Alexiadou & Anagnostopoulou (1998) to be characteristic of null-subject languages; the EPP strictly requires merge of a phrasal category with TP (see also Sheehan, Chapter 6). Definite subject pronouns are, then, attracted by the EPP, as they do not form a chain with T. They may value the f-features of T and be assigned a case by T, but as they have features not shared with T, they are heads of their own chain, and are thereby accessible to the EPP. This is the case even though language-particular conditions may allow them to end up not pronounced, if they find an antecedent in a higher clause.22

21

22

Roberts (forthcoming b) claims that in order for incorporation to happen the probe must not have an EPP-feature, since the same goal cannot be a copy of the probe and check its EPP-feature. However, if the EPP-feature in question can be checked by other categories than the goal, perhaps as a last resort when the goal is not available, as seems to be the case with the EPP of T in Finnish and also in other languages, for example Icelandic (see Holmberg 2000), then an EPP-feature is compatible with incorporation. According to Holmberg (Chapter 2), Holmberg and Sheehan (Chapter 3), and Holmberg, Nayudu & Sheehan (2009), the pronouns which are controlled are also deficient, having an unvalued D-feature, uD. Even an unvalued D-feature is a feature which T cannot copy, triggering movement to SpecTP.

The null generic subject pronoun in Finnish

219

We have thus achieved the goal of characterising the difference between the ‘lexically null’ G-pronoun (as in (19)) and the optionally null definite pronouns (as in (20)) in such a way that we can also explain why the former cannot check the EPP while the latter can. The question of what the exact feature composition is of the G-pronoun becomes highly relevant in this theory, since in order to be a non-head copy in a chain headed by T, the pronoun must not have any features that are not also represented in T. As mentioned, it is widely assumed that generic pronouns have a feature [human], on the basis of the observation that they can only refer to humans (Rizzi 1986a; Egerland 2003; Sigurðsson and Egerland, 2009). Since the feature [human] is a grammatical feature in the pronoun system of Finnish, distinguishing between the 3rd person pronouns se (SG) and ne (PL), which are unmarked for human, and the pronouns ha¨n (SG) and he, which are specified [human] (see Helasvuo & Laitinen 2006), this would entail no addition to the grammar of Finnish. However, we then have to assume that T also includes (unvalued) [human] among its unvalued f-features, since only then could a [human]-marked G-pronoun be a copy of T. This does not seem like an implausible extension of the feature matrix of T, even though it is not morphologically reflected (the expectation is that it would be, in some languages).23’24 Recall that Moltmann (2006) argues that one also has (a) an uninterpretable feature [generic], which means that it needs to be bound by a generic operator, and (b) an interpretable feature responsible for the ‘first person orentation’ of the pronoun. If we apply Moltmann’s theory of one to the Finnish G-pronoun, we are led to assume that T can have an uninterpretable feature [generic]. In that way the [generic]-marked pronoun can be a copy of T, and the chain (T, G-pronoun) can be bound by the generic operator in SpecCP.25 Similarly, in the case of the ‘first person orientation’ property, if we accept Moltmann’s theory, and consider first person orientation, i.e. inclusiveness, as a grammatically represented feature of the Finnish G-pronoun, then we have to assume that this feature has an unvalued counterpart in T. The alternative is to reject the analysis where inclusiveness is encoded as a feature, and derive it instead from other properties of the 23

24

25

Moltmann (2006), in fact, argues that ‘human’ is not the right notion for the G-pronoun one, but rather something like ‘conscious being’. This is a complication, if it can be shown that this does not hold true of the grammatical feature [human] in general that it distinguishes conscious from unconscious beings. It is possible that [human] as a feature of pronouns is in fact person. Only human-referring 3rd person pronouns would actually be 3rd person, while non-human referring pronouns would be personless. Alternatives may be considered, such as postulating an interpretable feature [generic] as a feature of (a variety of) C, which is passed down from C to T along with other formal features, assuming a relation between C and T as in Chomsky (2008); see text below.

220

Anders Holmberg

construction, presumably including the (assumed) fact that the pronoun is a D-less category bound by a generic operator. For example, the inclusiveness might arise as a default in this situation. I will have to leave this important issue for future research, though. 9

Non-nominative G-pronouns

We still have not got an account of G-pronouns which do not agree with T, and do not get their case from T. In the following I will deal specifically with G-pronouns with genitive case. We saw in Section 5 and in (28) that the G-pronoun can be assigned genitive case under certain conditions. The functional heads which are involved in case and agreement with the subject (SUB) and the object (OBJ) are shown in (30): (30)

C . . . T . . . Voice . . . SUB . . . v . . . OBJ

Following Chomsky (2008) I assume the f-features, the Case-feature and the EPP-feature of T derive from C, the head of the CP-phase,26 being ‘passed down’ from C to T. Following Roberts (forthcoming b) I assume that the f-features and Case-feature of v derive from Voice, the head of the vP-phase. Specifically, when Voice has the feature Active, the f-features and the Casefeature of Voice are passed down to v, the head assigning a y-role to the subject, which can then probe for the object, have its f-features valued by the object (though with no morphological reflex in Finnish), and assign accusative Case to it. This is shown schematically in (31), the lower line showing the situation after the features of active Voice are passed down to v, and v has assigned accusative to the object, and had its f-features valued by the object.27 (31)

Voice[ACT, ACC, uf] . . . SUB [uCase, 1SG] . . . v . . . OBJ[uCase, 3SG] ! Voice[] . . . SUB [uCase, 1SG] . . . v[ACT, ACC, 3SG] . . . OBJ[ACC, 3SG]

Next, the subject is probed by T, assigns values to T’s f-features, has its uCase-feature valued nominative, and is usually attracted by the EPP of T to merge once more with TP. This would be part of the derivation of the sentence (32): (32)

26 27

Mina¨ ostin auton. I-NOM bought car-ACC ‘I bought a car.’

A phase is a syntactic substructure which, when derived, is spelled out and interpreted. The phases are vP and CP, and arguably also DP and PP (see Svenonius 2004). Rather than ascribing an inherent ACC feature to active Voice, we might assume that accusative case is the nominal counterpart of the feature ACT, by analogy with NOM being the nominal counterpart of tense.

The null generic subject pronoun in Finnish

221

In the necessive construction Voice does not have the feature [ACT(ive)] but the feature [NEC(essive)]. This feature does not allow passing the f-features and case-feature of Voice down to v, but instead Voice probes for the closest nominal argument, the subject. The f-features of Voice are valued (with no morphological reflex) and the uCase-feature of the subject is assigned the value Genitive. The head v does not have Case-assigning capacity and T is too far away to probe for the object, so the object is assigned default Nominative value.28 (33)

Voice[NEC, GEN, Voice[NEC, GEN,

uf]

. . . SUB [uCase, 3SG] . . . v . . . OBJ[uCase, 3SG] ! . . . SUB [GEN, 3SG] . . . v . . . OBJ[NOM, 3SG]

3SG]

If the subject is a D-less 3SG pronoun, Voice and the G-pronoun form a chain, and the G-pronoun consequently remains not spelled out, as in (34) (¼ (28)). (34)

Nyt kannattaa vuokrata auto. now should-3SG rent car-NOM ‘It’s good value renting a car now.’

If the subject has a D-feature or other features not copied onto Voice, it will be spelled out, and usually be attracted by the EPP-feature of T to move, i.e. merge a second time with TP. This would be (part of) the derivation of (35), for example. (35)

Minun kannattaa vuokrata auto. I-GEN should rent car-NOM ‘I should rent a car.’

On a descriptive level, the necessive verb intervenes between T and the subject, assigning genitive case to the subject, leaving T without a goal to value its f-features. Unlike the situation in many other languages, for example Icelandic (see Sigurdsson 2004a) and the many ergative languages where T agrees with the object, Finnish T is not able to probe the object,

28

Voice with a Passive feature also does not allow passing of the f- and case-feature of Voice down to v, with a similar effect as in the necessive construction: T does not agree with the subject or the object, and the latter is assigned default nominative. Viewing [Necessive] as a Voice feature makes sense in that Passive and Necessive have complementary distribution (the inflection on the passive verb form is a morphological finiteness affix, a realisation of the category F, according to Holmberg, Nikannen, Oravita, Reime & Trosterud (1993)). (i)

(ii)

Ta¨ma¨ kirja oste-ta-an heti. this book-NOM buy-PASS-F at.once ‘This book is bought at once.’ Ta¨ma¨ kirja pita¨a¨ ostaa / *oste-ta heti. this book-NOM should buy-INF/ buy-PASS at.once ‘We should buy this book at once/This book should be bought at once.’

222

Anders Holmberg

either, and so must take recourse to default valuing of its f-features. However, the fact that there is a ‘necessive GSC’ shows that the necessive verb is more than just a genitive-assigning verb: it is a functional head which, like T, can form a chain with the subject. This is what is formally expressed by the theory articulated above. Within this theory the prohibition against a null 3rd person pronoun in Finnish cannot be seen as an effect of deficient agreement, as suggested above in connection with the paradigm (24). In the present theory agreement is a set of inherently unvalued f-features which do not affect interpretation. But in fact there is no prohibition. A 3rd person pronoun can be, and often is, null when it has a local enough antecedent in a higher clause, as in (36) (see Holmberg 2005; Chapter 2; Holmberg & Sheehan, Chapter 3; Holmberg, Nayudu & Sheehan 2009).29 (36)

Kalle oli iloinen kun (ha¨n) sai vapaapa¨iva¨n. Kalle was pleased when he had off-day ‘Kalle was pleased when he had a day off.’

Following Holmberg (2005) and Vainikka & Levy (1999) I assume that the null 1st and 2nd person pronouns in Finnish are, in fact, also null on account of having a local antecedent, the speaker and the addressee. I suggest implementing this idea along the lines of Sigurðsson (2004b), who argues that there is a syntactic representation of the speaker and the addressee as features in the C-domain. These are interpretable features, whose semantic value obviously depends on the speech situation, which can control, or bind, the subject pronoun. As Finnish allows null subjects in principle, it allows leaving a 1st or 2nd person pronoun unpronounced in this situation. There are important differences between licensing 1st/2nd person pronouns and 3rd person pronouns which I ignore in the present context.30 Why do English, French, German and Mainland Scandinavian not also have a null G-pronoun? The following is an explanation made possible by Roberts’ (forthcoming b) theory of incorporation: assume that a G-pronoun minimally has a number and a person feature. Then consider a language which lacks either an unvalued number or an unvalued person feature in T. In such a language a subject G-pronoun can never be a copy of T, but will be the head of a chain, and as such will be spelled out. This tallies with the observation that languages which have an overt G-pronoun are non-null-subject languages, which typically have a poorly specified agreement paradigm. 29 30

See Helasvuo & Laitinen (2006) for certain other cases of null 3rd person definite pronouns. Particularly under Sigurdsson’s (2004b) theory control of 1st and 2nd person null pronouns is a ‘phase-internal’ affair, while control of a 3rd person null pronoun is, at least in some cases, an ‘extra-phasal’ affair. See Holmberg and Sheehan (Chapter 3). It is not a discourse grammar phenomenon, though, as the antecedent has to be found in a higher clause.

The null generic subject pronoun in Finnish

223

But this does not explain why these languages also resist deletion of a subject pronoun in a finite clause when it is controlled by a higher DP. According to the theory articulated here, the (optional) null subject in (36) is a DP, the head of an A-chain which has undergone movement to specIP, and is deleted under identity with the antecedent DP in the higher clause. As discussed by Holmberg (Chapter 2) and Holmberg and Sheehan (Chapter 3), it is typical of languages which have null G-pronouns that they also allow null subjects derived in this way by deletion, while none of the languages with overt G-pronouns listed above allow this. Furthermore, although Finnish has a rich subject–verb agreement paradigm, it is not the case across the board that the languages which have a null G-pronoun have agreement paradigms that are richer than the languages which have an overt G-pronoun (see Holmberg, Chapter 2; Holmberg, Nayudu & Sheehan, 2009). Both French and German have reasonably well articulated agreement paradigms, distinguishing person as well as number (although not person and number for every verb form). An alternative explanation, following Holmberg (Chapter 2), is that nonnull-subject languages have an additional feature in T which triggers pronunciation of SpecTP (not just spell-out in the sense of assignment of a vocabulary item which can be deleted under appropriate conditions, but pronunciation). More precisely, other properties of the grammar conspire to force movement of the category probed by T to SpecTP, and then, in some languages, this category must be pronounced in this position. Even a bare, D-less pronoun will therefore invariably be pronounced in, for example, English either as the generic pronoun one, or as expletive it or there, depending on other properties of the construction (including whether the subject has a y-role or not).

10

Generic non-subjects

Finnish has a null generic direct object as well, with properties similar to the null arbitrary objects in Italian discussed in Rizzi (1986a). (37)

a. Ta¨ma¨ pa¨a¨to¨s ei ilahduta. this decision not happy.make ‘This decision doesn’t make one happy.’ b. La¨a¨ka¨rit kehottavat syo¨ma¨a¨n va¨hemma¨n rasvaa. doctors urge eat-INF less fat ‘Doctors urge people to eat less fat.’

The formal account is the same as for subjects, except that the head of the generic pronoun chain is v, not T. The direct object is probed by v (Chomsky 2000; 2001). If the object is a bare, D-less f-pronoun, v and the object will be

224

Anders Holmberg

copies after Agree, and form a chain headed by v, with the result that the object will not be pronounced. One context where the G-pronoun is not possible in Finnish is as possessor of NP. This is a case where the parallellism with generic one breaks down. Below, the ungrammatical possessive GSC is compared with the grammatical counterpart with a definite possessive pronoun. (38)

a. * lapsensa tuottaa aina huolia. children-PX cause always worries Intended reading: ‘One’s children are always a cause of worry.’ b. Ha¨nen lapsensa tuottaa aina huolia. he-GEN children cause always worries ‘His children are always a cause of worry.’

(39)

a. *Sita¨ ei voi ymma¨rta¨a¨ etta¨  lapsensa EXP not can understand that

ovat kasvaneet aikuisiksi. children-PX have grown adults-TRA

Intended: ‘One can’t understand that one’s children have become adults.’ ovat kasvaneet aikuisiksi. b. Ha¨n ei voi ymma¨rta¨a¨ etta¨ ha¨nen lapsensa he not can understand that he-GEN children-PX have grown

adults-TRA

‘He doesn’t understand that his children have become adults.’

The Finnish possessive NP with a pronominal possessor consists of a head noun with a possessive suffix attached plus a possessor pronoun with genitive case. The possessive suffix (-nsa in the 3rd person) is an anaphor subject to Principle A (see Trosterud 1993). First and 2nd person possessors are optionally null. A 3rd person (singular or plural) possessor is obligatorily overt. Another way to put it is that a 3rd person possessor is a pronoun subject to Principle B when overt, but an anaphor when null. (40) shows the singular paradigm. (40)

(minun) kirja-ni ‘my book’ (sinun) kirja-si ‘your book’ *(ha¨nen) kirjansa ‘his/her book’

The following is a reasonably plausible analysis of the possessive NP, in the framework of the present theory. (41)

[PossP Poss[uf, GEN, EPP] [NP [D, 3SG, uCase] N]] ! [PossP [D, 3SG, GEN] [Poss’ [Poss N Poss[3SG, GEN, EPP] [NP < [D, 3SG, GEN] > < N>]]]]

There is a functional head, here called Poss, with uf-features and a GEN case to assign. It receives f-feature values from a possessor DP or D-pronoun. Poss also has an EPP-feature triggering movement, i.e. second-merge of the possessor with PossP. The head noun moves and ends up adjoined to Poss, spelled out as shown in (40). A 1st and 2nd person pronoun can be deleted (left unpronounced); a 3rd person pronoun cannot.

The null generic subject pronoun in Finnish

225

The absence of a generic possessive pronoun cannot be explained in terms of case. As shown in Sections 4 and 5 a genitive-marked null pronoun can be generic. The absence of a generic possessive pronoun can be explained as an effect of morphological blocking: a D-less, 3rd person possessor will be analysed as forming a chain with Poss, and will therefore be obligatorily null, but this chain is interpreted as an anaphor, not as a generic pronoun. (42)

Marja kadotti [ hattunsa]. Marja lost hat-PX ‘Marja lost her hat.’

The anaphoric interpretation is not possible in (38a) and (39a) because it violates Principle A. The null 3rd person possessive pronoun would effectively be a null counterpart of the possessive reflexive pronoun sin in Scandinavian languages. (43)

Marjai tappade sini/*j hatt. Marja lost SIN hat

See, however, Trosterud (1993) for arguments against this type of analysis of possessive NPs, in favour of an analysis where the possessive suffix is itself the anaphoric argument (an analysis corresponding to the analysis of the subject pronoun presented, and rejected, in Section 5). As usual, an explanation of the absence of an interpretation on the basis of avoidance of ambiguity is not quite satisfactory, since ambiguity occurs in many other places in the grammar. We would presumably postulate a feature to distinguish between generic and anaphoric D-less f-complexes. The question is why Finnish grammar does not tolerate a generic-marked null fP, say, in complementary distribution with the anaphoric-marked null fP, in which case (38a) and (39a) would be grammatical precisely because in those contexts the anaphoric reading is ruled out.

11

On the Finnish passive

As mentioned, Finnish has a (so-called) passive construction which covers some of the meanings of impersonals which the GSC does not cover, namely, the arbitrary or quasi-existential meaning and the exclusive generic meaning. (44)

a. Sinne rakenne-tt-i-in silta. there build-PASS-PST-F bridge ‘They built a bridge there/A bridge was built there.’ b. Ita¨vallassa puhu-t-a-an saksaa. Austria-INE speak-PASS-PRS-F German ‘They speak German in Austria.’

226

Anders Holmberg

The passive verb form consist of a passive suffix, usually, -t, followed by the tense suffix, followed by a suffix -Vn, which takes the ‘morphological slot’ of the subject agreement suffix, but is invariant (except that the vowel copies the features of the preceding vowel). Holmberg, Nikannen, Oravita, Reime & Trosterud (1993) argue that it is a finiteness suffix (hence the label F) (see Helasvuo 2006; Helasvuo & Laitinen 2006). An interesting question is whether the passive also has a syntactically projected null subject. I will only give the barest sketch of an answer to this question here (see Shore 1988; Blevins 2003; Manninen & Nelson 2004; Helasvuo 2006 for discussion. Applying the same tests as in the case of the GSC, the answer is negative, albeit not unambiguously. First, the passive does support an agent-oriented adverbial. (45)

Ta¨a¨lla¨ lue-t-a-an ahkerasti kirjoja. here read-PASS-PRS-F eagerly books ‘Books are read eagerly here.’

The passive also supports, for example, a result clause with a PRO subject apparently controlled by the subject of the passive. (46)

Ta¨nne tul-t-i-in [PRO ostamaan keramiikkaa] here come-PASS-PST-F buy-INF pottery ‘People came here to buy pottery.’

But the passive does not support a reflexive or other anaphor, for example a possessive suffix (see Section 3). (47)

Shelliasemalla pes-t-a¨-a¨n auto (*-nsa). Shell-station-ADE wash-PASS-PRS-F car PX Intended: ‘One’s car can be washed at the Shell station.’

Nor does the verb agree with the subject; the passive form does not agree overtly, and the object has nominative case (which is an indication of nonagreement between subject and T). This can be understood if the construction, in the case of transitive verbs, has an agent as part of its argument structure, supporting an agent-oriented adverbial and an infinitival result clause with a PRO subject, but this agent is not syntactically projected, hence cannot trigger agreement or bind an anaphor. Furthermore, unlike the GSC, a predicative adjective or secondary predicate have plural form. (48)

a. Kesa¨lla¨ ol-t-i-in ilois-i-a summer-ADE be-PASS-PST-F happy-PL-PAR ‘In the summer people/we were happy.’ b. Sinne tul-t-i-in na¨lka¨is-i-na¨. there come-PASS-PST-F hungry-PL-ESS ‘People came there hungry.’

The null generic subject pronoun in Finnish

227

Recall that a predicative adjective or secondary predicate in the GSC has singular form, even though the entity referred to is a plural one (roughly ‘people including me’). This is evidence that the construction contains a structurally projected or represented subject with valued grammatical features, namely 3SG, determining the form of the predicate. If so, the fact that the predicates in the passive are plural can be regarded as an indication that the passive does not contain a syntactically projected subject, so that the form of the predicates is determined purely by the semantic properties of the agent. Finally, the fact that the Finnish passive does not support an agent phrase in the manner of ‘canonical passives’ can be taken as evidence that the construction does not have a structurally projected subject. (49)

Sinne rakennettiin silta (*englantilaisi-lta) (englantilais-ten toimesta) there build-PASS bridge (English-ABL) (English-GEN action-ABL) ‘A bridge was built there, by the English.’

The ‘natural’ form of an agent phrase would seem to be ablative, but this is completely impossible. The only way to express an agent overtly is by using an expression which literally translates as ‘by the action of x’. On the other hand, Manninen & Nelson (2004) point out that there is no principled reason why a proper agent phrase could not be expressed as in Finnish. More generally, it is unclear what conclusions to draw from (49) with regard to the syntactic status of the subject in passives. I leave the subject of passives with these sketchy remarks.

12

Other languages

Other languages that have a null generic pronoun include Marathi, Hebrew, Icelandic and Brazilian Portuguese. (50)

Hya khurchi- war aaramani bushushakto. this chair -on comfort-with sit-PRES.3SG ‘One can sit comfortably in this chair.’

(Marathi)

(51)

Yxolim la-sˇevet be-noxiout ba-kise ha-ze. can-3PL to-sit in-comfort in-the-chair the-this ‘One can sit comfortably in this chair.’

(Hebrew)

(52)

Nu´ ma´ fara að dansa. now may go to dance ‘One may begin to dance now.’

(53)

Aqui na˜o pode nadar. here not can swim ‘One can’t swim here.’

(Icelandic)

(Brazilian Portuguese)

228

Anders Holmberg

What Marathi, Hebrew, Icelandic, Brazilian Portuguese and Finnish have in common is that they are partial null-subject languages: in addition to the null G-pronoun subject, they all have non-referential null subjects (for instance with weather verbs), and, with the exception of Icelandic, they all allow null subjects in finite clauses that are controlled by a c-commanding argument in a higher clause (see Holmberg and Sheehan, Chapter 3). Some of them (Finnish and Hebrew) allow null 1st and 2nd person subject pronouns even without a linguistic antecedent. None of them allows a 3rd person definite null subject in an embedded finite clause without a local linguistic antecedent.31 Consistent null-subject languages, somewhat surprisingly, do not have a null generic pronoun, in the strict sense of a null ‘one’ with 3SG or 3PL agreement and no other morphology. Instead they use a variety of other strategies. One is using generic ‘you’, with 2SG agreement, as in (54a,b). Another is using some overt morphology in addition to 3SG or 3PL agreement; Romance languages and (some) Slavic languages make use of (cognates of) the overt reflexive morpheme se (Cinque 1988; Egerland 2003; Rivero & Sheppard 2003), as in (55a,b). Hausa, a consistent pro-drop language (see Jaggar 2001: 415), has a special impersonal pronoun a´ (with several allomorphs) with either arbitrary or generic reference, referred to as the ‘4PL pronoun’ in Jaggar (2001: 207f.); see (56). Standard Arabic makes use of an impersonal passive, as in (57). (54)

(55)

a. xsse-k texdem htta l-xamsa-w-settin need-you work-2SG until the-five-and-sixty ‘One has to work until the age of 65.’ b. Den mporeis na empistefteis kanenan not can-2SG to trust-2SG anyone ‘One can’t trust anyone.’ a. Si deve lavorare fino all’eta` di 65 anni. ‘One has to work until the age of 65.’ b. Aqui na˜o se pode nadar. ‘One can’t swim here.’

(56)

a` mai da` hanka`lı¯. ‘One should be careful.’

(57)

La y-usbah-u hunaa. not 3-swim.PASS-PL here ‘One can’t swim here.’

31

(Moroccan Arabic)

(Greek)

(Italian; Egerland 2003) (European Portuguese) (Hausa; Jaggar 2001: 208) (Standard Arabic)

Finnish, unlike the other partial null-subject languages, also allows control into a finite clause under connectivity in specificational predication, similar to control into non-finite clauses; see Holmberg, Nayudu & Sheehan (2009) and Holmberg & Sheehan (Chapter 3).

The null generic subject pronoun in Finnish

229

Yet another option is to make use of an overt indefinite quantificational expression meaning something like ‘anyone’, or a noun meaning ‘people’. (58)

Ewaru-aynaa ii kurcii loo sukham gaa kuurcoo waccu. who even this chair in comfortably sit may ‘One (anyone) can sit comfortably in this chair.’

(Telugu)

The reason why consistent null-subject languages do not have a Finnish-style incorporated generic subject is that T in these languages has an unvalued D-feature, [uD], which in the typical null subject construction is valued definite; see Holmberg (Chapter 2) for details of this valuation process. When a D-less fP subject is probed by T, and forms a chain with T, as described above, this chain will be definite (i.e. ‘he’ or ‘she’, if it is 3SG), by virtue of the definite-marked D-feature in T. One way to get a generic subject interpretation is then to have a proper indefinite QP subject, which values the uD-feature, and does not get incorporated (being too complex for incorporation; see Holmberg (Chapter 2)), and gets spelled out as the highest link in its chain. This is clearly the derivation of (58), with an overt QP subject. It is also a possible analysis of (55a) and (55b): the impersonal pronouns value T’s uD-feature, perhaps as impersonal, neither definite nor indefinite, but have some feature which prevents their incorporation in T. The effect of the passive morphology in (57) is to prevent the passing of unvalued f-features down from Voice to v (see Section 6). The subject will therefore be probed by Voice, not T (as in the case of the necessive construction in Finnish). In this way T’s uD-feature does not affect the interpretation of the subject (although exactly how the postulated uD-feature is valued or eliminated remains unclear).

13

Conclusions

Finnish has a null generic pronoun (G-pronoun) with formal and semantic properties closely resembling those of English one:  It has unambiguously inclusive reference;  It is specified 3SG, and as such triggers 3SG agreement on T (if it is probed by T), and singular agreement on secondary predicates;  It needs a Case value like arguments, and is assigned a Case value like other subjects: nominative, genitive, adessive, or partitive, depending on other properties of the sentence;  It is a feature complex made up of interpretable f-features (3SG) and an unvalued Case feature. We left open the possibility that it also has an uninterpretable generic feature and an inclusiveness feature (following Moltmann 2006);

230

Anders Holmberg

 The features of the G-pronoun are a subset of the features of T (when probed by T). This entails that the G-pronoun is formally a copy of T as a result of Agree, and consequently T and the subject form an A-chain headed by T.  As a consequence, the subject is not spelled out, since (normally) only the head of a chain is spelled out.  Another consequence is that the G-pronoun cannot check the EPP of T, not being the head of a chain;  In the non-nominative subject case discussed, the necessive construction, the subject G-pronoun is not probed by T but by (necessive) Voice, and receives genitive Case value from it.

6

‘Free’ inversion in Romance and the Null Subject Parameter Michelle Sheehan

1

Introduction

It has long been observed that the Romance null-subject languages permit ‘free inversion’ whereby a subject appears post-verbally, either on the right periphery of the sentence, or preceding the object but “obligatorily following participles and infinitives” Hulk and Pollock (2001: 3). Consider the following illustrative examples: (1)

E` arrivato Gianni. Is arrived Gianni ‘Gianni has arrived.’

(2)

Ayer gano´ Juan la loterı´a Yesterday won Juan the lottery ‘Yesterday Juan won the lottery.’

(Spanish)

(3)

Ha ganado la loterı´a Juan. Has won the lottery Juan ‘Juan has won the lottery.’

(Spanish)

(Italian)

These kinds of sentences superficially pose a problem for the Extended Projection Principle (EPP) (cf. Chomsky 1981: 26), which claims that all sentences have a ‘subject requirement’ for spec IP to be filled.1 In fact, the EPP is challenged doubly by NSLs: (i) where subjects are (sometimes) null, and also (ii) where subjects (sometimes) fail to raise to spec IP. In his classic analysis of the Null Subject Parameter (NSP), Rizzi (1982) attributes the post-verbal subjects in (1)–(3) to the presence of a null expletive in preverbal subject position, co-indexed with the post-verbal lexical subject.2 In more recent work, Rizzi (2006b) has developed this analysis

1 2

Of course, this is by no means accepted by all generativists. See Svenonius (2002a), Epstein and Seely (2006) and Biberauer (Chapter 4) for alternative views. The real power of Rizzi’s analysis was that this availability of post-verbal subjects was also responsible for another of the NSP correlates: the apparent violation of the that-trace filter, which made sentences such as (i) legitimate in Italian but not in English:

231

232

Michelle Sheehan

further, adopting the proposal outlined fully in Belletti (2004; 2005) that the low subject also moves to a low (narrow) Focus position in the vP periphery, thus recognising the oft-cited connection between post-verbal subjects and focus (cf. Lambrecht 2000). With the advent of the Minimalist Program, and the rejection of interfaceinvisible elements such as null pro, an alternative explanation of this phenomenon has been proposed by Barbosa (1995; 2009) and developed by Alexiadou and Anagnostopoulou (A&A) (1995; 1998). In their terms no null expletive is present in structures such as (1)–(3), but rather the agreement morphology in NSLs has the status of a pronoun and so verb-movement alone to the inflectional domain (IP or TP) serves to satisfy the EPP. As such, in line with economy, the post-verbal subject remains low in the absence of any attracting feature, because according to the notion of economy put forth in Chomsky (1995b: 234), “Nothing can join to a non-projecting category. [. . .] if a strong feature does not project.” One of the most appealing aspects of this analysis is that it seems to predict the occurrence of post-verbal subjects ‘for free’. In fact, if verbal morphology in Romance NSLs (and Greek) is always pronominal, satisfying the EPP, then there can be no preverbal subject A-position, and the true A-position of subjects must always be post-verbal.3 This analysis therefore makes the following two robust predictions about NSLs: (i) all preverbal subjects will have an A-bar status; (ii) the ‘neutral’ A-position of subjects will be postverbal.4 In this chapter I will consider to what extent this account of the NSP can explain a number of inversion structures in Romance.

(i)

3

4

Chi hai detto che ha chiamato? Who have-2s said that has called? Lit. ‘Who did you say that called?’

It seems clear that extraction from the post-verbal position is obligatory in Romance NSLs (and perhaps French; Taraldsen 2002). As such, this seems to support a parameterised approach to the EPP. However, work by Nicolis (2005) and Menuzzi (2004) suggests that even languages without free inversion extract wh-subjects from the post-verbal position to the CP-layer to avoid that-trace effects. A&A also discuss, but stop short of adopting, a possible extension of this idea, whereby pronominal Agr also absorbs the subject f-role, precluding the need for a null referential pro. In fact, this stronger version of their thesis makes radically different predictions. If agreement morphology were to absorb the f-role then there could be no A-position for overt subjects in NSLs. Rather, all subjects would be dislocated, in a manner akin to Baker’s (2000) account of non-configurational languages. Manzini and Savoia (2002) propose a version of this stronger thesis, rejecting structural f-assignment and advocating f-assignment via Agree. Barbosa (1995) and A&A (1998) present a range of facts to support the claim that preverbal subjects are CLLD in NSLs. These arguments are not without problems as discussed by Sheehan (2006). However, for reasons of space, this chapter is concerned with the status of post-verbal subjects only.

‘Free’ inversion in Romance and the Null Subject Parameter

233

First, in Section 2, I make clear my theoretical perspective regarding spellout, PF and the relationship between information structure, word order and stress. Taking inversion in wide and narrow focus contexts to be distinct phenomena, I deal with them separately. In Section 3, I give details of the inversion allowed in ‘wide focus environments’, and other contexts where focus can project (in Reinhart’s (1995; 2006) terms). I argue that, with the exception of certain structures in Spanish, inversion in these contexts is limited to locative inversion and expletive-associate inversion with the relevant semantic/Definiteness Effects. Even in Spanish, all inverted structures which allow focus to project require the merging of an XP in preverbal position. In Section 4, I discuss narrow subject focus, where subject focus fails to project, and how this relates to the EPP. In Section 5, I discuss the implications for parametric theory of these facts and in particular how these data relate to the NSP. 2

Background clarifications

2.1

Syntax and information structure

I adopt the basic tenets of Chomsky’s Minimalist Program (1995b; 2000; and some aspects of 2005) coupled with Reinhart’s (1995; 2006) view of ‘interface economy’. As such, uninterpretable features drive the syntactic derivation but PF considerations may also have an effect upon the form of a given utterance. I further assume, following Cinque (1993) and Reinhart (2006) that “the focus constituent is coded at PF” rather than in the Narrow Syntax: Since main stress is a requirement of the computational system (a derivation cannot be pronounced without main stress) an optimal language system would make use of this visible property of derivations to code information needed at the interface. (Reinhart 2006: 10)

As such, a derivation is only assigned an information structure at PF. This information structure is then tested against the real-life context for felicity. I assume, as is standard in the literature, that, for poorly understood reasons, the highest copy of a moved element will be spelled out in the absence of conflicting PF requirements.5 However, where PF requires otherwise, a lower copy can be targeted (cf. Bosˇkovic´ 2001; Bobaljik 2002; Bosˇkovic´ and Nunes 2007). Like Stepanovic´ (2003), I propose that prosody is a relevant PF consideration in this regard. 5

This is not an uncontroversial assumption. Note that Landau (2006b) denies that such a principle exists, as discussed by Holmberg (Chapter 2).

234

Michelle Sheehan

2.2

Contrastive/Identificational focus and information focus

For the purposes of this chapter, I naively accept the basic distinction described by Kiss (1998) between identificational (otherwise called contrastive) focus and informational focus, and her observation that while identificational focus is often associated with movement, information focus is not.6 Interestingly, in Spanish and Italian, there appears to be a preverbal position specialised for identificational focus, whereas the same does not seem to hold for EP (c.f. Barbosa 1995, inter alia). A possible reason for this will be proposed in Section 3.4.1. In the main, I will not discuss identificational focus here, but will limit myself to information focus. This is the focus used to introduce new un-presupposed information into the discourse, marked with a pitch accent. Since Chomsky (1971) and Jackendoff (1972), the test for information focus has been the wh-question: (4)

a. b. c. d.

Who invited John to the party? Mary invited John to the party. There is an x such that x invited John to the party x ¼ Mary

In (4), the answer to question (a) involves a focused subject (Mary), which receives intonational prominence. This answer (b), like the question (a), has the presupposition (c). Thus the ‘information’ contained in the utterance is twofold, comprising (c) and (d), but the information focus is simply Mary. 2.3

Nuclear stress and focus projection

Following work by Bresnan (1971) and Cinque (1993), I assume that default/ unmarked stress is assigned at PF via the Nuclear Stress Rule (NSR). In these cases, stress is unmarked because focus can ‘project’. Consider the following from Reinhart (1995) as an illustration, where small caps indicate nuclear stress: (5)

My neighbour is building a desk.

Although the nuclear stress falls on the word desk, according to the NSR any constituent containing the word desk can be the focus of the sentence, including the whole sentence (the wide focus), where underlining indicates focus: (6)

What’s that noise? [My neighbour is building a desk.]

(7)

What’s your neighbour doing? My neighbour [is building a desk].

6

Though see Samek-Lodovici (2006) for the claim that contrastive focus is also derived by something akin to the nuclear stress rule and remnant movement.

‘Free’ inversion in Romance and the Null Subject Parameter (8)

235

What’s your neighbour building? My neighbour is building [a desk].

With this stress the sentence is not, however, felicitous in answer to the following subject wh-question: (9)

Who is building a desk? #[My neighbour] is building a desk.

This is because the word desk is not contained in the subject constituent. In order for the subject to receive narrow focus, a marked stress rule is required in English, which assigns stress inside the subject: (10)

Who is building a desk? [My neighbour] is building a desk.

In this case the focus cannot ‘project’ to wide (sentence) focus, even though the stressed element is contained within the sentence: (11)

What’s that noise? #[My neighbour is building a desk].

For this reason, this stress pattern is considered ‘marked’. In short, for our purposes, a marked stress pattern is one in which focus cannot project. This will be used as a diagnostic in the determination of the unmarked subject position in Romance.

2.4

The importance of phases

Chomsky (2000; 2001) proposes that in order to reduce the ‘computational burden’ derivations might proceed by phase. If this is the correct approach, then the null hypothesis is that PF operations should also apply cyclically, at the phase level. Following Adger (2007), I propose that stress assignment also applies cyclically by phase, in conjunction with copy deletion, via the NSR, or a marked stress rule (MSR) if available: (12)

spell-out ! PF copy deletion ! NSR (or MSR if available)

Adger (2007) proposes an implementation of the NSR whereby the most stressed, non-deleted, syntactic object (SO) is strengthened at the phase level. Consider the following illustration from English. Lexical (non-functional) words enter the derivation with word stress (x). At the phase level the NSR strengthens the highest stress mark to the next level. Taking DP, vP and CP to be phases, the simple transitive sentence ‘The baby ate the gloop’ proceeds as shown in Figure 6.1, where elements in parenthesis are not spelled out as they are contained in the edge of the phase.

236

Michelle Sheehan X X Phase 1 = [DP (the) gloop]

X X Phase 2 = [DP

(the) baby]

X X

X

Phase 3 = [vP (the

X

X

X

baby)

(ate) the gloop

X X X

Phase 4 = [CP the

X

X

X

baby

ate

X the

gloop]

Figure 6.1 Stress assignment

Adger shows that an approach along these lines can derive the English facts as well as the main patterns in Irish, a VSO language. Note that under Adger’s approach objects can only escape being assigned nuclear stress by the NSR where they scramble/move out of VP. Note, however, that where the object has scrambled, we need some extra condition on focus projection, to choose between the subject and object at the completion of the CP phase. Both are phases and so would be equally likely to ‘project’. The evidence from Spanish and Italian suggests that the correct notion might be linear order (with a sensitivity to phrase boundaries), rather than c-command. This is welcome as there is no reason to believe that PF processes should have access to syntactic notions such as c-command after spell-out has occurred.

‘Free’ inversion in Romance and the Null Subject Parameter

2.5

237

The split CP

Since Rizzi (1997), it has been widely accepted that the C-system consists of a number of (optionally present/active) functional heads in a fixed universal order: (13)

Force Top

Foc Top

Fin

Note, crucially, that the Foc head is connected with fronted elements which receive contrastive focus and not with narrow/information focus. While in Italian all of these features, where present, seem to be associated with distinct syntactic positions, in other languages, such as Spanish, there seem to be fewer positions and this can be attributed to the proposal by Giorgi and Pianesi (1997) that a ‘feature scattering principle’ operates as part of UG: (14)

Feature Scattering Principle ‘Each feature can head a projection.’

Thus, while the order of features is universal, as careful work by Cinque (1999) has shown, I accept the proposal that the bundling of features into heads within the functional layers is parameterised (cf. the Introduction). There are strong empirical reasons even within Romance to take such a stance, as will be shown below. 3

Focus projection orders in Romance

3.1

SVO word order

In Spanish, Italian and EP, only SVO order is felicitous with transitive verbs in wide focus contexts, as the following examples from Italian illustrate: (15)

a. Cos’ e` successo? What is happened ‘What happened?’ b. Gianni ha dato un libro a Maria Gianni has given a book to Maria ‘Gianni gave a book to Maria.’ c. #Ha dato un libro a Maria Gianni has given a book to Maria Gianni d. *Ha dato Gianni un libro a Maria has given Gianni a book to Maria

In such cases, post-verbal subjects are not felicitous in Italian and EP. Spanish behaves slightly differently, as discussed in Section 3.4. Note that the preverbal subject must be included in the focus of the sentence in these cases, and

238

Michelle Sheehan

so cannot be considered a CLLD topic (as has been pointed out by Costa (2001a; 2001b; 2004)).7 This word order is straightforwardly derived if there is an EPP of the familiar English-type in Romance, triggering subject movement to spec IP. If, on the other hand, the EPP is satisfied by verb-movement, as proposed by Barbosa (1995) and Alexiadou and Anagnostopoulou (1998), then an alternative explanation of these facts must be found.8 3.2

SV and locative inversion/presentational sentences

In the case of intransitive verbs, matters are more complex and a distinction must be made between verbs which select a loco-temporal argument, of the type described by Pinto (1994; 1997), and those which do not. The former permit projection to wide focus with VS orders and the latter do not. Thus with ‘inversion verbs’, both SV and VS orders are felicitous in wide focus contexts in Spanish, Italian and EP. Consider the following from Italian: 7

8

As a further illustration that the contrast is not merely between preverbal ‘topics’ and post-verbal ‘foci’, consider the behaviour of negative words, which cannot be topics, as outlined by Costa (2004) for EP: (i)

a. Quem e´ que chegou? Who is that arrived? b. Na˜o chegou ningue´m Not arrived nobody ‘Nobody did.’ b0 . #Ningue´m chegou. Nobody arrived

(ii)

a. O que e´ que ningu´em fez? The what is that nobody did ‘What did nobody do?’ b. Ningue´m chegou. Nobody arrived b0 . #Na˜o chegou ningue´m not arrived nobody ‘Nobody arrived.’

These examples (which are paralleled in Italian and Spanish), illustrate that the preverbal subject in wide focus contexts is not a ‘topic’. This in turn serves to suggest that the preverbal position is unmarked, whereas the marked post-verbal/low position encodes focus. This is the opposite to the prediction made by A&A’s and Barbosa’s analyses, which predict that the default position of subjects is post-verbal. Alexiadou and Anagnostopoulou (2001; 2007) provide a possible account of this, based on the idea that only one argument can remain in vP for case reasons. However, as will be illustrated below, ‘free’ inversion is not readily available with all intransitive verbs in null subjects Romance, so this cannot be the whole story, though it is certainly an interesting proposal. They also discuss the problematic cases in Spanish where both the object and subject remain inside vP, attributing them to the availability of clitic doubling in that language. This account cannot explain, however, the fact that Spanish VSO order seems to require that another XP be merged preverbally. In this chapter it will be argued that the the subject leaves vP because of the EPP and the object leaves it to avoid nuclear stress.

‘Free’ inversion in Romance and the Null Subject Parameter (16)

239

a. Cos’e` successo? What is happened ‘What happened?’ b. E` arrivato Gianni. Is arrived Gianni ‘Gianni arrived.’ b0 . Gianni e` arrivato. Gianni is arrived ‘Gianni arrived.’

Interestingly, the group of verbs labelled ‘inversion verbs’ by Pinto (1997) cuts across the ‘unaccusative/unergative’ distinction: (17)

a. Cos’e` successo? What-is happened ‘What happened?’ b. Ha telefonato Gianni has telephoned Gianni ‘Gianni phoned.’ b0 . #Ha riso Gianni Has laughed Gianni ‘Gianni has laughed.’

(Italian)

Example (17) shows that inversion with wide focus is legitimate with the verb telefonare (‘to telephone’), but not with the verb ridere (‘to laugh’). While arrivare passes all the unaccusativity tests in Italian, telefonare and ridere do not. Pinto (1994) has shown that, in Italian, this inverted wide scope order is associated with a specific semantics. According to her, the location of the action/event is associated with that of the speaker in VS order, but not in SV. For example, the grammatical assertions in (16) and (17) are all possible answers to the question ‘what happened?’, however only the SV answers are location neutral, having the meaning ‘Gianni arrived somewhere’ and ‘Gianni phoned somewhere/someone’. The VS answers, on the other hand, have the implicit meaning ‘Gianni arrived here’ and ‘Gianni phoned here’. In these terms, (17b0 ) differs from the other examples in that ridere does not select a locative argument. Similar effects hold for EP: VS order is also possible with wide focus in presentational contexts, only where a locative argument is implicit. In such cases, inversion is optional, but corresponds to a subtle change in meaning similar to that described for Italian, as described by Costa (2004): (18)

a. O que e´ que aconteceu? the what is that happened? ‘What happened?’

(EP)

240

Michelle Sheehan b. Chegou o Joa˜o. Arrived the Joa˜o ‘Joa˜o arrived (here).’ b0 . Telefonou a tua ma˜e. Telephoned the your mother ‘Your mother called (here).’ b00 . #Tossiu a tua ma˜e. Coughed the your mother ‘Your mother coughed.’

Given the fact that both of these languages permit locative inversion (LI) (c.f. Ambar 1992; Pinto 1994; 1997), it seems logical to argue that these are instances of LI with a null locative, as proposed by Pinto (1997) for Italian. It is proposed, following Pinto, that in the VS structures in (16)–(18) a null loco-temporal argument raises to a preverbal position, satisfying the EPP in a way to be specified in Section 3.4.1.9 This allows the subject to remain low and receive nuclear stress via the NSR. Because stress is assigned in an unmarked way by the NSR, it can project to include the predicate. I further argue that, because of interface economy, derivations involving an in-situ null locative will be blocked as they will be indistinct at PF from those derivations lacking an argumental PP. For this reason only the VS order will have the ‘deictic’ interpretation described above. Where the PP is overt, on the other hand, either option will be possible, depending on whether the PP gets a topic interpretation (standard locative inversion) or not. However, crucially, if the PP fails to raise, then the subject must, in order to satisfy the EPP. Consider the following data from Ambar (1992): (19)

LOC chegaram os te´cnicos ontem LOC arrived the technicians yesterday ‘The technicians arrived (here) yesterday.’

(20)

A Lisboa chegaram os te´cnicos ontem. to Lisbon arrived the technicians yesterday Lit. ‘At Lisbon arrived the technicians yesterday.’

9

Fronted PPs behave like subjects in many ways, permitting raising, triggering that-trace effects Bresnan (1994: 96) and not permitting I-to-C movement (Hoekstra and Mulder 1990: 32): (i)

On that hill appears to be located a cathedral

(ii)

It’s in these villages that we all believe *that can be found the best examples of this cuisine.

(iii)

Out of which barn ran a horse?

(iv)

*Out of which barn did run a horse?

I therefore assume that it is the PP which satisfies the EPP in cases of LI.

‘Free’ inversion in Romance and the Null Subject Parameter (21)

Os te´cnicos chegaram a Lisboa ontem. the technicians arrived to Lisbon yesterday

(22)

*?Chegaram os te´cnicos a Lisboa ontem. Arrived the technicians to Lisbon yesterday ‘There arrived the technicians in Lisbon yesterday.’10

241

On these assumptions, (22) is bad because nothing raises to satisfy the EPP. Under an alternative analysis whereby verb-movement satisfies the EPP such intricacies remain unexplained. Note, crucially, that in all three languages inversion is not permitted in wide focus contexts with a definite post-verbal subject and an overt post-verbal (argumental) locative PP, as in (22). I return to the precise nature of the EPP, and how and why it can sometimes be satisfied by a PP, in Section 3.4.1. 3.2.1 Locative inversion in Spanish Zubizarreta (1998: 119) claims that Spanish differs from the other Romance languages in that it allows wide focus VS order with any intransitive verb. Consider, for example the following examples, some of which are infelicitous in the same context in Italian and EP: (23)

a. Que´ paso´? What happened ‘What happened?’ b. Se rio´ Juan. Himself laughed Juan ‘Juan laughed.’ b0 . Han estornudado tres leones. Have sneezed three lions ‘Three lions have sneezed.’

(Spanish)

However, it is not the case that Spanish allows VS order without restrictions, even with transitive verbs. Sun˜er (1982) convincingly shows that where the semantics of an utterance are not those of a presentational context, involving an implicit locative, then VS order is infelicitous in wide focus contexts. In fact, where the verb is modified, inversion becomes marginal: (24)

?*Llego´ tarde el tren Arrived late the train ‘The train arrived late.’

(25)

?*Crece el pasto sin cesar Grows the grass without to-cease ‘The grass grows non-stop.’

10

This is marked as ungrammatical rather than infelicitous as EP requires an XP to appear preverbally in order for inversion to take place, even where the subject receives narrow focus.

242

Michelle Sheehan

It would appear, then, that Spanish is less restricted in its ability to form presentational sentences than Italian and EP. This can be attributed to the fact that the EPP can be satisfied by non-argumental XPs in Spanish, whereas in EP and Italian it cannot, as discussed in Section 3.4.1. As such, a null locative adverbial can potentially be included with any intransitive verb, to force a presentational interpretation, even if the argument structure of the verb does not select a locative argument. Presentational constructions of this kind seem to be limited to intransitive verbs for some poorly understood semantic reason (cf. Lambrecht 2000). As such V-initial structures in Spanish with transitive verbs almost always seem to involve narrow subject focus (but see also footnote 12). 3.3

Definiteness effects

Pinto’s analysis predicts that where an overt post-verbal locative argument is present with an inversion verb such as morire ‘to die’, arrivare ‘to arrive’ or continuare ‘to continue’, the subject should surface in a preverbal position. This prediction appears to be largely true: a survey of newspaper titles in Spanish, Italian and EP found very few exceptions to this prediction.11 The exceptions were all of the following type, with an indefinite subject: (26)

Llegan 400 inmigrantes a las costas espan˜olas Arrive-pl 400 immigrants to the coasts Spanish-pl ‘400 immigrants arrive on Spanish coasts.’

These structures seem to be equivalent to there þ associate structures in English. Further data, based on native speaker intuitions, confirms that in Spanish, Italian and EP, subject inversion is licensed in wide focus contexts subject to Definiteness Effects with VSPP orders and unaccusative verbs: (27)

a. O que e´ que aconteceu? The what is that happened ‘What happened?’ b. Chegou algue´m ao cole´gio Arrived someone to-the school ‘Someone arrived at school.’ b0 . *Chegou o Joa˜o ao cole´gio Arrived the Joa˜o to-the school ‘Joa˜o arrived at school.’

(EP)

These are reminiscent of the effects reported for Italian by Belletti (1988). She showed that in unaccusative and passive structures where an overt locative PP follows the subject, Definiteness Effects are observed in Italian:

11

Note also that inversion was far more common in Spanish and Italian than in EP in the sample of titles collected, however this may be due to journalistic style.

‘Free’ inversion in Romance and the Null Subject Parameter (28)

All’improvviso e` entrato un uomo/*l’uomo dalla finestra. Unexpectedly is entered a man/the man of-the window Lit. ‘There entered a man unexpectedly through the window.’

(29)

E` stato messo un libro/ *il libro sul tavolo. Is been put a book the book on-the table Lit. ‘There has been put a book on the table.’

243

(from Belletti 1988: 9–10)

Crucially, examples (28) and (29) are possible in out-of-the-blue contexts. All of this suggests that Definiteness Effects can be observed in NSLs only where: (i) the verb is unaccusative/passive; (ii) the (argumental) locative PP is overt and fails to undergo movement; and (iii) the subject also remains postverbal without receiving narrow focus. Matters are slightly more complex in Spanish and EP. In the former a greater variety of XPs can satisfy the EPP and for this reason, Definiteness Effects can only be observed in verb-initial utterances where unaccusative (or passive) verbs surface with a post-verbal subject and PP. In EP, likewise, the effects are only seen in verb initial orders for more opaque reasons, to be discussed in Section 4.4. These data might be taken as evidence against an active EPP of the merge XP type in Romance NSLs. However, the striking similarities with English and French remain unexplained if such a line is taken. Rather, I will argue that these kinds of structures are identical to the English there þ associate structure, with the exception that NSLs have a null version of there. It is not necessarily the case that there and its null equivalent are inserted only to satisfy the EPP. Rather, any account of these expletives must capture their limited distribution: the fact that they only occur with indefinite associates and unnaccusative verbs. I leave this matter to one side here, as the primary focus of this study is to assess the relation of the EPP to inversion structures. If there were no EPP and hence no expletives in Romance then Definiteness Effects of this type would be unexpected.

3.4

Spanish XPVSO order

Note that there is a slight complication relating to transitive verbs in Spanish, which has led many to argue that Spanish is the type of language described by A&A (1998). It has been widely recognised, since Torrego (1984), that Spanish licenses inversion where certain XPs are fronted. Thus in Spanish, though not in Italian or EP, subject-verb inversion is optional where the IP is focused if and only if another XP topic surfaces before the finite verb. In these cases, the subject can only very marginally receive main stress and narrow focus. Usually, the object is stressed and this focus can project up to the IP level. Consider the following examples:

244

Michelle Sheehan

(30)

a. Que´ paso´ ayer? What happened yesterday? b. Ayer, Juan gano´ la loterı´a Yesterday Juan won the lottery ‘Yesterday Juan won the lottery.’ c. Ayer gano´ Juan la loterı´a Yesterday won Juan the lottery ‘Yesterday Juan won the lottery.’ d. #Gano´ Juan la loterı´a won Juan the lottery

(Spanish)

(from Zubizarreta 1998: 125) (Ordo´n˜ez 2000: 60, fn 3)

Note, crucially, that the presence of the preverbal XP is necessary in order for the subject to surface post-verbally in these contexts, as noted by Zubizarreta (1998) and Ordo´n˜ez (2000).12 Zubizarreta (1998) and Ordo´n˜ez and Trevin˜o (1999), following work by Torrego (1984), show that, in Spanish, these fronted XPs (foci, topics or emphatic phrases) can, and sometimes must, trigger subject–verb inversion. Where the fronted element is a topic, they claim that the effect is optional. However, it seems that where the fronted element is a locative, inversion is obligatory, irrespective of verb type:13 (31)

a. Que´ paso´ en el parque´? What happened in the park? b. En el parque me regalo´ Juan el anillo. In the park me gave Juan the ring c. # En el parque Juan me regalo´ el anillo In the park Juan me gave the ring

Likewise, where a robustly non-referential ‘emphatic phrase’ is fronted, the subject cannot intervene between that XP and the finite verb. Rather, obligatory subject–verb inversion is triggered:14

12

Note, however, that in a small number of contexts verb-initial VSO utterances are permitted in Spanish, as Zubizarreta (1998: 108) notes. These either involve special contrastive focus on the subject or are ‘eventive’: (i)

13 14

Acaba de ganar Espan˜a el Mundial de football. Recent past.3s of win Spain the world.cup of football ‘Spain just won the World Cup.’

It might be the case that the EPP is satisfied by a null adverbial in these sentences as Zubizarreta proposes. I leave these kinds of sentences to future research. It is expected that there will be subtle differences in interpretation associated with these two kinds of topicalisation. I leave this matter to future research. Catalan is like Spanish in this respect (Vallduvı´ 1990). Both Italian and EP allow the fronting of these so-called ‘emphatic phrases’ (c.f. Zubizarreta 1998), however, unlike in Spanish (and Catalan) subject inversion is not triggered in such contexts (Pilar Barbosa, p.c.):

‘Free’ inversion in Romance and the Null Subject Parameter

245

(32)

Nada les debe Juan a sus amigos. Nothing cl owes Juan to his friends ‘Juan owes nothing to his friends.’

(33)

*Nada Juan les debe a sus amigos. Nothing Juan cl owes to his friends (examples from Ordo´n˜ez and Trevin˜o 1999)

Note that this XP must (eventually) move to a CP position visible to a higher phase as cyclic movement is possible.15 As also observed by Torrego (1984), in such cases, inversion is obligatory in all intermediate clauses: (34)

Un viaje a las Canarias dice Juan que querı´a la gente que hiciera A trip to the Canaries said Juan that wanted the people that did Antonio este an˜o. Antonio this year ‘A trip to the Canaries, Juan said that people wanted Antonio to take this year.’

Note, on the other hand, that where a preverbal topic and subject cooccur, the subject always surfaces adjacent to the verb (as noted by Zubizarreta (1998)): (35)

Ayer, Maria presento´ su renuncia. Yesterday Maria presented her resignation ‘Yesterday Maria resigned.’

(Spanish)

(36)

??Maria, ayer, presento´ su renuncia. Maria yesterday presented her resignation ‘Yesterday Maria resigned.’

(Spanish)

Moreover, if no other XP fronts, recall that subject-raising is obligatory. The special subject status is reminiscent of a well-known fact about V2 languages.

15

(i)

Alguma coisa os miudos lhe disseram, mas na˜o sei o queˆ. Some thing the kids them gave but not know.1SG the what ‘Something the kids gave them, but I don’t know what.’

(ii)

*Alguma coisa lhe disseram os miudos, mas na˜o sei o queˆ. Some thing them said the kids but not know1sg the what

(iii)

Nenhuma resposta os miudos me deram ate´ hoje. No reply the children me gave until today ‘The children gave me no reply until today.’

(iv)

*Nenhuma resposta me deram os miudos ate´ hoje. No reply me gave the kids until today

This is not the same conclusion reached by Zubizarreta (1998), who argues that these XPs all target spec TP.

246

Michelle Sheehan

In these languages SVO is the unmarked order found in out-of-the-blue contexts (Zwart 1997), however, XPVSO is also legitimate where a certain XP is given special discourse status.16 Where a topic or focus is fronted in Spanish, it receives special discourse status, but where the subject raises (as a last resort) to satisfy the EPP, it can either be a topic, contrastive focus or part of the wide focus.17 I therefore propose that XPVSO is the word order triggered in Spanish where an XP other than the subject serves to satisfy the EPP. Note that, in Italian, VSO order is not only infelicitous, it is actually ungrammatical, even where another XP surfaces before the verb, as Belletti and Shlonsky (1995: 510) show: (37)

a. Cos’ e` successo ieri? What is happened yesterday ‘What happened yesterday?’ b. *Ieri ha dato Gianni un libro a Maria Yesterday has given Gianni a book to Maria ‘Yesterday, Gianni gave a book to Maria.’

(Italian)

It would appear, then, that in Spanish, though not in Italian, any topicalised or focused fronted XP can potentially satisfy the EPP. The behaviour of EP involves further complications, to be discussed in Section 4.4. 3.4.1 A parametric characterisation For conceptual reasons, Chomsky (2008) proposes that T inherits its Agree features from C, the phase-head that selects it. Adopting Rizzi’s (1997) split CP, the lowest C head is ‘Fin’, so let us assume that it is Fin that bears all T’s (uninterpretable) features until merger, including the EPP. The fact that, in Spanish, (non-DP) elements other than the subject can satisfy the EPP is informative: it suggests that the EPP cannot always be the D-feature of Chomsky (1995b). Rather, the EPP is the requirement to fill the specifier position of a head which independently 16 17

For this very reason, Zwart proposes that the subject remains in spec IP in subject-initial matrix clauses in V2 languages. This raises the question of whether the XPVSO order might be a V2 effect, with the verb occupying a position in C in Spanish. The behaviour of auxiliaries suggests that this is not the case: (i)

*A todo el mundo ha la prensa aceptado. To everybody have-3ps the press accepted ‘The press have accepted everybody.’

Moreover, one of the properties of Germanic inversion is that it is typically limited to matrix clauses and the complements of epistemic verbs such as ‘say’ and ‘believe’. The same is not true of Spanish, as Zubizarreta (1998: 100–103) notes: (ii)

Lamento que a NADIE le haya devuelto Marı´a su manuscrito. Regret:1s that to nobody cl have returned Maria his manuscript ‘I regret that Maria has not returned anyone’s manuscript to them.’

‘Free’ inversion in Romance and the Null Subject Parameter

247

establishes a relation with a phrase which it c-commands. As Chomsky (2001: 42) notes “the EPP-feature alone is not sufficient to identify a target”. Rather “the f-set (or comparable features, for other probes) is required to determine what kind of category K is sought”. In other words the EPP is really a feature of a feature. In order to identify a goal to attract, it must associate itself with another feature. Suppose, moreover, that there is a general economy consideration meaning that an EPP-feature must attach to a discourse-relevant feature over an uninterpretable feature.18 Now suppose that in Spanish, ‘Fin’ also (optionally) bears Focus/Emphatic/Topic features as these features are bundled together in this language, following Giorgi & Pianesi’s (1997) feature scattering approach discussed above. An EPP-feature is represented throughout as *: (38)

CP in Spanish Force Top Fin[(Foc)/(Top)/(Emp), uf, *. . .]

The effect of this will be that where a discourse feature is present on Fin, the EPP-feature will automatically associate itself with that feature, rather than with uninterpretable phi-features. This in turn means that in such cases, the EPP serves to trigger movement of the Top/Foc/Emp XP to spec FinP, rather than subject movement to spec IP.19 In Italian and EP, on the other hand, Fin does not bear any discourse features. Rather these features are scattered and occupy higher functional heads in the CP layer, most of which have their own EPP-feature, where the head/feature is active/present: 18

19

An alternative way to capture these facts is to argue that T is actually a phase-head in Spanish, and so SpecTP has edge properties. See Gallego (2008) for an implementation of this idea. Despite the attractiveness of such an approach, it faces many possible objections, not least the fact that while the verb moves to a position in I in both Italian and EP (cf. Sheehan 2006: chapter 3), both languages fail to display the optional ‘edge’ characteristics of spec TP in Spanish. Goodall (2001) argues that elements that can satisfy the EPP in Spanish are actually nominal in some way, giving evidence that coordinated locatives can trigger plural agreement in Spanish: (i)

Aquı´ y allı´ me gustan. Here and there me please ‘I like it here and there.’

He believes the alternative, that the EPP in Spanish is not limited to nominals, is less interesting. However, this kind of approach fails to explain why agreement always targets subjects in Spanish, even where they remain low. It is only in these cases, where the locative is actually the subject argument that it serves to trigger agreement: (ii)

*Aqui me dio mis hermanos el libro Here me gave my brothers the book

In actual fact it seems more problematic to assume that individual lexical items differ in this way in Spanish, than to assume a parametric difference like that outlined above.

248

Michelle Sheehan

(39)

CP-layer in Italian Force (Top*) (Foc/Emp [*])

(Top*)

Fin [uf*. . .]

CP-layer in EP Force (Top*) (Foc) (Emp*)

(Top*)

Fin [uf*. . .]

(40)

Note that empirical facts force us to posit additional differences between the split CPs of Italian and EP, as EP is generally accepted to lack contrastive focus movement to the CP-layer (Kato & Raposo 1994; 2006). One way to capture this is to posit Foc as a separate functional head which lacks an EPP-feature in EP. Because the features of C are more scattered in Italian and EP, the fronting of contrastive foci and emphatic phrases in Italian, or emphatic phrases in EP, does not trigger subject inversion, as the EPP on Fin operates independently of these higher functional heads (as described in unpublished work by Pilar Barbosa): (41)

Alguma coisa os miudos lhe disseram, mas na˜o sei o queˆ. Some thing the kids to-him said but not know-1s what ‘The kids said something to him but I don’t know what.’

(42)

*Alguma coisa lhe disseram os miudos, mas na˜o sei o queˆ. Some thing to-him said the kids but not know-1s what

(EP)

As such, in EP and Italian, even where an XP moves to the C-layer for discourse reasons the EPP on Fin still requires the subject to raise to spec TP. The only exceptions to this would appear to be locative inversion, where the feature which attracts the PP must be on Fin.20 This feature appears to be some kind of [deictic] topic feature, distinct from a standard topic feature. If this approach is correct then it is predicted that other SVO languages will display XPVS orders, where the EPP gets associated with a feature other than phi-features, despite having an active EPP in more neutral contexts. This approach also captures the fact that subject movement is the default (last resort) case, which is free from discourse effects in many languages. Languages such as English, Italian and EP thus represent one end of the spectrum, where the EPP has become practically divorced from discourse effects as a result of the scattering of features in CP (subject prominent languages). Spanish is somewhere in between: the EPP is associated with discourse features on Fin where they are present, otherwise it triggers subject movement as a default. At the other end of the scale are languages like Hungarian, which are often labelled ‘discourse configurational’. In these languages, the EPP is always associated with a discourse feature, suggesting that these features are always present on Fin. A significant part of the word order variation in Romance can therefore be attributed to the different feature scattering in the CP-layer, which while connected to the status of the EPP, does not relate directly to the NSP. 20

Similar effects can be observed in English in cases of locative inversion, where movement of a PP to the C-layer blocks subject movement to spec TP.

‘Free’ inversion in Romance and the Null Subject Parameter

4

249

Narrow focus and subject inversion in Romance

Subject–verb inversion is also possible in Romance NSLs where the subject receives narrow focus. In such cases, the subject receives nuclear stress via the NSR but, crucially, focus cannot project in such cases and so this must be considered a marked word order: (43)

a. Quie´n te regalo´ la botella de vino? Who cl-you gave the bottle of wine ‘Who gave you the bottle of wine?’ b. Me regalo´ la botella de vino Marı´a cl-me gave the bottle of wine Maria ‘Maria gave me the bottle of wine.’

(44)

a. Que´ paso´? What happened? b. #Me regalo´ la botella de vino Marı´a cl-me gave the bottle of wine Maria ‘Maria gave me the bottle of wine.’

(Spanish)

By now it should be clear that the claim that the post-verbal position is the true A-position for NSLs is highly problematic. Nonetheless, the data from narrow subject focused orders across Romance present a potential problem for the claim that Romance NSLs have an active EPP requiring move/merge XP. In fact, superficially, in Spanish and Italian, and to a lesser extent EP, it appears that the prosodic requirement to keep a focused subject low overrides the EPP. Hence in Spanish both XPV(O)S and V(O)S orders are felicitous and in Italian (XP)V(O)S is the norm in narrow focus contexts.21 In what follows I argue that these inversion structures are the result of the spelling out of the vP-internal subject at PF for prosodic reasons. As such the EPP is satisfied in the Narrow Syntax but ‘undone’ at PF (cf. Landau (2006b) for the suggestion that any copy of a moved element can be targeted at PF). 4.1

V(O)S order in Spanish and Italian

The unmarked order with narrow subject focus in both Spanish and Italian is VS, where any direct or indirect objects appear as clitics on the verb. However, if the object is strongly destressed then VOS order is also possible in both languages, though more marginally in Italian (cf. Belletti 2001; 2004; 2005). It has been shown that the following three properties hold of V(O)S structures in Spanish and Italian:

21

OT can deal very easily with the tension between discourse requirements and NarrowSyntactic requirements by making them relative rather than absolute and ranking them differently in different languages. Costa (2001b; 2004) and Szendro˝i (2002) both provide interesting analyses of EP and Italian respectively along these lines.

250

Michelle Sheehan

Subjects appear to remain low (cf. Belletti 2001; 2005; Ordo´n˜ez 2000; Zubizarreta 1998) (ii) Subjects do not have all the properties associated with low subjects (Belletti 2001; 2004). (iii) Objects c-command subjects in VOS in Spanish but not in Italian (Ordo´n˜ez 1998; 2000; Belletti 2004; Zubizarreta 1998) (i)

4.1.1 Subjects remain low There is evidence that the subject remains low in V(O)S with a narrow focused subject from the placement of adverbs in the clause. Belletti (2001: 61) shows that in inverted structures in Italian, the subject obligatorily follows low adverbs such as bene and completamente as well as quantifiers such as tutto. Adopting the position for such adverbs proposed by Cinque (1999), she takes this as evidence that the subject is in a position below the lowest functional projection of IP:22 (45)

?Capira` completamente/bene Maria23 will-understand completely/well Maria ‘Maria will understand completely/well.’

(46)

*Capira` Maria completamente/bene Will-understand Maria completely/well

(47)

Capira` tutto Maria Will-understand everything Maria ‘Maria will understand everything.’

(48)

*Capira` Maria tutto Will-understand Maria everything

This suggests that, in Italian, the subject occupies a vP-internal position. The same pattern is also true of Peninsular Spanish: (49)

Entendera´ bien Juan. Understand-fut-3s well Juan ‘Juan will understand well.’

(50)

*Entendera´ Juan bien. Understand-fut-3s Juan well ‘Juan will understand well.’

Additional evidence for the low position of the subject comes from the incompatibility of floating quantifiers with VOS order in both languages:

22 23

I abstract away from the facts regarding placement of focused pronouns in inverted orders, which seem to occupy a higher position, preceding low adverbs. See Belletti (2001: 62) for a possible reason for the slight marginality of this sentence.

‘Free’ inversion in Romance and the Null Subject Parameter (51)

*Se han comprado todos un coche los chicos cl-self have bought all a car the boys

(52)

*Hanno mangiato tutti una mela i ragazzi Have-3PL eaten all an apple the boys

251

If quantifiers are stranded via movement of the subject, as proposed by Sportiche (1988), then the ungrammaticality of (51) and (52) supports the idea that the subject is in-situ. 4.1.2 Subjects do not behave like low subjects Although there is evidence that post-verbal focused subjects are in a low position, there is also evidence suggesting that they do not have all the properties of low subjects. Belletti (2001) shows that the in-situ subjects of unaccusative verbs (which are subject to Definiteness Effects) permit ne-extraction: (53)

E` arrivato uno studente al giornale Is arrived a student to-the newspaper

(54)

Ne e` arrivato uno al giornale Of-them is arrived one at-the newspaper

Uncontroversially, Beletti takes this as evidence that the post-verbal subject is in vP in (53) and (54), as we have also proposed. In VOS orders, on the other hand, which are not subject to Definiteness Effects, ne-extraction is not grammatical: (55)

E` arrivato al giornale uno studente Is arrived to-the newspaper a student ‘It was a student who arrived at the newspaper.’

(56)

??Ne e` arrivato al giornale uno Of-them is arrived at-the newspaper one

It appears then that focused post-verbal subjects also have properties which set them apart from unfocused low subjects. Belletti (2001; 2005) takes this as evidence that these subjects raise to a low focus position in the vP periphery. However, I would like to propose that it might be evidence, rather, of the fact that the subject moves to spec TP but is spelled out low for prosodic reasons. 4.1.3 C-command relations between the subject and the object It has been argued in the literature that objects c-command the subject in VOS orders in Spanish but not Italian. Ordo´n˜ez (2000) gives two pieces of evidence in favour of this position for Spanish and Belletti (2004) and Zubizarreta (1998) show that the structure of Italian is different. The arguments are summarised below.

252

Michelle Sheehan

4.1.3.1 Variable binding

The first piece of evidence Ordo´n˜ez (1998; 2000) gives to show that the object asymmetrically c-commands the subject in VOS comes from quantifier-variable binding. Assuming that binding is subject to a c-command restriction (as proposed by Reinhart (1983)), Ordo´n˜ez attributes the following grammaticality facts to the hypothesis that the object c-commands the subject at the level of Narrow Syntax in (59) but not in (57) or (58): (57)

*Su1 madre le presento´ a [cada nin˜o]1 al director His mother cl-him presented A each child to-the director ‘His mother introduced each child to the headteacher.’

(58)

*Aquı´ presento´ su1 madre a [cada nin˜o]1 Here presented his mother A each child ‘Here his mother introduced each child.’

(59)

Aquı´ presento´ a [cada nin˜o] su1 madre Here presented A each child his mother ‘Here his mother introduced each child.’

However, previous work by Zubizarreta shows that this is not the whole truth, as Ordo´n˜ez partly acknowledges in a footnote.24 In fact, it seems that binding is very heavily related to information structure. Even in English, a pronoun contained in a focused DP is much more easily bound by a non-c-commanding quantifier than a pronoun contained in a non-focused DP, as Zubizarreta (1998) shows: (60)

*His1 mother will accompany [every boy]1 on the first day of school

(61)

[F His1 mother] will accompany [every boy]1 on the first day of school

The same facts are true of Spanish. A contrastively focused subject in SVO or VSO order behaves exactly like the post-verbal subject in (59): (62)

El primer dı´a de escuela acompan˜ara´ su MADRE a cada nin˜o The first day of school accompany-fut-3s his mother A each child y no su padre and not his father ‘On the first day of school his MOTHER will accompany each child and not his father.’

(63)

El primer dı´a de escuela su MADRE acompan˜ara´ a cada nin˜o The first day of school his mother accompany- fut-3s A each child y no su padre and not his father ‘On the first day of school his MOTHER will accompany each child and not his father.’

24

Moreover, as Alexiadou & Anagnostopoulou (2001) point out, the same facts do not hold with negative quantifiers (attributed to Cecilia Poletto, Giuseppe Longobardi and Maria-Luisa Zubizarreta).

‘Free’ inversion in Romance and the Null Subject Parameter

253

This suggests that it is the fact that the subject must be focused in VOS orders which makes pronominals contained in it so easy to bind. Focus clearly feeds binding, but this does not always appear to correlate with a difference in syntactic position.25 4.1.4 Condition C The second piece of evidence provided by Ordo´n˜ez (2000) relates to condition C. He shows that a condition C violation obtains between an R-expression contained in a DP subject and a coreferential pronominal in VOS order in Spanish. The same is not true of VSO order: (64)

Los hermanos de Eva1 le compraron el libro a ella1 The brothers of Eva cl-her bought the book for her ‘Eva’s brothers bought the book for her.’

(65)

*El libro se lo compraron a ella1 los hermanos de Eva1 The book cl-her cl-it bought for her the brothers of Eva ‘The book, Eva’s brothers bought it for her.’

(66)

*Esta man˜ana lo1 castigo´ a e´l1 la madre de Juan1 This morning cl-him punished A him the mother of Juan ‘This morning, Juan’s mother punished him.’

(67)

Esta man˜ana lo1 castigo´ la madre de Juan1 a EL1 This morning cl-him punished the mother of Juan A him ‘This morning, Juan’s mother punished him.’

Assuming that Condition C is also dependent on c-command as is standard (cf. Chomsky 1993), this appears to be evidence that a copy of the object c-commands a copy of the subject. If condition C applies to the PF-form, this might be evidence that the overt copy of the object c-commands that of the subject in Spanish VOS. In Italian, this effect is difficult to test due to the more marginal nature of VOS with a non-clitic object. Note also that even in Spanish this test is not conclusive as a doubled object pronoun of the type in examples (64)–(67) necessarily receives narrow focus and for this reason might only be felicitous in final position (for reasons of stress assignment). 4.1.5 Condition A Belletti (2004) provides evidence that the opposite c-command relations actually hold in Italian, inasfar as VOS is actually acceptable. An anaphor contained within an object can be bound by a subject

25

Zubizarreta (1998: 13) attributes this to the fact that “focused subjects (like indefinite subjects) are reconstructed to their vP-internal position” and that quantified objects raise to a higher position at LF. The proposal to be made here is that Spanish (and Italian) differ from languages like English in that they also ‘reconstruct’ their focused subjects at PF, in contexts of narrow focus.

254

Michelle Sheehan

in VOS order, whereas an anaphor contained within a subject cannot be bound by a preceding object: (68)

a. Chi ha salutato Gianni? Who has greeted Gianni b. *Hanno salutato Giannii i proprii genitori have(pl) greeted Gianni the own parents

(69)

a. Chi ha salutato i propri genitori? Who has greeted the own parents b. Ha salutato i proprii genitori Giannii has(s) greeted the own parents Gianni

These facts clearly suggest that the subject asymmetrically c-commands the object at a certain level of representation in VOS order in Italian. Judgements for Spanish are subtle and mixed, but for many speakers the contrast is reversed: (70)

a. Quie´n ha ayudado a Juan? Who has helped A Juan ‘Who has helped Juan?’ b. Le han ayudado a Juan sus propios padres. CL have helped A Juan his own parents ‘His own parents have helped Juan.’

(71)

a. Quie´n ha ayudado a sus padres? Who has helped A his parents ‘Who has helped his parents?’ b. ??Les ha ayudado a sus propios padres Juan. CL has helped A his own parents Juan ‘Juan has helped his own parents.’

Despite superficial similarities, this contrast appears to be evidence for a difference in the c-command relations between the subject and the object in Spanish and Italian. In Italian, the object, while it precedes the subject, does not c-command it. It seems to be fronted as part of a larger constituent (VP). Zubizarreta (1998: 137) arrives at the same conclusion, based on the fact that Italian, unlike Spanish, is sensitive to a Relative Weight Constraint, similar to that described for French Stylistic Inversion (cf. Kayne & Pollock 2001). This is why VOS structures with non-cliticised objects are much more marginal in Italian than in Spanish. In Spanish, the object seems to move alone to a position c-commanding the subject. 4.2

Spanish V(O)S

Following Zubizarreta (1998) and Ordo´n˜ez (1998; 2000) I argue that VOS structures are derived via scrambling of the object over the subject in Spanish, and remnant VP-movement, but by movement of VP over the subject in Italian and spelling out of the low subject copy. In fact, Ordo´n˜ez (1998) has shown that there are similarities between object movement in Spanish and

‘Free’ inversion in Romance and the Null Subject Parameter

255

scrambling in languages such as Korean and German. Moreover, recall from Section 2.4 that the only way for an object to avoid being assigned nuclear stress via a phasal NSR is for it to scramble out of the VP. Reinhart (1995; 2006) also claims a similar prosodic motivation for scrambling in Dutch.26 I argue that this scrambling operation is available, not in the PF component (contra Zubizarreta) but in the Narrow Syntax. The derivation proceeds as shown in Figure 6.2.27 Stress assignment proceeds as follows: Phase 1 spellout VP ¼ [V object] delete copies ¼ [V object] NSR applies vacuously Phase 2 Spellout TP ¼ [subject VþvþT object[–stress] subject Vþv] Delete copies ¼ [subject VþvþT object[–stress] subject Vþv] ¼ [subject VþvþT object] NSR applies ¼ *subject VþvþT object[–stress] This PF form will crash, as the object bears some feature to trigger movement to the edge of vP specifically to avoid being assigned NS. We can think of this as a [–stress] feature for our purposes.28 If scrambling is driven by PF-features, then this feature is visible to PF, and incompatible with stress. The only way for the derivation not to crash is for the lower copy of the subject to be targeted for spell-out. As discussed by Bosˇkovic´ (2001), there is ample evidence in natural languages that this kind of lower copy spell-out is possible where a syntax–PF mismatch occurs. Deletion of copies other than the highest is dispreferred by economy, but not ruled out if necessary for convergence.29 Spelling out the low copy gives us the following: Phase 2b Spell-out TP ¼ [subject VþvþT object[–stress] subject Vþv] Delete copies ¼ [subject VþvþT object[–stress] subject Vþv] ¼ [VþvþT object subject] NSR applies ¼ VþvþT object[–stress] subject

26 27 28 29

Neeleman & Reinhart (1998) actually propose that scrambling is a base-generated effect rather than the result of movement. Evidence from adverb placement suggests that the verb actually targets a higher head in Italian than in EP and Spanish (cf. Sheehan 2006). I gloss over these facts here. In fact exactly how movement to the phase edge is triggered is the subject of ongoing debate: see Felser (2004) for a discussion. As Chomsky (2005: 3) notes: “We might discover that SMT [the strong minimalist thesis – MS] is satisfied by phonological systems that violate otherwise valid principles of computational efficiency, while doing the best it can to [. . .] map to the SM interface syntactic objects generated by computations that are ‘well-designed’ to satisfy C-I conditions.”

256

Michelle Sheehan

Phase 2

CP

C

TP

Subject

T⬘

V +v + T

vP

Object

Phase 1

vP

v⬘

Subject

V +v

VP

V

Object

Figure 6.2 Derivation of VOS in Spanish

Because the lower copy of the subject follows the object copy and the verb complex, it is targeted by the NSR. Note also that the object is forced to leave VP where it bears the [–stress] feature as otherwise the derivation will crash where the NSR applies to phase 1.30 According to Cinque (1993) and Zubizarreta (1998), marked stress should not be able to project.31 As expected this is a marked option and, as such, focus cannot project, meaning that the only reading for the sentence is one where the subject receives narrow focus.32 30

31 32

Potentially problematic for this are structures decribed by Cardinaletti (2002) as ‘marginalisation’ where a destressed object apparently remains in-situ. I leave this further level of complication to future research. But see Neeleman & Reinhart (1998) for a different view, differentiating anaphoric destressing from focus strengthening. Bosˇkovicˇ (2002) shows that similar effects can be seen in Romanian with wh-questions. While multiple wh-movement is required in Romanian, a low copy of a wh-phrase may be spelled out if required by PF.

‘Free’ inversion in Romance and the Null Subject Parameter

4.3

257

Italian V(O)S

In Italian, the whole VP is attracted to the edge of the vP phase, to avoid nuclear stress on the object. The derivation proceeds the same as that described for Spanish, modulo this difference. Note that were the NSR subject to c-command rather than linear precedence, then it is not clear why the subject would necessarily receive NS in such cases, as no c-command relation would hold between subject and object at PF. The NSR therefore applies vacuously at the first phase as although the object is inside VP, the whole VP escapes application of the NSR. It remains unclear why the object cannot scramble in Italian. One possibility is the stricter dependency between V and the object, perhaps for case reasons (cf. Belletti 2004). I leave this matter to future research. 4.4

(O)VS and XPVSO order in European Portuguese

In EP, the facts are slightly more complex. Post-verbal subjects are only permitted where another element is fronted via Topicalisation and VOS order is not felicitous in contexts of narrow subject focus (Ambar 1992):33 (72)

a. Quem comeu o bolo (ontem)? who ate the cake ‘Who ate the cake?’ b. (?O bolo), comeu a Maria The cake ate the Maria ‘Maria ate the cake.’ b0 . Ontem comeu a Maria o bolo yesterday ate the Maria the cake ‘Yesterday, Maria ate the cake.’ b00 . #Comeu o bolo a Maria Ate the cake the Maria ‘Maria ate the cake.’

(EP)

In such cases, the subject receives obligatory focus, as shown by the fact that the object cannot be contrastively stressed: (73)

*Desta vez falo eu com a Maria, e na˜o contigo. This time speak I with the Maria and not with-you

(74)

*Ontem comeu a Joana a sopa, e na˜o o frango yesterday ate the Joana the soup, and not the chicken

33

There is some disagreement in this regard, probably due to dialectal variation. For Costa (2004) VOS order is fully acceptable but for other speakers, notably Ambar (1992), it is not. For those dialects that allow VOS, the facts seem to be similar to those in Spanish. I focus on the variant of EP which is different from Spanish and Italian here, as my northern informants also had these preferences.

258

Michelle Sheehan

Moreover, unlike the XPVSO examples in Spanish, (76) is perfect in answer to a wh-subject question: (75)

Quem e´ que falou com a Maria desta vez? Who is that spoke with the Maria this time ‘Who has spoken to Maria this time?

(76)

Desta vez falou o Joa˜o com a Maria Of-this time spoke the Joa˜o with the Maria ‘This time Joa˜o spoke to Maria.’

(77)

*Desta vez falou com a Maria o Joa˜o Of-this time spoke with the Maria the Joa˜o

This order in EP is therefore equivalent in information structure terms to VOS in Spanish and Italian. Note that without the presence of the XP, VSO is infelicitous in this context in EP, as the following examples indicate: (78)

#falou o Joa˜o com a Maria.34 Spoke the Joa˜o with the Maria

(79)

Desta vez falou o Joa˜o com a Maria Of-this time spoke the Joa˜o with the Maria ‘This time Joa˜o spoke to Maria.’

Despite certain superficial similarities, it seems that XPVSO in EP has a different information structure from that in Spanish. In EP, only the subject can receive main stress in this word order, via an MSR. For some reason it is not possible to stress the object, via the NSR in such cases. It seems necessary to conclude again, then, that the same surface order in two closely related languages has two vastly different underlying structures. Despite the apparent similarities between the two languages, in EP, inversion is due to I-to-C movement.35 Evidence for this position comes from the fact that the most natural subject position is between the auxiliary and the subject, contrary to the facts for Spanish: (80)

34

35

Nesse dia, tinha a Maria emprestado os seus melhores discos ao Pedro. On-that day had the Maria lent the her best CDs to Pedro ‘On that day Maria had lent her best CDs to Pedro.’

Costa’s (2004) careful study of EP word order reveals that VSO order with no fronted XP is felicitous where the subject and object are both focused. I do not discuss such intricacies here for reasons of space. See also Ambar (1992) for further evidence that inversion in EP is triggered by verb fronting in such cases.

‘Free’ inversion in Romance and the Null Subject Parameter (81)

259

?Nesse dia, tinha emprestado a Maria os seus melhores discos ao Pedro. On-that day had lent the Maria the her best CDs to Pedro ‘On that day Maria had lent her best CDs to Pedro.’ (examples from Ambar (1992: 80))

This is difficult to capture if the subject remains in-situ in vP, as EP past participles clearly raise out of vP, as illustrated by the following sentence, taking bem/bien ‘well’ to mark the edge of vP (Costa 2004): (82)

O Paulo tinha lido bem os livros. The Paulo had read well the books ‘Paulo had read some books well.’

(83)

*O Paulo tinha bem lido os livros.36 The Paulo had well read the books

Note also that Topicalisation triggers enclisis in EP, whereas non-topicalised subjects/fronted elements trigger proclisis as Pilar Barbosa has pointed out. Thus the subject in (84) has been topicalised but the non-referential subject in (85) has not, and nor has the embedded subject in (86): (84)

O Joa˜o mandou-te uma carta ontem. The Joa˜o sent-cl-you a letter yesterday ‘Joa˜o sent you a letter yesterday.’

(85)

Algue´m te mandou uma carta ontem. Someone cl-you sent a letter yesterday ‘Someone sent you a letter yesterday.’

(86)

Acho que o Joa˜o te mandou uma carta. Find-1s that the Joa˜o you-cl sent a letter ‘I think that Joa˜o sent you a card.’

These patterns are also apparent where XPs other than the subject are fronted. Thus in (87), where the subject is focused and the adverbial ontem ‘yesterday’ is topicalised, the pattern of cliticisation reflects this fact: (87)

36

a. Quem te mandou uma carta ontem? Who cl.2s sent a letter yesterday ‘Who sent you a letter yesterday?’ b. *Ontem me mandou algue´m uma carta Yesterday cl.1s sent someone a letter ‘Someone sent me a letter yesterday.’ b0 . Ontem mandou-me algue´m uma carta Yesterday sent-cl.1s someone a letter ‘Someone sent me a letter yesterday.’

A reviewer points out to me that this might be an effect of the verb ler. To clarify, the same word order effects apply to all past participles in EP.

260

Michelle Sheehan

Assuming, following Kayne (1991) that enclicis is the result of verb movement, I therefore propose that enclisis in EP is the result of verb movement to the Top head. On the way to Top, the finite verb, which is otherwise spelled out in T, raises successively through all intervening heads including the Cl head causing the clitic to suffix onto the verb (see Figure 6.3) Verb-movement to Top removes the finite verb from the possible focus set of the sentence. This movement seems to preclude the possibility of object shift/scrambling in EP. Either the object can be topicalised, elided or as a last resort destressed. While EP does not allow discourse features to be bundled onto Fin, the (optionally present) head Top is strong, meaning that it is a spellout point for the finite verb.37 It also bears an obligatory EPP-feature, where present, triggering movement of a topic to spec TopP. There is thus reason to believe that the inversion seen in EP in instances of narrow subject focus is independent of the NSP, and is rather due to facts about Topicalisation in that language. In such structures the subject occurs neutrally in spec TP but can also surface in spec vP, suggesting some marginal interaction with the EPP. I leave this added complexity to future research. 5

Implications for parametric theory

It has been argued that the kinds of inversion seen in Romance NSLs are not those predicted by a parameterised EPP of the type proposed by Barbosa and A&A. The availability of inversion in wide focus contexts can be attributed to the availability of null locatives and expletives in a given language. While this fails to fall out from a parameterised EPP, it can be attributed to the NSP if the latter relates to the possibility of filling spec TP with non-PF interpretable material (cf. Holmberg 2005; Chapter 2). One major difference distinguishing the three languages is that features are scattered differently in the CP-layer. This has the effect of triggering subject inversion as a side effect where certain discourse-related movement occurs. This kind of inversion is therefore totally independent of the licensing of null subjects, though closely connected to the status of the EPP, and in some ways akin to the kind of inversion seen in certain V2 structures, and discourse configurational languages where the EPP is always associated with discourse features (see also Holmberg (Chapter 2), who shows that this property does not fit into a structured null subject parameter hierarchy). The word orders associated with narrow subject focus in Spanish and Italian are distinct again. The possibility of VOS order can be indirectly linked to the NSP as arguably only NSLs permit the deletion of copies in spec IP (as they 37

The idea would be that certain aspectual heads such as ja´ and sempre as well as negation block this verb-movement triggering proclisis.

‘Free’ inversion in Romance and the Null Subject Parameter

261

TopP Topic position Top⬘ FinP

Top* Fin

AgrP38 Subject

Agr⬘ CIP

Agr CI

TP T⬘ T

vP

Figure 6.3 Verb movement in EP

lack what Holmberg (Chapter 2) calls the P-property). In fact these properties seem even more closely related if we adopt a deletion analysis of null subjects (cf. Roberts, Chapter 1). However, the fact that (some dialects of) EP lack such a spell-out process indicates that it is not a necessary effect of the NSP.39 Languages use different strategies to bridge the gap between syntax and information structure. Some languages like English, have very restricted syntax, and so have to resort to destressing in order to alter focus sets. Other languages, such as EP, have null topics and topicalisation which they use to alter the assignment of nuclear stress. Spanish and Italian allow removal of the object by scrambling or VP-fronting, but cliticisation of the object is much preferred in such contexts.

38

39

I use the label AgrP here merely to reflect the fact that there are at least two different positions in IP which can be targeted by the finite verb. See Sheehan (2006) for evidence that the verb raises to Agr in Italian, but to T in Spanish and EP in non-topicalisation environments. This implies a different status for clitics in Italian and other uniformly proclitic languages (cf. Roberts forthcoming b). If Samek-Lodovici (2006) is correct and contrastive focus is really akin to narrow focus with extraction of topics followed by remnant movement, then the fact that EP lacks both kinds of focus strategies would find a principled explanation.

262

Michelle Sheehan

All of this seems surprising as movement is often considered to be a last resort, occurring obligatorily to save a derivation from crashing. The interaction with information structure is more complicated than this. One way to think of it is that a destressing is a marked PF rule, and is to be avoided at all costs. Although it is used (in many languages) as a last resort, the syntactic freedom of a given language is manipulated to prevent it having to apply. This has the effect of making certain word orders specialised for certain information structure contexts in languages with a more flexible syntax. 6

Conclusions

It has been shown that while Romance languages have a number of different inversion strategies, ranging from the fronting of locatives and extraposition through I-to-C movement and even spelling out of low copies for prosodic reasons, none of these instances of inversion fall out from a parameterised EPP of the type proposed by A&A. Crucially, the inversion seen in wide focus contexts is not licensed via verb-movement to the inflectional domain, but rather via XP-movement which satisfies the EPP and allows the subject to remain insitu. In narrow subject focus contexts inversion is due to (i) verb-movement to Top in EP or (ii) the spelling out of a low copy in Spanish and Italian. A unified analysis of these types of inversion does not seem possible at this stage.

7

Subjects, Tense and verb-movement* Theresa Biberauer and Ian Roberts

1

Introduction

The goal of this chapter is to provide a more complete basis for T’s attraction properties by exploiting a neglected property of T, its Tense features. In Section 2, we will argue on cross-linguistic grounds that the properties triggering V-to-T movement and licensing null subjects should be kept distinct. In other words, we reject the idea, first proposed by Roberts (1985), that ‘strong’ agreement of some kind is always responsible for both. Instead, in Section 3, we suggest distinguishing T’s tense (i.e. ‘verbal’ or Vrelated) properties from its agreement/j (i.e. ‘nominal’ or D-related) properties and propose a correlation between richness of tense inflection and V-to-T movement on the one hand and, following numerous researchers since Rizzi (1982), between richness of agreement inflection and the availability of null subjects on the other (see Roberts (Chapter 1) and Holmberg (Chapter 2) for recent proposals in this respect). We then construct a new typology on the basis of this proposal. In Section 4, we first show how Modern English fits into this typology; in this connection, the auxiliary system is crucial, and we make some novel proposals in this regard (Section 4.1). Section 4.2 deals with relevant aspects of the historical development of this system, focusing on the development of systematic V-to-T movement from the loss of V2, and showing how this was a rather short-lived option, as predicted by the typology of Section 3. Section 5 extends the typology by looking at the Celtic languages, and further compares these with the Polynesian V-initial languages, giving rise to a clear distinction between the two groups, showing that there is more than one way in which consistent V-initial order can be derived (cf. the contributions in Carnie & Guilfoyle (2000) and Carnie, Dooley & Harley (2005) for earlier, rather different arguments in favour of this conclusion). Section 6 concludes the chapter. *

This chapter is a longer version of that which appeared in the Cambridge Occasional Papers in Linguistics (COPiL) 3. We thank the GLOW V in Asia audience for their comments, in particular Pino Longobardi and Marc Richards. We also thank Marı´a-Luisa Zubizarreta for comments on an earlier version of this chapter and Neil Myler for his help with Spanish data.

263

264

Theresa Biberauer and Ian Roberts

2

Triggering V-to-T and licensing null subjects: agreement-centred approaches

The first proposal regarding the relation between the strength of agreement and null subjects was made by Taraldsen (1978), and has become known as Taraldsen’s Generalisation. This can be summarised as follows: “[t]he intuitive idea is that where there is overt agreement, the subject can be dropped” (Chomsky 1981: 241). Similarly, V-to-T movement has been related to relatively ‘rich’ verbal agreement inflection (Roberts 1985: 32; 1993b; 1999; Vikner 1997; 2001; Rohrbacher 1999; Kosmeijer 1986; Bobaljik & Thra´insson 1998; Koeneman 2000). As Koeneman & Neeleman (2001) point out, combining these two ideas leads to the postulation of three classes of languages as far as inflectional richness is concerned. These are given in (1): (1)

a. the rich: V-to-T and null subjects (Italian, Greek, Spanish, etc.); b. the middle class: enough ‘wealth’ for V-to-T but not enough for null subjects (French, Middle English); c. the impoverished: neither V-to-T nor null subjects (Modern English, Mainland Scandinavian).

The implicational corollary of (1) is that no null-subject language lacks V-toT movement. The connection between V-to-T movement and null subjects becomes still closer in the context of proposals such as those in Barbosa (1995) and Alexiadou & Anagnostopoulou (1998) to the effect that in nullsubject languages T’s EPP-property is satisfied by movement of a D-featurebearing V (this is a variant of Holmberg’s (2005) Hypothesis A; see Holmberg (Chapter 2), Roberts (Chapter 1)). Against this background, it is a tempting move to establish a relation between V-to-T movement and the Null Subject Parameter by supposing that wherever V raises to T, null subjects are licensed and overt expletives barred. The problem with this idea is that there is well-known empirical evidence, notably from French, that V-to-T movement can exist in a non-null-subject language (cf. also Sheehan (2006) and Biberauer (Chapter 4) for discussion of overt expletives in full NSLs, i.e. in V-to-T raising languages). Well-known examples like (2a) illustrate the operation of V-to-T movement in French (see Emonds 1978; Pollock 1989), while the obligatory expletive in what is clearly a non-V2 language illustrated in (2b) shows that French is not a null-subject language:1 (2)

1

a. Jean embrasse souvent Marie John kisses often Mary ‘John often kisses Mary.’

There is an enduring question as to whether Modern French is in fact a null-subject language. See the discussion in Roberts (Chapter 8).

Subjects, Tense and verb-movement

265

b. *(Il) y a une licorne dans le jardin there is a unicorn in the garden ‘There is a unicorn in the garden.’

Standardly, directly relating V-to-T movement to the licensing of null subjects is ruled out by appeal to the relative weakness of French agreement marking, while at the same time positing that this inflection is strong enough to trigger V-to-T movement. Another possibility is to consider the relative strength of specific agreement features of T, e.g. Person vs. Number as in Roberts (1993b). Such approaches seem to us to border on stipulation, and in any case to miss the potential relevance of another independently required property of T, namely its Tense features. 3

An alternative: T and tense inflection

In this section we will make a proposal regarding the relation between (finite) T and (finite) V. The proposal is open to a simple form of parametrisation, which we will illustrate by presenting a typology of the ways in which T’s verbal (V-) as opposed to its nominal (D-) features can be satisfied. The discussion in this section is simplified in that we omit reference to v, in order to indicate more clearly the ways in which our proposals regarding T’s V-features are related to properties of lexical verbs in various languages. In our discussion of the auxiliary system of Modern English in the next section, we will introduce v. We propose as a universal property of simple finite clauses that T and V enter into an Agree relation. It is plausible to think that T has an unvalued V-feature since we take T to be inherently verbal (it is commonly thought of as the position into which auxiliaries are merged; see Section 4.1 below, and cf. also proposals in terms of which the extended projections of a given head project successively fewer features – cf. i.a. Cecchetto & Donati 2007), but to lack the fundamental semantic property of verbs, namely argument structure. Since finite verbs may bear tense morphology but have no temporal content of their own, we can think that V has an unvalued Tense feature. On the other hand, (main) verbs have argument structure, which we take to be intrinsically connected to having a valued V-feature (see Baker 2003a). These universal properties of T and V suffice to cause T to function as a probe and V as a goal in a simple clause, following the standard conception of this relation in Chomsky (2000; 2001). How the probe–goal relation results in feature valuing varies parametrically. In English, Agree licenses V’s tense morphology; this is how we conceive the traditional ‘Affix Hopping’ relation of Chomsky (1957) and much subsequent work. Since we assume that in Continental Germanic V only moves in verb-second environments, and therefore only to C and never just to T, T and V Agree with no V-movement taking place, in non-V2 environments in these languages (see for example

266

Theresa Biberauer and Ian Roberts

Vikner 2005).2 In Romance languages (both in French and the null-subject languages Italian, Spanish, etc.), this Agree relation is associated with an ‘EPP’-like feature on T which triggers V-movement, whose precise nature we will describe in more detail below. We propose that the difference between Germanic and Romance is correlated with the richness of the inflectional (i.e. synthetic) marking of tense distinctions. The Germanic and Romance languages differ noticeably in the number of synthetic tense paradigms that are typically found. As the following contrast shows, the Romance languages are considerably richer, in a clear intuitive sense, than the Germanic ones in this respect: (3)

Romance: French: parle (present indicative/subjunctive), parlerai (future), parlerais (conditional), parlais (imperfect), [parlai (preterite), parlasse (past subjunctive)]3 Italian: parlo (present), parlero` (future), parlerei (conditional), parlavo (imperfect), parli (present subjunctive), parlassi (past subjunctive), parlai (preterit) Spanish: hablo (present), hablare´ (future), hablarι´a (conditional), hablaba (imperfect), hable´ (preterite), hable (present subjunctive), hablase (past subjunctive I), hablara (past subjunctive II)

(4)

Germanic: German: spreche (present indicative/subjunctive), sprach (past), spra¨che (past subjunctive) English: speak (present), spoke (past) Swedish: snakker (present), snakket (past)

Our approach thus postulates that there are two quite distinct types of ‘richness’ of verbal inflection: agreement inflection and tense inflection. We take tense inflection to include the marking of notional mood and aspect, as in the Romance subjunctive, futures, conditionals and imperfect forms; it is possible that the relevant features are associated with modal and aspectual heads, but crucially they are associated with a single inflection or set of inflections on the verb, and the verb moves to T consistently, hence these features must be implicated in the probe–goal relations realised by the morphology and systematically connected to verb-movement. The contrast between Romance and Germanic then really lies in the inflectional realisation of these features, since Germanic does realise a basic tense opposition. ‘Rich’ agreement inflection has many of the properties which are standardly attributed to it: it triggers movement of a D-bearing category (perhaps 2

3

This statement implies that we disagree with the view that in subject-initial V2 clauses V moves only to T (cf. Travis 1984; Zwart 1997). See Schwartz & Vikner (1996) for arguments for why V must leave TP in V2 clauses. The last two forms are in brackets since they are not part of spoken French, existing only in the literary language and mainly learned in school.

Subjects, Tense and verb-movement

267

V in a null-subject language) and it licenses null subjects. ‘Poor’ agreement inflection, on the other hand, determines the presence of overt expletives, perhaps for the reasons given in Richards & Biberauer (2005). On the other hand, ‘rich’ tense inflection triggers V-movement and is irrelevant to subjectlicensing. As we will see in Section 4, it may also determine the possibility of VP-fronting and ellipsis. Our proposal entails a new typology, and eliminates the ‘middle class’ of semirich T noticed by Koeneman & Neeleman (2001). The typology is given in (5): (5)

a. Rich agreement and rich tense inflection: hence V-to-T and null subjects, e.g. Italian, Greek, Spanish, etc. b. Poor agreement but rich tense: hence V-to-T, but no null subjects, e.g. French (and see Section 4.2 on Middle English). c. Poor tense and poor agreement: hence no V-to-T and no null subjects, e.g. Modern English, Mainland Scandinavian (see Section 4.1 on Modern English). d. Rich agreement and poor tense: null subjects, but no V-to-T; no clear example, but see below.

Before proceeding further, we need to understand more clearly in what sense the ‘richness’ of inflection may be connected to movement. In the context of a feature-strength-oriented model of syntax such as that of Chomsky (1995b), the connection would be straightforward: rich inflection would correlate with a strong feature and, consequently, with the occurrence of overt/pre-spell-out movement. In an Agree-based model such as that of current minimalism, which entails numerous conceptual advantages over its checking-based predecessor (notably, not necessitating the postulation of non-interpretively motivated covert movement triggered by weak features), the connection is, however, less readily captured: in the Agree model, there is no longer any specific reason to assume a correlation between inflectional richness and the presence of a movement trigger (EPP-feature). It could, of course, be the case that inflectional richness constitutes an ‘extra’ morphological cue (alongside the usual positional cues) as to the presence of an EPP-feature. Here we suggest a relatively informal, non-technical answer to this question which we, however, think is on the right track. Suppose that the presence of ‘rich’ tense implies that finite verbs in the languages in question are not categorially simple Vs, but are instead compound elements consisting of V and a fully-specified T (cf. also Koeneman (2000) for a very similar proposal regarding VþAgr). This compound is formed in the Numeration; forming it is part of the process of pre-syntactic word-formation. Such a compound element must merge with both the V-complement, in order to form a VP, and the T-complement, forming a TP. The only way to satisfy the properties of the compound VþT element is to first merge it as V (thereby satisfying its V-/ thematic component; cf. also Koeneman 2000; Neeleman & van de Koot 2002; Cecchetto & Donati 2007) and then to raise it, following the standard movement

268

Theresa Biberauer and Ian Roberts

process, to project as what we usually think of as T. We are therefore proposing a form of partial reprojection, in that the T-features of the compound element determine the formation of the TP (the V-features do not, as these have played their role in forming the thematic domain of VP, although they must move with T as part of the compound VþT element). Movement is thus triggered by the inherent features of the compound VþT element, and ‘richness’ of tense morphology is what underlies the lexical requirement for the formation of such an element in the Numeration in the first place: the individual reflexes of ‘rich’ tense morphology are assumed to be stored in the Lexicon as independent lexical items (cf. Bobaljik & Thra´insson’s (1998) parallel assumptions about the storage of rich agreement morphology in null-subject languages). A consequence of the proposal we have outlined here is that we do not in fact need to postulate an EPP-feature on T, V-to-T movement being triggered thanks to the intrinsic nature of the lexically formed compound VþT. The above sketch does not explain why the line between richness and poverty of tense morphology lies exactly where it does; we suspect that this has to do with the presence of aspectual distinctions, at least in past tenses, in Romance. This therefore suggests that v-features may be relevant (see Section 4.1 on the idea that aspectual features are associated with v). However, we will not speculate further on this here, but instead proceed to consider more closely the implications of our typology. 4

Extending the approach: parametric variation in respect of the position of tense features

In the previous section, we established that the Romance languages, as a family, have rich tense inflection and V-to-T movement, and suggested why this may be. We implied that Germanic lacks both of these properties. Of course, all the Germanic languages except (Modern) English are verb-second (V2) languages. If we leave Modern English to one side for a moment, then, we observe that the Germanic languages have impoverished tense inflection, but that they do show V-to-C movement in the relevant ‘root’ environments. Suppose then that this is related to a further aspect of the V-T Agree relation. More precisely, suppose that the V-movement component of V2 is triggered by a T-related feature of C and that full V2 languages in some sense involve a ‘hybrid’ C, which is also the locus of Tense.4 We suggest that Tense morphology is not ‘richly’ realised on C in these languages because C is a phase-head. Arguably there is no need 4

Cf. Vikner (1995) for overview discussion of proposals along these lines. We take it that the XP-movement component of V2 is triggered by an Edge Feature in the sense of Chomsky (2008); for arguments that V2 has these two components, pace Mu¨ller (2004a), see Biberauer & Roberts (2004).

Subjects, Tense and verb-movement

269

for a phase-head to morphologically realise its uninterpretable features, since the presence of these features is guaranteed by UG (see Chomsky 2008); phaseheads may, however, realise these features, either independently or along with one or more of the non-phasal heads to which it has ‘spread’ features (cf. Section 5 below for more concrete discussion of the options that appear to be available). On the other hand, non-phase-heads realise uninterpretable features overtly; this gives an overt realisation of their status as ‘derived probes’. This leads to the prediction that ‘rich’ morphology is only systematically realised on non-phase-heads that are probes, such as Romance T. We are now in a position to explain the gap in the typology in (5): no system can have rich agreement (where we take this to mean rich agreement inflection) and lack V-to-T movement, since the agreement features will have to be realised in the Numeration as part of the ‘Tense’ element that is merged with V and gives rise to movement/reprojection (cf. again Koeneman 2000). All of this leads us to conclude that T is not involved in the V-movement that gives rise to V2 (since we do not assume the Head Movement Constraint, V-movement to C is not required to pass through T.)5 Hence, V2 is quite independent of the ‘richness’ of tense marking in Germanic. Germanic V is therefore either in C (V2 clauses) or in v/V (elsewhere), but never in T, just as the inflectional richness-oriented proposal outlined in Section 3 would lead us to expect. The question we wish to concentrate on in this section is where English fits into this picture. Earlier stages of English (up to roughly 1450; Fischer, Kemenade, Koopman & van der Wurff 2000: 133ff.) were V2, as is well known (see the references given by Fischer et al.). Presumably, then, we can treat English up to this period on a par with the rest of Germanic.6 Early Modern English, on the other hand, appears to have had V-to-T movement that has often been likened to that in Modern French (Roberts 1985; 1993b: 1999; Pollock 1989), while Modern English, since the seventeenth century, lacks V-movement. In this section, we begin by providing a sketch of an analysis of the Modern English situation, which entails partial analysis of the auxiliary system, negation and the conditions determining the appearance of 5

6

The Head Movement Constraint is incompatible with the probe–goal system generally, in that it is in general not possible to formulate an ‘absolute intervention’ constraint of this type, the probe–goal relation being subject to a ‘relativised intervention’ constraint holding between heads. Head-movement itself may be incompatible with this system, as argued by Chomsky (2001: 37–38), although it is possible that a form of head-movement is compatible with current general assumptions, as we are assuming here (see the above comment on reprojection; see also Lechner 2006; Matushansky 2006; Roberts forthcoming b; Vicente 2006). Old English and at least Southern dialects of Middle English allowed ‘verb third’ orders if the item intervening between the initial XP and the verb was a pronoun and if the initial XP was not a wh-phrase, the negative ne or the discourse adverb þa (van Kemenade 1987: chapter 4). Following Roberts (1996a), we take it that the verb is in C in these contexts (pace i.a. Haeberli 1999; Fuss 2003).

270

Theresa Biberauer and Ian Roberts

auxiliary do (Section 4.1). We then proceed in Section 4.2 to an analysis of the development of verb-movement in English since 1450.

4.1

‘Affix Hopping’, ‘do-deletion’ and T-V Agree in English

In this section we present our analysis of Modern English (NE). Here we want to propose that NE represents a third option in relation to those of Romance and the V2 Germanic languages as just described. In NE, Tense features are lexicalised in T as auxiliaries, there is no V-movement to T or C, and V’s unvalued T-feature is licensed by Agree with T, as mentioned in the previous section. In other words, the T-V Agree relation does not involve V-movement in NE. Neither does it involve T-to-V movement (“Affix-Hopping”), as has frequently been proposed (Chomsky 1957; 1991; Pollock 1989; Lasnik, Depiante & Stepanov 2000; Bobaljik 1995; Freidin 2004). Instead, we propose that the close relation between an auxiliary and the form of the non-finite verb or auxiliary which immediately follows it is an instance of Agree. In (6) we give some examples of the Agree relation which holds between an auxiliary and a main verb: (6)

a. b. c. d. e.

John has eaten John is eating John was eaten John must eat-ø For John to eat-ø . . .

In (6a), we take have (or, more precisely, the complex of formal features that corresponds to the lexeme HAVE) to be merged as a T bearing a valued Perfect feature, while V bears the unvalued counterpart of this feature. For concreteness, we take formal features to be attribute–value pairs of the general form [Attribute: Value]. Unvalued features lack a value, i.e. they have the general form [Attribute: __]. For this reason they cannot be interpreted at the interfaces. In (6a), then, T has the feature [Aspect: Perfect] and V has the feature [Aspect: __]. The Match component of Agree simply copies the value from the probe to the empty slot in the goal (the unvalued Aspect feature on the V-goal can be thought of as the feature that makes it ‘active’ here, parallel to the manner in which unvalued Case renders DPs ‘active’, and T’s feature matrix will also, as outlined in Section 3, include an unvalued V-feature, which is the basis for its probe status). In (6b), T is specified as [Asp: Progressive] and V once again Agrees, taking on the Progressive value of this feature. In (6c), we can take T to be [Voice: Passive] and in (6d) and (6e) T has some Mood specification, with which V will Agree. Clearly, there is much more to say about the features we are working with here, but our main concern is not with the content of the

Subjects, Tense and verb-movement

271

features but with the dependencies they are associated with and which are mediated by the Agree operation. We will look a little more closely at the features themselves when we introduce v at the end of this section. It is clear that this approach to the dependency between the form of the auxiliary (i.e. the form of T) and the form of the main verb in V almost exactly replicates the effects of traditional Affix Hopping, without recourse to downward movement or special morphological operations (beyond the usual conventions of phonological realisation of feature bundles which may be partially created in the syntax). This relation has been observed since Chomsky (1957) to be highly local (see in particular Emonds 1976) in that each auxiliary determines the form of just the following verb or auxiliary; there is no possibility of a relation holding between Aux1 and V in a sequence like Aux1-Aux2-V, as shown by the ungrammaticality of examples like (7): (7)

a. *John is[Prog] been[Pass] smoking[Prog] b. *John has[Perf] being[Prog] smoked[Perf]

This locality is guaranteed by the non-intervention clause in the definition of Agree. We follow standard assumptions (Chomsky 2000; 2001) in taking the structural environment in which Agree holds to be defined as follows: (8)

a Agrees with b where: (i) a asymmetrically c-commands b. (ii) there is no g non-distinct in formal features from a such that g c-commands b and a c-commands g.

If we assume iterations of auxiliaries to involve at least one occurrence of vP, then (7a) would have a structure like (9): (9)

TP T [Asp: Prog]

vP

v [Voice: Pass]

VP V [Asp:—]

In (9) we can see that v is a structural intervener between T and V, in the sense that v c-commands V and T c-commands v. Thus, T corresponds to a in the definition in (8), V to b and v to g. However, for v to function as an intervener as defined in (8ii), it must be non-distinct in formal features from T. We therefore must assume that formal features such as Aspect and Voice are not distinct from one another in the relevant sense, i.e. that they are all V- or T-features (we will clarify which of these is the correct option below).

272

Theresa Biberauer and Ian Roberts

We will not go into the question of the order of NE auxiliaries here. Cinque (1999: 153ff.) shows that the basic order Mood > Tense > Perfect > Progressive > Voice is cross-linguistically very widespread, and, with the proviso that Tense and Mood are apparently collapsed into a single position (in that modals are inherently finite), this order is followed by the NE auxiliaries. If we take the Aspect and Voice features to be features of v, then this is arguably a case of the situation described by Chomsky (2005: 18) in his remark that “the more elaborate structures revealed by the cartographic inquiries are based on linearization of features in [phase] labels”. Where an auxiliary appears V does not have an unvalued Tense feature. If it did, it would be too far from T to be valued under Agree, still assuming that v has features which make it an intervener (see above). This is in fact the correct result, since, whenever an auxiliary appears, V shows up in some non-finite form (participle or ‘bare infinitive’). We take it that these forms lack the unvalued Tense feature. These remarks give an outline analysis of the basic cases of auxiliary-verb dependencies in NE, involving the aspectual and modal auxiliaries (and infinitival to). ‘do-support’ is of course a more complex phenomenon, and we now present an outline analysis of this. The first observation is that overt do Agrees with a bare verb (like modals and to, see (6d,e)). We suggest that the appearance of overt do is regulated by the presence of an ‘extra’ feature on T, in addition to Tense- and j-features, with this ‘extra’ feature – call it [þAffect] – necessarily resulting in a specific discourse effect (cf. the so-called Fox–Reinhart intuition on optionality further discussed below). Let us look first at negative do. In this connection, we wish to argue, following Zwicky & Pullum (1983) and Williams (1994: 168), that the auxiliaryþn’t combinations of NE are lexical items, i.e. forms such as won’t, can’t, shan’t and, most importantly for our purposes, don’t, are synchronically lexical items distinct from the positive auxiliaries will, can, shall and do. We will argue for this view by showing that the obvious alternative, cliticisation of n’t to the auxiliary, is not a viable option. One possibility would be to claim that n’t cliticises to the auxiliary, with the auxiliary in T. The problem with this is that it violates the generalisation, originally due to Kayne (1991), that cliticisation always involves adjunction to the left of the host (in Kayne (1994), this is made to follow from the Linear Correspondence Axiom). An alternative is to propose movement of the auxiliary (presumably from v, see below) to n’t (where n’t occupies either Neg or T). An objection to this is that the presence of n’t is not sufficient to trigger auxiliary-movement, as we can see in ‘subjunctive’ examples like (10):

Subjects, Tense and verb-movement (10)

273

a. I require that he not have left when I get back b. *I require that hen’t have left when I get back c. * I require that he haven’t left when I get back

(10a) illustrates the not-auxiliary order, the input to the putative cliticisation process. (10b) shows that not cannot contract onto the subject (this is quite independent of segmental phonology, cf. I require that Mr Wood*n’t/not do it). (10c) shows that the auxiliary-movement cannot be triggered purely by the presence of not; this further shows that the ungrammaticality of (10b) is not simply due to lack of auxiliary-movement. Another objection to the cliticisation analysis is that certain auxiliaryþn’t combinations are not allowed in inverted position. This is true of counterfactual conditionals featuring what are historically past subjunctive auxiliaries (the observation that only had, were and should may invert in this context is originally due to Pesetsky (1989)): (11)

a. *Hadn’t I done that, everything would have been fine b. Had I not done that, everything would have been fine c. If I hadn’t done that, everything would have been fine

These examples show that it is the combination of contraction and inversion that is not allowed in this context, which we will refer to as involving ‘counterfactual C’. It is of course possible for forms such as hadn’t to appear in C in interrogatives: (12)

Hadn’t he finished when you got back?

One might interpret the contrast between (11a) and (11b) as indicating that some auxiliaryþn’t combinations (those which are permitted there) are formed in C, while others (those which cannot appear in C) are not. This would, however, have to be a kind of ‘long-distance’ cliticisation of a type absolutely unattested elsewhere in NE, as the fact that non-contracted not is not allowed in or adjacent to C attests: (13)

*Did not John leave?

So, in order to maintain that auxiliaryþn’t combinations are formed by auxiliary-movement to n’t in T, we have to maintain that T is the only position in which the combination can be formed, but that auxiliary movement is not driven by the need to form the combination and that some auxiliaryþn’t combinations cannot move on to C. We take these facts to show that the auxiliary-movement analysis cannot be maintained. On the other hand, these facts can be straightforwardly captured if we treat negative auxiliaries as directly merged in T as a single lexical item. The facts in (11) can be captured in these terms if we say that a negative auxiliary can only receive an interrogative interpretation when it raises to C, with the

274

Theresa Biberauer and Ian Roberts

counterfactual interpretation not being available. In fact, this is true of all negative auxiliaries, not just those whose positive forms are able to appear in ‘counterfactual C’ (i.e. C with an ‘extra’ movement-triggering feature; see below). Distinguishing negative from positive auxiliaries makes this observation easier to state; if there were simply a syntactic auxiliary-movement rule, giving rise to blind, syntactic formation of auxiliaryþn’t, it would be very difficult to rule out T-to-C movement in examples like (11a). A final argument in favour of a lexical treatment of negative auxiliaries comes from Zwicky & Pullum (1983), who observe that inflections, but not clitics, trigger stem allomorphy. Since n’t triggers stem allomorphy on the auxiliaries it combines with (e.g. will/won’t), we must therefore conclude that n’t is an inflection which attaches to auxiliaries (see also Spencer 1991: 381ff.). Our analysis therefore implies that NE has a class of negative auxiliaries, similar to the Uralic languages, Latin, Old English and various other languages. Negative do (i.e. the forms don’t, doesn’t, didn’t) is thus the form that corresponds to negative, non-modal, finite T with various Tense and f-feature specifications. If we claim that auxiliaryþn’t combinations are lexically formed negative auxiliaries, then we must provide an analysis of ‘non-contracted’ not, which we now see as an element synchronically independent of n’t. In many contexts, not has an interpretation and a syntax distinct from n’t. In fact, whenever not is non-adjacent to the auxiliary, constituent negation results: (14)

a. b. c. d.

John has always not smoked The kids have all not done their homework The kids must all not go out tonight *The kids need all not do their homework

The constituent-negation interpretation is clear in (14a), which can only mean ‘At all times, it is the case that John engaged in not smoking’. Negation only applies to the VP (or perhaps vP) smoked. Similarly, in (14b) and (14c), the relative scope of negation and the floated universal quantifier follows the surface order. The ungrammaticality of (14d) can be attributed to the fact that modal need is a kind of Negative-Polarity Item. As such, it must be in the scope of negation or similar affective operator, which it is not in (14d) since not occupies a structurally lower position than T and has scope only over its c-command domain, vP/VP, and all has scope over not here. In all these examples, then, not does not have clausal scope, having instead a narrower scope, which we take to be a form of constituent negation. Contrast the examples in (15), where not has been replaced by n’t throughout: (15)

a. b. c. d.

John hasn’t always smoked The kids haven’t all done their homework The kids mustn’t all go out tonight The kids needn’t all do their homework

Subjects, Tense and verb-movement

275

Here n’t has clausal scope, entailing that it has wider scope than always in (15a) and than all in (15b–d). Furthermore, as with other possibility modals, it has scope over need in the context of the lexical form needn’t (15d) (see Williams (1994: 168), for further discussion of the relative scope of modals and negation). The scope contrasts in (14) and (15) support the idea that n’t has not moved from a position following the adverb or floated quantifier, since, under that proposal, some special stipulation must be made in order to prevent ‘reconstruction’ of this movement at LF, which would give rise to the same scope readings in (14) and (15). But not can, of course, have clausal scope. This happens when it appears adjacent to the auxiliary: (16)

John must/does not smoke

What forces do to appear in (16) (when must is absent)? We suggest that there is a ‘negative-concord’ relation between T and not here. More precisely, T in this case is comparable to French ne and not to pas: T has an ‘affective’ (polarity-related) feature which can act as a scope marker and license not, while not is the true negation. In NE this ‘affective’ T, which we call T[þAff] from now on, has no specific realisation comparable to ne, but it does have the requirement that the feature bundle of which it is part must be overtly realised as some form of auxiliary. This analysis is supported by the ungrammaticality of constituent negation without do-support or some other auxiliary being present: (17)

*John always not smokes

In (18), we give some examples of possible feature compositions of T (prior to Agree), and the various auxiliaries that realise them (once the relevant unvalued features have been valued): (18)

a. [uV, uj, iTense, iAsp, (Aff)]: have, has, had, be, is, am, are, was b. [uV, uj, iTense, iM, (Aff)]: can, will, must, might, etc. c. [uV, uj, iTense, iAsp, Neg, (Aff)]: haven’t, hasn’t, hadn’t, isn’t, aren’t, wasn’t d. [uV, uj, iTense, iM, Neg, (Aff)]: won’t, can’t, shan’t, etc. e. [uV, uj, iTense, Neg, (Aff)]: don’t, doesn’t, didn’t f. [uV, uj, iTense, Aff]: do, does, did g. [uV, uj, iTense]: 

In (18a–d) we are using the features ‘Asp’ and ‘M’ as covers for a more finegrained analysis of aspectual and modal features, a matter which we leave aside here (but cf. the brief discussion of Perfect and Progressive features above). (17) is ruled out, assuming that T must have an (activated) affective feature here in order to license not; an overt auxiliary must therefore be inserted. Furthermore, the [þAff]-feature can be optionally activated in all

276

Theresa Biberauer and Ian Roberts

cases in (18a–e), whereas positive forms of do ((18f)) are obligatorily specified as [þAff]. As briefly described earlier, aspectual auxiliaries Agree with the first verb or auxiliary in their complement. Similarly, modals and do, both in their positive and negative forms, Agree with a bare-stem form of the verb. In (18g), we have posited a non-overt ‘do’ (see Emonds 1976; Pollock 1989), which we represent as . This element selects a V bearing Tense- and j-features in addition to its thematic content and which has these features valued under Agree with T and consequently inflects for them.7  therefore appears where we have both the most impoverished feature content of T, as (18) clearly shows, and it also imposes the weakest requirement on the form of the head of its complement. There are four further contexts where do obligatorily appears: interrogatives, emphasis, ellipsis and VP-fronting. (19)

a. b. c. d.

Does John smoke? John does (so/too) smoke John smokes Gauloises and Sam does __ too He threatened to smoke Gauloises and [smoke Gauloises] he did __

For the interrogative case (19a), we follow the standard analysis, which has its roots in den Besten (1983), in assuming that T moves to C. The movement is triggered by the presence of an optional (tense-related) move diacritic (perhaps a kind of EPP-feature) on C. Because of its optional status, this feature must trigger an interpretive effect; cf. Chomsky’s (2001: 34) remark that “[o]ptional operations can apply only if they have an effect on outcome.” Here the ‘new interpretation’ which arises from movement is interrogativity, a canonical property of C. C is always overtly realised (via do-support in the absence of a more elaborately specified T; cf. (18)) owing to the presence of the diacritic on C; the maximally underspecified auxiliary, namely (18g), is not an option owing to the fact that the outcome of movement must be visible – hence the moved element must be lexically realised – in order for the interpretive effect to hold. In (19b), T is obligatorily realised as do because [þAff] is part of T’s feature bundle here, giving rise in this case to what we loosely call the

7

English has another null T, the ‘subjunctive’, as in: (i)

I require that he T not be disturbed until I get back.

(See Emonds (1976) on the ‘null-modal’ analysis of such examples.) This T certainly has some further mood feature and, like a modal, selects/Agrees with a bare verb. See Roberts (1993b: 324) for discussion of the apparent lack of otherwise obligatory inversion in: (ii)

I require that under no circumstances he be disturbed.

Subjects, Tense and verb-movement

277

‘emphatic’ interpretation. Thus one of the forms in (18f) is required; (18g) cannot be selected as a potential spell-out form since it is a less specific form (cf. Panini’s Principle). We propose essentially the same for (19c) and (19d). Here again, the null form is barred owing to the presence of the optional [þAff]-feature on T. The auxiliary in VP-ellipsis and VP-fronting contexts is emphatic (i.e. underscores positive polarity), and contracted auxiliaries cannot be emphatic (cf. the fact that emphatic do [did] can never be reduced to ’d), hence the ungrammaticality of (20): (20)

a. *John has smoked Gauloises and Fred’s __ too b. *He said he had smoked Gauloises and [smoked Gauloises] he’d __

So we see that where [þAff] is part of the feature bundle on T, the auxiliary must be spelled out as a lexically full form.8 VP-ellipsis and VP-fronting pattern alike as far as the auxiliaries are concerned, although it is well known that they differ, particularly in that VP-ellipsis can affect any constituent part of a string of auxiliaries headed by a non-finite auxiliary or main verb, while VP-fronting may only affect the lowest VP, as long as it is not passive (Ross 1969; Akmajian, Steele & Wasow 1979; Baker, Johnson & Roberts 1989): (21)

a. We thought the students might have been being arrested and . . . [being arrested] they might have been __ *[arrested] they might have been being __ *[been being arrested] they might have __ *[have been being arrested] they might __ b. The students might have been being arrested, and the workers definitely could ((have (been (being arrested))).

This concludes our analysis of ‘do-support’. The final question we address here brings v fully into the picture. Up to now we have been assuming for simplicity that all auxiliaries are merged in T. We have already mentioned that sequences of auxiliaries involve v; see (7) and (9). But, if the second auxiliary is merged in v when more than one is

8

There are (northern) varieties of British English in which an auxiliary can contract to the subject with negation non-contracted, as in (i): (i)

I’ve not finished yet

In Standard English, this is not possible, the relevant form being (ii): (ii)

I haven’t finished yet

We observe that the relation of the [þAff]-feature to auxiliary contraction is different in these varieties. This is probably linked to the fact that negative auxiliaries of the haven’t, didn’t type are not really found in the relevant varieties. We might therefore conjecture that Negation is always realised on a head distinct from T in these varieties.

278

Theresa Biberauer and Ian Roberts

present, what about the single auxiliary when only one is present? We consider it very plausible that aspectuals and, possibly, root modals are always merged in v. They Agree with V and (in finite contexts) move to T, via reprojection as described for V-movement in Romance in Section 3.9 The Agree relations we discussed in connection with (6) are now more properly seen as T-v-V relations. A v which has M or Asp features reprojects higher up in the clausal domain, plausibly also as a species of T. Do, have and epistemic modals, by contrast, are merged in T in NE. A consequence of all this is that ‘VP-fronting’ is really vP-fronting. Evidence for this comes from the anti-reconstruction effects discussed by Huang (1993). Assuming, as is standard, that the subject is merged in a specifier of vP, then the fact that a trace of the subject appears to bind an anaphor under vP-fronting can be accounted for:10 (22)

a. Which pictures of himself does John think Bill will like best? b. Vote for himself, John thinks Bill might

The general conclusion of this section is that in NE T is always realised by an auxiliary, either null or overt, subject to the conditions given in (18). We have seen how this approach gives rise to a general, and rather novel, account of ‘do-support’ and we have sketched an outline analysis of the rest of the auxiliary system, as well as of negation. The NE system therefore represents a third option in relation to Romance and Germanic as described in the previous section. In this system, T’s uninterpretable V-feature is licensed by the realisation of a verbal element (i.e. an auxiliary) in T, either as a consequence of merger or movement. 4.2

Middle English

One language that appears problematic for the proposals we have made here is Middle English (ME). This is because it is usually thought that ME had general V-to-T in finite clauses, this option having been lost in Early Modern English (ENE) (cf. Roberts 1985; 1993b; 1999; Pollock 1989). (23)

9

10

a. My wyfe rose nott (Mosse´ 1968, cited in Roberts 1985: 23) b. Se ye not how his herte is endured . . . ? see you not how his heart is hardened . . . ? (Anon. The Examination of Master William Thorpe, 44, cited in Roberts 1993b: 239)

In fact, at least aspectuals plausibly select another vP. See Biberauer & D’Alessandro (2006) and D’Alessandro & Roberts (2006) on this. In a way, this is the obvious updating of the approach to auxiliaries in Ross (1969). Heycock (1995) points out certain empirical difficulties for Huang’s approach and proposes an alternative analysis.

Subjects, Tense and verb-movement

279

ME was, however, like NE in distinguishing only two synthetic tenses, present and past.11 Therefore, according to our proposals, V-to-T should not have existed. One possibility would be to argue that ME was like the other Germanic languages we have discussed in having V2, but not V-to-T. However, whilst this is probably correct for Old English (OE) and Early ME, there are two reasons to think that this is not the case for Late ME. One is that V2 was lost in the Late ME period; this is in fact usually thought to have happened in the fifteenth century (van Kemenade 1987: 219ff.; Fischer et al. 2000: 133ff.). Another is that we find subordinate clauses which are known to resist embedded V2 with the order Subject-Finite Verb-not, such as the following: (24)

And gif he be noght so, then . . . and if he be not so, then . . . (1420s: James I, Kingis Quair, 62 – Gray 1985: 73, cited in Roberts 1993b: 323)

If V2 is not possible in clauses introduced by a complementiser (gif here), and if the relative positions of the finite verb and noght are taken as a diagnostic for V-movement to a position external to vP/VP (see Emonds 1978; Pollock 1989), then V must have moved to T to create the order seen here. Similarly, V3 orders featuring a lexical verb preceding a floated quantifier signal the same movement operation: (25)

a. In doleful wise they ended both their days (1589: Marlowe, The Jew of Malta, III, iii, 21, cited in Roberts 1993b: 253) b. for the panes and pottes garnisseth wel the ketchyn . . . (1531: Elyot, The Boke Named the Gouernour, cited in Roberts 1993b: 258)

We therefore conclude that Late ME, at least, had V-to-T movement independently of V2. So how should we analyse this in relation to the proposals made above? The answer, we suggest, lies in a closer consideration of the diachronic facts. As just mentioned, OE fits fairly straightforwardly with our proposals. We analyse this stage of English as being, in the relevant respects, exactly like Modern German and Dutch. Thus we take it that OE lacked V-to-T 11

The subjunctive was arguably still available in Late ME and ENE, as seen in (24). However, the present indicative and present subjunctive were barely distinct (only in the 2SG and 3SG forms, in fact), and so we feel justified in collapsing these as we did in the case of both French and German. The past subjunctive may have still been productive, but was morphologically distinct from the indicative only in the singular of be, which was were. For this reason, we leave this form aside here.

280

Theresa Biberauer and Ian Roberts

(see Biberauer & Roberts (2005) for a detailed analysis of OE word order, and see again Vikner (2005) on German and Dutch). By Late ME, as we have just seen, English had V-to-T. So how did this operation emerge historically? We propose that V-to-T emerged with the reanalysis of subject-initial V2 orders. The reanalysis is schematised in (26) (see also Adams 1987b; Roberts 1993b on Old French; Willis 1998 on Middle Welsh): (26)

[CP DP [C V] [TP tDP [T tV] [vP tDP [v tV] VP]]] ! [TP DP [TV] [vP tDP [v tV] VP]]

So we propose that V-to-T emerged as a consequence of the loss of V2, through the reanalysis of subject-initial V2 clauses as shown in (26). The result of this reanalysis was a system with V-to-T movement and relatively ‘impoverished’ tense inflection. As we have seen, French has between four and six synthetic tense forms (see (3)), while Late ME had only two (see footnote 11). We are forced to conclude that in the period following the reanalysis in (26), English verbs, despite their relatively impoverished tense inflection, were analysed as compound VþT elements in the manner described in Section 3. The consequence of this is that very frequently the postulated Tense element had a null realisation, and it always had an impoverished feature content (marking only past vs. present; compare the Romance languages where, in addition to past vs. present, aspectual distinctions are made among past tenses). The evidence for the compound VþT element in Late ME/ENE was therefore clearly rather slight. We suggest that this contributed to the reanalysis of the modals and do as auxiliaries, which took place in the early sixteenth century (it is usually said to have taken place by 1530: see Lightfoot 1979; Roberts 1985; 1993b; Warner 1997). Following Roberts & Roussou (2003: 41–42) and Biberauer & Roberts (2005; 2006), we take the reanalysis of the modals and do to have involved the reduction of a formerly biclausal structure in which the modal/do acted as a restructuring verb with a defective (TP) complement to a monoclausal structure in which the modal/do is merged outside of VP (here we have indicated the merge position of the new auxiliary as T, but see below for a refinement): (27)

[TP Subj [T0 T [VP VR (¼modal) [TP tSubj [T0 [T Vþv T] [vP tSubj [v tV] VP]]]]]] ! [TP Subj [T0 [T modal] [vP tSubj v VP]]]

Roberts & Roussou (2003: 41–42) suggest that the loss of infinitive morphology around 1500 meant there was no trigger for V-movement in the lower clause, or indeed for the lower functional T-v system; hence the reanalysis in (27) in the early sixteenth century.

Subjects, Tense and verb-movement

281

In terms of the idea put forward at the end of the previous section that root modals and aspectual auxiliaries are merged in v and raise to T in NE, we could reconstrue (27) as reanalysing some auxiliaries, in particular root modals, as merged in the matrix v (aspectual auxiliaries may have been matrix v-elements prior to the reanalysis). After the reanalysis in (27), English therefore had a class of auxiliaries, in that it had a class of elements, including both root and epistemic modals, aspectuals and do, which were merged in either v or T and which ultimately had to surface in T. This reanalysis contributed to the conditions for the loss of V-to-T movement later in ENE (from around 1575 onwards; see Kroch 1989; Roberts 1993b). However, V-to-T movement still applied to main verbs in clauses where no auxiliary was present for at least this period (and it has been suggested that in fact it survived through the seventeenth century; see Warner 1997). Roberts (1993b: 249), following Jespersen (1909–49, VI: 502), suggests that in the seventeenth century this was “conditioned by extragrammatical factors”, in that V-to-T movement was used by certain authors, e.g. Milton, as a conscious index of ‘high’ style, probably in imitation of Shakespeare and the 1611 King James Bible. It is also worth emphasising here how different the sixteenth-century English situation was from the Modern French one. One important factor concerns the apparent general availability of do after the reanalysis in (27); do could seemingly appear in any context, except where a modal or aspectual was present. Thus, do was always available as an alternative to verbmovement. It is unclear, in fact, why do did not emerge as the general default lexicalisation of T in all environments, giving rise to a situation in NE in which all tenses are periphrastic and there is never any form of tense or agreement marking on the verb (some south-western dialects of English may instantiate this system; see Roberts 1993b: 307). A second factor was the availability of Subject-Adverb-Verb orders, something which is impossible in French (see Pollock 1989). Such orders are attested in ME (see Kroch 1989: 226; Roberts 1993b: 254), where they have often been analysed as instances of Icelandic-style stylistic fronting (see for example Platzack 1995).12 By the middle of the sixteenth century, when stylistic fronting-type structures are no longer attested (cf. Biberauer & Roberts 2006; Biberauer, Chapter 4), Adv-V order is quite common: Kroch (1989: 226) quotes the following figures from Ellega˚rd’s (1953: 184) study of the relative order of never and a main verb:

12

Biberauer & Roberts (2006; 2008) present an alternative analysis, which they argue to be empirically superior to the stylistic-fronting one.

282

Theresa Biberauer and Ian Roberts

(28)

Period Period Period Period

3 (1475–1500) 4 (1500–1525) 5 (1525–1535) 6 (1535–1550)

80 never-V 80 never-V 151 never-V 125 never-V

102 V-never 28 V-never 16 V-never 13 V-never

We observe a dramatic decline in the relative occurrence of V-never order after 1500. If we follow Kroch in taking V-never order to be an indication of V-to-T movement (i.e. we take never to be adjoined to vP), then this indicates a decline in V-to-T movement during the early sixteenth century. This is consistent with the predictions made here, and again indicates a major difference as compared with Modern French (where the order jamais – finite verb is completely impossible).13 We have seen how do became an auxiliary in the 1520s owing to the reanalysis in (27). It is important to keep in mind that this gave rise to the optional, ‘exuberant’ do, found also in positive declaratives in the sixteenth century. The following are examples of this kind of do from Shakespeare (metrical stress marked): (29)

a. Where e´yes did o´nce inha´bit (from Richard III, cited in Barber 1976: 164) b. Rough wı´ndes do sha´ke the da´rling bu´ds of Ma´ie (from Sonnet 18)

But the modern do-support system does not emerge until later. In fact, it does not appear until after the development of contracted negation. This last development took place around 1600, according to Jespersen, who says: The contracted forms seem to have come into use in speech, though not yet in writing, about the year 1600. In a few instances (extremely few) they may be inferred from the metre in Sh[akespeare], though the full form is written. (Jespersen 1909–49, V: 429, cited in Roberts 1993b: 305).

In the late sixteenth and early seventeenth centuries, do-insertion is not connected to negation. In negative clauses, V-to-T and do-insertion are apparently options. It is also possible for neither of these operations to take place in a negative clause: (30)

13

a. Or if there were, it not belongs to you. (1600: Shakespeare 2 Henry IV, IV, i, 98, cited in Battistella & Lobeck 1988: 33) b. Safe on this ground we not fear today to tempt your laughter by our rustic play. (1637: Jonson Sad Shepherd, Prologue 37, cited in Kroch 1989: 235)

Strikingly, however, French resembles English in no longer permitting the stylistic-fronting operation that was possible at an earlier stage. Clearly then, stylistic-fronting cannot be connected to V-to-T movement, as has often been assumed (cf. Mathieu (2006) for discussion and also Biberauer (Chapter 4), and Holmberg (Chapter 2) for consideration of the parametric cluster that stylistic fronting appears to be associated with).

Subjects, Tense and verb-movement

283

Probably by the middle of the seventeenth century (Roberts 1993b: 308), sequences of auxiliaryþcontracted negation are reanalysed as negative auxiliaries of the NE kind (see the discussion in the previous section). Once the negative auxiliaries, crucially including doesn’t, don’t, didn’t, are established as the unmarked expression of clausal negation, the modern system of dosupport comes into being. The above accounts for the development of dosupport in negative contexts, but what about the other NE contexts discussed in the previous section? We assume that do-support became obligatory in questions where no other auxiliary is present as a consequence of the loss of V-to-T movement, making V-movement to C impossible, the existence of a dummy auxiliary and the continued lexicalisation requirement associated with a movement-triggering C. In such a system, the only option for lexicalising C involves the dummy auxiliary. In terms of the idea that the reanalysis in (27) gave rise to a subclass of auxiliaries that was merged in v, we can maintain that, although V-to-v-to-T was lost by the end of the sixteenth century, v-to-T remained. We can thus think of the development of do-support in the seventeenth-century as a shift from the earlier obligatory v-to-T movement (first fed by V-to-v movement, and as such moving a main verb to T, but later only moving an auxiliary merged in v) to optional v-to-T movement creating a discourse effect (cf. Chomsky 2001: 34). The difference between the two systems concerns the status of phonologically empty v, which in the earlier grammar, until the seventeenth century, moved to T (i.e. in examples like (30)). In the later grammar, only v containing an auxiliary moved to T. This is a natural simplification of the grammar, given that movement of empty v to T could never be directly observed in the input available to acquirers. In order to summarise our proposals regarding the development of the auxiliary system, let us reconsider the list of features of the NE auxiliaries given in (18) above: (18)

a. [uV, uj, iTense, iAsp, (Aff)]: have, has, had, be, is, am, are, was b. [uV, uj, iTense, iM, (Aff)]: can, will, must, might, etc. c. [uV, uj, iTense, iAsp, Neg, (Aff)]: haven’t, hasn’t, hadn’t, isn’t, aren’t, wasn’t d. [uV, uj, iTense, iM, Neg, (Aff)]: won’t, can’t, shan’t, etc. e. [uV, uj, iTense, Neg, (Aff)]: don’t, doesn’t, didn’t f. [uV, uj, iTense, Aff]: do, does, did g. [uV, uj, iTense]: 

Prior to the reanalysis in (27) only the aspectual auxiliaries in (18a) existed. The reanalysis in (27) created the modals and do, as in (18b) and (18f), although they, like the aspectuals, lacked the modern [þAff]-feature. The post-1600 reanalysis of auxiliaryþn’t as negative auxiliaries created the

284

Theresa Biberauer and Ian Roberts

classes in (18c–e). Finally, the loss of V-to-T in the latter part of the sixteenth century gave rise to a null (v-)auxiliary of the kind represented in (18g); in the seventeenth century, when the modern system of do-support emerged, the [þAff]-feature was added to (18a–f), and the null auxiliary was reanalysed as a (discourse-neutral) T-element, where it had previously been a v-element. 4.3

Summary and an interim typology

In the preceding section we sketched an account of the development of the NE auxiliary system, building on and slightly refining the proposals in Biberauer & Roberts (2008). The main conclusion in the present context is that we are able to integrate the various stages of English into the system for the realisation of T-features which we exploited in Section 3 to account for what seems to be a general difference between Germanic and Romance. Following Ouali’s (2008) terminology of KEEP (phase-head retains uninterpretable features), SHARE (uninterpretable features are copied to a phase-internal non-phase-head) and DONATE (uninterpretable features are given to a phase-internal non-phase-head) and focusing on the functional head that we are concerned with here (T), we can phrase the typology as follows: (31)

(i) C DONATES j- and T-features to T: Romance, English (ii) C KEEPS j- and T-features: Continental Germanic14

(31) is a typology of functional elements in the C-phase. The differences between English and Romance emerge in the v-phase, since, as we have seen, English lacks V-to-v movement while the Romance languages have this. English in other words represents an ‘analytic’ version of Romance (see Huang (2007) and the discussion of analyticity in the Introduction). This is in turn connected to the fact that Romance verbs are associated with T-features which are capable of reprojecting, and English main verbs are not (although auxiliaries are, and can be merged in v and raised to T). Options of the kind seen in (31) extend to movement-triggering features such as the EPP-features assumed in current minimalism too. In the Mainland Scandinavian languages, the EPP-feature with which C is endowed as a parametric option (see Biberauer, Chapter 4) is SHAREd with T, giving rise to the combination of V2 and an English-style expletive/subject requirement in SpecTP (cf. again Biberauer, Chapter 4). In English and French, this feature 14

In the case of the West Germanic varieties permitting so-called complementiser agreement (cf. West Flemish and Bavarian German, for example), C may in fact KEEP its T-features, but SHARE its j-features. If this is the correct characterisation of the varieties in question, we see that it is possible for different options to apply to the various sub-features of a given phase-head (cf. also the discussion of Mainland Scandinavian and Celtic in the main text, and Ouali (2008) on Berber).

Subjects, Tense and verb-movement

285

is DONATEd, hence V2 is absent (in declarative C), but there is a subject requirement associated with SpecTP. On German and Icelandic, see Section 5.2 below. Importantly, consistent null-subject languages fit into this typology too; Italian, for instance, has the same property as English and French, with the subject requirement being met by a deleted pronoun in null-subject structures (see Roberts, Chapter 1; Sheehan, Chapter 6). From the perspective of EPP-feature distribution, then, we obtain the following (interim) partial typology: (32)

(i) C SHAREs its EPP-feature: Mainland Scandinavian (ii) C DONATEs its EPP-feature: English, Romance

Taking into account the T- and j-distribution typology illustrated in (31), we therefore see that the English and Romance Cs are formally identical in respect of the formal feature-retention vs. sharing options that they employ: both involve Cs which DONATE their T-, j- and EPP-features. Mainland Scandinavian, by contrast, differs from these languages not only in opting for different feature-distribution choices (KEEP in the case of T- and j-features and SHARE for C’s EPP-feature instead of the consistent DONATE option employed by English and Romance), but also in that it does not choose the same distribution option for all the features concerned. We will discuss what appear to be further examples of this parametrically available option in Section 5 below. We now turn to extending the typologies we have considered thus far to a range of other languages, before attempting to unify them. 5

Extending the typology

Here we will look first at Icelandic; we will argue that this system instantiates the SHARE option as regards the relationship between the T- and j-features of C and T. Second, we look briefly at some of the Celtic languages; here we will see that DONATE and KEEP can be related to different (sub-)sets of C’s features (cf. also footnote 14). Finally, we consider the case of non-Celtic V-initial languages. 5.1

Icelandic

Given what we have seen so far, we might expect that Icelandic, as a V2 language, would instantiate the KEEP option. The KEEP option, as we have seen, is incompatible with V-to-T movement. However, it is generally assumed that Icelandic has V-to-T in embedded wh-clauses, relative clauses, some adverbial clauses and also in certain non-finite clauses, at least as an option (cf. Thra´insson 2003; 2007: 62–63; Johnson & Vikner 1994):

286

Theresa Biberauer and Ian Roberts

(33)

a. E´g spurði af hverju Helgi hefði ekki (*hefði) lesið þessa bo´k I asked why Helgi had not had read this book ‘I asked why Helgi had not read this book’ (Vikner 1995: 145 (31c)) b. . . . hið eina sem e´g hlaut aldrei (*hlaut) var . . . . . . the only that I got never got was . . . ‘. . . the only thing that I never got was . . .’ (Thra´insson 2003: 182 (54a)) c. . . . af þvı´ að hann vill ekki (*vill) fylgja honum . . . because that he wants not wants follow him ‘. . . because he doesn’t want to follow him’ (Thra´insson 2003: 157 (12a)) d. . . . ef við lju´kum ekki verkinu ´ı dag . . . if we finish not job-the today ‘. . . if we don’t finish the job today’ (Thra´insson 2003: 182 (55a)) e. Að lesa ekki (*lesa) Chomsky varð henni að falli to read not read Chomsky became her-dat to fall ‘Not reading Chomsky caused her downfall.’ (Thra´insson 2007: 417 (8.64b))

The question, then, is why does Icelandic allow V-to-T? This question becomes more acute when we consider the nature of tense inflection in Icelandic, since it only has four synthetic tenses (present and past subjunctive and indicative), just like German (see Thra´insson 1994: 159; 2007: 395). One possibility would, of course, be to ascribe V-movement to Icelandic’s oftdiscussed rich agreement. This rich agreement inflection can be observed in paradigms such as the following: (34)

e´g kasta við ko¨stum (“I throw”, etc.) þu kastar þið kastið hann kastar þeir kasta

In Section 3 above, we, however, proposed that rich agreement should not be viewed as a possible trigger for V-movement, rich agreement being the source of null subjects, with rich tense underlying V-movement. Specifically, recall (5d), where we suggested that the combination of rich agreement and poor tense would deliver null subjects in the absence of V-to-T movement, a state of affairs which is not, to our knowledge, attested. Notice, however, that Icelandic can, in our terms, be described as a rich agreement and poor tense language (as pointed out to us by M. Richards, p.c.). The question, then, is why this language should exhibit the very opposite properties – lack of null subjects and V-to-T movement – to those predicted in (5d)? Three possibilities suggest themselves: first, that our proposal that V-movement is necessarily driven by tense richness is in fact incorrect and that agreement can also trigger reprojection of the type we have described for rich tense; second, that Icelandic tense is in fact rich enough to trigger reprojection and that Icelandic is thus a Germanic language which is formally

Subjects, Tense and verb-movement

287

identical to the Romance languages on the V-movement front; and finally, that V-to-T movement may in fact be the consequence of vP-remnant movement to SpecTP (cf. Bentzen 2007 and also Biberauer, Chapter 4).15,16 Given the, in our view, very positive consequences that appear to follow from distinguishing the triggers for V-to-T movement and null-subject-licensing, we reject the first possibility. In this connection, it is also worth noting the empirical fact that Yiddish, which also has V-movement in ‘non-V2’ contexts (cf. Vikner 2001), distinguishes four different person–number forms, just like German, which lacks V-to-T (cf. once again Vikner (2001; 2005), who also argues convincingly that Yiddish is an OV language, thereby removing the possibility that the verb-movement difference simply correlates with the OV/VO distinction). Agreement richness therefore does not seem like a feasible V-to-T trigger, independently of our theoretical concerns here. The second possibility may be more promising, however, as it is noteworthy that Icelandic differs starkly from German in respect of the way in

15

We leave aside here the possibility that the clauses in question are in fact V-in-C V2 clauses, involving CP-recursion of the type that is generally assumed for embedded declaratives in Icelandic (cf. Schwartz & Vikner (1996) for discussion; Vikner & Johnson (1994) suggest a CP-recursion analysis for the infinitival structures featuring V-movement, but cf. Bobaljik & Thra´insson (1998) and Thra´insson (2003) for critical discussion of this idea). This may, however, be the correct analysis for at least some of the structures in question here. Consider, for example, the fact that embedded wh-interrogatives are compatible with overt expletives, despite the fact that these expletives are C-expletives which cannot therefore surface in SpecTP (cf. i.a. Holmberg & Platzack 1995; Richards & Biberauer 2005; Biberauer, Chapter 4; the same is true for Yiddish – cf. Vikner (1995: 87–77) for discussion of both languages). As the examples below show, overt expletives are barred from the corresponding matrix structures, a state of affairs which is accounted for if we analyse the wh-element in (i) as occupying the specifier of the higher CP, with the expletive in the specifier of the lower C. In matrix V2 clauses, only a single CP is present, with the result that these structures cannot accommodate two SpecCP elements (the wh-element and það): (i)

(ii)

16

Jo´n vissi ekki hvernig það hefðu komist svona margir ´ı mark John knew not how there had come so many in goal ‘John didn’t know how so many goals had been scored.’ (cf. Vikner 1995: 86 (32)) Hvernig (*það) hefðu komist svona margir ´ı mark how there have come so many in goal

Given the above, we leave open the possibility that embedded wh-interrogatives in Icelandic and Yiddish may in fact involve V-in-C V2. We also leave aside the possibility that C in the relevant clausal contexts distributes its features differently to V2 C (all the structures in which V-to-T movement occurs represent clear non-V2 contexts in other V2 languages, and may therefore involve different Cs). Although theoretically possible and empirically attested in other contexts – consider, for example, the different verb-attraction properties of declarative and interrogative matrix C in English – it is not clear to us that this possibility would deliver any genuine insight into Icelandic’s verb-movement patterns: we could postulate that non-V2 C SHAREs all its features, rather than KEEPing T as V2 C does, but this would simply raise the question why non-V2 Cs in other V2 languages cannot employ SHAREing of this kind.

288

Theresa Biberauer and Ian Roberts

which it employs mood, specifying the use of indicative vs. subjunctive forms in non-matrix clauses in a manner that German does not (cf. Thra´insson (2007: chapter 8) for discussion, and recall that we are using ‘tense’ here as a cover-term for tense, mood and aspect). If the fact that Icelandic matrix verbs and complementisers introducing clausal complements select for CPs with specific mood properties requires mood features to be saliently represented as part of an independently stored AgrþT bundle of the sort that we have proposed will reproject, it may be the case that Icelandic consistently exhibits V-to-T movement for the same reason that Romance does. We will leave open this possibility here. It is worth noting, however, that the same account seems more problematic for Yiddish, which is undoubtedly tensepoor, although it has developed an aspectual system absent in other Germanic languages (cf. Gold 1999). The final possibility also seems feasible, particularly given recent work on Northern Norwegian dialects, which suggests that what appears to be V-to-T movement is also a possibility in certain inflectionally impoverished Norwegian dialects (cf. Bentzen (2007) for references). We leave aside the technical details of this type of proposal (see Bentzen (2007) and Biberauer (Chapter 4) for some discussion), but note only that the approach we are proposing here is compatible with the possibility that what appears to be V-to-T movement in Icelandic (and possibly also Yiddish) is in fact remnant vP-raising to SpecTP. As will become clearer in our discussion of tense-poor VP-raising languages in Section 5.3, we predict V-movement via VP-raising to be possible in precisely those languages which lack rich tense agreement. If the speculations in the preceding paragraph ultimately prove to be on the wrong track and Icelandic tense does indeed turn out to be poor (in our terms), the appearance of V-raising may thus be the consequence of VP-movement. Leaving unresolved the question of why Icelandic exhibits what appears to be V-to-T movement, we turn to the much-discussed issue of why it lacks referential null subjects. This question becomes particularly acute if the tenserich analysis of Icelandic postulated above, in terms of which it is formally identical to the Romance null-subject languages, is correct. We propose that the difference between Icelandic and Romance lies in the nature of the movement trigger (EPP-feature) on T. Recall from the previous section that we take such features to be inherently associated with phaseheads, with inheritance by T subject to parametric variation following Ouali’s options KEEP, SHARE and DONATE. There we also pointed out that consistent null-subject languages are like English and French in inheriting both j-features and the EPP-feature from C. In other words, the SpecTP position has the same abstract properties in consistent null-subject languages as in English and French (on this, see Holmberg 2005; Roberts, Chapter 1; Sheehan,

Subjects, Tense and verb-movement

289

Chapter 6). This is exactly where the difference lies between Icelandic and consistent null-subject languages. We propose that T has an optional EPPfeature (see also Biberauer & Richards 2006; Biberauer, Chapter 4). We once again follow the previously mentioned Fox–Reinhart intuition regarding optionality, which we, following Chomsky (2001: 34), state as (35): (35)

“Optional operations can apply only if they have an effect on outcome.”

The principal consequence is that D-attraction to SpecTP in Icelandic, unlike in English or Romance, has a discourse effect (as has also been observed by Jonas 1996; Bobaljik & Jonas 1996; Bobaljik & Thra´insson 1998; Thra´insson 2003; and Wurmbrand 2006; see Biberauer, Chapter 4, for detailed discussion). There are various indications of this in Icelandic. First, definite/specific and indefinite/non-specific subjects occupy different positions, with the former raising higher than the latter, as seen in (36): (36)

a. I´ gær kla´raði mu´sin sennilega ostinn yesterday finished mouse-the probably cheese-the ‘Yesterday the mouse probably finished the cheese.’ b. *I´ gær kla´raði mu´s sennilega ostinn yesterday finished mouse probably cheese-the (Jonas 1996: 112 (47))

Second, subject DPs that undergo raising to SpecTP are interpreted as ‘old’ information. In (37a), in the context given, the definite subject margar af þessum bo´kum (“many of these books”) is raised to SpecTP and is fully grammatical and appropriate. In (37b), on the other hand, this DP arguably remains in its vP-internal first-merged position; as such, it must be read as new information, which is infelicitous in this context: (37)

Við settum allar bækurnar fyrir ma´lstofuna ´ı lestrarsalinn . . . we put all books-the for seminar-the in reading room-the (a) . . . en það hafa margar af þessum bo´kum verið lesnar a´ður . . . but there have many of these books been read before ‘. . . but many of these books have been read before’ (b) *. . . en það hafa verið lesnar margar af þessum bo´kum a´ður . . . but there have been read many of these books before (Jonas 1996: 63 (27))

Similar effects have been observed in German (see Diesing 1992; Rosengren 2002; Biberauer 2004; Chapter 4; cf. also the more general ‘defocusing’ effect of associated with the optional movement operation labelled scrambling in German).

290

Theresa Biberauer and Ian Roberts

Third, the phenomenon of ‘null expletives’ shows that SpecTP is not filled as a result of an obligatory EPP-feature (in contrast with English and the Mainland Scandinavian languages see Biberauer, Chapter 4): (38)

a. I gær hefur komið stra´kur yesterday has come boy ‘Yesterday a boy came.’ b. I gær hefur verið dansað yesterday has been danced ‘Yesterday there was dancing.’

Fourth, stylistic fronting shows that the category which fills SpecTP does not have to be a DP subject (cf. Holmberg & Platzack 1995; Holmberg 2000; 2006): (39)

Fundurinn, sem fram hafði farið ´ı Oslo, var skemmtilegur meeting-the that on had gone in Oslo was fun ‘The meeting that had gone on in Oslo was fun.’

As argued in Biberauer (Chapter 4), the above facts establish that SpecTP is filled just where T bears an optional EPP-feature, which, in line with the Fox– Reinhart intuition in (33), has the consequence that moving a DP (or other category) there must have a discourse effect. We suggest that the deletion operation which derives consistent null subjects is incompatible with the obligatory discourse effect. This is supported by the fact that, in consistent null-subject languages, focused subjects are overtly realised, as in the following examples repeated from the introduction: (40)

a. Lui parla italiano. HE speaks Italian. b. E´l habla espan˜ol. HE speaks Spanish. c. Aftos mila ellinika. HE speaks Greek.

(Italian) (Spanish) (Greek)

Note that this account gives us an explanation for why neither Icelandic nor German are consistent null-subject languages, despite their relatively ‘rich’ agreement inflection: T in both of these languages is only optionally associated with an EPP-feature, with the result that the element that raises to SpecTP must be overtly realised (see Biberauer (Chapter 4) for further discussion; see Mu¨ller (2005) for a different account for German). Some support for our approach comes from diachronic considerations, particularly from the history of French (see Dupuis (1989) and Mathieu (2006) on stylistic fronting in Old French). Old French also had optional

Subjects, Tense and verb-movement

291

expletive pronouns in subordinate clauses and showed some variation in the position of indefinite subjects (see Vance 1997). However, V-to-T in French survived the changes in the nature of the subject position (which probably took place in the sixteenth and seventeenth centuries – see once again Vance (1997)), owing to the richness of the tense system, which has remained constant throughout the history of the language. So we see that the diachronic evidence supports our analysis of the trigger for V-to-T movement in Modern French, and also our central contention that the tense (V-) and agreement (D-) features of T play distinct parametric roles. 5.2

Celtic

In this section we briefly extend the typology to some of the Celtic languages. We will argue that in Irish C SHAREs T-features in subordinate clauses, while in Welsh certain T-features are KEPT on C, while others are DONATEd to T. As with parametric options in general (see the Introduction), we observe that different (sub-) features may act similarly or differently in relation to the formal options made available by the system. Irish has a class of tense-marked complementisers, illustrated in (41): (41)

Present declarative C: go Past declarative C: gur Present negative C: nı´ Past negative C: nı´or Present interrogative C: an Past interrogative C: ar

(Cottell 1995; McCloskey 2001: 77)

As has often been observed, these complementisers realise T-features on C. Tense is also realised on V, which moves to T as in Romance (McCloskey 1996): (42)

Deir siad gur ghoid na sı´ogaı´ ´ı say they C[þPast] stole the fairies her ‘They say that the fairies stole her away.’

(McCloskey 2001: 67)

So here we see SHARE of the T[þPast] feature between C and T, resulting in C being realised as the past-tense complementiser gur and V-to-T movement of ghoid. In Welsh, the situation is a little more complex. In the contemporary language, there are two tenses which are only realised on the auxiliary bod (‘be’). We see this in the fact that bod has distinct paradigms in the present and imperfect according to the clause type:

292 (43)

Theresa Biberauer and Ian Roberts Present tense: Affirmative: dw i Interrogative: ydw i Negative: (dy)dw i ti wyt i dwyt ti mae o/hi ydy o/hi dydy o/hi dan ni ydan ni (dy)dan ni dach chi ydach chi (dy)dach chi maen nhw ydyn nhw dydyn nhw ‘I am’ etc. ‘Am I?’ etc. ‘I’m not’ etc. Imperfect tense: Affirmative: roeddwn i Interrogative: oeddwn i Negative: doeddwn i roeddet ti oeddet ti doeddet ti roedd o/hi oedd o/hi doedd o/hi roedden ni oedden ni doedden ni roeddech chi oeddech chi doeddech chi roedden nhw oedden nhw doedden nhw ‘I was’ etc. ‘Was I?’ etc. ‘I wasn’t’ etc.

In terms of the general reprojection-based approach to head-movement outlined above, we treat bod as the single lexical item able to combine with present/imperfect features in the Numeration. Moreover, bod is the only Welsh verb able to raise to C. There are three pieces of evidence for the latter assertion. First, bod is in complementary distribution with many preverbal particles which are best treated as C-elements, e.g. the whparticle a/y:17 (44)

a. Y dynion sydd wedi gwerthu’r ci the men prt asp sell the dog ‘It’s the men who have sold the dog.’ b. Y dynion a werthodd y ci the men prt sold the dog ‘It’s the men who have sold the dog.’

If bod raises to C, we immediately understand why it is in complementary distribution with C-particles. Second, Rouveret (1996) argues that identificational clauses are V2 clauses in the sense that the precopular XP is in SpecCP and the copula in C. Only bod can appear in such clauses, again suggesting that it alone raises to C: (45)

17

a. Y brenin ydy Arthur the king is Arthur ‘Arthur is the king.’

Sydd in (44a) is a further suppletive form of bod which appears when the local subject has been extracted; see Hendrik (1996).

Subjects, Tense and verb-movement

293

b. Arthur ydy’r brenin. Arthur is the king ‘It is Arthur who is the king.’ c. *Ydy Arthur y brenin is Arthur the king

Third, exactly where the embedded clause contains one of the tenses of bod that is specific to this verb (the present or the imperfect), the non-finite form bod appears instead of the finite form: (46)

a. Mae’n deud bod nhw fan hyn is asp say bod they place this ‘He says that they’re here.’ b. Dywedodd hi bod y treˆn yn hwyr said she BOD the train is late ‘She said that the train was late.’

(King 1993: 304)

(King 1993: 304)

(See Tallerman (1998), Borsley, Tallerman & Willis (2007: 78ff.) for evidence that these clauses are finite.) Roberts (2005: 36) argues that the two observations that bod has tenses that no other verb has and that it has a different distribution from all other verbs can be accounted for by saying that the relevant features, the T-features present and imperfect, are located in C. Bod alone has the morphological capacity to realise them, but, in terms of our reprojection approach, in order to do this it must raise to the position that probes for these features. If we now say that the position in question in Welsh is C, then we explain the unique distribution of bod and also its position-dependent forms. In turn, if this is correct, then we have evidence for C KEEPing these particular T-features, but not others. Old Irish and Scots Gaelic are of some interest in the present context, since it seems that in these languages V-movement to C is morphologically marked. However, the root-embedded asymmetry persists, which implies that the inability of embedded C to host incorporation is due to the presence of embedded complementisers, along the lines of den Besten’s (1983) original insight. Carnie, Harley & Pyatt (2000) argue that Old Irish has a ‘filled-C’ requirement, although it is not verb second (see also Doherty (1996; 1997) Adger (2006b) and Newton (2006) on the Old Irish ‘verbal complex’; Doherty in fact argues that Old Irish is a residual V2 language). Old Irish, then, may be a case of a language with just one part of the verb-second constraint, the part that relates to C (Ferraresi 2005; Longobardi 1991 argue that Gothic also has a ‘filled-C’ requirement, but only in embedded clauses). The element in a matrix C is therefore in absolute first position. Various other kinds of material can satisfy this constraint in a root clause: ‘conjunct particles’ (negation,

294

Theresa Biberauer and Ian Roberts

question markers, complementisers), preverbs. Most interestingly, when the verb itself occupies C, it takes on a specific morphological form, traditionally known as the absolute form (the non-initial form is known as the conjunct form). This alternation is illustrated in (47): (47)

a. Beirid in fer in claideb carries (absolute) the man the sword ‘The man carries the sword.’ b. Nı´ beir /*beirid in fer in claideb. not carries (absolute)/(conjunct) the man the sword ‘The man doesn’t carry the sword.’ (Carnie, Harley & Pyatt 2000: 45)

As Carnie, Harley & Pyatt suggest, the alternation in form can be interpreted as a morphological reflex of the features that attract the verb to C. Clearly, it is extremely tempting to analyse this morphological distinction in terms of KEEP (absolute) vs DONATE (conjunct) of C’s T-features. Scots Gaelic retains the Old Irish distinction between absolute and conjunct verb forms. The distinction remains productive in the future tense, where we observe examples like the following (from Calder 1990: 223): (48)

a. o`laidh mi drink I b. o`l mi drink I

(absolute present/future tense) (conjunct present/future tense)

The conjunct forms occur when there is a particle in C (e.g. gun ‘that’). The independent forms appear when there is no particle (see Calder 1990: 202ff.). It may be more difficult to maintain that the verb is in C in (48a) and in T in (48b), but what could be maintained is that the conjunct tense appears just where C’s features are SHAREd with a complementiser, while the absolute form is reflex of KEEP where no complementiser is present. Here, as in Old Irish, the KEEP vs SHARE option gives rise to a morphological difference in the features in question. Modern Irish retains the same distinction, but just with a handful of verbs. The historically conjunct forms are found after a C-particle (except for the leniting particle aL – see McCloskey 2001); the historically absolute forms are found after aL and in absolute initial position: (49)

a. Creidim go bhfaca me´ do nighean I-believe that saw-dep i your daughter ‘I believe that I saw your daughter.’ b. an bean a chonaic tu´ the woman aL saw-indep you ‘the woman that saw you.’

Subjects, Tense and verb-movement

295

c. Chonaic tu´ ´ı saw-indep you her ‘You saw her.’

It is unlikely that these forms are synchronically associated with V-movement to C; I am aware of no independent evidence for this, and Doherty (2000: 28–31) argues that the correlation between independent morphology and initial position (i.e. C) was weakening as early as Classical Old Irish (ninth century AD). But we can maintain that the morphological difference reflects whether the features are DONATEd or SHAREd (note in this connection that aL is invariant). 5.3

Other V-initial languages

In this context, it is interesting to contrast the verb-movement in the Celtic VSO languages with VP-movement of the kind seen in various other V-initial languages, according to recent analyses (Massam & Smallwood (1997), Massam (2000; 2005), Rackowski & Travis (2000), and many of the papers in Carnie, Dooley & Harley (2005)). In her study of VOS and VSO in Niuean, for example, Massam (2000; 2005) argues that there is an operation fronting an apparently verbal constituent, and that this constituent is fronted to a position within TP (we return to this assumption below). She then shows that there is a general operation which fronts non-verbal predicates which are clearly larger than heads, e.g. relative clauses. Thirdly, Massam shows that what has been called noun-incorporation in Niuean (e.g. by Baker 1988) cannot be movement of N into V (pace Baker) since there are clear cases where a constituent larger than N undergoes this operation. She proposes instead that putative noun-incorporation is really the absence of object-shift to a VP-external position. In that case, the fact that the apparently incorporated noun moves with the verb shows that what is moved is VP rather than V. VOS order is thus derived by VP-fronting, and VSO by object-shift to a VP-external position combined with remnant VP-fronting, as shown in (50):18 (50)

a. [TP [VP V O] T [vP S v . . . (VP)]] –VOS b. [TP [VP V (O)] T [vP S v [AbsP O (VP)]]] –VSO

As (50) shows, the landing-site of VP-fronting is taken to be SpecTP. Massam (2000) argues that this is motivated by essentially the same property as that which causes the subject to raise to SpecTP in languages like English, French and Mainland Scandinavian: the operations “can be seen as two reflections of 18

AbsP here stands for Absolutive Phrase, which Massam suggests may correspond to AgrOP in more familiar languages. Our summary here glosses over the complication that Niuean is an ergative language and Massam’s treatment of the assignment of ergative and absolutive case.

296

Theresa Biberauer and Ian Roberts

a single EPP predication feature” (Massam 2000: 111). This type of analysis, first put forward by Massam & Smallwood (1997), and developed by Rackowski & Travis (2000) as well as several of the papers in Carnie, Dooley & Harley (2005), has been applied to a number of languages which display both VOS and VSO orders (mainly, but not exclusively Polynesian and Mayan languages). We turn to consideration of why these languages should be able to satisfy T’s EPP-requirements in this manner below. Assuming that this general type of analysis is correct for verb-initial languages like Niuean, which exhibit both VSO and VOS orders (unlike rigidly VSO languages such as the Celtic languages, where it is at the very least much harder to motivate a remnant VP-fronting analysis), we can draw some interesting conclusions in relation to the approach to verb-movement sketched above. In terms of our proposal that verb-movement to T is necessarily connected to richness of tense inflection, we arrive at the prediction that VOS/VSO languages of the Niuean type will have impoverished tense morphology. If these languages had ‘rich’ tense inflection on verbs, then this would have to be realised by a VþT compound in the Numeration, which would, in turn, give rise to head-movement of the VþT complex from V to T, thereby ruling out the VP-movement containing an unmoved verb that Massam proposes (cf. also Massam 2005: 239). It certainly seems to be the case that tense inflection is not found in Niuean, where tense/aspect information is entirely carried by sentence-initial particles; Massam (2000: 101) concludes that such particles represent “a portmanteau Complementiser/Tense element”. In the terms described above, we can simply take these elements to be realisations of T-features KEPT in C, the absence of tense inflection on verbs suggesting that the presence of Tense features on C is not the consequence of the SHARE option.19 Both Tongan and Ma¯ori appear to pattern the same way (Churchward 1953; Chung & Ladusaw 2003, cited in Chung 2005; see also Otsuka (2005) on Tongan). Recall that we have proposed that T-features are also KEPT on C in Germanic. The question that therefore arises is why Niuean and languages of this type do not exhibit V-movement to C – why do we have VP-movement to SpecTP and not V2-style V-movement to C? Here another observation of Massam’s becomes crucial, namely that “the Niuean verb is not a morphosyntactic verb, but is more participial in nature, bearing no features for finiteness or tense” (Massam (2005: 227); see also Massam (2005: 240), 19

Massam (2005: 240), in fact, observes that “Niuean does not exhibit any finiteness features in Infl (i.e. T, TB/IR)”, suggesting that it is not only the lexical component (or first phase – cf. Ramchand 2008) of the clausal structure that may be deficient in verbal properties (see following discussion in the main text), but also at least the inflectional (i.e. non-phase-head) component of the higher phase. In our terms, it may, then, be the case that C is the sole locus of V-features, i.e. that it has KEPT all of these features.

Subjects, Tense and verb-movement

297

where it is allowed that Niuean verbs may be “participial or even nominal in nature”; cf. also footnote 19). This observation follows on from a well established tradition in Polynesian linguistics advocating the absence of a (lexical) noun–verb distinction (cf. Massam (2005: 230–31) for brief overview discussion). If it is indeed the case that Niuean verbs are, in some sense, non-verbal and that T, likewise, lacks a V-probe, we can understand why this language fails to exhibit V2, this crucially being a phenomenon requiring a probed-for verbal element. What we still need to understand, however, is why T should attract VP rather than, say, the subject.20 As noted above, Massam (2000: 111) proposes that VP-movement in Niuean parallels subject movement in languages like English, being the consequence of EPP-driven movement. In our terms, this entails that Niuean differs from non-English Germanic in that C DONATEs its EPP-feature to T. This accounts for the absence of XP-movement into the C-domain in this language (but see below on other non-Celtic V-initial languages), but still does not account for the fact that T, on Massam’s analysis, attracts VP rather than the closest potential XP, the subject in SpecvP. We have seen that Niuean VP cannot be regarded as ‘properly verbal’; therefore it does not seem feasible to propose that VP-fronting is the consequence of a V-oriented EPP-feature (cf. i.a. Davies & Dubinsky (2001) and the discussion in Biberauer, Chapter 4). The fact that T does not appear to bear any V-features (cf. footnote 19) further underscores the likelihood that the trigger for VP-movement is not a V-oriented EPP-feature. If, on the other hand, we assume that T’s inherited EPP-feature is a ‘freestanding’ one which has not been inherited from C in conjunction with a specific substantive feature of C, we would still expect it to target the subject rather than VP since ‘free-standing’ features are assumed to operate ‘blindly’, targeting the closest available XP (cf. Biberauer (Chapter 4), and Holmberg (Chapter 2), on Icelandic, a language in which T’s inherited EPP-feature is also ‘free-standing’). Assuming that the theoretical assumptions outlined here are correct, Massam’s appealing proposal therefore appears to face a problem. If we take into account two as yet unmentioned properties of V-initial structures in Niuean and Austronesian languages more generally, two potential solutions, however, suggest themselves. Let us consider these briefly as they may point to respects in which the proposals outlined here may fruitfully be pursued in future research. A peculiar property of the predicate fronting found in Austronesian V-initial structures is that it results in predicate focus, rather similar to that

20

Cf. also Chung (2005) for discussion of the trigger behind VP-movement in VOS languages.

298

Theresa Biberauer and Ian Roberts

found in English structures of the kind illustrated in (52) (cf. i.a. Aldridge (2004), Paul (2008) for discussion): (51)

Formidable player, that Nadal!

Aldridge (2006: 39) notes that “there is some evidence that languages with VP-fronting use the vP edge as a focus position”. If this is indeed the case, it may be that (50) in fact represents simplified structures which do not indicate the (obligatory) focus-driven movement of VP that has taken place (under the influence of an Edge Feature21) in the lower phase; the correct structures would, instead, be those in (52): (52)

a. [TP [VP V O] T [vP (VP) S v . . . (VP)]] –VOS b. [TP [VP V (O)] T [vP (VP) S v [AbsP O (VP)]]] –VSO

As (52) clearly shows, focus-driven VP-movement places VP at the edge of the vP-phase, with the result that it will be the XP that we would expect T’s EPP-feature to target. If this proposal is correct, VP-movement to SpecTP is then not entirely parallel to English subject-fronting as it does not involve Agreedriven movement. Niuean can, however, be said to feature ‘an EPP-requirement’ in the sense that SpecTP must be projected and filled (cf. Biberauer (Chapter 4), for further discussion; cf. also Aldridge (2006: 43), who explicitly states that “predicate fronting is not driven by lexical features like [V] or [T]”). A second possibility (which need not be incompatible with that outlined above) is that the higher (CP) phase in Niuean is in fact slightly more elaborated than indicated in (50). One striking and oft-noted property of Niuean and Austronesian languages more generally that suggests that this may be correct is the restriction that the clause-final subjects in these languages are obligatorily definite or generic (cf. i.a. Aldridge (2004) Paul (2000; 2008) and Pearson (2001), who also refer to the ‘topicality’ or necessary status as old information of subjects in these languages). If some version of Diesing’s (1992) Mapping Hypothesis is correct, we are led to expect that subjects in the languages under discussion have moved out of the domain of existential closure, i.e. vP in the context of the phrase structure assumed here (cf. Biberauer, Chapter 4). If that is correct, it may, in turn, be the case that the clause-final subject occupies SpecTP, just as it does in English, with the fronted VP occupying a higher position, either a higher specifier of TP or a CP-related specifier (cf. the above-cited authors for various proposals along these lines). If the higher position is C-related, it might in fact be the case that 21

Edge Features may in fact correspond to ‘free-standing’ EPP-features that have been retained by phase-heads, or they may be ‘extra’ movement diacritics that are only available to phase-heads and that cannot therefore be passed on. We leave this matter aside here, although it should be noted that our earlier characterisation of Germanic V2 equates C’s Edge Feature with a ‘free-standing’ version of the EPP-feature that has been SHAREd with T.

Subjects, Tense and verb-movement

299

T’s inherited f- and EPP-features are ‘bundled’ together, as in English, and that the VP-triggering feature is a ‘free-standing’ EPP-feature retained by a lower C-head. We leave all these matters for further research as our major concern here is with the availability of VP-fronting in VOS languages.22 In this connection, we actually find a piece of indirect evidence that not all cases of verb-movement should be handled as remnant vP/VP-movement. If it is true that systems of the Niuean type are incompatible with ‘rich’ tense marking, then we predict, as already noted, that VSO/VOS languages should not have this kind of rich tense marking. But strict VSO languages, i.e. those which exclude VOS, may, if VSO can also be derived by V-v movement to T (or perhaps C). The Celtic languages are, as is well-known, strict VSO languages. Welsh, in fact, has a rather Romance-like set of synthetic verb tenses: (53)

22

We do note in passing, however, that evidence from another VOS language, Malagasy, suggests that an analysis of the sort proposed here, in which subject-raising is followed by VP-fronting around the subject, with the latter movement targeting a higher projection, is superior to earlier Spec-final accounts in which the initial VP was dominated by the subject (cf. Guilfoyle, Hung & Travis (1992) for an influential proposal; cf. also Aldridge (2004; 2006) for a range of arguments against the Spec-final analysis). Paul (2007) observes that all-rhematic/topicless structures of the type found in ‘headlinese’ consistently exhibit SVO ordering in Malagasy. If it is correct that ‘headlinese’ generally involves reduced syntactic structure (cf. i.a. Vinet 1993; Paesani 2006), the SVO ordering of Malagasy headlines can be viewed as the consequence of the fact that these structures lack the functional structure containing the VP-fronting trigger. Strikingly, the subject in Malagasy headlines is necessarily indefinite (since it represents new information), as shown below (examples from Paul 2007): (i)

(ii)

23

Welsh (forms of prynu ‘to buy’): brynith (present/future), brynodd (past), prynai (imperfect/conditional), [prynasai (pluperfect), pryno (present subjunctive)]23

Lehilahy nangalatra bisikileta man PST.AT.steal bicycle ‘Man stole bicycle’ *Ny lehilahy nangaltra bisikileta DET man PST.AT.steal bicycle

If these structures are TPs – something that may be suggested by the presence of the verbal inflection (although it is worth noting Dahl’s (1996: 168) observation that “[t]he Malagasy verb (with tense inflection) is used both as a finite verb and also, without any change, as we use infinitives and participles in European languages”; cf. our discussion of Niuean verbs in the main text) – SpecTP may in fact not be limited to definite/old-information subjects; instead, it may be the case that the information status interpretation assigned to the subject is determined on the basis of the presence vs absence of VP-fronting (cf. Neeleman & van de Koot (2008) for discussion of this type of ‘discourse-templatic’ approach to notions like ‘topic’ and ‘focus’). The variation in the initial consonant is an aspect of the consonant mutation system: here /b/ replaces /p/ as a function of the characteristic sentence-initial particle found with the relevant tense. As in the French examples in (3), the tenses in parentheses are part of the literary language only. Similarly, the present and imperfect interpretations of the future and conditional forms are either literary or restricted to certain subclasses of verbs. Literary Welsh also has a synthetic impersonal passive, or autonomous, form: prynir (present/future), prynwyd (past), pryner (subjunctive).

300

Theresa Biberauer and Ian Roberts

If Welsh is Romance-like in its tense inflection, as would appear to be the case from (53), then T-features combine in the Numeration with V-features in the way described above, giving rise to ‘reprojective’ V-to-T movement. This will then force verb-movement to T and thereby prevent VP- or vP-movement to SpecTP to satisfy the EPP. In fact, the standard analyses of Welsh clause structure and VSO order involve verb-movement to T (Borsley, Tallerman & Willis 2007; Roberts 2005; Tallerman 1998). It is well known that Irish resembles Welsh in having unmarked VSO order. This can be seen in both the matrix and the embedded clause in (42), repeated here: (54)

Deir siad gur ghoid na sı´ogaı´ ´ı. say they C[þPast] stole the fairies her ‘They say that the fairies stole her away.’

Irish is also similar to Welsh in its inventory of synthetic tenses, in that it has a present, future, past and conditional (cf. Mac Eoin 2002). We therefore expect Irish to pattern with Welsh in not allowing VOS orders, which is also empirically correct. In general, then, and perhaps not too surprisingly, Irish patterns with Welsh in relation to the typology of verb-initial languages sketched above. We would expect the same to be true of Breton (although here the obligatory XP-fronting in main clauses adds a further complication; see Jouitteau 2005) and Scots Gaelic. Similarly, Austronesian languages which do exhibit tense inflection, e.g. Tagalog, are predicted to permit V-movement to a higher verbal projection. Aldridge (2006) argues that this is indeed the correct analysis for Tagalog, a VSO language which permits various types of A0 -fronting, but not VOS orders. A further question concerns the status of the VSO Semitic languages (Classical Arabic, Egyptian/Coptic and Biblical Hebrew), but we will leave this matter for further research (cf., however, i.a. Ouhalla (1994), Shlonsky (1997) and Doron (2000), who have previously related Semitic to Celtic, and see Ouali (2008) for discussion of f-feature spreading in Berber). 6

Conclusion

In this chapter, we have proposed a new account of the cross-linguistic trigger for V-to-T movement, focusing on tense rather than agreement morphology as a potential trigger. We have seen how this simplifies the relation of V-to-T movement to the Null Subject Parameter, and how it gives a relatively clean distinction between the Romance languages and the Germanic languages and also seems to facilitate insights into V-initial languages. More specifically, our consideration of Germanic, Romance and what appear to be two distinct types of V-initial languages suggests that the

Subjects, Tense and verb-movement

301

following parametric considerations play a role in determining whether a language will have verb-movement or not: (i) rich tense morphology, where richness of tense is defined in relation to the number of tense/mood/aspect paradigms a verb has (cf. Section 3 above). If a language is tense-rich, it necessarily has V-to-T movement; if it is not, it will lack V-to-T movement. It may, however, exhibit verbmovement to a phase-head, either to v or to C, subject to the consideration in (ii) (cf. Afrikaans and Mainland Scandinavian), and verbs may also move as part of larger XPs (e.g. VPs, as in Niuean); and (ii) phase-head-driven distribution of uninterpretable (i.e. probing) features. If one or both of the clausal phase-heads, v and C, have retained their V-features (Ouali’s KEEP option), verb-movement to the relevant phasehead will take place; by contrast, V-to-T movement and movement to non-phase-heads is ruled out under these circumstances. Our approach thus predicts a unidirectional implication between tenserichness and V-movement: if a language is tense-rich, it must exhibit V-movement to one or more non-phase-heads. Further, it predicts that tense-poor languages may exhibit movement to phase-heads (v or C) and/or that verbs may move as part of larger constituents. On the empirical side, we have tried to show that (i) accurately predicts V-to-T movement in Romance and Celtic, while (ii) correctly predicts the absence of this movement in Germanic and Niuean-style V-initial languages. The difference between the inflectionally impoverished Germanic languages (Mainland Scandinavian and Afrikaans) and Niuean-style languages appears to reside in the fact that the lower phase in Germanic is unambiguously verbal, with the consequence that v is able to KEEP its V-features, thereby triggering phase-head-driven V-movement; in Niuean, by contrast, the clausal functional structure does not appear to be verbal, with the result that we would not expect v (or its notfully-verbal first-phase counterpart) to be able to attract V. Clearly, the precise mechanics of phase-head-driven head-movement remain to be worked out. The fact that it involves an interplay between the featural make-up of functional categories and the composition of lexical items, however, points to another instance of parametric variation ‘located’ in the Lexicon (cf. the socalled Borer-Chomsky conjecture discussed in Baker (2008a; 2008b) and also in the Introduction to this volume). On the subjects front, we have likewise considered the role both of morphology and of the feature-specification associated with relevant functional categories. Following a long tradition of researchers, we have adopted the idea that rich agreement morphology is a necessary condition to license null subjects. We have, however, also shown that this morphological consideration is not sufficient to ensure null subjects, its interaction with the features that

302

Theresa Biberauer and Ian Roberts

T inherits from C being crucial in determining whether null subjects will in fact be attested or not. In particular, we have shown that V2 languages like Icelandic and German, in which C only optionally SHAREs its EPP-features with T, cannot license null subjects. The prediction is that the same will be true for other languages in which C either SHAREs or DONATEs its EPP-features with T on an optional basis.

8

Varieties of French and the Null Subject Parameter1 Ian Roberts

Introduction This chapter is an attempt to apply the analysis to pro presented in Roberts (chapter 1) to French, a language whose status in relation to the Null Subject Parameter has been debated. We will see that some aspects of the questions concerning the nature and status of null subjects in French are clarified by this approach to null subjects. In particular, I propose a revision and updating of the analysis of French subject enclitics proposed in Sportiche (1999), exploiting the idea that only phase-heads can be first-merged as active probes proposed in Chomsky (2008). I will also attempt a rough characterisation of some aspects of register variation in French, following the proposals in Zribi-Hertz (1994). This leads to a fairly neat and interesting characterisation of French in relation to the Null Subject Parameter. 1

Is French a null-subject language?

It is usually thought that French is not a null-subject language (Burzio 1986: 135ff.; Perlmutter 1971; Rizzi 1982; 1986b: 400f.; Kayne 1983; Taraldsen 1978). The following contrasts with Italian motivate this ((1b) is a grammatical imperative, but ungrammatical as a declarative): (1)

1

a. Parla italiano. b. *Parle italien. ‘(He/she) speaks Italian.’

I’d like to thank the other members of the null-subject project group – Theresa Biberauer, Anders Holmberg, Chris Johns, Michelle Sheehan and David Willis – for their comments on various earlier versions of this work. Preliminary versions of this chapter were presented at the Null Subject Workshop held at the Universities of Geneva, Leipzig and Cambridge, at the Encontro Lingua Falada e Escrita V, Maceio´, Brazil, November 2005, and at the Romance– Japanese Comparative Syntax Workshop held at the University of Siena in May 2007. I’d like to thank the participants at those meetings, especially Claire Blanche-Benveniste, Joa˜o Costa, Eric Haeberli, Ad Neeleman, Liliane Haegeman, Luigi Rizzi and Ur Shlonsky, for their comments and questions. All errors are my own.

303

304

Ian Roberts

(2)

a. Chi hai detto che – ha scritto questo libro? b. *Qui as-tu dit qu’ – a e´crit ce livre? Who have-2SG said that – has written this book

(3)

a. E’ arrivato Gianni. b. *Est arrive´ Jean. Has arrived John. ‘John has arrived.’

(4)

a. Hanno telefonato molti studenti. b. *Ont te´le´phone´ beaucoup d’e´tudiants. Have telephoned many students. ‘Many students have telephoned.’

Regarding the three main properties identified by Rizzi (1982) as constituting evidence for the setting of the Null Subject Parameter, then, French appears to pattern consistently as a non-null-subject language. However, it has often been suggested that the subject clitics of French are really agreement markers of some kind, i.e. a realisation of the j (and, given the analysis in Roberts (Chapter 1), D) features of T (Hulk 1986; 1991; Jaeggli 1982; Roberge 1986; di Sciullo 1990; Sportiche 1983; 1999; see also, in a typological-descriptive framework, Harris 1978; 1988). According to this approach, we would have a null subject in an example like (5): (5)

pro je mange la pomme. I eat the apple ‘I eat the apple.’

The contrast between the behaviour of subject clitics in French and in many (but not all) Northern Italian dialects has been taken to indicate that this is not the correct analysis of French subject clitics (in proclisis; I will turn to the enclitics which arise under subject-clitic inversion in the next section). Rizzi (1986b) and Brandi & Cordin (1989) argue that subject clitics in many Northern Italian dialects are not subject pronouns but markers of subject agreement, while those of French are subject pronouns, on several grounds. First, there are frequently gaps and syncretisms in the subject-clitic paradigms in Northern Italian dialects but not in French, and pronominal paradigms do not normally show such gaps.2 Second, in many (but not all) Northern

2

In this connection it is important to bear in mind the role of final-obstruent deletion in French. The 3SG and 3PL subject pronouns, in both masculine and feminine, are often pronounced identically in the singular and in the plural: il(s) as /i(l)/ and elle(s) as /el/. However, if the following verb begins with a vowel, the final /z/ of the plural appears, giving il aime ‘he loves’ /ilem/ vs. ils aiment ‘they love’ /i(l)zem/. It is thus justified to treat the underlying forms of the plural pronouns as /ilz/ and /elz/, with regular final-obstruent deletion deleting /z/ where there is no following vowel (i.e. in non-liaison contexts; see below). I will return to the topic of final-obstruent deletion in French below.

Varieties of French and the Null Subject Parameter

305

Italian dialects, the subject clitic must always appear, and, therefore ‘doubles’ a nominal subject. In many dialects (see below), it appears with the kind of non-referential subject which cannot be left-dislocated. This contrasts with subject left-dislocation of the type found in French: (6)

a. Nessuno gl’ha detto nulla. No-one SCL.has said nothing b. *Personne il n’a rien dit. No-one he not.has said anything ‘No-one said anything.’

(Florentine) (French)

Third, some subject clitics in some varieties follow the preverbal negation, and as such appear to be more fully integrated into the clitic cluster proclitic to the finite verb than those of French, for example: (7)

a. No te ghe l’hai dit. Not you to-him it.have said b. Un tu gliel’ha detto. Not you to-him-it.have said ‘You have not said it to him.’

(Trentino) (Florentine)

Fourth, under coordination, the clitics must be repeated in both conjuncts in Northern Italian dialects, while this is not the case in French: (8)

a. Il chante et danse. He sings and dances b. La canta e *(la) balla She sings and she dances

(French) (Trentino)

Poletto (2000: 18–30) gives a very detailed and critical review of these tests, showing that Northern Italian subject clitics do not constitute a homogeneous set of entities; similarly Manzini & Savoia (2005, I: Chapter 1) subject these tests to close scrutiny, and conclude that they are not decisive for determining the nature of the subject clitics. Nevertheless, with the possible exception of those found in certain Veneto dialects, the Northern Italian subject clitics contrast in their distribution with French subject pronouns along the lines just illustrated, and this is what is most important here. It is on the basis of this contrasting behaviour that Rizzi (1986b), Brandi & Cordin (1989), and, with considerably more empirical breadth and depth, Poletto (2000) have concluded that French subject clitics are pronouns which cliticise in PF to the finite verb they precede (following the original proposal in Kayne (1983)). According to Rizzi and Brandi & Cordin (see also Manzini & Savoia 2005, I: 196ff.), subject clitics and subject-agreement inflection in many Northern Italian dialects are both realisations of j- or D-features which license the subject argument; we can restate their conclusions in current terms by saying that these subject clitics instantiate uninterpretable

306

Ian Roberts

j-features, while their French counterparts instantiate interpretable j-features.3 In this connection, it is worth taking a brief look at how this kind of analysis of Northern Italian subject clitics interacts with Mu¨ller’s pro-generalisation discussed in Roberts (Chapter 1, Section 2.5), repeated here as (9): (9)

An argumental pro DP cannot undergo Agree with a functional head a if a has been subjected (perhaps vacuously) to a j-feature neutralising impoverishment in the Numeration.

As already observed by Renzi & Vanelli (1983) and Poletto (2000), there is a tendency for subject clitics to appear with those persons of the verb which are not distinguished by the inflection on the verb. The forms of the Carrara dialect given in (10) illustrate the common situation (this is the present tense of the verb ‘to sleep’; see Manzini & Savoia 2005, I: 80): (10)

a dɔrmə, tə dɔrmə, i/al dɔrmə, a durmiŋ, durmitə, i/al dɔrmənə

Looking at the verbal inflection, we observe syncretisms in the three persons of the singular, but here the three clitics are distinct. Conversely, the 1SG and 1PL clitics are the same, but the verbal inflection differs. So, taken together, the subject clitics and the verbal inflection provide distinct agreement marking for each person. Hence, if we treat the two as instantiating T’s j-set, the systems are predicted to allow consistent null subjects by Mu¨ller’s pro generalisation. However, as Manzini & Savoia (2005, I: 118ff.) point out, (10) is not the only situation found in the dialects; some have fully differentiated verbal inflection, but syncretisms and gaps in the clitic paradigm, for example the Como dialect (Manzini & Savoia 2005, I: 100): (11)

dɔrmi, ta dɔrmat, al/la dɔrma, dɔrmum, dɔrmuf, dɔrmaŋ

But what does seem to hold however, at least across a very high proportion of cases, is the observation that, taken together, the subject clitics and the verbal inflection provide distinct agreement marking for each person (this is true for all but three of the 180 dialects whose paradigms are given by Manzini & Savoia (2005, I: 72–117)). Furthermore, no variety shows a fully differentiated paradigm of subject clitics, with no gaps and no syncretisms, while this is of course the situation in French. For example, contrast the forms in the Carrara and Como varieties above, as given by Manzini & Savoia (2005, I: 80, 100), with the corresponding French forms: (12)

3

je dors, tu dors, il/elle dort, nous dormons, vous dormez, ils/elles dorment

Cardinaletti & Repetti (2006a; 2006b) dissent from this general view. They argue that subject clitics, at least in certain Veneto dialects, are subject pronouns.

Varieties of French and the Null Subject Parameter

307

French has syncretic verbal inflection throughout the singular forms (in nearly all varieties of French dors and dort are pronounced identically as /d r/, but see the discussion of orthography and liaison below), but that in Carrara there is syncretism between the 1SG and 1PL clitic forms and there is a gap at 2PL. Not one of the 180 dialects reported by Manzini & Savoia has exactly the French pattern of partial syncretism in the verb endings and total differentiation of the subject pronouns; this is, of course, the typical pattern of nonnull-subject languages (e.g. English, German, etc.). So this is a further indication that Northern Italian dialects are different from French in this respect, and in fact that the former are consistent null-subject languages while the latter is not. It has, however, been argued that French allows expletive null subjects in Stylistic Inversion (Styl-Inv) constructions: c

(13)

a. A qui a parle´ ton ami? To whom has spoken your friend? ‘Who did your friend speak to?’ b. l’ homme a` qui a parle´ ton ami the man to whom has spoken your friend ‘the man who your friend spoke to’ c. Je souhaiterais que vienne ton ami I wish that come(subjunc) your friend ‘I wish that your friend would come.’

Kayne & Pollock (1978) and Pollock (1986) essentially treated this construction as a highly restricted occurrence of ‘free inversion’, with a null expletive in the preverbal subject position (what we would now call SpecTP) and the overt subject in a lower position. In their more recent treatment, however, Kayne & Pollock (2001/2005) argue that the post-verbal subject is in a ‘high’ subject position and that Styl-Inv involves remnant-TP movement to a still higher position. Regarding the expletive null subject, they conclude that “no instance of SI is an expletive il sentence with il unpronounced” (Kayne & Pollock 2001/2005: 23, note 31). On the other hand, Kayne & Pollock posit a silent subject clitic in Styl-Inv constructions, one which is capable of doubling a definite subject like ton ami in (13a). They give three pieces of evidence for this. First, in certain noncolloquial registers 3rd person strong pronouns seem to appear as both preverbal and post-verbal subjects, but not 1st or 2nd person ones, which must cooccur with an overt subject clitic: (14)

a. LUI a mange´/Qu’ a mange´ LUI? HE has eaten/what has eaten HE? ‘HE has eaten/what has HE eaten?’ b. *TOI as mange/*Qu’as mange´ TOI? YOU have eaten/what have eaten YOU? ‘YOU have eaten/what have you eaten?’

308

Ian Roberts c. TOI tu as mange´. YOU you have eaten ‘YOU have eaten.’

Kayne & Pollock suggest that there is a 3rd person null subject in (14a) in the position occupied by tu in (14c). (14b) is ungrammatical because 2nd person null subjects are not allowed in French.4 Second, indefinite subjects are not good in Styl-Inv: (15)

*Quel gaˆteau a mange´ quelqu’un? Which cake has eaten somebody?

If there is a weak subject pronoun in Styl-Inv, this can be assimilated to cases like (6b), indicating the impossibility of dislocating a non-referential quantifier doubled by a pronoun of this kind. Third, weak subject pronouns can neither be dislocated nor can they be post-verbal in Styl-Inv: (16)

a. *Il, il mangera ce gaˆteau. He, he will-eat this cake b. *Quand a parle´ il? When has spoken he?

Again, these reduce to a single fact if il is ‘dislocated’ in (16b) and there is a null subject in the true subject position. As a weak pronoun, il cannot appear in a dislocated position. Kayne & Pollock’s arguments that there may be an argumental 3rd person null subject at least in certain non-colloquial registers of French in leftdislocation and Styl-Inv constructions are quite convincing. In the light of the approach to null subjects in Roberts (Chapter 1), I will treat this element as a weak pronoun, in fact pro, rather than as a subject clitic. We can then conclude that the relevant register of French is a highly restricted variety of partial null-subject language, in the sense of Holmberg, Nayudu & Sheehan (2009); it allows expletive and 3rd person null subjects just under the highly restricted conditions just described. Let us now consider more closely how French complies with Mu¨ller’s pro-generalisation in (9). Recall that this generalisation implies that, where the j-features of a given functional head have undergone impoverishment, that functional head cannot license a null argument. The sign of impoverishment is system-defining syncretism. So our question now is: Are there cases of system-defining syncretism in the French verbal-agreement paradigm?

4

In more colloquial registers, a weak subject pronoun (il) obligatorily appears in (14a). Kayne & Pollock observe that registers which require the 3rd person subject clitic here do not allow Styl-Inv at all (these include Quebec French).

Varieties of French and the Null Subject Parameter

309

Harris (1988: 224) gives the following summary of French verbal conjugation: (17)

Present indicative by conjugation type: donner ‘to give’ donn-e donn-es donn-e donn-ons donn-ez donn-ent

(18)

voir ‘to see’ voi-s voi-s voi-t voy-ons voy-ez voi-ent

rompre ‘to break’ romp-s romp-s romp-t romp-ons romp-ez romp-ent

finir ‘to finish’ fini-s fini-s fini-t fin-iss-ons fin-iss-ez fin-iss-ent

Other tenses (all conjugations):5 Imperfect donn-ais donn-ais donn-ait donn-ions donn-iez donn-aient

Present Subjunctive donn-e donn-es donn-e donn-ions donn-iez donn-ent

Future donne-r-ai donne-r-as donne-r-a donne-r-ons donne-r-ez donne-r-ont

Conditional donne-r-ais donne-r-ais donne-r-ait donne-r-ions donne-r-iez donne-r-aient

As is well known, “[t]o say that French orthography is less than ideal would be something of an understatement” (Harris 1988: 215). Accordingly, we need to take certain orthographic and phonological factors into consideration before we can fully evaluate the above paradigms in relation to impoverishment. Furthermore, I will concentrate on the person and number features, without speculating as to the best morphological analysis of the tense markers (and the theme vowels in the present). The most salient fact is that word-final obstruents are generally not pronounced. In fact the final orthographic <s> in all the 2SG forms and all the 1SG forms where it appears is never pronounced. The final <s> of the 1PL, on the other hand, is pronounced, but only in liaison contexts (where the following word in the same prosodic domain begins with a vowel). This probably justifies positing it as part of the relevant vocabulary item, and allowing it to be deleted in non-liaison contexts by the very general process of final-obstruent deletion (see Anderson (1982) and Dell (1985) for details). The same is arguably true of the final of the

5

There are two further tenses which are not part of spoken French, the preterit (or passe´ simple) and the imperfect subjunctive. I will follow Harris in leaving these aside, as they are confined to the written language, and the latter is very rarely used even there.

310

Ian Roberts

3PL and the final of the 2PL. These person endings are therefore clearly distinct in all tenses; we may tentatively assign them the following exponence, following the feature system in Mu¨ller (2005):6 (19)

a. /onz/ $ [þpl, þ1] b. /ez/ $ [þpl, –1, þ2] c. /ət/ $ [þpl, –1, –2]

There is clearly no impoverishment here. The singular forms are more interesting, however. Here again, the status of the orthographic final consonants in relation to liaison, in particular in ‘high’ registers, is crucial. Armstrong (1962: 165) states that optional liaison is found “[a]fter the third person of verbs ending in the letter t, except when a personal pronoun follows, in which case liaison is compulsory”.7 There is nonetheless no syncretism between 3SG and 3PL in the present tense, since the 3PL ending is /ət/, as we have seen, and the 3SG ending is /t/.8 It is possible that what we have there is a syncretism between 3SG and 3PL in verbs of all conjugations in the imperfect and conditional tenses; however, we can take the ‘imperfect’ ending (common to both tenses) as /e  i/ (perhaps more abstractly a non-low front vowel or glide) and then, as with the present, treat the 3SG ending as /t/ and the 3PL as /ət/, with a regular phonological rule deleting /ə/ after /e/. In that case, the singular forms are as in (17): (20)

a. b. c. d.

 $ [–Pl, aPersi, –aPersj] (i 6¼ j) /t/ $ [–Pl, –1, –2] (see footnote 8) /e/ $ [–Pl, þFuture, þ1] /a/ $ [–Pl, þFuture, –1]

(It is unclear whether the homophony of the 2SG and 3SG future endings really represents a system-defining syncretism in Mu¨ller’s sense.) In all tenses except the future, there is syncretism between 1st and 2nd person. This latter point creates a difficulty for Mu¨ller’s feature system for persons, since, by positing the features [1] and [2], the prediction is made that

6

7

8

These are somewhat approximate phonological forms: /onz/ for example is realised with nasal vowel /o˜/ and no nasal consonant. This results from regular phonological processes, however (see in particular Dell 1985). Thanks to Adam Ledgeway for pointing this quotation out to me. The ‘compulsory liaison’ with an inverted pronoun alluded to here is arguably a different phenomenon involving the interrogative marker /t/. I will look at this in detail in the next section. This does not apply to 1st conjugation verbs, as shown in the paradigm for donner in (17), or for the subclass of 4th conjugation verbs illustrated by ouvrir ‘to open’, which have 3SG present ouvre, 3PL ouvrent. In the 3SG of donner, there is an underlying /ə/, but this occurs throughout the singular, and is the 1st conjugation theme vowel (see Dell (1985: 213–14) for arguments that /ə/ is the theme vowel in the 1st conjugation, while there is a zero theme vowel in the 3rd conjugation).

Varieties of French and the Null Subject Parameter

311

impoverishment creates syncretism between 1st and 3rd person forms (specified as [–2]) and between 2nd and 3rd person forms, (specified as [–1]), but not between 1st and 2nd person forms. I have indicated this in using anotation in (20a), intended to indicate that the positive value of one of the person features, [þ1] or [þ2], must be chosen, along with the negative value of the other one. Leaving aside the difficulty of formulating this rule given the feature system Mu¨ller adopts (which does not in itself pose a problem for the general idea of system-defining syncretism), it is clear that there is systemwide syncretism here. So it seems clear that French conforms to Mu¨ller’s pro generalisation, and that we therefore predict that it cannot be a consistent null-subject language (whatever its possible status as a (very) partial nullsubject language, given Kayne & Pollock’s observations reported above).9 So we conclude that (certain registers of) French may constitute a very limited null-subject system. All registers show instances of system-defining syncretism in Mu¨ller’s sense, and so none is a consistent null-subject language. Therefore, the standard observation concerning the difference between French and Italian is confirmed, although we arguably have a slightly deeper understanding of it now. There is one very important proviso to this conclusion, however, concerning what happens in cases of inversion, other than Stylistic Inversion, in French. I now turn to this. 2

Null subjects and subject-clitic inversion

In this section, I will consider the status of subject clitics in enclisis in French, and compare it with Franco-Provenc¸al Valdoˆtain and some Veneto varieties. In all three cases, the enclitic subject clitics are best analysed as agreement markers, forming an special interrogative conjugation. This is true despite the differences in the status of the preverbal clitics, which in Valdoˆtain and Veneto, but not, as we have seen, in French, are agreement markers. 2.1

French

Here I want to consider the status of null subjects in subject–clitic inversion in French. I will argue that in this context French is a consistent null-subject 9

The 3rd person liaison described by Armstrong is characteristic of rather careful, non-colloquial registers. Arguably, this kind of liaison is not found in more colloquial registers. Where this kind of liaison is not found, there is general syncretism of singular forms in all tenses except the future, where there is syncretism of 2nd and 3rd person. Clearly, then, we have different and more generalised impoverishment in these registers. These may also be the same registers as those which lack examples like (14a), although further research on the precise nature of the register variation is required in order to establish this. See Zribi-Hertz (1994) and the comments in Section 3 below for more on register variation.

312

Ian Roberts

language, and that the subject clitic apparently enclitic to C in these contexts is in fact a realisation of the j-features of C; in these respects, my analysis is similar to those in Zribi-Hertz (1994) and Sportiche (1999), although it differs from them in detail. Subject–clitic inversion is illustrated in (21): (21)

As-tu vu Marie? Have you seen Mary?

Standardly, the verb is thought to have moved through T to C (this analysis originated in den Besten (1983), was developed in Kayne (1983) and Rizzi & Roberts (1989), and has been challenged in Poletto & Pollock (2004) and Pollock (2006); see footnote 13). The subject clitic is clearly enclitic to the verb in C here. This can be seen from the fact that no material of any kind may intervene between the inverted verb and the subject. In English parenthetical material can be marginally inserted between an inverted auxiliary and the subject (although this deteriorates where the subject is pronominal): (22)

a. ?Has, by the way, John seen Mary? b. ?*Have, by the way, you seen Mary?

In French, interpolation is impossible here: (23)

**As, a` propos, tu vu Marie? (¼(22b))

In general, non-clitic subjects are impossible in the position occupied by the clitic in (21): (24)

*A Jean vu Marie? Has John seen Mary?

Furthermore, although it is well-established that subject pronouns generally cliticise in French (Kayne 1972; 1975); Cardinaletti & Starke (1999) treat the subject proclitics as weak pronouns which cliticise/prosodically restructure at PF), both Cardinaletti & Starke (1999) and Sportiche (1999: 202) point out that enclitic subjects are more restricted in distribution than proclitic ones in certain ways. For example, coordinated subject pronouns are possible in preverbal position, but not where the verb is inverted: (25)

a. Il ou elle connait bien le proble`me. He or she knows well the problem. ‘He or she knows the problem well.’ b. *Mange-t- il ou (t-)elle? Eats he or she?

Varieties of French and the Null Subject Parameter

313

Further, as can in fact be seen in (23), subject–clitic inversion with a 3SG clitic is associated with a specific phonological operation, here insertion of an epenthetic /t/.10 Cardinaletti & Starke (1999: 167) observe that subject clitics in enclitic position cannot be omitted in the second conjunct of a coordinate structure: (26)

a. Il aime les choux, mais – ne les mange que cuits? He likes the cabbages, but – not them eats but cooked b. *Aime-t-il les choux, mais – ne les mange que cuits? Likes he the cabbages, but – not them eats but cooked? ‘Does he like cabbage, but only eats it cooked?’

This, in their terms, suffices to classify enclitic il as a clitic, rather than a weak pronoun. Finally, in Modern French there are rather heavy restrictions on inversion over a 1SG subject. Pollock (2006: 651) observes that inversion over 1SG je is only possible where the inverted element is a modal or aspectual auxiliary, or the verb is in the future or conditional form; forms such as arrive-je? ‘arrive I?’ and comprends-je ‘understand I?’ are highly marginal at best. We have mentioned that Cardinaletti & Starke (1999) analyse the ‘subject clitics’ of French as weak pronouns. But this only applies to subject clitics in proclisis to the verb. In the enclisis environment, as we can see from the above, these elements behave differently. In terms of the general account of cliticisation in Roberts (forthcoming b), C is a target for cliticisation, being a phase-head. We could regard subject pronouns as cliticising directly to C from their first-merged position in SpecvP. However, this idea is rather problematic. If subject j-features are really features of C, as proposed in

10

The epenthetic /t/ is not a liaison consonant. This can be seen by contrasting it with the underlying final /t/ of the 3PL ending with a 1st conjugation verb. As we mentioned in Section 1, in very careful speech the 3PL form can give rise to liaison in non-inversion contexts, leading to a /t/ being pronounced in the onset of a following vowel-initial syllable, as in: (i)

Ils jouent a` la poupe´e. They play with the doll

(/izˇutalapupe/)

The analogue to (i) is completely impossible with a 3SG 1st conjugation verb (or a 4th conjugation verb of the ouvrir subclass): (ii)

Elle/il joue *-t- a` la poupe´e. S/he plays with the doll.

On the other hand, the epenthetic /t/ is obligatory in inversion in the 3SG in all registers which allow inversion, as we saw in the quotation from Armstrong (1962) given in Section 1. Pollock (2006: 627–30) gives an analysis of epenthetic /t/ which makes its appearance central to the analysis of subject–clitic inversion, and which entails the postulation of a null variant of this element where the subject is 1PL, 2SG and 2PL.

314

Ian Roberts

Chomsky (2008), and C ‘withholds’ these features from T in the residual V2 environment (see below), there is no reason to think that T has a V-attracting feature in this environment either. If T does not attract V then V moves directly from v to C (there is no reason to assume the Head Movement Constraint; see Roberts (forthcoming b: chapter 5)) and, by the Strict Cycle, moves before the subject moves from SpecvP. This would give rise to proclisis of the subject to the verb in C, assuming that head-movement is always left-adjunction (Kayne 1991; 1994). On these assumptions, it is difficult to see how the subject could be enclitic to the verb. Instead, still following Chomsky’s (2008) suggestion that T’s j-features are actually features of C, I propose what amounts in all respects except one to a version of the analysis in Sportiche (1999) (my analysis is also close to those put forward in Zribi-Hertz (1994) and Poletto (2000: chapter 3) for Northern Italian dialects, in that I take “subject–clitic inversion” to be “a morphological process of affixation” which “always implies syntactic movement of the inflected verb” (Poletto 2000: 45)). In the relevant contexts in French (essentially a class of residual V2 contexts rather similar to those of English: root interrogatives, counterfactual conditionals, quotes (optionally), and clauses beginning with certain adverbs, e.g. peut-eˆtre ‘perhaps’), C simply does not pass its features to T. C’s j-features are then realised as enclitics in C. In other words, I propose a variant of the idea that French has a special ‘interrogative conjugation’ (conjugaison interrogative; see also Pollock 2006: 628f.). Forms such as enclitic -tu, -t-il, -t-elle, etc., are realisations of the j-set of residual V2 C; we can think of them as realisations of C[þQ], or whatever feature best characterises residual V2 C. We can clearly capture the presence of epenthetic /t/ in this way (see footnote 7 and 10 above). It seems that there is no 1SG form in the majority of verbs, unsurprising if this is an inflection class, but surprising if we dealing with a pronominal paradigm (see the discussion of French as opposed to Northern Italian subject clitics in Section 1, and Rizzi (1986b)). Moreover, there is some evidence that the presence of an interrogative ending of this class causes stem allomorphy on the verb, thereby showing a typical property of an inflection (see Zwicky & Pullum 1983): the modal verb pouvoir ‘can’ allows a 1SG ‘enclitic’ or affix, but the suppletive and otherwise obsolete form puis surfaces as the verb stem instead of the expected peux: puis-je? ‘can I?’, but not *peux-je? The properties noted in (22–25) follow straightforwardly on this analysis: interpolation, coordination and ellipsis are all operations which cannot affect affixes independently of stems, and yet this is what must be happening in these examples. Concerning (22), we can assume that, since there are no j-features in T, there is no Agree relation between T and the subject, and therefore no reason for the subject to raise to SpecTP. Instead, the relevant

Varieties of French and the Null Subject Parameter

315

Agree relation holds between C’s f-set and the subject, and indeed the subject may be attracted to SpecCP, giving rise to complex inversion. The grammatical version of (24) is thus: (27)

Jean a-t-il vu Marie? John has-3SG seen Mary? ‘Has John seen Mary?’

As I mentioned above, this analysis is very close to that proposed in Sportiche (1999: 206ff.). The principal difference is that Sportiche proposes that the interrogative conjugation is formed at the T-level rather than the C-level (Sportiche proposes that it moves to C covertly; in this way, he captures the well-known root-embedded asymmetry affecting this construction (den Besten 1983; Rizzi & Roberts 1989)). Sportiche points out that there is evidence for this view from the fact that adverbial material which would normally be analysed as appearing at the edge of TP precedes the interrogative verb in complex inversion rather than following it. This is the case of temporal adjuncts such as quand le vote a eu lieu ‘when the vote had taken place’, which can appear in pre-subject position (following dans quelle ville in (28b)), but not readily between the auxiliary and the participle in examples like the following: (28)

a. Les e´lecteurs sont ??(quand le vote a eu lieu) alle´s a` la peˆche. The voters are when the vote has had place gone to the fishing ‘The voters, when the vote had taken place, went fishing.’ b. Dans quelle ville, les e´lecteurs sont-ils ??(quand le vote a eu lieu) alle´s a` la peˆche? In which town the voters are they when the vote has had place gone to the fishing. ‘In which town did the voters, when the vote had taken place, go fishing?’

It may be possible, however, to think that this material is licensed by features of C which are inherited by T in non-residual-V2 contexts, but ‘withheld’ in residual V2 contexts.11 Again like Sportiche (1999), the analysis of complex inversion and subject–clitic inversion just sketched predicts that where we have subject– clitic inversion we have a null subject (see also Zribi-Hertz 1994: 140): 11

The difficulty of combining subject extraction with an apparent enclitic subject pronoun might be seen as a further problem for our analysis of complex inversion: (i)

*Qui a-t-il mange´ de la soupe? Who has-3SG eaten of the soup

However, Sportiche (1999: 215) points out that examples of this type are attested, citing Gre´visse (1980). Sportiche suggests that the operative restriction has to do with the specificity of the wh-expression, rather than with the possibility of extraction itself.

316

Ian Roberts

(29)

pro A-t-il vu Marie? has-3SG seen Mary? ‘Has he seen Mary?’

The null subject occupies the specifier of the position occupied by the verb bearing interrogative inflection; it is attracted there by the EPP feature associated with residual V2 C (a further feature withheld from T in this context). Unlike Sportiche, I take this to be SpecCP. In fact, the interrogative conjugation does not show any person–number syncretism, being of the following form with main verbs (using Mu¨ller’s feature system; these are realisation of j-features on C rather than T, hence there is a further contextual restriction here that is not specified in (30)):12 (30)

a. b. c. d. e. f. g. h.

[þ1, –2, –pl] [–1, þ2, –pl] [–1, –2, –pl, –fem] [–1, –2, –pl, þfem] [þ1, –2, þpl] [–1, þ2, þpl] [–1, –2, þpl, –fem] [–1, –2, þpl, þfem]

$ $ /ty/ $ /til/ $ /tel/ $ /nuz/ $ /vuz/ $ /tilz/ $ /telz/

(The final /z/ in the plural forms only surfaces in careful liaison contexts, e.g. ont-ils a` faire cela? /o˜tilzafersla/ ‘Do they have to do that?’). Given these forms, we expect C to be able to have a D-feature, given the suggestion in Roberts (Chapter 1) that a category with fully specified j-features can bear such a feature; it therefore should be able to delete a weak pronoun and thereby give rise to a null subject. Sportiche (1999: 214) gives some evidence for the presence of a null subject in cases of subject–clitic inversion from the following binding facts:

12

There are two further forms which need to be considered here: the generic element on, which surfaces here as /to˜/, and the ‘demonstrative’ ce. The former can be integrated into the paradigm with the relevant feature specification (whatever characterises an arbitrary pronoun able also to receive a 1PL interpretation; see Cinque (1988), Holmberg (Chapter 5)). The latter can, at a first approximation, be seen as an inanimate 3rd person ending, although much more needs to be said about enclitic ce in questions (particularly of the qu’est-ce que ‘what is it that’ variety); see Munaro & Pollock 2005). This idea does not account for the fact that -ce is possible in subject–clitic inversion, but not in complex inversion: (i)

Est-ce correct?

Is it correct? (ii)

*Cela est-ce correct? That is it correct?

See Zribi-Hertz (1994: 141–43) for a proposal and references. Thanks to Claire BlancheBenveniste for drawing my attention to these forms.

Varieties of French and the Null Subject Parameter (31)

317

a. Si Jeani e´tait venu, ili aurait de´cide´. If John was come, he would-have decided. ‘If John had come, he would have decided.’ b. *Ili aurait, si Jeani e´tait venu, de´cide´ he would-have, if John were come, decided c. *Aurait-ili, si Jeani e´tait venu, de´cide´ would-have-he, if John were come, decided

As Sportiche points out, (31b) is a Principle C effect, since il binds Jean. The effect is also found in (31c); here this cannot plausibly be attributed to the interrogative inflection -t-il, but may instead be attributed to pro. Pollock (2006: 622f.) points out some asymmetries in the distribution of preverbal pronominal subjects in complex inversion. First, as also pointed out by Kayne (1983) and Rizzi & Roberts (1989), a subject clitic is not allowed: (32)

*Ou` il est-il alle´? Where he is-3SG gone?

This relates to the distribution of weak, as opposed to strong pronouns. It appears from (32) that overt weak pronouns cannot appear in SpecCP, but are restricted to SpecTP. This follows from the fact that in French C can trigger null-subject deletion, but not T, as described above. Thus, we can take it that proclitic il must delete in this context. Second, only 3rd person strong pronouns are allowed: (33)

a. Quel livre eux ont-ils apporte´? Which book they have-3PL brought? ‘Which book have they brought?’ b. *Quel livre toi as-tu apporte´? Which book you did-2SG bring?

This contrast parallels that in (14) of the previous section, as Pollock points out. However, given our analysis, the structure cannot be the same, since 2nd person null subjects are allowed in SpecCP (see (21)). It seems, then, that we have to conclude that 3rd person strong pronouns are allowed in SpecCP. Being strong pronouns, they are not subject to deletion in this position (see Roberts (Chapter 1: footnote 14) for a speculation regarding the difference between strong and weak pronouns). This in turn implies that the weak pronouns il/ils and their strong counterparts lui/eux are not purely contextual variants, pace Cardinaletti & Starke (1999). If it is also the case that lui/ eux can appear in SpecTP, then this weakens Kayne & Pollock’s argument for a null subject in contexts like (14a) above, but this does not fully undermine their case for a null subject in Styl-Inv contexts. Clearly, there is

318

Ian Roberts

in any case a contrast between 3rd person and other persons, in that non-3rd person strong pronouns cannot appear in this position, as (33b) shows.13 Our preliminary conclusion regarding French, then, is that this language is fully null-subject at the C-level, but not at the T-level. This is due to the 13

Pollock in fact argues against T-to-C movement in the various inversion constructions in French (subject–clitic inversion, complex inversion and Stylistic Inversion), positing instead remnant XP-movement into the CP-field. This does not materially affect the analysis of subject enclisis and null subjects, either Pollock’s (which involves cliticisation to AgrS) or that proposed here; we could consider the interrogative conjugation to be a phrasal affix attached to the right of the fronted remnant XP; this is very close to what Pollock (2006: 628) proposes for epenthetic /t/. However, the arguments given by Pollock (and Poletto & Pollock 2004) against a headmovement analysis of verb-movement into the C-system do not in fact apply, given the assumptions about morphology, stemming from Mu¨ller (2005), adopted in Roberts (Chapter 1). These arguments are based on Kayne’s (1994: 42–46) discussion of the landing site of clitics in Romance. Kayne considers a basic example with a direct-object clitic: (i)

Vous le voyez. You(pl) him/it see ‘You see him/it.’

Kayne adopts three postulates. First, that morphologically derived forms such as voyez are syntactically formed, possibly by syntactically combining the root voi/voy- with the ending -ez. Second, that the LCA applies to sub-word-level operations, and third, that the LCA bans multiple head-adjunction. Given these three postulates, the clitic would have to adjoin to the verb root, followed by adjunction of [ le voy-] to (the functional head occupied by) -ez. Where the verb bears a prefix, as in vous le pre´voyez ‘you foresee it’, the clitic would have to first attach to the prefix. Kayne goes on to suggest that it is more plausible to assume that clitics adjoin to empty functional heads. Kayne further observes enclisis to infinitives of the type found in Italian and Spanish (e.g. farlo ‘to do it’, etc.), along with the well-known facts of enclisis to imperatives as in fais-le ‘do it’. Kayne concludes from these cases that in general verb-movement to C does not ‘carry along’ clitics. It then follows that, in standard cases of subject–clitic inversion, the cliticþverb combination has not moved to C as a single element. Kayne follows Sportiche’s (1999) proposal that there may be V-movement to C at LF, hence accounting for the root nature of the construction (he suggests that the clitic may delete at LF; see his note 16). Finally, Kayne observes that high-register examples such as . . . le bien faire . . . ‘to do it well’ support the idea that the clitic and the verb do not have to combine. Poletto & Pollock (2004) and Pollock (2006) endorse Kayne’s general conclusion that clitics and verbs cannot and do not combine in syntax, but propose that, instead of covert verbmovement into the C-system in subject–clitic inversion, there is overt remnant movement. The derivation of an example like l’as-tu fait ‘Have you done it?’) would proceed as follows: (ii)

! (movement of ZP) a. Tu [XP le [YP as [ZP fait]]] b. Tu [ZP fait] [XP le [YP as (ZP)]]] ! (remnant movement of XP) c. [XP le [YP as (ZP)]] tu [ZP fait] (XP)

Remnant XP-movement is triggered by the interrogative feature of the attracting head, which is part of an articulated C-system, and which is realised as epenthetic /t/ in the 3rd person (see Footnote 11). It is unclear what the trigger (or the landing site) of ZP-movement is, and what the nature of XP is, as well as the cliticisation operation itself. Given the general approach to cliticisation adopted here, and outlined in detail in Roberts (forthcoming b), we do not need to draw the conclusions drawn by Kayne and adopted by (Poletto and) Pollock. Although I assume a version of the LCA, I do not assume that all morphology is syntactic affixation, and therefore that the LCA necessarily applies at the

Varieties of French and the Null Subject Parameter

319

impoverished nature of verbal inflection, in contrast with the fully specified j-features associated with the conjugaison interrogative given in (30). Of course, if the possibility of null subjects/arguments ultimately depends on impoverishment of j-features, then it follows that different probes may differentially license null subjects/arguments (in fact, this is also true of Italian: T licenses consistent null subjects, while v only allows a restricted type of partial null object; see Rizzi 1986a). 2.2

Franco-Provenc¸al Valdoˆtain

Valdoˆtain has cases like (34), discussed in Roberts (1993a) and Pollock (2006: 632–36), which appear to be instances of the same string as in (32): (34)

a. Cen que dz’i dzo fe? What that I-have I done? ‘What have I done?’ b. Dze medzo-dzo` an pomma? I eat I an apple ‘Do I eat an apple?’

Similar examples, from various Provenc¸al varieties, are mentioned by Poletto (2000: 54–55), and a wide range of apparently similar cases are reported in Manzini & Savoia (2005, I: 551ff.). One possibility is to treat the enclitic as a realisation of C’s j-features, just as in French, and the proclitic as attracted there from SpecvP. The proclitic must be a clitic, rather than a weak pronoun as in French. Now, recall the interaction of EPP features with cliticisation postulated in Roberts (forthcoming b), as follows: (35)

A probe P can act as in incorporation host for a goal G only if P lacks an EPP feature capable of attracting G.

Cliticisation of the proclitic to C therefore implies, given (35), that C in this variety does not have an obligatory EPP feature. Assuming that expletives are merged directly in their surface positions purely in order to satisfy the EPP, this then implies that we not find the equivalent of (35) with an expletive pronoun. However, examples such as the following show that we do find expletive proclitics in the construction analogous to (35) in Valdoˆtain: sub-word level. Moreover, I do not adopt Kayne’s specific conclusion, from his formulation of the LCA, that multiple head-adjunction is impossible. In the terms of Roberts (forthcoming b), minimal phases can have multiple specifiers at their edges just as maximal phases can. In fact, I have been assuming that cliticisation is movement to a functional head, v in the case of Romance. A complement clitic and verb can and must combine at v, with only ‘excorporation’ of the highest specifier of the minimal phase v*min allowed, giving rise to clitic-climbing, as argued in detail in Roberts (forthcoming b: Section 3.4). Thus we are not forced to a remnant-movement analysis of verb-movement into the C-system (see Roberts (forthcoming b: chapter 4) on how full and residual V2 should be handled).

320

Ian Roberts

(36)

a. Cenque l’y est-e¨ ba-le´? What it-there is-it over-there ‘What is there over there?’ (“Qu’y a-t-il la`-bas?”) b. Que l’est-e¨ que te va trove´? What it-is-it that you will find? ‘What will you find?’

(Chenal 1986: 389))

(Chenal 1986: 388)

Furthermore, Valdoˆtain has complex inversion, where the preverbal subject cooccurs with a subject clitic: (37)

a. Ton ommo l’ est-e¨ dza torna`? Your man he is-he already come-back ‘Has your man already come back?’ b. Pierre l’at- e¨ manda` euna lettra a Caterine? Pierre he-has-he sent a letter to Catherine? ‘Has Pierre sent a letter to Catherine?’

(Chenal 1986: 350)

(Chenal 1986: 373)

Finally, Valdoˆtain subject clitics are of the type that appear with subjects of all kinds, including extracted wh-phrases and non-referentially quantified subjects, as shown in Roberts (1993a):14 (38)

a. Gnunc l’a viu-me. No-one he-has seen me ‘No-one saw me.’ b. Ki y et vinya`? Who he is come? ‘Who has come?’

(Roberts 1993a: 330)

(Roberts 1993a: 328)

For all these reasons, it seems best to regard the apparent proclitic subject in (34) and (35) as a realisation of j-features (this is really the analysis in Rizzi (1986b) and Brandi & Cordin (1989); Poletto (2000: 140f.) shows that this is one type of ‘subject clitic’ among several found in Northern Italy). This in fact implies that interrogative C realises j-features twice over, with the interrogative conjugation the syncretic realisation of the Q-morpheme and j-features, while the ‘proclitic’ in these cases is the realisation purely of 14

(38a) is an example of another phenomenon of interest in Valdoˆtain (also found in some Piedmontese varieties; Parry 1994), which Roberts (1993a: 329) calls ‘OCL-for-SCL’. Here it seems that there is just one morphological ‘slot’ for a proclitic. Where there is more than one clitic, objects are enclitic to the past participle, as here. In some varieties (as in the Val d’Ayas, from which this example is taken), object-enclisis to the participle is optional, and the object can seemingly replace the subject clitic, as in (i): (i)

Gnunc m’a viu. No-one me-saw ‘No-one saw me.’

In other varieties, e.g. Chaˆtillon, this is not possible, and objects must be enclitic to the participle in the presence of a subject clitic.

Varieties of French and the Null Subject Parameter

321

j-features. A more attractive alternative is to regard the ‘proclitic’ as a realisation of the D-feature (this is proposed by Manzini & Savoia (2005, I: 57f.)), with the interrogative conjugation realising Q and j-features.15 So we see that Valdoˆtain differs from French in the nature of its proclitic subject clitics, and hence allows examples like (34) and (35). On the other hand, Valdoˆtain is like French in having the interrogative conjugation. Hence this is a null-subject language at both the T- and C-levels. 2.3

Veneto and other Northern Italian dialects

The idea that the apparent enclitic subject pronoun in contexts of subject– clitic and complex inversion is really a manifestation of the j- or D-features associated with an inversion-triggering ‘residual V2’ C, which I will continue to refer as the interrogative-conjugation analysis, carries over quite successfully to a number of Northern Italian dialects, notably the Veneto dialects. Here, as was observed by Renzi & Vanelli (1983), we find a distinct clitic paradigm in inversion contexts. Poletto (2000: 53f.) argues in detail for an analysis along these lines (but see Cardinaletti & Repetti (2006b) for a different perspective). Poletto (2000: 53f.) observes, across a range of Northern Italian dialects, differences in the form of proclitic vs. enclitic subject clitics, differences in the number of gaps in the paradigm, differences in the apparent ability of proclitics and enclitics to be associated with null subjects, and the possibility of cooccurrence of proclitic and enclitic subject clitics. All of these, she argues, indicate that the best synchronic analysis involves treating the two series as structurally distinct, with, in the terms adopted here, the enclitics being a realisation of interrogative conjugation. (Note, in fact, that three of Poletto’s criteria are satisfied in French, as analysed above: there are differences in the form of the 3SG clitics, since the enclitics have an initial /t-/ (the ‘epenthetic’ /t/); there is a (near) gap in the 1SG slot of the enclitic paradigm, but no such gap in the proclitic paradigm, and the enclitics license null subjects but not the proclitics; on the other hand, there is no possibility of cooccurrence of proclitics and enclitics.) The Veneto variety of Loreo illustrates differences in form between proclitics and enclitics: (39) Proclitic: Enclitic:

15

1 a ia

2 a te to

3 el/la lo/la

4 a ia

5 a o

6 i/le li/le

(Poletto 2000: 54)

Poletto (2000: 59–62) also gives examples of what she calls deictic proclitics along with subject enclitics; she treats these as the realisations of different heads in the C-system.

322

Ian Roberts

Paduan illustrates clearly the difference in the number of subject clitics that are realised as a function of proclisis or enclisis: (40) Proclitic: Enclitic:

1  i

2 te to

3 el/la lo/la

4  i

5  o

6 i/le li/le

(Poletto 2000: 54)

In this connection, one of the generalisations originally made by Renzi & Vanelli (1983) is relevant: If interrogative sentences are formed via subject inversion, (i) the number of enclitic pronouns found in interrogative sentences is equal to or greater than the number of proclitic pronouns in declarative sentences, and (ii) the subject pronouns found in proclitic position are also found in enclitic position. (Renzi & Vanelli’s (1983) Generalisation 9)

This generalisation has proven fairly robust (although not without counterexamples, see below; see also Cardinaletti & Repetti 2006b).16 In terms of Chomsky’s (2008) proposal for feature inheritance from C to T, we could perhaps interpret it to mean that the realisation of j-features in the phase-head is consistently ‘richer’, in the sense that more features have a distinct realisation, than in the inheriting position. Why this should be remains unclear, however.17

16

17

A further possibility is that some or all persons of the putative interrogative conjugation show syncretism. Manzini & Savoia (2005, I: 564ff.) report a good number of cases of this type. Where person distinctions have been entirely lost in the enclitic, we can assume that it has been analysed as a ‘pure’ Q-morpheme. This has been proposed several times for Colloquial French tu/ti (Roberge 1986; Roberge & Vinet 1989; Roberts 1993b); there are also a number of Franco-Provenc¸al varieties which show this (see Roberts 1993a). In these varieties, the Q-marker appears in T and there is no T-to-C movement and therefore no subject–clitic inversion of any kind. Whether this analysis can be maintained for all the varieties discussed by Manzini & Savoia remains to be seen. This interpretation of Renzi & Vanelli’s generalisation may seem to contradict the often-made observation that agreement between a verb in C and a subject in TP is frequently less specified than that between a verb in T and a subject in TP (see, for example, Guasti & Rizzi 1999). But the latter observation may fall under a proposal made by D’Alessandro & Roberts (2008) regarding the morphophonological realisation of Agree relations: (i)

Overt morphophonological agreement between a and b is realised iff a and b are contained in the complement of the minimal phase-head H.

Where the verb is in C and the subject in TP, the agreeing elements are not both contained in the complement of the minimal phase-head, since for the subject this is C. Hence agreement may not be realised (or, weakening (i) slightly, may be realised in a more impoverished fashion). Where both the subject and the verb are inside TP, they are in the complement of the same minimal phase-head C, and hence ‘full’ overt agreement may be realised. The greater ‘richness’ of agreement in C can thus only be seen, given (some version of) (i), where the subject is raised into the C-system. If we compare the French interrogative conjugation, as seen in complex inversion and subject-clitic inversion as analysed above, with the ‘standard’ TP-internal subject–verb agreement, we witness a dramatic difference in

Varieties of French and the Null Subject Parameter

323

Poletto (2000: 54) gives evidence that dialects vary in the association of enclitic and proclitic subject clitics with apparent null subjects, as follows: (41)

Montesover Trentino: a. Piove. Rains ‘It’s raining.’ b. Piove-l? rains it ‘Is it raining?’

(42)

Friulian (Collina): a. Al plof. It rains ‘It’s raining.’ b. Plof? Rains? ‘Is it raining?’

Poletto (2000) shows that in Montesover Trentino subject (pro)clitics only appear with a definite, referential subject; hence there is no subject proclitic in (41a), but presumably just a quasi-argumental pro, attracted to SpecTP by the j-features and EPP-features T inherits from C. In (41b), on the other hand, -l is a realisation of the 3SG features interrogative C does not pass to T. Here, presumably, quasi-argumental pro is licensed by these features, either in SpecCP or in SpecvP (but not in SpecTP, as there is no EPP feature associated with this position where there is no feature inheritance from C). The Friulian examples in (42) appear to show realisation of the 3SG features T inherits from C as al, and no realisation of these features in C, apparently in violation of part (ii) of Renzi & Vanelli’s generalisation (and therefore inconsistent with the speculation in footnote 17 as to what this may follow from). There are also cases of stem allomorphy on the verb. In Paduan, the final -a of the 3SG present tense of a first-conjugation verb raises to /e/ when a subject enclitic is present: el magna (‘he eats’) vs. magne-lo (‘eats he?’). Given Zwicky & Pullum’s (1983) observation that such stem allomorphy is characteristic of affixation rather than cliticisation, this supports the idea that subject clitics are to be analysed as interrogative conjugation rather than as actual clitics. Cardinaletti & Repetti (2006b) give further examples of this from Bellunese, Veronese and Venetian.18

18

‘richness’, to the point where interrogative conjugation licenses referential null subjects, while the ‘declarative conjugation’ realised on T at best only partially allows them. Cardinaletti & Repetti (2006b: 38–39) point out that a similar alternation is found with enclitic object pronouns in imperatives in some of these dialects, at least. Assuming that the verb moves to C in (positive) imperatives in these varieties, this suggests that the verb has a distinct inflection when in C. This weakens the argument that the enclitics are inflections, but is

324

Ian Roberts

Cardinaletti & Repetti (2006b) in fact argue that there is no need for a ‘twoparadigm’ analysis of subject proclitics and enclitics of the type sketched above and argued for in Sportiche (1999), Poletto (2000) and others, as we have seen. They show that the differences in form between proclitics and enclitics (at least in the Emilian dialect of Donceto and in Paduan) can be handled by regular phonological processes, and argue that the differing occurrence of proclitics and declarative clauses and enclitics in interrogatives can be keyed to the extent of verb-movement (essentially, clitics are preferred to null subjects, given Cardinaletti & Starke’s (1999) Minimise Structure principle combined with the idea that clitics are heads while pro is a DP, and cliticisation will occur whenever the verb moves to a structurally high enough position). This last idea cannot be stated given the approach to cliticisation in Roberts (forthcoming b), adopted here: on this approach cliticisation is in principle independent of verb-movement, although the two often interact. Thus an analysis of the kind put forward by Cardinaletti & Repetti is not available to us. The fact that the proclitic and enclitic forms may be derivable from a single underlying form by regular phonological processes does not in itself argue against a ‘two-paradigm’ approach, at least of the kind suggested above. The proclitic and enclitic forms are realisations of the same j-sets; all that is proposed here is that the proclitics are the realisation of a j-set associated with a D-feature, while the enclitics are a realisation of the same j-set associated with a Q-feature. Moreover, it is not in dispute that the two paradigms have a single diachronic origin, and Cardinaletti & Repetti do suggest that some of the phonological processes, notably the syncretisms found in certain dialects, are diachronic. A final question concerning subject enclisis in French and Northern Italian dialects concerns the general absence of complex inversion of the type illustrated by the French sentence in (27) in Northern Italian dialects, as the following Veneto and Friulian examples, from Poletto (1990), illustrate: (43)

a. *Quando Gianni vienlo? When John will-come.he? b. *Quant Gianni isal vegnut? when John is-he come

(Veneto) (Friulian)

These should be contrasted with the grammatical French sentence in (42): (44)

Quand Jean est-il venu? When John is-he come? ‘When did John come?’

consistent with what is found in various Germanic V2 languages; see for example Zwart (1997) on Dutch.

Varieties of French and the Null Subject Parameter

325

The central plank of our analysis of the French construction is the idea that, in the relevant residual V2 contexts, C does not transfer its features to T. Therefore, there are no j-features in T, there is no Agree relation between T and the subject, and no reason for the subject to raise to SpecTP. Instead, the relevant Agree relation holds between C’s j-set and the subject, and the subject may be attracted to SpecCP, giving rise to complex inversion. The morphophonological reflex of the retention of C’s j-features is their realisation as the apparent subject enclitic, giving rise to the ‘interrogative conjugation’. Since many Northern Italian dialects show the interrogative conjugation, as we have just seen, why do they lack complex inversion? The only speculation that I have to offer relies on the separation of Agree from Move, the latter occurring only where the probe of an Agree relation bears an EPP feature. Implicit in the account of French complex inversion sketched above was the idea that C retains the EPP feature normally associated with T, and hence there is effectively a subject position in SpecCP (along with a wh-position, as (44) illustrates). It may be that Northern Italian dialects lack an EPP feature on C; it is also lacking on T in interrogatives (if not, it could never be satisfied as T cannot act as a probe since it has no j-features in these contexts). Therefore the subject cannot raise to either SpecTP or SpecCP. It may be that it can Agree with C’s j-set and remain in SpecvP; this may be the key to the analysis of ‘free inversion’ in these varieties, if we allow for massive leftward-movement of (at least) VP (see Kayne & Pollock (2001/2005), where a slightly different version of this idea is developed for Stylistic Inversion in French). 2.4

Conclusion

What I have tried to show here is that French, Valdoˆtain and certain Veneto varieties analysed by Poletto all show interrogative conjugation; that is, ‘rich’ agreement on interrogative C in residual V2 environments, the environments where C withholds its j-features from T in Chomsky’s (2008) terms. This analysis accounts best for the facts of French subject–clitic and complex inversion, largely following Sportiche (1999). A consequence of this analysis of French enclitic subject pronouns is that French allows referential null subjects in all persons in this context. Valdoˆtain and Veneto allow referential null subjects also with proclitic subject pronouns, unlike French. 3

Conclusion

Concerning the specific question of the status of null subjects in French, we conclude that French is a (very) partial null-subject language at the TP-level in registers which allow (14a) and Stylistic Inversion; otherwise it

326

Ian Roberts

is a non-null-subject language at this level. In interrogatives and other residual-V2 environments, it is a consistent null-subject language, owing to the interrogative conjugation. There are registers, including probably Quebec French, in which French has no null-subject property at the TP-level, but retains consistent null subjects in the residual-V2 environment. Furthermore, there are very colloquial registers in which complex inversion and subject– clitic inversion are lacking: these are fully non-null-subject systems. Finally, according to a number of authors (see Lambrecht 1981; Roberge & Vinet 1989; Zribi-Hertz 1994; and the references given there) in the variety which Zribi-Hertz (1994: 137) refers to as franc¸ais tre`s e´volue´ (FTE, or ‘very evolved French’) subject proclitics are to be analysed as realising j-features of T rather as in certain Northern Italian dialects and Valdoˆtain, as we saw in Section 2.2 and 2.3. This is shown by the fact that they cooccur with nonreferentially quantified subjects, as in the attested examples in (45): (45)

a. Tout le monde il est beau, tout le monde il est gentil. (film title) Everyone he is handsome, everyone he is nice ‘Everyone is handsome, everyone is nice.’ b. Personne il fiche rien, a` Toulon. No-one he does anything at Toulon ‘No-one does anything in Toulon.’ (Zribi-Hertz 1994: (19a, e), 137); (67b) from P. Mille Barnavaux et quelques femmes, 1908)

Further evidence for the clitic status of these pronouns comes from the fact that they cannot be dropped under coordination: (46)

a. Il mangera beaucoup de viande et *(il) boira du bon vin. He will-eat much of meat and (he) will-drink of-the good wine ‘He will eat a lot of meat and drink good wine.’ b. Il a mange´ beaucoup de viande et ** (il) a bu du bon vin. He has eaten much of meat and (he) has drunk of-the good wine ‘He has eaten a lot of meat and drunk good wine.’ (Zribi-Hertz 1994: (21), 138)

As Zribi-Hertz points out, if il realises T’s j-features here, rather than being a pronoun, then an example like il parle must feature a null subject. Hence FTE is a null-subject language at the TP-level. We see, then, that there is a range of synchronic variation across dialects and sociolects of French. There are at least four register levels: (47)

(i) ‘high’ registers: allowing (14a) and Stylistic Inversion: (very) partial null-subject system in TP, fully null-subject in CP; (ii) registers not allowing (14a) and Stylistic Inversion: non-null-subject in TP, consistently null-subject in CP;

Varieties of French and the Null Subject Parameter

327

(iii) colloquial registers in which complex inversion and subject–clitic inversion are lacking: fully non-null-subject systems; (iv) the variety/varieties seen in (43) and (44) in which subject proclitics are to be analysed as realising j-features of T rather as in some Northern Italian dialects and Valdoˆtain: fully null-subject in TP.

The variety in (47i) is formal literary French, or perhaps franc¸ais standard moderne (FSM) in the terminology of Zribi-Hertz (1994: 136), which she describes as “the productive standard language, without archaisms, i.e. wellformed constructions of Classical French which nowadays are no longer productive”.19 This variety also makes use of the cases of careful, optional liaison discussed in Section 1. The variety in (47ii) may correspond to Quebec French (Kayne & Pollock 2001/2005: 13), as well as slightly less formal varieties than (47i) spoken elsewhere. Both (47iii) and (47iv) represent the colloquial spoken French often referred to as franc¸ais avance´. More precisely, we can perhaps identify (47iii) with Zribi-Hertz’ (1994: 137) franc¸ais parle´ courant (‘common spoken French’). Zribi-Hertz points out that franc¸ais avance´ lacks enclitic subject pronouns, since it lacks subject–clitic and complex inversion. Finally, variety (45iv) is Zribi-Hertz’ FEC. This synchronic variation across dialects and sociolects of French clearly requires further investigation, particularly in relation to the status of optional liaison, given the importance of this phenomenon to determining the underlying phonological forms of certain inflections and the implications of those forms for determining system-defining syncretisms in the sense discussed in Section 1 above. It seems likely that this kind of optional liaison is unavailable in franc¸ais avance´, and possibly in variety (47ii). Other syntactic features may correlate with these distinctions: franc¸ais avance´ lacks the preverbal negative element ne (as can in fact be observed in (45b)), and may have systematic wh-in-situ in root clauses, for example. The approach to null subjects outlined in Roberts (Chapter 1) allows us to clarify the situation in French in a useful way. We can see that our results are fully consistent with the general deletion approach to null subjects and Mu¨ller’s pro-generalisation. 19

“la langue standard productive, c’est-a`-dire de´barrasse´e des ses archaı¨smes, i.e. des tournures bien forme´es du franc¸ais classique devenues improductives aujourd’hui” (Zribi-Hertz 1994: 136); my translation.

Bibliography

Abels, K. 2008. Towards a restrictive theory of (remnant) movement. Linguistic Variation Yearbook 7: 53–120. Abels, K. and A. Neeleman. 2006. Universal 20 without the LCA. Unpublished ms: UCL. (Available on-line at: http://www.phon.ucl.ac.uk/home/klaus/papers/ KlausandAdGLOWv2.pdf) Abraham, W. 1993. Null subjects in the history of German: from IP to CP. Lingua 89: 117–42. Adams, J. 2003. Bilingualism and the Latin language. Cambridge: Cambridge University Press. Adams, M. 1987a. From Old French to theory of pro-drop. Natural Language and Linguistic Theory 5: 1–32. 1987b. Old French, null subjects and Verb-Second phenomena. PhD dissertation: UCLA. Adger, D. 2006a. Combinatorial variation. Journal of Linguistics 42: 503–30. 2006b. Post syntactic movement and the Old Irish Verb. Natural Language and Linguistic Theory 24: 605–54. 2007. Stress and phasal syntax, Linguistic Analysis 33: 238–66. Adraga˜o, M. and J. Costa. 2004. On the status of preverbal subjects in null subject languages: evidence from acquisition. In J. van Kampen and S. Baauw (eds) Proceedings of GALA 2003. Utrecht: LOT Occasional Series, 69–80. Akmajian, A., S. Steele and T. Wasow 1979. The category AUX in Universal Grammar. Linguistic Inquiry 10:1–64. Aldridge, E. 2004. Ergativity and word order in Austronesian languages. PhD dissertation: Cornell. 2006. The heterogeneity of VOS and extraction in Austronesian languages. Unpublished ms: Northwestern University. (Available on-line at: http://faculty. washington.edu/eca1/VOS.pdf) Alexiadou, A. 1997. Adverb placement: a case study in antisymmetric syntax. Amsterdam: Benjamins. Alexiadou, A. and E. Anagnostopoulou. 1995. SVO and EPP in null subject languages and Germanic. FAS Papers in Linguistics 4:1–21. 1998. Parametrizing AGR: word order, V-movement and EPP checking. Natural Language and Linguistic Theory 16: 491–539. 2001. The subject-in-situ generalization and the role of case in driving computations, Linguistic Inquiry 32: 193–231.

328

Bibliography

329

2007. The subject-in-situ generalization revisited. In U. Sauerland and H-M. Ga¨rtner (eds) Interfaces þ Recursion ¼ Language? Chomsky’s minimalism and the view from syntax and semantics. Berlin: Mouton de Gruyter, 31–60. Alexiadou, A. and G. Mu¨ller. 2005. Class features as probes. Ms, Universita¨t Stuttgart and Universita¨t Leipzig. Allen, C. 1995. Case marking and reanalysis. Grammatical relations from Old to Early Modern English. Oxford: Oxford University Press. Ambar, M. 1992. Para uma sintaxe da invers~ ao sujeito verbo em portugu^ es. Lisbon: Edicoes Colibri. Anderson, S. 1982. The analysis of French schwa, or how to get something for nothing. Language 58: 534–73. Ariel, M. 1990. Accessing noun phrase antecedents. London/New York: Routledge. Armstrong, L. 1962. The phonetics of French. A practical handbook. London: Bells & Son. Baker C. 1970. Notes on the description of English questions: the role of an abstract question morpheme. Foundations of Language 6: 197–219. Baker, M. 1988. Incorporation: a theory of grammatical function changing. Chicago: University of Chicago Press. 1996. The polysynthesis parameter. Oxford: Oxford University Press. 2000. The natures of nonconfigurationality. In M. Baltin and C. Collins (eds) The handbook of contemporary syntactic theory. Oxford: Blackwell, 407–438. 2001. The atoms of language. Oxford: Oxford University Press. 2003a. Lexical categories. Verbs, nouns and adjectives. Cambridge: Cambridge University Press. 2003b. Building and merging, not checking: the non-existence of (Aux)-S-V-O languages. Linguistic Inquiry 33: 321–29. 2008a. The syntax of agreement and concord. Cambridge: Cambridge University Press. 2008b. The macroparameter in a microparametric world. In T. Biberauer (ed.) The limits of syntactic variation. Amsterdam: Benjamins, 351–74. Baker, M., K. Johnson and I. Roberts 1989. Passive arguments raised. Linguistic Inquiry 20: 219–51. Barber, C. 1976. Early Modern English. London: Deutsch. Barbiers, S. 2005. Word order variation in three-verb clusters and the division of labour between genarative linguistics and sociolinguistics. In L. Cornips and K. Corrigan (eds) Syntax and variation. Reconciling the biological and the social. Amsterdam: Benjamins, 233–64. Barbosa, P. 1995. Null subjects. PhD dissertation: MIT. 2009. Two kinds of subject pro. Studia Linguistica 63: 2–58. Basilico, D. 1997. The topic is ‘there’. Studia Linguistica 51: 278–316. Battistella, E. and A. Lobeck. 1988. An ECP account of verb second in Old English. Proceedings of the Conference on the Theory and Practice of Historical Linguistics. Bayer, J. 1984. COMP in Bavarian syntax. The Linguistic Review 3: 209–74. Belletti, A. 1988. The case of unaccusatives. Linguistic Inquiry 19: 1–34. 2001. Inversion as focalisation. In A. Hulk and J-Y. Pollock (eds) Subject inversion in Romance and the theory of Universal Grammar. New York: Oxford University Press, 60–91.

330

Bibliography

2004. Aspects of the low IP area. In L. Rizzi (ed.) The structure of CP and IP. The cartography of syntactic structures. New York: Oxford University Press, 16–51. 2005. Extended doubling and the VP periphery. Probus 17: 1–35. Belletti, A., E. Bennati and A. Sorace. In press. Theoretical and developmental issues in the syntax of subjects: Evidence from near-native Italian. Natural Language and Linguistic Theory. Belletti, A. and U. Shlonsky. 1995. The order of verbal complements in Italian and in Hebrew, Natural Language and Linguistic Theory 13: 489–526. Beninca`, P. 1988. Soggetto Posverbale. In L. Renzi (ed.) Grande grammatica italiana di consultazione. Bologna: Il Mulino, 122–26. Beninca`, P. and C. Poletto. 2005. On some descriptive generalizations in Romance. In G. Cinque and R. Kayne (eds) The Oxford handbook of comparative syntax. Oxford: Oxford University Press, 221–58. Bennis, H. 1986. Gaps and dummies. Dordrecht: Foris. Bentzen, K. 2007. Order and structure in embedded clauses in Northern Norwegian. PhD dissertation: Troms. Berwick, R. 1985. The acquisition of syntactic knowledge. Cambridge, MA: MIT Press. den Besten, H. 1983. On the interaction of root transformations and lexical deletive rules. In W. Abraham (ed.) On the formal syntax of the Westgermania. Amsterdam: Benjamins, 47–131. Bhatt, R. and R. Pancheva. 2004. Late merger of degree clauses. Linguistic Inquiry 35: 1–47. Biberauer, T. 2003. Verb second (V2) in Afrikaans: a minimalist investigation of word-order variation. PhD dissertation: University of Cambridge. 2004. Reconsidering Spec-TP in Germanic. Cambridge Occasional Papers in Linguistics 1: 15–40. 2008a. Introduction. In T. Biberauer (ed.) The limits of syntactic variation. Amsterdam: Benjamins, 1–72. (ed.) 2008b. The limits of syntactic variation. Amsterdam: Benjamins. 2009. How “well-behaved” is Afrikaans? V2 in Modern Spoken Afrikaans. In H. den Besten, F. Hinskens and J. Koch (eds) Ein Tryptichon. Amsterdam: Voortgang, 176–208. Biberauer, T. and R. D’Alessandro. 2006. Syntactic doubling and the encoding of voice in Eastern Abruzzese. In D. Baumer, D. Montero and M. Scanlon (eds) Proceedings of the 25th West Coast Conference on Formal Linguistics. Somerville, MA: Cascadilla Press, 87–95. Biberauer, T., A. Holmberg and I. Roberts. 2007. Disharmonic word-order systems and the Final-over-Final-Constraint (FOFC). In A. Bisetto, A. Barbieri and F. Barbieri (eds) Proceedings of XXXIII Incontro di Grammatica Generativa. (Available on-line at:http://amsacta.cib.unibo.it/archive/00002397/01/ PROCEEDINGS_IGG33.pdf) Biberauer, T. and M. Richards. 2006. True optionality: when the grammar doesn’t mind. In C. Boeckx (ed.) Minimalist essays. Amsterdam: Benjamins, 35–67. Biberauer, T. and I. Roberts. 2004. Evidence that V2 involves two movements: a reply to Mu¨ller. Cambridge Occasional Papers in Linguistics 1: 41–62. 2005. Changing EPP-parameters in the history of English: accounting for variation and change. English Language and Linguistics 9, 1: 5–46.

Bibliography

331

2006. The loss of V-Aux orders and remnant fronting in Late Middle English: causes and consequences. In J. Hartmann and L. Molna´rfi (eds) Comparative studies in Germanic Syntax: From A(frikaans) to Z(€ urich German). Amsterdam: Benjamins, 263–97. 2008. Cascading parameter changes: internally driven change in Middle and Early Modern English. In Th. Eytho´rsson (ed.) Grammatical change and linguistic theory: the Rosendal papers. Amsterdam: Benjamins, 79–113. 2009. The return of the subset principle. In P. Crisma and G. Longobardi (eds) Historical syntax and linguistic theory. New York: Oxford University Press, 8–74. Blevins, J. 2003. Passives and impersonals. Journal of Linguistics 39: 473–520. Bobaljik, J. 1995. Morphosyntax: the syntax of verbal inflection. PhD dissertation: MIT. 2002. A-chains at the PF-interface: copies and ‘covert’ movement. Natural Language and Linguistic Theory 20: 197–267. Bobaljik, J. and D. Jonas. 1996. Subject positions and the roles of TP. Linguistic Inquiry 27: 195–236. Bobaljik, J. and I. Landau. 2007. Fact and fiction in Icelandic control. To appear in Linguistic Inquiry. Bobaljik, J. and H. Thra´insson. 1998. Two heads aren’t always better than one. Syntax 1: 37–71. Boden, M. 2006. Mind and machine: The history of cognitive science, Volume I. Oxford: Oxford University Press. Boeckx, C. 2003. Islands and chains. Resumption as derivational residue. Amsterdam: Benjamins. Forthcoming. Approaching parameters from below. In A-M. Di Sciullo and C. Boeckx (eds) Biolinguistic approaches to language evolution. Oxford: Oxford University Press. Boeckx, C. and N. Hornstein. 2003. Reply to ‘Control is not Movement’. Linguistic Inquiry 34: 269–80. 2004. Movement under control. Linguistic Inquiry 35: 431–52. Bolinger, D. 1977. Meaning and form. London: Longman. Borer, H. 1984. Parametric syntax. Case studies in Semitic and Romance languages. Dordrecht: Foris. 1986. I-Subjects. Linguistic Inquiry 17: 375–416. 1989. Anaphoric Agr. In O. Jaeggli and K. Safir (eds) The null subject parameter. Dordrecht: Kluwer, 69–109. Borsley, R., M. Tallerman and D. Willis. 2007. The syntax of Welsh. Cambridge: Cambridge University Press. Bosˇkovic´, Zˇ. 2001. On the nature of the syntax-phonology interface: cliticization and related phenomena. Amsterdam: Elsevier. 2002. On multiple wh-fronting. Linguistic Inquiry 33: 351–83. Bosˇkovic´, Zˇ. and J. Nunes. 2007. The copy theory of movement: A view from PF. In N. Corver and J. Nunes (eds) The copy theory of movement. Amsterdam: Benjamins, 13–74. Brandi, L. and P. Cordin. 1989. Two Italian dialects and the null subject parameter. In O. Jaeggli and K. Safir (eds) The null subject parameter. Dordrecht: Kluwer, 111–42. Breivik, L. 1990. Existential ‘there’: a synchronic and diachronic study. Oslo: Novus Press.

332

Bibliography

Bresnan, J. 1971. Sentence stress and syntactic transformations. Language 47: 257–81. 1994. Locative inversion and the architecture of universal grammar. Language 70: 72–131. Broekhuis, H. 2000. Against feature strength: the case of Scandinavian object shift. Natural Language and Linguistic Theory 18: 673–721. Broekhuis, H., N. Corver, R. Huybregts, U. Kleinhenz and J. Koster (eds.). 2005. Organising grammar: linguistic studies in honor of Henk van Riemsdijk. Berlin: De Gruyter. Burridge, K. 1983. Syntactic change in Germanic. Aspects of language change in Germanic with particular reference to Middle Dutch. Amsterdam: Benjamins. Burton-Roberts, N. and G. Poole. 2006a. Syntax vs. phonology: a representational approach to stylistic fronting and verb-second in Icelandic. Lingua 116: 562–600. 2006b. ‘Virtual conceptual necessity’, feature dissociation, and the Saussurian legacy in generative grammar. Journal of Linguistics 42: 575–628. Burzio, L. 1981. Intransitive verbs and Italian auxiliaries. PhD dissertation: MIT. 1986. Italian syntax: a Government-Binding approach. Dordrecht: Kluwer. Cabredo-Hofherr, P. 2004. Impersonal pronouns in Somali, German and French. Paper presented at Syntax of the World’s Languages 1, Leipzig. Calder, G. 1990. A Gaelic grammar. Glasgow: Gairm Press. Cardinaletti, A. 1990. Impersonal constructions and sentential arguments in German. Padua: Unipress. 1997. Subjects and clause structure. In L. Haegeman (ed.) The new comparative syntax. London: Longman, 33–63. 2002. Against optional and zero clitics. Right dislocation and marginalization. Studia Linguistica 56: 29–57. 2004. Toward a cartography of subject positions. In L. Rizzi (ed.) The structure of CP and IP. The cartography of syntactic structures. Oxford: Oxford University Press, 115–65. Cardinaletti, A. and L. Repetti. 2006a. Proclitic vs. enclitic pronouns in Northern Italian dialects. Talk given at the 1st Cambridge Italian Dialect Syntax Meeting. 2006b. Subject clitics in interrogative sentences: phonology, syntax and microvariation. Unpublished ms: SUNNY Stony Brook/University of Venice. Cardinaletti, A. and M. Starke. 1999. The typology of structural deficiency: A case study of the three classes of pronouns. In H. van Riemsdijk (ed.) Clitics in the languages of Europe. Berlin: Mouton de Gruyter, 145–235. Carilho, E. 2008. Beyond doubling: overt expletives in European Portuguese dialects. In S. Barbiers, M. van der Ham, O. Koeneman and M. Lekakou (eds) Microvariations in syntactic doubling. Bingley: Emerald, 299–322. Carnie, A., S. Dooley and H. Harley (eds). 2005. Verb first. On the syntax of verb-initial languages. Amsterdam: Benjamins. Carnie, A. and E. Guilfoyle (eds). 2000. The syntax of verb-initial languages. New York: Oxford University Press. Carnie, A., H. Harley and E. Pyatt. 2000. VSO order as raising to Comp. In A. Carnie and E. Guilfoyle (eds) The syntax of verb-initial languages. Oxford: Oxford University Press, 39–60. Cecchetto, C. 1999. A comparative analysis of left and right dislocation in Romance. Studia Linguistica 53: 40–67.

Bibliography

333

2000. Doubling structures and reconstruction. Probus 12: 93–126. Cecchetto, C. and C. Donati. 2007. On labelling: Principle C and head movement. Studies in Linguistics, Volume 1: 16–37. Chenal, A. 1986. Le franco-provenc¸al vald^ otain. Aosta: Musumeci Editeur. Chierchia, G. 1998. Reference to kinds across languages. Natural Language Semantics, 6(4): 339–405. Chomsky, N. 1957. Syntactic structures. The Hague: Mouton. 1964. Current issues in linguistic theory. The Hague: Mouton. 1971. Deep structure, surface structure, and semantic interpretation. In D. Steinberg and L. Jacobovits (eds) Semantics, 183–216. 1973. Conditions on transformations. In S. Anderson and P. Kiparsky (eds) A Festschrift for Morris Halle. New York: Holt, Reinhart and Winston, 232–86. 1981. Lectures on Government and Binding. Dordrecht: Foris. 1982. Some concepts and consequences of the theory of Government and Binding. Cambridge, MA: MIT Press. 1986. Knowledge of language: its nature, origin, and use. Praeger: New York. 1991. Some notes on economy of derivation and representation. In R. Freidin (ed.) Principles and parameters in comparative grammar. Cambridge, MA: MIT Press, 117–154. 1993. A minimalist program for linguistic theory. In K. Hale and J. Keyser (eds) The view from Building 20: essays in Linguistics in honor of Sylvain Bromberger. Cambridge, MA: MIT Press, 1–52. 1995a. Bare phrase structure. In G. Webelhuth (ed.) Government and Binding theory and the minimalist program. Oxford: Blackwell, 383–439. 1995b. The minimalist program. Cambridge, MA: MIT Press. 2000. Minimalist inquiries: the framework. In R. Martin, D. Michaels and J. Uriagereka (eds) Step by step: essays on minimalist syntax in honor of Howard Lasnik. Cambridge, MA: MIT Press, 89–156. 2001. Derivation by phase. In M. Kenstowicz (ed.) Ken Hale: a life in language. Cambridge, MA: MIT Press, 1–53. 2002. On nature and language. Cambridge: Cambridge University Press. 2004. Beyond explanatory adequacy. In A. Belletti (ed.) Structures and beyond: the cartography of syntactic structures, Volume 3. Oxford: Oxford University Press, 104–13. 2005. Three factors in language design. Linguistic Inquiry 36: 1–22. 2007. Approaching UG from below. In U. Sauerland and G.-M. Ga¨rtner (eds) Interface þ Recursion ¼ Language? Chomsky’s minimalism and the view from syntax and semantics. Berlin: Mouton de Gruyter, 1–29. 2008. On phases. In R. Freidin, C. Otero and M-L. Zubizarreta (eds) Foundational issues in linguistic theory. Cambridge, MA: MIT Press, 133–66. Chomsky, N. and M. Halle. 1968. The sound pattern of English. New York: Harper and Row. Chomsky, N. and H. Lasnik. 1977. Filters and control. Linguistic Inquiry 8: 425–504. Chung, S. 2005. What fronts? On the VP raising account of verb-initial order. In A. Carnie, H. Harley and S. Dooley (eds) On the syntax of verb-initial languages. Amsterdam: Benjamins, 9–29.

334

Bibliography

Chung, S. and W. Ladusaw. 2003. Restriction and saturation. Cambridge, MA: MIT Press. Churchward, C. 1953. Tongan Grammar. Oxford: Oxford University Press. Cinque, G. 1988. On si constructions and the theory of ARB. Linguistic Inquiry 19: 521–81. 1993. A null theory of phrase and compound stress. Linguistic Inquiry 24: 239–97. 1999. Adverbs and functional heads. A crosslinguistic perspective. New York: Oxford University Press. Clackson, J. 2004. Latin. In R. Woodward (ed.) The Cambridge encyclopedia of the world’s ancient languages. Cambridge: Cambridge University Press, 789–811. Clark, R. and I. Roberts. 1993. A computational approach to language, learnability and language change. Linguistic Inquiry 24: 299–345. Cole, M. Forthcoming. Null Subjects: A reanalysis of the data. Costa, J. 2001a. Marked versus unmarked inversion and optimality theory. In A. Hulk and J-Y. Pollock (eds) Subject inversion in Romance and the theory of Universal Grammar. Oxford: Oxford University Press, 91–106. 2001b. Postverbal subjects and agreement in unaccusative contexts in European Portuguese. The Linguistic Review 18: 1–17. 2004. Subject positions and interfaces: the case of European Portuguese. Berlin: Mouton de Gruyter. Cottell, S. 1995. The representation of tense in modern Irish. Geneva Generative Papers 3: 105–24. Cowper, E. and D. Currie Hall. 2008. Argumenthood, pronouns and nominal feature geometry. To appear in J. Ghomeshi, I. Paul and M. Wiltschko (eds) Determiners: variation and universals. Amsterdam: Benjamins. Croft, W. 2003. Typology and universals. Cambridge: Cambridge University Press. Csato´, E. and L. Johansen. 1998. Turkish. In L. Johansen and E. Csato´ (eds) The Turkic languages. London: Routledge, 203–35. Culicover, P. 1999. Syntactic nuts. Oxford: Oxford University Press. Culicover, P. and R. Jackendoff. 2001. Control is not movement. Linguistic Inquiry 32: 493–512. Dahl, O. 1996. Predicate, subject and topic in Malagasy. Oceanic Linguistics 35, 2: 167–79. D’Alessandro, R. and I. Roberts. 2006. Split ergativity in Abruzzese and the null-subject parameter. Paper given at the 34th Romance Linguistics Seminar, University of Cambridge. (Available on-line at: http://www.robertadalessandro.it/ files/DAlessandro_Roberts_Abruzzese.pdf) 2008. Movement and agreement in Italian past participles and defective phases. Linguistic Inquiry 39: 477–91. Davies, W. and S. Dubinsky. 2001. Functional architecture and the distribution of subject properties. In W. Davies and S. Dubinsky (eds) Objects and other subjects: grammatical functions, functional categories, and configurationality. Dordrecht: Kluwer, 247–79. 2004. The grammar of raising and control. Oxford: Blackwell. Deal, A. 2008. The origin and content of expletives: evidence from ‘selection’. To appear in Syntax.

Bibliography

335

De´chaıˆne, R. and M. Wiltschko. 2002. Decomposing pronouns. Linguistic Inquiry 33: 409–42. Dell, F. 1985. Les r egles et les sons. Paris: Hermann. Diesing, M. 1992. Indefinites. Cambridge, MA: MIT Press. Diesing, M. and E. Jelinek. 1995. Distributing arguments. Natural Language Semantics 3: 123–76. den Dikken, M. 1995. Particles. On the syntax of verb-particle, triadic, and causative constructions. New York: Oxford University Press. 2007. Questionnaire study on Dutch that-trace effects: stimuli and results. Unpublished ms: CUNY. [Available on-line at: http://web.gc.cuny.edu/dept/ lingu/dendikken/docs/dutch_that_trace_results.pdf] Di Sciullo, A-M. 1990. On the properties of clitics. Binding in Romance. Carleton University, Ottawa: Canadian Linguistics Association. Doherty, C. 1996. Clausal structure and the modern Irish copula. Natural Language and Linguistic Theory 14: 1–46. 1997. Predicate initial constructions in Irish. In B. Agbayani and S-W. Tang (eds) Proceedings of the Fifteenth West Coast Conference on Formal Linguistics. Stanford: Center for the Study of Language and Information, 81–95. 2000. Residual verb second in Early Irish: on the nature of Bergin’s construction. Diachronica 17: 5–38. Donaldson, B. 1993. A grammar of Afrikaans. New York: Mouton de Gruyter. Doron, E. 2000. VSO and left-conjunct agreement: biblical Hebrew vs. modern Hebrew. In A. Carnie and E. Guilfoyle (eds) The syntax of verb initial languages. Oxford: Oxford University Press, 75–96. Dresher, E. 1992. A learning model for a parametric theory in phonology. In R. Levine (ed.) Formal grammar: theory and implementation. New York: Oxford University Press, 290–317. 1999. Charting the learning path: cues to parameter setting. Linguistic Inquiry 30: 27–68. Dryer, M. 1992. On the Greenbergian word-order correlations. Language 68: 81–138. 2005. Relationship between the order of object and verb and the order of adposition and noun phrase. In M. Haspelmath, M. Dryer, D. Gil and B. Comrie (eds) The world atlas of language structures, Oxford: Oxford University Press, 386–89. 2007. Review of Frederick J. Newmeyer, Possible and probable languages: a generative perspective on linguistic typology. Journal of Linguistics 43: 244–52. Duarte, E. 1995. A perda do principio evite pronome no portugu^ es brasileiro. PhD dissertation: Instituto de estudos da linguagem, UNICAMP. 2000. The loss of the ‘Avoid Pronoun’ principle in Brazilian Portuguese. In M. Kato and E. Negra˜o (eds) Brazilian Portuguese and the null subject parameter. Frankfurt: Vervuert, 17–36. Dupuis, F. 1989. L’expression du sujet dans le subordonn ees en ancien franc¸ais. PhD dissertation: Montreal. Egerland, V. 2003. Impersonal pronouns in Scandinavian and Romance. Working Papers in Scandinavian Syntax 71: 75–102. Ellega˚rd, A. 1953. The auxiliary ‘do’: the establishment and regulation of its use in English. Stockholm: Almqvist and Wiksell.

336

Bibliography

Emonds, J. 1976. A transformational approach to English syntax: root, structurepreserving and local transformations. New York: Academic Press. 1978. The verbal complex V-V’ in French. Linguistic Inquiry 9: 151–75. Epstein, S. and D. Seely. 2002. Derivation and explanation in the minimalist program. Malden, MA: Blackwell. 2006. Derivations in minimalism. Cambridge: Cambridge University Press. Evers, A. 1975. The transformational cycle of Dutch and German. PhD dissertation: University of Utrecht. Falk, C. 1993a. Non-referential subjects in the history of Swedish. PhD dissertation: University of Lund. 1993b. Non-referential subjects and agreement in the history of Swedish. Lingua 89: 143–80. Fassi Fehri, A. 1993. Issues in the structure of Arabic clauses and words. Dordrecht: Kluwer. Forthcoming. Arabic silent pronouns, person and voice. Forthcoming in Transactions of the Philological Society. Felser, C. 2004. Wh-copying, phases, and successive cyclicity. Lingua 114: 543–74. Ferraresi, G. 1991. Word order and phrase structure in Gothic. PhD dissertation: University of Stuttgart. 2005. Word order and phrase structure in Gothic. Leuven: Peeters Publishers. Ferreira, M. 2004. Hyperraising and null subjects in Brazilian Portuguese. MIT Working Papers in Linguistics 46: 57–85. Figueiredo Silva, M. 2000. Main and embedded null subjects in Brazilian Portuguese. In M. Kato and E. Negra˜o (eds) Brazilian Portuguese and the null subject parameter. Frankfurt: Vervuert, 127–45. Fischer, O., A. van Kemenade, W. Koopman and W. van der Wurff (eds). 2000. The syntax of early English. Cambridge: Cambridge University Press. Fitzpatrick, J. 2006. Deletion through movement. Natural Language and Linguistic Theory 24: 399–431. Fox, D. 2000. Economy and semantic interpretation. Cambridge, MA: MIT Press. Franks, R. 1995. Parameters of Slavic morphosyntax. New York: Oxford University Press. Frascarelli, M. 2007. Subjects, topics, and the interpretation of referential pro. Natural Language and Linguistic Theory 25: 691–734. Frascarelli, M. and R. Hinterho¨lzl. 2007. Types of topics in German and Italian. In S. Winkler and K. Schwabe (eds) On information structure, meaning and form. Amsterdam: Benjamins, 87–116. Freeze, R. 1992. Existentials and other locatives. Language 68: 553–95. 2001. Existential constructions. In M. Haspelmath, E. Ko¨nig, W. Oesterreicher and W. Raible (eds) Language typology and language universals, Volume 2. Berlin: Mouton de Gruyter, 941–53. Freidin, R. 2004. Syntactic structures redux. Syntax, 7,2: 101–27. Frey, W. 2001. About the whereabouts of indefinites. Theoretical Linguistics 27: 137–61. Frey, W. and K. Pittner. 1998. Zur Positionierung von Adverbialen im deutschen Mittelfeld. Linguistische Berichte 176: 489–534. Fuss, E. 2003. On the historical core of V2 in Germanic. Nordic Journal of Linguistics, 26,2: 195–231.

Bibliography

337

2005. The rise of agreement. A formal approach to the syntax and grammaticalization of agreement. Amsterdam: Benjamins. ´ . 2008. Phase theory and parametric variation. PhD dissertation: Gallego, A Universitat Auto`noma de Barcelona. van Gelderen, E. 1993. The rise of functional categories. Amsterdam: Benjamins. Gianollo, C., C. Guardiano and G. Longobardi. 2008. Three fundamental issues in parametric linguistics. In T. Biberauer (ed.) The limits of syntactic variation. Amsterdam: Benjamins, 109–42. Gildersleeve, B. and G. Lodge. 1895/1997. Latin grammar. Bristol: Bristol Classical Press. Gilligan, G. 1987. A cross linguistic approach to the pro-drop parameter. PhD dissertation: University of Southern California. Giorgi, A. and F. Pianesi. 1997. Tense and aspect. Oxford: Oxford University Press. Gold, E. 1999. Aspect, tense and the lexicon: expression of time in Yiddish. PhD dissertation: Toronto. Goodall, G. 2001. The EPP in Spanish. In W. Davies and S. Dubinsky (eds) Objects and other subjects: grammatical functions, functional categories, and configurationality. Dordrecht: Kluwer, 193–223. Gray, D. 1985. The Oxford book of Late Medieval prose and verse. Oxford: Oxford University Press. Greenberg, J. 1963/2007. Some universals of grammar with particular reference to the order of meaningful elements. In J. Greenberg (ed.) Universals of language. Cambridge, MA: MIT Press, 58–90. Reprinted in I. Roberts (ed.) 2007. Comparative grammar, Volume 1, 41–74. Gre´visse, M. 1980. Le bon usage. Paris: Glembloux. Grewendorf, G. 1989. Small pro in German. In G. Grewendorf and W. Sternefeld (eds) Scrambling and barriers. Amsterdam: Benjamins, 294–315. Grimshaw, J. and V. Samek-Lodovici. 1998. Optimal subjects and subject universals. In P. Barbosa, D. Fox, P. Hagstrom, M. McGinnis and D. Pesetsky (eds) Optimality and competition in syntax. Cambridge, MA: MIT Press, 193–219. Groat, E. 1999. Raising the case of expletives. In S. Epstein and N. Hornstein (eds) Working minimalism. Cambridge, MA: MIT Press, 27–44. Guasti, M.-T. and L. Rizzi. 1999. Agreement and tense as distinct syntactic positions: evidence from acquisition. Unpublished ms: Universita` di Siena. Guilfoyle, E., L. H. Hung and L. Travis. 1992. Spec of IP and Spec of VP: two subjects in Austronesian languages. Natural Language and Linguistic Theory 10: 375–414. Gutman, E. 2004. Third person null subjects in Hebrew, Finnish and Rumanian: an accessibility theoretic account. Journal of Linguistics 40: 463–90. Haeberli, E. 1999. On the word order ‘XP-subject’ in the Germanic languages. Journal of Comparative Germanic Linguistics 3: 1–36. 2002. Features, categories and the syntax of A-positions. Cross-linguistic variation in the Germanic languages. Dordrecht: Kluwer. Haegeman, L. 1990. Subject pronouns and subject clitics in West Flemish. The Linguistic Review 7: 333–63. 1992. Theory and description in generative syntax. Cambridge: Cambridge University Press.

338

Bibliography

2000. Adult null subjects in non pro-drop languages. In M.-A. Friedemann and L. Rizzi (eds) The acquisition of syntax: studies in comparative developmental linguistics. London: Longman, 129–69. 2006. Subject omission in present-day written English: on the theoretical relevance of peripheral data. Rivista di grammatica generativa 31. Haider, H. 1993. Deutsche Syntax Generativ. Tu¨bingen: Gunter Narr. 2000. OV is more basic than VO. In P. Svenonius (ed.) The derivation of VO and OV. Amsterdam: Benjamins, 45–67. 2005. Mittelfeld phenomena. Scrambling in Germanic. In M. Everaert and H. van Riemsdijk (eds) The Blackwell companion to syntax. Oxford: Blackwell, 204–74. Haider, H. and I. Rosengren. 1998. Scrambling. Sprache und Pragmatik 49: 1–104. 2003. Scrambling: non-triggered chain formation in OV languages. Journal of Germanic Linguistics 15,3: 203–67. Hakulinen, A. 1975. Suomen sita¨: pragmatiikan heijastuma syntaksissa. Sananjalka 17: 25–41. [Finnish sit€ a: A reflection of pragmatics in syntax.] Hakulinen, A. and L. Karttunen. 1973. Missing persons: on generic sentences in Finnish. Papers from the Ninth Regional Meeting of the Chicago Linguistic Society: 157–71. Hale, K. and S. Keyser. 2000. Prolegomenon to a theory of argument structure. Cambridge, MA: MIT Press. Hall, R. A. 1950. The reconstruction of Proto-Romance. Language 26: 6–27. Halle, M. and A. Marantz. 1993. Distributed morphology and the pieces of inflection. In K. Hale and J. Keyser (eds) The view from Building 20. Cambridge, MA: MIT Press, 111–76. Harris, M. 1978. The evolution of French syntax: a comparative approach. London: Longman. 1988. French. In M. Harris and N. Vincent (eds) The Romance languages. London: Routledge, 209–45. Hartmann, J. 2008. Expletives in existentials. PhD dissertation: Tilburg University. Haspelmath, M. 2001. The European linguistic area: Standard Average European. In M. Haspelmath, E. Ko¨nig, W. Oesterreicher and W. Raible (eds) Language typology and language universals. Berlin: Mouton de Gruyter, 1492–510. Haspelmath, M., M. Dryer, D. Gil and B. Comrie (eds) 2005. World atlas of language sructures. Oxford: Blackwell. Hauser, M., N. Chomsky and T. Fitch. 2002. The faculty of language: what is it, who has it, and how did it evolve? Science 298: 1569–79. Hawkins, J. 1983. Word order universals. New York: Academic Press. 1990. A parsing theory of word order universals. Linguistic Inquiry 21: 221–62. 1994. A performance theory of order and constituency. Cambridge: Cambridge University Press. 2004. Efficiency and complexity in grammars. Oxford: Oxford University Press. Hazout, I. 2004. The syntax of existential constructions. Linguistic Inquiry 35, 3: 393–430. Helasvuo, M-L. 2006. Passive – personal or impersonal. In M.-L. Helasvuo and L. Campbell (eds) Grammar from the human perspective. Amsterdam: Benjamins, 233–55.

Bibliography

339

Helasvuo, M-L. and L. Laitinen. 2006. Person in Finnish. Paradigmatic and syntagmatic relations and interactions. In M.-L. Helasvuo and L. Campbell (eds) Grammar from the human perspective. Amsterdam: Benjamins, 173–207. Hendrik, R. 1996. Some syntactic effects of suppletion in the Celtic copulas. In R. Borsely and I. Roberts (eds) The syntax of the Celtic languages. Cambridge: Cambridge University Press, 75–96. Hermon, G. and J. Yoon. 1989. The licensing and identification of pro and the typology of AGR. Chicago Linguistic Society 25,1. 174–92. Heycock, C. 1995. Asymmetries in reconstruction. Linguistic Inquiry 26: 547–70. Heycock, C. and A. Kroch. 1999. Pseudocleft connectivity: implications for the LF interface level. Linguistic Inquiry 30: 365–97. Heycock, C. and A. Sorace. 2007. Verb movement in Faroese: new perspectives on an old question. Nordlyd 35. Hinterho¨lzl, R. 2005. Scrambling, remnant movement, and restructuring in West Germanic. New York: Oxford University Press. Hoekstra, T. and R. Mulder. 1990. Unergatives as copular verbs; locational and existential predication. The Linguistic Review 7: 1–79. Holmberg, A. 2000. Scandinavian stylistic fronting: how any category can become an expletive. Linguistic Inquiry 31: 445–83. 2001. The syntax of yes and no in Finnish. Studia Linguistica 55: 141–75. 2002. Expletives and agreement in Scandinavian passives. Journal of Comparative Germanic Linguistics 4: 85–128. 2003a. Topic-drop or VP-focus. In L-O. Delsing et al. (eds) Grammar in focus. Festschrift for Christer Platzack 18 November 2003. Lund: Department of Scandinavian Languages, 159–66. 2003b. Yes/no-questions and the relation between tense and polarity in English and Finnish. In P. Pica (ed.) Linguistic Variation Yearbook 3: 43–68. 2005. Is there a little pro? Evidence from Finnish. Linguistic Inquiry 36: 533–64. 2006. Stylistic Fronting. In M. Everaert and H. van Riemsdijk (eds) The Blackwell companion to syntax. Oxford: Blackwell, 530–63. 2007. Null subjects and polarity focus. Studia Linguistica 61: 212–36. (ed.). 2009. Studia Linguistica: special issue on partial pro-drop 63, 1. Holmberg, A. and Th. Hro´arsdo´ttir. 2004. Agreement in dative constructions in Icelandic. Lingua 113: 997–1019. Holmberg, A., A. Nayudu and M. Sheehan. 2009. Three partial null-subject languages: a comparison of Brazilian Portuguese, Finnish and Marathi. In Studia Linguistica 63: 59–97. Holmberg, A. and U. Nikanne. 2002. Expletives, subjects and topics in Finnish. In P. Svenonius (ed.) Subjects, expletives and the EPP. New York: Oxford University Press, 71–105. Holmberg, A., U. Nikanne, I. Oraviita, H. Reime and T. Trosterud. 1993. The structure of INFL and the finite clause in Finnish. In A. Holmberg and U. Nikanne (eds) Case and other functional categories in Finnish syntax. Berlin: Mouton de Gruyter, 175–206. Holmberg, A. and C. Platzack. 1991. On the role of inflection in Scandinavian syntax. In W. Abraham, W. Kosmeijer and E. Reuland (eds) Issues in Germanic syntax. Berlin: Mouton de Gruyter, 93–118.

340

Bibliography

1995. The role of inflection in Scandinavian syntax. New York: Oxford University Press. de Hoop, H. 1996. Case configuration and noun phrase interpretation. New York: Garland. Hornstein, N. 1999. Movement and control. Linguistic Inquiry 30: 69–96. 2000. Move! A minimalist theory of construal. Oxford: Blackwell. 2003. On control. In R. Hendrick (ed.) Minimalist syntax. Oxford: Blackwell, 6–81. Horvath, J. 2000. Interfaces vs. the computational system in the syntax of focus. In H. Bennis, M. Everaert and E. Reuland (eds) Interface strategies. Amsterdam: Royal Netherlands Academy of Arts and Sciences, 183–206. 2006. Pied-piping. In M. Everaert and H. van Riemsdijk (eds) The Blackwell companion to syntax. Oxford: Blackwell, 569–630. 2007. Separating ‘focus movement’ from focus. In S. Karimi et al. (eds) Phrasal and clausal architecture. Amsterdam: Benjamins. Householder, F. 1981. Syntax of Apollonius Dyscolus. Amsterdam: Benjamins. Hrafnbjargarson, G. 2004a. Stylistic fronting. Studia Linguistica 58, 2: 88–134. 2004b. Oblique subjects and stylistic fronting in the history of Scandinavian and English: the role of IP-Spec. PhD dissertation: University of Aarhus. Huang, J. 1982. Logical relations in Chinese and the theory of grammar. PhD dissertation: MIT. 1984. On the distribution and reference of empty pronouns. Linguistic Inquiry 15: 531–74. 1989. Pro-drop in Chinese: a generalized control theory. In O. Jaeggli and K. Safir (eds) The null subject parameter. Dordrecht: Kluwer, 185–214. 1993. Reconstruction and the structure of VP. Linguistic Inquiry 24:103–38. 2007. The macro-history of Chinese syntax and the theory of language change. Talk given at the University of Chicago. Huang, Y. 1994. The syntax and pragmatics of anaphora: a study with special reference to Chinese. Cambridge: Cambridge University Press. 2000. Anaphora. A cross-linguistic study. New York: Oxford University Press. Hulk, A. 1986. Subject clitics and the pro-drop parameter. In P. Coopmans, Y. Bordelois and B. Dotson-Smith (eds) Going Romance: formal parameters of generative grammar II. Dordrecht: Foris, 107–20. 1991. Les pronoms clitiques sujet et la the´orie linguistique. Actes du XVIIIe congr es international de linguistique et de philologie romanes. Tu¨bingen: Niemeyer, 504–14. Hulk, A. and J-Y. Pollock. 2001. Subject positions in Romance and the theory of Universal Grammar. In A. Hulk and J-Y Pollock (eds) Subject inversion in Romance and the theory of Universal Grammar. Oxford: Oxford University Press, 3–19. Ihsane, T. and G. Puskas. 2001. Specific is not definite. Generative grammar in Geneva 2: 39–54. Jackendoff, R. 1972. Semantic interpretation in generative grammar. Cambridge, MA: MIT Press. 2002. Foundations of language. Oxford: Oxford University Press.

Bibliography

341

Jacobs, J. 1997. I-Topikalisierung. Linguistische Berichte 168: 91–133. Jaeggli, O. 1982. Topics in Romance syntax. Dordrecht: Foris. Jaeggli, O. and K. Safir. 1989. The null subject parameter. Dordrecht: Foris. Jaggar, P. 2001. Hausa. Amsterdam: Benjamins. Jayaseelan, K. 1999. Empty pronouns in Dravidian. In K. Jayaseelan (ed.) Parametric studies in Malayalam syntax. New Delhi: Allied Publishers, 14–25. Jespersen, O. 1909–49. A modern English grammar on historical principles, I–VII. London/Copenhagen: Allen and Unwin. 1924. The philosophy of grammar. London: Allen and Unwin. Johns, C. 2007. Interpreting agreement. PhD dissertation: Durham University. Johnson, K. and S. Vikner. 1994. The position of the verb in Scandinavian infinitives. Working Papers in Scandinavian Syntax 53: 61–84. Jonas, D. 1996. Clause structure and verb syntax in Scandinavian and English. PhD dissertation: Harvard. Jo´nsson, J. 1991. Clausal architecture and case in Icelandic. PhD dissertation: University of Massachusetts, Amherst. Jouitteau, M. 2004. Gestures as expletives, multichannel syntax. In B. Schmeiser, V. Chand, A. Kelleher and A. Rodriguez (eds) Proceedings of WCCFL 23. Somerville, MA: Cascadilla Press, 422–35. 2005. Nominal properties of vPs in Breton: a hypothesis for the typology of VSO languages. In A. Carnie, H. Harley and S. Dooley (eds) Verb first. On the syntax of verb-initial languages. Amsterdam: Benjamins, 265–80. Julien, M. 2002. Syntactic heads and word formation. New York: Oxford University Press. Kanerva, J. 1987. Morphological integrity and syntax: the evidence from Finnish possessive suffixes. Language 63: 498–501. Kato, M. 1999. Strong and weak pronominals in the null subject parameter. Probus 11:1–37. 2000. The partial pro-drop nature and the restricted VS order in Brazilian Portuguese. In M. Kato and E. Negra˜o (eds) Brazilian Portuguese and the null subject parameter. Frankfurt: Vervuert, 223–58. Kato, M. and E. Negra˜o (eds). 2000. The null subject parameter in Brazilian Portuguese. Frankfurt: Vervuert-Iberoamericana. Kato, M. and E. Raposo. 1994. European and Brazilian Portuguese word order: questions, focus and topic constructions. In C. Parodi, C. Quicoli, M. Saltarelli and M.-L. Zubizarreta (eds) Aspects of Romance Linguistics: Selected Papers from the LSRL XXVI. Washington, DC: Georgetown University Press, 267–78. 2006. Topicalization in European and Brazilian Portuguese. In J. Camacho, N. Flores-Ferra´n, L. Sa´nchez, V. De´prez and M.-J. Cabrera (eds) Romance Linguistics 2006: Selected papers from the 36th Linguistic Symposium on Romance Languages. Amsterdam: Benjamins, 199–212. Kauffman, S. 1995. At Home in the Universe. London: Viking. Kayne, R. 1972. Subject inversion in French interrogatives. In J. Casagrande and B. Saciuk (eds) Generative studies in Romance linguistics. Rowley, MA: Newbury House, 70–126. 1975 French syntax: the transformational cycle. Cambridge, MA: MIT Press.

342

Bibliography

1983. Chains, categories external to S, and French complex inversion. Natural Language and Linguistic Theory 1: 107–39. 1991. Romance clitics, verb movement and PRO. Linguistic Inquiry 22: 647–86. Reprinted in Kayne (2000). 1994. The antisymmetry of syntax, Cambridge, MA: MIT Press. 2000. Parameters and universals. New York: Oxford University Press. 2005a Some notes on comparative syntax, with special reference to English and French. In G. Cinque and R. Kayne (eds) The Oxford handbook of comparative syntax. Oxford: Oxford University Press, 3–69. 2005b. Movement and silence. New York: Oxford University Press. 2008a. Expletives, datives, and the tension between morphology and syntax. In T. Biberauer (ed.) The limits of syntactic variation. Amsterdam: Benjamins, 175–218. 2008b. Antisymmetry and the lexicon. Unpublished ms: New York University. Kayne, R. and J-Y. Pollock. 1978. Stylistic inversion, successive cyclicity and move NP in French. Linguistic Inquiry 9: 595–621. 2001/2005. New thoughts on stylistic inversion. In A. Hulk and J.-Y. Pollock (eds) Subject inversion in Romance and the theory of Universal Grammar. New York: Oxford University Press, 107–62. Reprinted in Kayne (2005b). van Kemenade, A. 1987. Syntactic case and morphological case in the history of English. Dordrecht: Foris. King, G. 1993. Modern Welsh: a comprehensive grammar. London: Routledge. Kiparsky, P. 1973. ‘Elsewhere’ in phonology. In S. Anderson and P. Kiparsky (eds) A Festschrift for Morris Halle. New York: Holt, Reinhart and Winston, 93–106. 2001. Structural case in Finnish. Lingua 111: 315–76. Kiss, K. 1995. Introduction. In K. Katalin (ed.) Discourse-configurational languages. New York: Oxford University Press, 3–27. 1996. Two subject positions in English. The Linguistic Review 13: 119–142. 1998. Identificational focus versus information focus. Language 74: 245–73. Koeneman, O. 2000. The flexible nature of verb movement. PhD dissertation: Groningen. Koeneman, O. and A. Neeleman. 2001. Predication, verb movement and the distribution of expletives. Lingua 111:189–233. Kornfilt, J. 1985. Case marking, agreement, and empty categories in Turkish. PhD dissertation: Harvard. 1997. Turkish. London: Routledge. 2003. Turkish. In W. Frawley (ed.) International encyclopedia of linguistics, Volume IV, 2nd edition. Oxford: Oxford University Press, 301–7. Kosmeijer, W. 1986. The status of the finite inflection in Icelandic and Swedish. Working Papers in Scandinavian Syntax 26: 1–41. Koster, J. 1987. Domains and dynasties: the radical autonomy of syntax. Dordrecht: Foris. Kratzer, A. and E. Selkirk. 2007. Default phrase stress, prosodic phrasing and the spellout edge. The Linguistic Review 24: 93–135. Kroch, A. 1989. Reflexes of grammar in patterns of language change. Language Variation and Change 1: 199–244.

Bibliography

343

1994. Morphosyntactic variation. In K. Beals, R. Knippen, L. Melnar, H. Suzuki and E. Zeinfeld (eds) Proceedings of the 30th Regional Meeting of the Chicago Linguistic Society. Chicago: University of Chicago Press, 180–201. 2000. Syntactic change. In M. Baltin and C. Collins (eds) The handbook of contemporary syntactic theory. Malden, MA: Blackwell, 629–739. Kroch, A. and A. Taylor. 1997. Verb movement in Old and Middle English: dialect variation and language contact. In A. van Kemenade and N. Vincent (eds) Parameters of morphosyntactic change. Cambridge: Cambridge University Press, 297–325. Kuroda, Y. 1988. Whether we agree or not: a comparative syntax of English and Japanese. Lingvisticae Investigationes 12: 1–47. Laitinen, L. 1995. Nollapersoona [the Null Person]. Viritt€ aj€ a 1995: 337–58. 2006. Zero person in Finnish. In M.-L. Helasvuo and L. Campbell (eds) Grammar from the human perspective. Amsterdam: Benjamins, 210–31. Laitinen, L. and M. Vilkuna. 1993. Case-marking in necessive constructions. In A. Holmberg and U. Nikanne (eds) Case and other functional categories in Finnish syntax. Berlin: Mouton de Gruyter, 23–48. Lambrecht, K. 1981. Topic, antitopic and verb agreement in non-standard French. Amsterdam: Benjamins. 2000. When subjects behave like objects: an analysis of the merging of S and O in sentence focus constructions across languages. Studies in Language 24: 611–82. Landau, I. 1999. Elements of control. PhD thesis, Cambridge, MA: MIT. 2000. Elements of control: structure and meaning in infinitival constructions. Dordrecht: Kluwer. 2003. Movement out of control. Linguistic Inquiry 34: 471–98. 2004. The scale of finiteness and the calculus of control. Natural Language and Linguistic Theory 22: 811–77. 2006a. Chain resolution in Hebrew VP-fronting. Syntax 9: 32–65. 2006b. EPP extensions. Linguistic Inquiry 38: 485–523. Larson, R. and G. Segal. 1995. Knowledge of meaning. Cambridge, MA: MIT Press. Lasnik, H. 1995. Case and expletives revisited: on greed and other human failings. Linguistic Inquiry 26: 615–33. Reprinted in H. Lasnik 1999. Minimalist Analysis. London: Blackwell, 74–96. Lasnik, H., M. Depiante and A. Stepanov. 2000. Syntactic structures revisited. Contemporary lectures on classic transformational theory. Cambridge, MA: MIT Press. Le´autaud, P. 1988. Le Fl eau. Journal particulier. 1917þ1939. Paris: Mercure de France. Lechner, W. 2006. An interpretive effect of head movement. In Frascarelli, M. (ed.) Phases of interpretation. Berlin: Mouton de Gruyter, 45–70. Li, C. and S. Thompson. 1976. Subject and topic: a new typology of language. In C. Li (ed.) Subject and topic. London/New York: Academic Press, 457–89. Lightfoot, D. 1979. Principles of diachronic syntax. Cambridge: Cambridge University Press. 1991. How to set parameters: arguments from language change. Cambridge, MA: MIT Press.

344

Bibliography

Lindseth, M. 1998. Null-subject properties of Slavic languages, with special reference to Russian, Czech and Sorbian. Munchen: Sagner. Lo¨flund, J. 1998. Suomen kirjoitetun kielen yleispassiivi. [The general passive in ˚ bo Akademi University Press. written Finnish.] Turku: A Lohndal, T. 2009. That-t in Scandinavian and elsewhere: variation in the position of C. Studia Linguistica 63: 204–32. Longobardi, G. 1991. Alcune riflessioni informali sulla posizione del verbo gotico e le prospettive di una sintassi comparata dei complementatori generici. Unpublished ms: University of Venice. 1994. Reference and proper names: a theory of N-Movement in syntax and Logical Form. Linguistic Inquiry 25: 609–65. Longobardi, G. and C. Guardiano. 2008. Evidence for syntax as a signal of historical relatedness. To appear in Lingua. Lyngfelt, B. 2002. OT semantics and control. Rutgers optimality archive 411. Lyons, C. 1999. Definiteness. Cambridge: Cambridge University Press. Mac Eoin, G. 2002. Irish. In B. Martin and J. Fife (eds) The Celtic languages. London: Routledge, 101–44. Mackenzie, D. 2003. Pashto. In W. Frawley (ed.) International encyclopedia of linguistics, Volume III, 2nd edition. Oxford: Oxford University Press, 251–57. Maling, J. 1980. Inversion in embedded clauses in Modern Icelandic. Islenskt mal og almenn marfrædi 2: 175–93. Reprinted in J. Maling and A. Zaenen (eds). 1990. Modern Icelandic syntax. San Diego, CA: Academic Press, 71–91. 1993. Of nominative and accusative: the hierarchical assignment of grammatical case in Finnish. In A. Holmberg and U. Nikanne (eds) Case and other functional categories in Finnish syntax. Berlin: Mouton de Gruyter, 49–74. Manninen, S. and D. Nelson. 2004. What is a passive? The case of Finnish. Studia Linguistica 58(3): 212–51. Manzini, R. and A. Roussou. 2000. A minimalist theory of A-movement and control. Lingua 110: 409–47. Manzini, R. and L. Savoia. 2002. Parameters of subject inflection in Italian dialects. In P. Svenonius (ed.) Subjects, expletives and the EPP. Oxford: Oxford University Press, 157–200. 2005. I dialetti italiani e romanci. Alessandria: Edizioni dell’Orso. Marantz, A. 2001. Words. WCCFL 20 handout: University of Southern California. 2006. Phases and words. Unpublished ms: New York University. Martin, R. 1996. A minimalist theory of PRO and control. PhD dissertation: University of Connecticut. 2001. Null Case and the distribution of PRO. Linguistic Inquiry 32: 141–66. Martins, A. 2006. Emphatic affirmation and polarity. Contrasting European Portuguese with Brazilian Portuguese, Spanish, Catalan and Galician. In J. Doetjes and P. Gonzales (eds) Romance languages and linguistic theory 2004. Amsterdam: Benjamins, 197–223. Massam. D. 2000. VSO and VOS: aspects of Niuean word order. In A. Carnie and E. Guilfoyle (eds) The syntax of verb-initial languages. New York: Oxford University Press, 97–116.

Bibliography

345

2001. On predication and the status of subjects in Niuean. In W. Davies and S. Dubinsky (eds) Objects and other subjects: grammatical functions, functional categories, and configurationality. Dordrecht: Kluwer, 225–46. 2005. Lexical categories, lack of inflection, and predicate fronting in Niuean. In A. Carnie, S. Dooley and H. Harley (eds) Verb first. On the syntax of verb-initial languages. Amsterdam: Benjamins, 227–42. Massam, D. and C. Smallwood. 1997. Essential features of predication in English and Niuean. In K. Kusumoto (ed.) Proceedings of NELS 27: 236–72. Masullo, P. 1993. Two types of quirky subjects: Spanish versus Icelandic. In A. Shafer (ed.) Proceedings of NELS 23: 303–17. Mathieu, E. 2006. Stylistic fronting in Old French. Probus 18: 219–66. Matushansky, O. 2006. Head movement in linguistic theory. Linguistic Inquiry 37,1: 69–110. Mavrogiorgos, M. 2006. Morphology is at the edge. Unpublished ms: University of Cambridge. May, R. 1985. Logical form. Cambridge, MA: MIT Press. Mayr, C. 2009. On the necessity phi-features: the case of Bavarian subject extraction. In P. Panagiotidis (ed.) The complementizer domain: subjects and wh-dependencies. Oxford: Oxford University Press. McCloskey, J. 1996. Subjects and subject positions in Irish. In R. Borsley and I. Roberts (eds) The syntax of the Celtic languages: a comparative perspective. Cambridge: Cambridge University Press, 241–83. 2001. The morphosyntax of wh-extraction in Irish. Journal of Linguistics 37: 67–100. McCloskey, J. and K. Hale. 1984. On the syntax of person-number inflection in modern Irish. Natural Language and Linguistic Theory 1: 487–553. McShane, M. 2009. Subject ellipsis in Russian and Polish. Studia Linguistica 63(1): 98–132. Menuzzi, S. 2004. A ordem verbo-sujeito no Portugueˆs do Brasil: para uma comparac¸a˜o das abordagens formalistas e funcionalistas. Revista da ANPOLL 16: 349–84. Milsark, R. 1974. Existential sentences in English. PhD dissertation: MIT. Published as: Milsark, R. 1979. Existential sentences in English. New York: Garland. Mobbs, I. 2008. ‘Functionalism’, the design of the language faculty, and typology. Unpublished ms: University of Cambridge. (Available on-line at: http://ling.auf. net/lingBuzz/000680) Modesto, M. 2000a. On the identification of null arguments. PhD dissertation: University of Southern Califonia. 2000b. Null subjects without ‘rich’ agreement. In M. Kato and E. Negra˜o (eds) The Null Subject Parameter in Brazilian Portuguese. Frankfurt: Vervuert-Iberoamericana, 147–74. 2008. Topic prominence and null subjects. In T. Biberauer (ed.) The limits of syntactic variation. Amsterdam: Benjamins, 375–410. Mohr, S. 2005. Clausal architecture and subject positions. Amsterdam: Benjamins. Moltmann, F. 2006. Generic one, arbitrary PRO, and the first person. Natural Language Semantics 14: 257–81. Montalbetti, M. 1984. After Binding. PhD dissertation: MIT.

346

Bibliography

Moro, A. 1997. The raising of predicates. Cambridge: Cambridge University Press. Morpurgo-Davies, A. 1998. History of linguistics, Volume IV: nineteenth-century linguistics (General Editor: G. Lepschy) London: Longman. Mosse´, F. 1968. Manual of Middle English. Baltimore: Johns Hopkins Press. Mu¨ller, G. 1998. Incomplete category fronting. A derivational approach to remnant movement in German. Dordrecht: Kluwer. 2004a. Verb-second as vP-first. Journal of Comparative Germanic Linguistics 7,3: 179–234. 2004b. Argument encoding and the order of elementary operations. Unpublished ms: Leipzig. (Available on-line at: http://www.uni-leipzig.de/ muellerg/mu56.pdf) 2004c. Phrase impenetrability and wh-intervention. In A. Stepanov, G. Fanselow and R. Vogel (eds) Minimality effects in syntax. Berlin: Mouton de Gruyter. 2005. Pro-drop and impoverishment. In P. Brandt and E. Fuss (eds) Form, structure and grammar. A Festschrift presented to G€ unther Grewendorf on the occasion of his 60th birthday. Tu¨bingen: Narr, 93–115. 2007. Some consequences of an impoverishment-based approach to morphological richness and pro-drop. Unpublished ms: Leipzig. (Available on-line at: http:// www.uni-leipzig.de/muellerg/mu228.pdf) Munaro, N. and J.-Y Pollock. 2005. Qu-est-ce-que (qu)-est-ce-que? A case study in comparative Romance interrogative syntax. In G. Cinque and R. Kayne (eds) The Oxford handbook of comparative syntax. Oxford: Oxford University Press, 542–606. Muysken, P. 1982. Parametrizing the notion ‘head’. Journal of Linguistic Research 2: 57–75. Nash, L. and A. Rouveret. 1997. Proxy categories in phrase structure theory. In K. Kusumoto (ed.) Proceedings of NELS 27: 287–304. Nayudu, A. 2008. Issues in the syntax of Marathi: A minimalist approach. PhD dissertation: Durham University. Neeleman, A. 1994. Complex predicates. PhD dissertation: Utrecht University. Neeleman, A. and H. van de Koot. 2002. The configurational matrix. Linguistic Inquiry 33: 529–74. 2006. Syntactic haplology. In M. Everaert and H. van Riemsdijk (eds) The Blackwell companion to syntax. Oxford: Blackwell, 684–710. 2008. Dutch scrambling and the nature of discourse templates. Journal of Comparative Germanic Linguistics 11(2): 137–89. Neeleman, A. and T. Reinhart. 1998. Scrambling and the PF-interface. In M. Butt and W. Geuder (eds) The projection of arguments: lexical and compositional factors. Chicago: CSLI Publications, 309–53. Neeleman, A. and K. Szendro˝i. 2007. Radical pro-drop and the morphology of pronouns. Linguistic Inquiry 38: 671–714. 2008. Case morphology and radical pro-drop. In T. Biberauer (ed.) The limits of syntactic variation. Amsterdam: Benjamins, 331–50. Neeleman, A. and F. Weerman. 1999. Flexible syntax. A theory of Case and arguments. Dordrecht: Kluwer. Negra˜o, E. 1997. Asymmetries in the distribution of overt and empty categories in Brazilian Portuguese. In J. Black and V. Motapanyane (eds) Clitics, pronouns and movement. Amsterdam: Benjamins, 217–35.

Bibliography

347

Negra˜o, E. and A. Mu¨ller. 1996. As Mudanc¸as no sistema pronominal do Portugueˆs do Brasil: substituc¸a˜o ou especializac¸a˜o de formas? DELTA 12: 125–52. Newmeyer, F. 2004. Against a parameter-setting approach to language variation. Linguistic Variation Yearbook 4: 181–234. 2005. Possible and probable languages. A generative perspective on linguistic typology. Oxford: Oxford University Press. 2006. A rejoinder to ‘On the role of parameters in Universal Grammar: a reply to Newmeyer’ by Ian Roberts and Anders Holmberg. Unpublished ms: University of Washington (Seattle). (Available on-line at: http://people.pwf.cam.ac.uk/mtb23/ NSP/Parametricdebate.html) Newton, G. 2006. The development and loss of the Old Irish double system of verbal inflection. PhD dissertation: University of Cambridge. Nicolis, M. 2005. On pro-drop. PhD dissertation: Universita` degli Studi di Siena. 2008. The null subject parameter and correlating properties: the case of creole languages. In T. Biberauer (ed.) The limits of syntactic variation. Amsterdam: Benjamins, 271–94. Niyogi, P. 2006. The computational nature of language learning and evolution. Cambridge, MA: MIT Press. Niyogi, P. and R. Berwick. 1995. The logical problem of language change. A. I. Memo No. 1516, MIT Artificial Intelligence Laboratory. 1997. A dynamical systems model for language change. Complex Systems 11: 161–204. Nunes, J. 1995. The copy theory of movement and linearization of chains in the minimalist program. PhD dissertation: University of Maryland, College Park. 2004. Linearization of chains and sideward movement. Cambridge, MA: MIT Press. O’Neil, J. 1997. Means of control: deriving the properties of PRO in the minimalist program. PhD dissertation: Harvard University. Ordo´n˜ez, F. 1997. Word order and clause structure in Spanish and other Romance languages. PhD dissertation: CUNY Graduate Center. 1998. Post-verbal asymmetries in Spanish. Natural Language & Linguistic Theory 16: 313–46. 2000. The clausal structure of Spanish. New York: Garland. 2006. The order of subjects in Spanish and Catalan. Talk given at the Encontro Lingua Falada e Escrita V, Federal University of Macei o. Ordo´n˜ez, F. and E. Trevin˜o. 1999. Left dislocated subjects and the pro-drop parameter: a case study of Spanish. Lingua 107: 39–68. Otsuka, Y. 2005. Two derivations of VSO: a comparative study of Niuean and Tongan. In A. Carnie, S. Dooley and H. Harley (eds) Verb first. On the syntax of verb-initial languages. Amsterdam: Benjamins, 65–90. Ouali, H. 2008. On C-to-T j-feature transfer: the nature of agreement and anti-agreement in Berber. In R. D’Alessandro, G. Hrafnbjargarson and S. Fischer (eds) Agreement restrictions. Berlin: Mouton de Gruyter 159–80. Ouhalla, J. 1991. Functional categories and parametric variation. London: Routledge. 1994. Verb movement and word order in Arabic. In D. Lightfoot and N. Hornstein (eds) Verb movement. Cambridge: Cambridge University Press, 41–72. ¨ ztu¨rk, B. 2001. Turkish as a non-pro-drop language. In E. Taylan (ed.) The verb in O Turkish. Amsterdam: Benjamins, 239–60.

348

Bibliography

2008. Non-configurationality: free word order and argument drop in Turkish. In T. Biberauer (ed.) The limits of syntactic variation. Amsterdam: Benjamins, 411–40. Paesani, K. 2006. The case of special registers. In L. Progovac, K. Paesani, E. Casielles and E. Barton (eds) The syntax of nonsententials. Multidisciplinary perspectives. Amsterdam: Benjamins, 147–82. Palmer, L. 1961. La lingua Latina. Bologna: Einaudi. [Italian translation of L. Palmer. 1954. The Latin language. London: Faber and Faber.] Parry, M. 1994. Posizione dei clitici complemento nelle costruzioni verbali perifrastiche del piemontese. In G. Clivio and C. Pich (eds) At d€ el (X) R€ escontr ant€ ernassional d€ e Studi an sla Lenga e la Literatura piemont eisa 8: 247–59. Paul, I. 2000. Malagasy clause structure. PhD dissertation: McGill University. 2007. Missing topics in Malagasy headlines. In M. Radisˇic´ (ed.) Proceedings of the 2007 Annual Conference of the Canadian Linguistic Association, 10 pages. 2008. Great coffee, that Maxwell House! In E. Efner and M. Walkow (eds) Proceedings of NELS 38. Amherst, Mass: GLSA Publications. Pearson, M. 2001. The clause structure of Malagasy: a minimalist approach. PhD dissertation: UCLA. Perlmutter, D. 1971. Deep and surface constraints in syntax. New York: Holt, Rinehart and Winston. Pesetsky, D. 1989. Language-particular processes and the Earliness Principle. Unpublished ms: MIT. (Available on-line at: http://web.mit.edu/linguistics/ people/faculty/pesetsky/earliness.pdf) Pesetsky, D. and E. Torrego. 2001. T-to-C movement: causes and consequences. In M. Kenstowicz (ed.) Ken Hale: a life in language. Cambridge, MA: MIT Press, 355–426. 2004. Tense, case and the nature of syntactic categories. In J. Gue´ron and J. Lecarme (eds) The syntax of time. Cambridge MA: MIT Press, 495–538. 2006. Probes, goals and syntactic categories. In Y. Otsu (ed.) Proceedings of the Seventh Tokyo Conference on Psycholinguistics. Amsterdam: Benjamins, 262–94. 2007. The syntax of valuation and the interpretability of features. In S. Karimi, V. Samiian and W. Wilkins (eds) Phrasal and clausal architecture: syntactic derivation and interpretation. Cambridge, MA: MIT Press, 193–220. Piattelli-Palmarini, M. and D. Vercelli. To appear. Language in an epigenetic framework. In M. Piattelli-Palmarini, J. Uriagereka and P. Salaburu (eds) Of minds and language: the Basque country encounter with Noam Chomsky. Oxford: Oxford University Press. Pinto, M. 1994. Subjects in Italian: distribution and interpretation. In R. Bok-Bennema and C. Cremers (eds) Linguistics in the Netherlands. Amsterdam: Benjamins, 175–87. 1997. Licensing and interpretation of inverted subjects in Italian. PhD dissertation: Utrecht. Plath, S. 1983. The journal of Sylvia Plath. Edited by T. Hughes and F. McCollough. New York: Ballantine Books. Platzack, C. 1985. The Scandinavian languages and the null subject parameter. Working Papers in Scandinavian Syntax 20: 1–65.

Bibliography

349

1987. The Scandinavian languages and the null-subject parameter. Natural Language and Linguistic Theory 5: 377–401. 1995. The loss of verb second in English and French. In A. Battye and I. Roberts (eds) Clause structure and language change. New York: Oxford University Press, 200–26. 2004. Agreement and the person phrase hypothesis. Working Papers in Scandinavian Syntax 73: 83–112. Poletto, C. 1990. Three types of subject clitics and the theory of pro. Unpublished ms: University of Venice. 2000. The higher functional field in the Northern Italian dialects. Oxford: Oxford University Press. Poletto, C. and J.-Y. Pollock. 2004. On the left periphery of some Romance wh-questions. In L. Rizzi (ed) The structure of IP and CP. Oxford: Oxford University Press, 251–96. Pollock, J.-Y. 1986. Sur la syntaxe de EN et le parameter du sujet nul. In M. Ronat and D. Couquaux (eds) La grammaire modulaire. Paris: Les editions de minuit, 211–46. 1989. Verb movement, Universal Grammar and the structure of IP. Linguistic Inquiry 20: 365–424. 1997. Langage et Cognition: introduction au programme minimaliste de la grammaire g en erative. Paris: Presses Universitaires de France. 2006. Subject clitics and complex inversion in French. In M. Everaert and H. van Riemsdijk (eds) The Blackwell companion to syntax. Oxford: Blackwell, 601–59. Ponelis, F. 1993. The development of Afrikaans. Frankfurt: Peter Lang. Postal, P. 1977. About a ‘nonargument’ for raising. Linguistic Inquiry 8: 141–54. Rackowski, A. and L. Travis. 2000. V-initial languages: X or XP movement and adverbial placement. In A. Carnie and E. Guilfoyle (eds) The syntax of verb-initial languages. New York: Oxford University Press, 117–41. Ramchand, G. 2008. Verb meaning and the lexicon. A first phase syntax. Cambridge: Cambridge University Press. Ramchand, G. and P. Svenonius. 2008. Mapping a parochial lexicon onto a universal semantics. In T. Biberauer (ed.) The limits of syntactic variation. Amsterdam: Benjamins, 219–46. Raposo, E. 1986. On the null object in European Portuguese. In O. Jaeggli and C. Silva-Corvala´n (eds) Studies in Romance linguistics. Dordrecht: Foris, 373–90. Rebuschi, G. and L. Tuller. 1999. The grammar of focus. Amsterdam: Benjamins. Reime, H. 1993. Accusative marking in Finnish. In A. Holmberg and U. Nikanne (eds) Case and other functional categories in Finnish syntax. Berlin: Mouton de Gruyter, 175–206. Reinhart, T. 1983. Anaphora and semantic interpretation. London: Croom Helm. 1995. Interface strategies. OTS Working Papers in Linguistics. Utrecht. 2006. Interface strategies: optimal and costly computations. Cambridge, MA: MIT Press. Renzi, L. and L. Vanelli. 1983. I pronomi soggetto in alcune varieta` romanze. In P. Beninca` et al. (eds) Scritti linguistici in onore di G.B. Pellegrini, Vol. 1. Pisa: Pacini, 121–45.

350

Bibliography

Rezac, M. 2003. The fine structure of cyclic Agree. Syntax 6: 156–82. 2004. Elements of cyclic syntax: agree and merge. PhD dissertation: University of Toronto. Richards, M. 2004. Object shift in North and West Germanic: optionality, scrambling and base-generated OV. PhD dissertation: University of Cambridge. 2007. On feature-inheritance: an argument from the Phase Impenetrability Condition. Linguistic Inquiry 38,3: 563–72. Richards, M. and T. Biberauer. 2005. Explaining Expl. In M. den Dikken and C. Tortora (eds) The function of function words and functional categories. Amsterdam: Benjamins, 115–54. Richards, N. 2001. Movement in language. Interactions and architectures. Oxford: Oxford University Press. Rivero, M-L. and M. Milojevic´ Sheppard. 2003. Indefinite reflexive clitics in Slavic: Polish and Slovenian. Natural Language and Linguistic Theory 21: 89–155. Rix, H. 2004. Etruscan. In R. Woodward (ed.) The Cambridge encyclopedia of the world’s ancient languages. Cambridge: Cambridge University Press, 943–66. Rizzi, L. 1978. Violations of the wh-island constraint and the subjacency condition. In C. Dubisson, D. Lightfoot and Y.-C. Morin (eds) Montreal Working Papers in Linguistics 11: 155–90. 1982. Issues in Italian syntax. Dordrecht: Foris. 1986a. Null objects in Italian and the theory of pro. Linguistic Inquiry 17: 501–57. 1986b. On the status of subject clitics in Romance. In O. Jaeggli and C. Silva-Corvala´n (eds) Studies in Romance linguistics. Dordrecht: Foris, 391–420. 1987. Three remarks on null subjects. Talk given at the Workshop on Dialectology, GLOW X, University of Venice. 1992. Early null subjects and root null subjects. In T. Hoekstra and B. Schwartz (eds) Language acquisition studies in generative grammar. Amsterdam: Benjamins, 151–76. 1997. On the fine structure of the left periphery. In L. Haegeman (ed) Elements of grammar. Dordrecht: Kluwer, 281–337. 2006a. Grammatically-based target-inconsistencies in child language. In K. Deen, J. Nomura, B. Schulz and B. Schwartz (eds) The Proceedings of the Inaugural Conference on Generative Approaches to Language Acquisition – North America (GALANA). UCONN/MIT Working Papers in Linguistics. Cambridge, MA: MIT Press. 2006b. On the form of chains: criteria positions and ECP effects. In L. Cheng and N. Corver (eds) Wh movement: moving on. Cambridge MA: MIT Press. Rizzi, L. and I. Roberts. 1989. Complex inversion in French. Probus 1: 1–30. Rizzi, L. & U. Shlonsky. 2005. Strategies of subject extraction. In H.-M. Ga¨rtner and U. Sauerland (eds) Interfaces þ Recursion ¼ Language? Chomsky’s Minimalism and the view from syntax-semantics. Berlin: Mouton de Gruyter, 115–60. Roberge, Y. 1986. The syntactic recoverability of null arguments. PhD dissertation: University of British Columbia. Roberge, Y. and M-T. Vinet. 1989. La variation dialectale en grammaire universelle. Montre´al: Presses Universitaires de Montre´al/Editions de l’Universite´ de Sherbrooke.

Bibliography

351

Roberts, I. 1985. Agreement parameters and the development of English modal auxiliaries. Natural Language and Linguistic Theory 3: 21–58. 1993a. The nature of subject clitics in Franco-Provenc¸al Valdoˆtain. In A. Belletti (ed.) Dialects of Italy. Turin: Rosenberg and Sellier, 319–53. 1993b. Verbs and diachronic syntax: a comparative history of English and French. Dordrecht: Kluwer. 1996a. Remarks on the Old English C-system and diachrony of V2. In E. Brandner and G. Ferraresi (eds) Linguistische Berichte 7: 154–67. 1996b. Comparative syntax. London: Edward Arnold. 1998. Have-be raising, move F, and procrastinate. Linguistic Inquiry 29: 113–25. 1999. Verb movement and markedness. In M. deGraff (ed.) Language creation and change. Cambridge, Mass: MIT Press, 287–328. 2001. Language change and learnability. In S. Bertolo (ed.) Parametric linguistics and learnability. Cambridge: Cambridge University Press, 81–125. 2004. The C-system in Brythonic Celtic languages, V2 and the EPP. In L. Rizzi (ed.) The structure of CP and IP. The cartography of syntactic structures, Volume 2. Oxford: Oxford University Press, 297–328. 2005. Principles and parameters in a VSO language. A case study in Welsh. New York: Oxford University Press. 2007a. Diachronic syntax. Oxford: Oxford University Press. 2007b. Introduction. In I. Roberts (ed.) Comparative grammar: critical concepts. Volume II: the null subject parameter. London: Routledge, 1–44. 2008. The mystery of the overlooked discipline: modern syntactic theory and cognitive science. Unpublished ms: University of Cambridge. Forthcoming a. Taraldsen’s Generalisation and language change: two ways to lose null subjects. In P. Svenonius (ed.) Festschrift for Tarald Taraldsen. Oxford: Oxford University Press. Forthcoming b. Agreement and head movement: clitics, incorporation and detective goals. To appear with MIT Press. Roberts, I. and A. Holmberg. 2005. On the role of parameters in Universal Grammar: a reply to Newmeyer. In H. Broekhuis, N. Corver, M. Everaert and J. Koster (eds) Organising grammar: a Festschrift for Henk van Riemsdijk. Berlin: Mouton de Gruyter, 538–53. Roberts, I. and A. Roussou. 2003. Syntactic change. A minimalist approach to grammaticalization. Cambridge: Cambridge University Press. Rodrigues, C. 2002. Morphology and null subjects in Brazilian Portuguese. In D. Lightfoot (ed.) Syntactic effects of morphological change. Oxford: Oxford University Press, 160–78. 2004. Impoverished morphology and A-movement out of Case-domains. PhD dissertation: University of Maryland. Ro¨gnvaldsson, E. and H. Thra´insson. 1990. On Icelandic word order once more. In J. Maling and A. Zaenen (eds) Modern Icelandic syntax. New York: Academic Press. Rohrbacher, B. 1999. Morphology-driven syntax. Amsterdam: Benjamins. Rosenbaum, P. 1967. The grammar of English predicate complement constructions. Cambridge, MA: MIT Press.

352

Bibliography

Rosengren, I. 2002. EPP: a syntactic device in the service of semantics. Studia Linguistica 562: 145–90. Ross, J. 1967. Constraints on variables in Syntax. PhD dissertation: MIT. 1969. Auxilaries as main verbs. In W. Todd (ed.) Studies in philosophical linguistics. Evanston, Il: Great Expectations Press. 1970. On declarative sentences. In R. Jacobs and P. Rosenbaum (eds) Readings in English transformational grammar. Waltham, MA: Ginn. Rouveret, A. 1996. Bod in the present tense and in other tenses. In R. Borsley and I. Roberts (eds) The syntax of the Celtic languages. Cambridge: Cambridge University Press, 125–70. Russell, B. 1905. On denoting. Mind 14: 479–93. Sabel, J. 2000. Expletives as features. In R. Billerey and B. Lillehaugen (eds) Proceedings of the 19th West Coast Conference on Formal Linguistics 19: 401–14. Saeed, J. 1996. Head-marking and pronominal clitics in Somali. In R. J. Hayward and I. M. Lewis (eds) Voice and power: The culture of language in north-east Africa: essays in honour of B. W. Andrzejewski, 37–49. Saito, M. 2007. Notes on East Asian argument ellipsis. Unpublished ms: Nanzan University and University of Connecticut. Samek-Lodovici, V. 1996. Constraints on subjects, an optimality theoretic analysis. PhD dissertation: Rutgers University. 2006. When right dislocation meets the left-periphery. Lingua 116: 836–73. Sapir, E. 1921. Language. New York: Harcourt Brace and Co. € Schlegel, A. 1817. Uber dramatische Kunst und Litteratur, Grundz€ uge einer Kultur- und V€ olkergeschichte Alteuropas, 2nd edition. 3 Volumes, Heidelberg: Mohr and Winter. Schleicher, A. 1861–2. Compendium der vergleichenden Grammatik der indogermanischen Sprachen. Kurzer Abriss einer Laut- und Formenlehre der indogermanischen Ursprache, des Altindischen, Altiranischen, Altgriechischen, Altitalischen, Altkeltischen, Altslawischen, Litauischen und Altdeutschen, 2 Volumes. Weimar: Bo¨hlau. Schwartz, B. and S. Vikner. 1996. The verb always leaves IP in V2 clauses. In A. Belletti and L. Rizzi (eds) Parameters and functional heads. New York: Oxford University Press, 11–62. Sells, P. 1987. Aspects of logophoricity. Linguistic Inquiry 18: 445–80. 2005. The peripherality of the Icelandic expletive. In M. Butt and T. King (eds) Proceedings of LFG-05. Stanford: CSLI Online Proceedings, 408–28. Sheehan, M. 2006. The EPP and null subjects in Romance. PhD dissertation: Newcastle University. Shlonsky, U. 1997. Clause structure and word order in Hebrew and Arabic: an essay in comparative Semitic syntax. New York: Oxford University Press. 2009. Hebrew as a partial null-subject language. Studia Linguistica 63: 133–57. Shore, S. 1988. On the so-called Finnish passive. Word 39: 151–76. Sigurðsson, H. 1989. Verbal syntax and case in Icelandic. PhD dissertation: Lund University. 2000. The locus of Case and agreement. Working Papers in Scandinavian Syntax 65: 65–108.

Bibliography

353

2004a. Icelandic non-nominative subjects: facts and implications. In P. Bhaskararao and K. Subbarao (eds) Non-nominative subjects. Amsterdam: Benjamins, 137–59. 2004b. The syntax of Person, tense, and speech features. Rivista di Linguistica/ Italian Journal of Linguistics 16: 219–51. 2006a. Agree in syntax, agreement in signs. In C. Boeckx (ed.) Agreement systems. Amsterdam: Benjamins, 201–37. 2006b. PF is more ‘syntactic’ than often assumed. Working Papers in Scandinavian Syntax 77: 101–28. In press. The case of PRO. To appear in Natural Language and Linguistic Theory. Sigurðsson, H. and V. Egerland. 2009. Impersonal null subjects in Icelandic and elsewhere. Studia Linguistica 63: 158–85. Sigurðsson, H. and A. Holmberg. 2008. Icelandic dative intervention: person and number are separate probes. In R. d’Alessandro, G. Hrafnbjargarson and S. Fischer (eds) Agreement restrictions. Berlin: Mouton de Gruyter, 251–80. Sigurðsson, H. and J. Maling. 2008. Argument drop and the Empty Left Edge Condition. Working Papers in Scandinavian Syntax 81:1–27. Sobin, N. 1987. The variable status of Comp-trace phenomena. Natural Language and Linguistic Theory 5: 33–60. Spencer, A. 1991. Introduction to morphological theory. Oxford: Blackwell. Sportiche, D. 1983. Structural invariance and symmetry in syntax. PhD dissertation: MIT. 1988. A theory of floating quantifiers and its corollaries for constituent structure. Linguisic Inquiry 19: 425–51. 1999. Subject clitics in French and Romance, complex inversion and clitic doubling. In K. Johnson and I. Roberts (eds) Beyond principles and parameters: essays in memory of Osvaldo Jaeggli. Dordrecht: Kluwer, 189–222. Stepanovic´, S. 2003. Multiple wh-fronting in Serbo-Croatian matrix questions and the matrix sluicing construction. In C. Boeckx and K. Grohmann (eds) Multiple wh-fronting. Amsterdam: Benjamins, 255–84. Stowell, T. 1981. The origins of phrase structure. PhD dissertation: MIT. Sun˜er, M. 1982. Syntax and semantics of Spanish presentational sentence-types. Washington DC: Georgetown University Press. 2002. The lexical preverbal subject in a Romance null-subject language: where art thou? In R. Nun˜ez-Ceden˜o, L. Lopez and R. Cameron (eds) A Romance perspective on language knowledge and use. Amsterdam: Benjamins, 341–58. Svenonius, P. 2002a. Introduction. In Svenonius, P. (ed.) Subjects, expletives and the EPP. Oxford: Oxford University Press, 3–28. 2002b. Subject positions and the placement of adverbials. In P. Svenonius (ed.) Subjects, expletives and the EPP. Oxford: Oxford University Press, 201–42. 2002c. Limits on P: filling in holes vs. falling in holes. Unpublished ms: University of Troms. 2004. On the edge. In D. Adger, C. de Cat and G. Tsoulas (eds) Peripheries Dordrecht: Kluwer, 261–87. Szabolcsi, A. 1983. The possessor that ran away from home. The Linguistic Review 3: 89–102. 1994. The noun phrase. In F. Kiefer and K. Kiss (eds) The syntactic structure of Hungarian. San Diego: Academic Press, 179–274.

354

Bibliography

Szendro˝i, K. 2002. Stress-focus correspondence in Italian. In C. Beyssade, R. Bok-Bennema, F. Drijkoningen and P. Monachesi (eds) Proceedings of Going Romance 2000. Amsterdam: Benjamins, 287–305. Tallerman, M. 1987. Mutation and the syntactic structure of modern colloquial Welsh. PhD dissertation: University of Hull. 1998. The uniform case-licensing of subjects in Welsh. Linguistic Review 15: 69–133. Tamburelli, M. 2006. Remarks on richness. UCL Working Papers in Linguistics 18: 439–55. Taraldsen, T. 1978. On the NIC, vacuous application and the that-trace filter. Unpublished ms: MIT. 1979. The theoretical implications of a class of marked extractions. In A. Belletti, L. Brandi and L. Rizzi (eds) Theory of markedness in generative grammar. Pisa: Scuola Normale Superiore, 475–516. 2002. Que/qui alternation and the distribution of expletives. In P. Svenonius (ed.) Subjects, expletives and the EPP. Oxford: Oxford University Press, 29–42. Taylan, E. (ed.). 2001. The verb in Turkish. Amsterdam: Benjamins. Thra´insson, H. 1994. Icelandic. In E. Ko¨nig and J. van der Auwera (eds) The Germanic languages. London: Routledge, 142–89. 2003. Syntactic variation, historical development and minimalism. In R. Hendrik (ed.). Minimalist syntax. Oxford: Blackwell, 152–91. 2007. The syntax of Icelandic. Cambridge: Cambridge University Press. Thra´insson, H., H. Petersen, J. Jacobsen and Z. Hansen (eds). 2004. Faroese. An overview and reference grammar. To´rshavn: Foroya Frodskaparfelag. Timberlake, A. 1975. The nominative object in Finnish. Lingua 35: 201–30. Tomioka, S. 2003. The semantics of Japanese null pronouns and its cross-linguistic implications. In K. Schwabe and S. Winkler (eds) The interfaces: deriving and interpreting omitted structures. Amsterdam: Benjamins, 321–40. Torrego, E. 1984. On inversion in Spanish and some of its effects, Linguistic Inquiry 15: 103–29. Travis, L. 1984. Parameters and effects of word order variation. PhD dissertation: MIT. 2006. VP-, D-movement languages. In R. Zanuttini, H. Campos, E. Herburger and P. Portner (eds) Negation, tense and clausal architecture. Georgetown: Georgetown University Press Publishers, 127–50. Trosterud, T. 1993. Anaphors and binding domains in Finnish. In A. Holmberg and U. Nikanne (eds) Case and other functional categories in Finnish syntax. Berlin: Mouton de Gruyter, 225–43. Uriagereka, J. 2007. Clarifying the notion ‘parameter’. Biolinguistics 1: 99–113. Vainikka, A. 1989. Deriving syntactic representations in Finnish. PhD dissertation: University of Massachusetts, Amherst. 1993. The three structural cases in Finnish. In A. Holmberg and U. Nikanne (eds) Case and other functional categories in Finnish syntax. Berlin: Mouton de Gruyter, 129–59. Vainikka, A. and Y. Levy. 1999. Empty subjects in Finnish and Hebrew. Natural Language and Linguistic Theory 17: 613–71.

Bibliography

355

Vallduvı´, E. 1990. The informational component. PhD dissertation: University of Pennsylvania. Vance, B. 1997. Syntactic change in medieval French. Verb-second and null-subjects. Dordrecht: Kluwer. Vanelli, L., L. Renzi and P. Beninca`. 1985/2007. Typologie des pronoms sujets dans les langues romanes. Actes du XIIe Congr es des Linguistique et Philologie Romanes. Aix-en-Provence. Published in Italian translation as Vanelli, L., L. Renzi and P. Beninca`. 1986. Tipologia dei pronomi soggetto nelle lingue romanze medievali. Quaderni Patavini di Linguistica 5: 49–66; and reprinted in Beninca`, P. (ed.). 1994. La variazione sintattica. Bologna: Il Mulino, 195–213. Published in English translation as Vanelli, L., L. Renzi and P. Beninca`. 2007. A typology of Romance subject pronouns. In I. Roberts (ed.) Comparative grammar, Volume II: the null subject parameter. London: Routledge, 234–45. Vangsnes, . 2002. Icelandic expletive constructions and the distribution of subject types. In P. Svenonius (ed.) Subjects, expletives and the EPP. Oxford: Oxford University Press, 43–70. Vicente, L. 2005. Towards a unified theory of movement: an argument from Spanish predicate clefts. In M. Salzmann and L. Vicente (eds) Leiden Papers in Linguistics 2: 43–67. 2006. The syntax of heads and phrases: a study of verb (phrase) fronting. PhD dissertation: Leiden. Vikner, S. 1995. Verb movement and expletive subjects in the Germanic languages. Oxford: Oxford University Press. 1997. V-to-I movement and inflection for person in all tenses. In L. Haegeman (ed.) The new comparative syntax. London: Longman, 187–213. 2001. Verb movement variation in Germanic and Optimality Theory. Habilitationschrift: Tu¨bingen. 2005. Immobile complex verbs in Germanic. Journal of Comparative Germanic Linguistics 8,1/2: 83–115. Vikner, S. and K. Johnson. 1994. The position of the verb in Scandinavian infinitives. Working Papers in Scandinavian Syntax 53: 61–84. Vinet, M-T. 1993. L’aspect et la copule vide dans la grammaire des titres. Langue franc¸aise 100: 83–100. Wali, K. and O. N. Koul. 1996. Kashmiri. A cognitive descriptive grammar. London/New York: Routledge. 1997. Kashmiri: a cognitive-descriptive grammar. London: Routledge. Wallace, R. 2004. Sabellian Languages. In R. Woodward (ed.) The Cambridge encyclopedia of the world’s ancient languages. Cambridge: Cambridge University Press, 812–39. Wanner, D. 1987. The development of Romance clitic pronouns from Latin to Old Romance. Berlin: Mouton de Gruyter. Warner, A. 1997. The structure of parametric change, and V-movement in the history of English. In A. van Kemenade and N. Vincent (eds) Parameters of morphosyntactic change. Cambridge: Cambridge University Press, 380–94. Wexler, K. 1998. Very early parameter setting and the unique checking constraint: a new explanation of the optional infinitive stage. Lingua 106: 23–79.

356

Bibliography

Williams, A. 2000. Null subjects in Middle English existentials. In S. Pintzuk, G. Tsoulas and A. Warner (eds) Diachronic syntax. Models and mechanisms. Oxford: Oxford University Press, 164–87. Williams, E. 1994. Thematic structure in syntax. Cambridge, MA: MIT Press. Willis, D. 1998. Syntactic change in Welsh. A study of the loss of verb-second. Oxford: Oxford University Press. Wurmbrand, S. 2004. Syntactic vs. post-syntactic movement. In S. Burelle and S. Somesfalean (eds) Proceedings of the 2003 Annual Meeting of the Canadian Linguistic Association (CLA), 284–95. 2006. Licensing Case. Journal of Germanic Linguistics 18,3: 175–236. Zribi-Hertz, A. 1994. La syntaxe des clitiques nominatifs en franc¸ais standard et en franc¸ais avance´. In G. Kleiber and G. Roques (eds) Travaux de linguistique et de philology. Strasbourg-Nancy: Klincksieck, 131–47. Zubizarreta, M.-L. 1998. Prosody, focus and word order. Cambridge, MA: MIT Press. Zwart, C. J.-W. 1990. Clitics in Dutch: evidence for the position of Infl. Groningen University. 1994. Dutch is head initial. The Linguistic Review 11: 377–406. 1997. The morphosyntax of verb movement: a minimalist approach to the syntax of Dutch. Dordrecht: Kluwer. 2007. Uncharted territory? Towards a noncartographic account of Germanic syntax. Groninger Arbeiten zur germanistischen Linguistik 45: 55–75. Zwicky, A. and G. Pullum. 1983. Cliticisation vs. inflection: English n’t. Language 59: 502–513.

Index

Abels, K., 179 Abraham, W., 170 acquisition, 15, 17, 18, 22, 25–6, 27, 32, 44, 47, 53 conservative, 40–1, 47 see also explanatory adequacy; poverty of the stimulus; primary linguistic data Adams, M., 196, 280 Adger, D., 170, 235–6, 293 Adraga˜o, M., 73 Afrikaans, 8, 19, 156, 167, 169–70, 171–4, 175, 180–2, 185–8, 189, 192, 195, 301 Agree, 32, 33, 39, 60–1, 66–8, 76, 81, 83, 89, 96, 97, 98, 104, 108, 109, 111, 120, 149, 150, 164, 166, 174, 178, 180, 182–3, 191, 194, 195, 197, 216, 217, 223, 230, 232, 265, 267, 268, 269, 270–2, 275–6, 278, 298, 306, 314, 322, 325 agreement, inflection, 2–3, 4, 5, 6, 8, 9, 10, 13–14, 17, 18, 19, 34, 37, 41, 49, 55, 62, 74, 85–6, 154, 165, 170, 201, 206, 214, 228, 232, 247, 263, 265, 300, 306–7 poor, 170, 180, 184, 266–7, 288, 301, 318 subject, 16, 23, 35, 42, 58, 60, 73, 75, 80, 130, 206, 226, 232 rich, 35, 107, 122, 154, 155, 159, 160, 170, 175, 176, 180, 183, 184, 196, 223, 263, 264–5, 266–7, 268, 269, 286, 287, 290, 301, 322, 325 see also Agree; features, f AGR head, 117, 214–16 Akmajian, A., 277 Aldridge, E., 198, 297–8, 299, 300 Alexiadou, A., 13, 34, 41, 58, 62, 75, 85, 87, 99, 100, 154, 156, 157, 161, 162, 165, 175, 191, 194, 214, 218, 232, 238, 252, 264 Allen, C., 184 Ambar, M., 240, 257, 258

Anagnostopoulou, E., 13, 34, 41, 58, 63, 75, 83, 85, 87, 99, 100, 154, 156, 157, 161, 162, 165, 175, 194, 214, 218, 232, 238, 252, 264 Anderson, S., 309 Arabic, 3, 91, 127, 132, 162, 198, 201 Classical, 300 Morrocan, 228 Standard, 228 Ariel, M., 119 Armstrong, L., 310, 311 Assamese, 93, 113 A-Topic, see Topic, Aboutness-shift Austronesian languages, 297–300 auxiliary, 20, 37, 68, 70, 98, 265, 270–8, 280, 283–4 selection of, 31 Avoid Pronoun Principle, 170 Baker, C., 77, 155 Baker, M., 31, 32, 33, 35, 36, 38–9, 41, 44, 45, 48, 49, 123, 165, 177, 194, 197, 232, 265, 277, 295, 301 Balkan languages, 135 Bantu, 68 see also Swahili Barbiers, S., 183 Barbosa, P., 13, 34, 41, 58, 85, 87, 99, 100, 103, 156, 157, 161, 214, 218, 232, 234, 238, 248, 259, 260, 264 Bare Phrase Structure, 63, 97 Basilico, 180 Basque 3, 86 Battistella, E., 282 Bayer, J., 193 Belletti, A., 119, 231, 242, 246, 249–50, 251, 253–4, 257 Bengali, 113, 127 Beninca`, P., 7, 37 Bennati, E., 116 Bennis, H., 156, 159, 167, 179, 187 Bentzen, K., 287, 288–5

357

358

Index

Berber, 284, 300 Berwick, R., 41, 53 Bhatt, R., 77 Biberauer, T., 1, 2, 8, 30, 40, 41, 43, 47, 55, 56, 68, 69, 74, 98, 108, 109, 112, 154, 157, 158, 159, 161, 163, 164, 165, 168, 169, 170, 171, 172, 173, 174, 175, 176, 178, 179, 180, 181, 182, 184, 186, 187, 188, 189, 192, 193, 197, 198, 199, 231, 264, 267, 268, 278, 279, 280, 281, 282, 283, 284, 287, 288, 289, 290, 297, 298 Blevins, J., 202, 226 Bobaljik, J., 98, 104, 154, 175, 183, 186, 189, 192, 217, 233, 264, 267, 270, 287, 289 Boden, M., 45 Boeckx, C., 54–6, 58, 104, 108, 163 Bolinger, D., 159, 191 Borer, H., 13, 32–5, 59, 87–8, 161 Borer–Chomsky Conjecture, 32–5, 39, 44, 50, 54, 197, 301 Borsley, B., 19, 293, 300 Bosˇkovic´, Z., 233, 255, 256 BPS, see Bare Phrase Structure Brandi, L., 304, 305, 320 Breivik, L., 170 Bresnan, J., 234, 240 Breton, 165, 198–9, 300 Broekhuis, H., 185 Burridge, K., 170 Burton-Roberts, N., 54 Burzio, L., 85, 94, 243, 303 Cabredo Hofherr, P., 132, 202 Calder, G., 294 Cape Verdean, 8, 22 Cardinaletti, A., 8, 13–14, 64, 66, 70–1, 72–3, 85, 87, 94, 99, 118, 156, 159, 163, 167, 186, 256, 306, 312–13, 317, 321, 322, 323–4 Carilho, E., 197 Carnie, A., 69, 157, 263, 293, 294, 295, 296 Carrara, 306–7 case, morphological, 10, 37, 42, 206–7, 209, 223, 226 Case, syntactic, 9, 60, 201, 209–10, 214, 215–16, 218, 220–1, 229, 238, 257, 270 see also features, Case Catalan, 244 Cecchetto, C., 163, 265, 267 Celtic, 5, 19, 38, 69, 165, 263, 284, 285, 291, 295, 296, 299, 300, 301 see also Irish; Welsh C-head, 20, 21–2, 33, 34, 43, 65, 72, 107, 108, 163, 178, 194, 195, 198, 211, 218, 219,

220, 246, 265, 268–9, 270, 273–4, 276, 283, 284, 287, 288, 291–5, 296–7, 298, 299, 301–2, 311–12, 313–16, 317, 318, 319–21, 322, 323, 325 Chain, 58, 76 Chamorro, 198 Chaˆtillon, 320 Chenal, A., 320 Chierchia, G., 9, 37 Chinese, 8, 9, 13, 29, 33, 37, 42–3, 62, 78, 82, 106, 122, 161 Cantonese, 130 Chomsky, N., 4, 13, 15, 18, 25–7, 32–5, 39–40, 43, 50, 54, 58, 59–60, 65, 66, 67, 76, 77, 78, 79, 84, 97, 108, 109, 110, 111, 116, 120, 133, 137, 156, 158, 160, 161, 163, 164, 165, 166, 170, 172, 175, 178, 179, 188, 191, 192, 194, 195, 204, 216, 219, 220, 223, 231, 232, 233, 234, 235, 246–7, 253, 255, 264, 265, 267, 268, 269, 270, 271, 272, 276, 284, 289, 303, 313, 314, 322, 325 Chung, S., 198, 297 Churchward, C., 296 Cinque, G., 42, 191, 201, 202, 228, 233, 234, 237, 250, 255, 272, 316 Clark, R., 41 clitics, 12, 13, 31, 37, 41, 45, 63–8, 69, 70, 71, 74, 76, 89, 133, 238 subject, 3, 37, 64, 65, 73, 82, 87, 303, 304–7, 308, 311, 318, 321, 326 object, 3, 37, 64, 65, 120, 249, 259–60, 261, 318, 320, 323 Cole, M., 13, 99, 119 contact, language, 38 control, object, 134, 142, 144–6, 149 subject, 4, 58, 87, 126, 134, 137, 139, 142, 143, 147, 150, 214, 226 non-obligatory, 126, 136 obligatory, 126, 134, 135, 136, 138, 139–41, 146, 149, 150–1, 152 complementiser-trace effects, 3, 16–18, 19–22, 34, 114, 193, 231, 240 Como, 306 Comrie, B., 5, 26 copy theory of movement, see movement; copy theory of Coptic, 300 Cordin, P., 304, 305, 320 Costa, J., 73, 100, 232, 237, 239, 249, 257, 258 Cottell, S., 291 Cowper, E., 163

Index creoles, 8, 17, 22, 116, 156, 170 see also Cape Verdean; Dutch, Berbice; Haitian; Jamaican; Kriyol; Mauritian; Papiamentu; Saramaccan Croft, W., 21 Csato´, E., 6 Culicover, P., 55, 58 D-head, 64, 65–6, 72, 73, 75, 121 D’Alessandro, R., 121, 278, 322 Dahl, O., 299 Danish, 106, 167, 168 Davies, W., 58, 157, 166, 167, 176, 297 Deal, A., 180 De´chaıˆne, R., 64, 94 defectivity, 14, 64, 66, 69, 70 De Hoop, H., 185 deletion, 9, 58, 59, 62, 63, 66, 68, 70, 75, 76, 77, 83, 84, 85, 87, 89, 90, 223, 261, 290, 316, 317 recoverability of, 84 see also null subject; analyses; deletion Dell, F., 309, 310 den Besten, H., 276, 293, 312, 315 den Dikken, M., 180, 193 descriptive adequacy, 15, 25, 27, 30, 31, 32, 35, 39, 50 Diesing, M., 160, 185, 188, 189, 289, 298 di Sciullo, A.-M., 304 Distributed Morphology, 169, 217 Doherty, C., 293, 295 Dooley, S., 69, 263, 295, 296 Donaldson, B., 170 Donati, C., 265, 267 Doron, E., 300 Dravidian, 122–3 Dresher, E., 44, 47 drift, typological, 38 Dryer, M., 5, 24, 25, 26, 29, 36, 40 Duarte, E., 11, 128 Dubinsky, S., 58, 157, 166, 167, 176, 297 Dupuis, F., 290 Dutch, 8, 19, 29, 153, 154, 155, 156, 167, 169, 170, 171, 172, 175, 180, 182–3, 185–8, 189, 190, 191, 193, 195–6, 197, 255, 279–80, 324 Berbice, 8, 22 Dyscolus, Apollonius, 3, 75 East Asian languages, 6, 27, 62, 78, 82 Edge Feature, see features, movement, Edge efficiency, computational, 29, 51–2 performance, 28–9

359 Egerland, V., 106, 128, 179, 201, 202, 208, 219, 228 Egyptian, see Coptic Ellega˚rd, A., 281 ellipsis, 9, 10, 64, 131 Elsewhere Condition, 9–10 Emonds, J., 264, 271, 275, 276, 279 empty categories, 17, 31, 58, 86 see also traces; pro; PRO English British, Northern, 277 Middle, 24, 176, 184, 264, 267, 269, 278–81 Modern, 3, 4, 5, 7, 8, 9, 12, 14, 16–18, 19, 20, 22, 33, 41, 42, 49, 64, 69, 82, 86, 88, 90, 91, 93, 105–6, 112, 113, 116, 127, 143, 155, 165, 166, 167, 170, 173–4, 175, 179, 184, 186, 187, 195, 197, 200, 202, 203, 216, 222, 231, 234, 235–6, 242, 243, 248, 252, 253, 261, 263, 264, 265, 266, 267, 268, 269, 270–8, 279, 281, 282, 283, 284–5, 287, 288–9, 290, 295, 297, 298, 307, 312, 314 Early Modern, 269, 278, 279, 281 Old, 74, 176, 184, 269, 274, 279 Ozark, 20 EPP feature, see also features, movement, EPP Epstein, S., 157, 231 Etruscan, 38 Evers, A., 173 explanatory adequacy, 15, 17, 25–7, 32, 34, 35, 39, 50 see also poverty of the stimulus expletives, 3, 8, 61–2, 94, 128, 132, 155, 156, 157–9, 162, 163, 166, 167, 183, 187, 190, 191, 192–3, 195–6, 198–9, 204, 209, 211, 223, 233, 243, 264, 267, 284, 287, 290, 319 null, see null subject, expletive Extended Projection Principle, 4, 13, 18, 20, 34, 43, 59, 62, 67, 69, 73, 74, 85, 97, 99, 100–1, 102, 103, 105, 108–9, 110–12, 117, 119, 122, 124, 128, 129, 130, 133–4, 156, 157, 158, 159, 161, 167, 171, 176, 177, 180, 184, 189, 195, 199, 201, 204, 209, 212, 218–19, 220, 231, 232, 233, 238, 240–1, 242, 243, 244, 246, 247–8, 249, 260, 262, 267, 290, 295–6, 297, 298, 300, 319 Extension Condition, 65 Falk, C., 153, 167, 170, 183, 196 Faroese, 156, 158, 167, 176, 184 Fassi Fehri, A., 13, 89, 93 Fava, 323

360

Index

features, formal, 33–4, 35, 265 anaphoric, 9, 58 Case, 9, 32, 66, 72, 79–80, 83, 117, 216–17, 220–1, 224 Categorial, 32 checking, 267 D, 14, 34–5, 62, 72, 75, 78, 79, 82, 85, 87, 94–5, 97, 98, 99, 101, 103, 104, 105, 106, 117, 119, 122, 123, 125, 130, 132, 133, 150, 151, 161, 163, 178, 196–7, 216, 218, 229, 246, 264, 265, 266, 291, 304, 305, 316, 321, 324 f, 32, 41, 43, 59–60, 61, 62, 65, 67, 75, 76, 78, 81, 82, 83, 85, 86, 87, 94–5, 97, 99, 101, 111–12, 117, 120, 122, 123, 130, 152, 163, 166, 175, 206, 208, 216, 217, 218, 220–2, 224, 229, 247, 248, 263, 272, 274, 275, 276, 282, 284, 285, 288, 298, 300, 304, 305–6, 308, 312, 313, 314, 316, 318–19, 320–1, 322, 323, 324, 325, 326–7 interpretability, 34–5, 43, 60–1 matrix, 60–1, 66 movement, 33 Edge, 32, 43, 65, 67, 78, 195, 268, 298 EPP, 13, 32, 39–40, 43, 46, 54, 58, 66, 67–8, 70, 75, 76, 78, 79, 116, 120, 161, 164, 175, 188, 216, 218, 221, 224, 230, 247–8, 260, 264, 267, 276, 284–5, 288, 290, 297–8, 298–9, 302, 316, 319, 323, 325 pronominal, 9, 14, 58, 88 see also features, D strength, 267 T, 68, 263, 265, 266, 268, 270, 271–2, 274, 275, 276, 282, 283, 284, 285, 287, 291, 293, 294, 296, 298, 300 V, 178, 265, 268, 270, 271, 275, 278, 282, 296, 297, 298, 300, 301 valuation, 59, 60–1, 66, 70, 265 see also Agree wh-, 33 Felser, C., 255 Ferraresi, G., 293 Ferreira, M., 139, 146 Finnish, 3, 11–12, 61–2, 63, 72, 75, 85, 88, 89–90, 91, 92, 93, 94, 95, 99, 101–2, 103, 104, 105, 113, 114, 116, 117, 118, 119, 121, 122, 123, 124, 125, 126, 127, 128–9, 130, 131, 132, 134, 135–41, 147, 151, 152, 153, 156, 200–16, 209–10, 218, 220–2, 223–7, 228, 229–30 Fischer, O., 269, 279 Fitch, T., 33

Fitzpatrick, J., 194 Flemish, West, 193, 284 Florentine, 305 Focus, 232, 243, 246, 252–3, 256, 258, 261, 290, 298, 299 Contrastive, see Focus, Identificational Identificational, 234, 237, 244, 252, 257, 261 Informational, 234, 237 Narrow, 233, 234, 237, 241, 242, 243, 249, 257, 260, 261 Wide, 233, 234, 237, 238, 239, 241, 262 Fox–Reinhart Intuition, 172, 188, 192, 272, 289, 290 Franco-Provenc¸al, 86, 311, 319, 322 Franks, R., 198 Frascarelli, M., 7, 95, 98, 99, 100, 104, 118, 126, 133, 134 Freeze, R., 94, 166 Freezing, 179 Freiden, R., 270 French Colloquial, 266, 308, 322 franc¸ais avance´, 327 franc¸ais parle´ courant, 327 franc¸ais standard moderne (FSM), 327 franc¸ais tre`s e´volue´ (FTE), 326 Modern, 3, 4, 5, 7, 17, 18, 20, 22, 33, 50, 64, 67, 69, 78, 86, 91, 112, 113, 121, 122, 124, 132, 155, 165, 196, 197, 200, 201–2, 222, 223, 232, 243, 264–5, 267, 269, 275, 279, 280, 281–2, 284, 285, 288, 290–1, 295, 299, 303–27 Old, 24, 114–15, 280, 290 Register variation in, 311, 326–7 Quebec, 308, 326, 327 Frey, W., 185, 186 Friulian, 323, 324 Fuss, E., 154, 269 Gaelic, Scots, 293–4, 300 Gallego, A., 247 GB Theory, see Government and Binding Theory Generalise Features 52 GenF, see Generalise Features German, 8, 19, 29, 40, 44, 81, 112, 113, 128, 153, 154, 155, 157–60, 162, 167, 168, 169, 170, 171, 172, 175, 176, 177, 179–80, 181, 184, 185, 186, 187, 188–9, 190, 191, 193, 194, 195–6, 197, 200, 217, 222, 223, 255, 266, 279–80, 285, 287–8, 289, 290, 302, 307 Bavarian, 153, 154, 193, 284

Index Germanic, 8, 19, 24, 26, 155, 156, 157, 162, 164, 166–96, 197, 201, 246, 266, 268–70, 278, 279, 283, 284, 286–8, 296, 297, 298, 300–1, 323 Continental, 265 West, 74, 284 see also Afrikaans; Dutch; English; German gerundive, 59 Gianollo, C., 26–7, 32, 34, 35, 44, 45, 46, 47, 54, 155 Gil, D., 5, 26 Gilligan, G., 5, 18, 21, 23, 26, 46, 54, 153, 170 Giorgi, A., 237, 247 Gold, E., 288 Goodall, G., 247 Gothic, 293 Government and Binding Theory, 58, 86, 87, 154, 155, 157, 164, 170 Greek, Ancient, 3, 38, 75 Modern, 4, 5, 6–7, 12, 16, 17, 62, 68, 91, 127, 129, 154, 175, 201, 228, 232, 264, 267, 290 Greenberg, J., 18, 19, 25–7, 31, 68 Grevisse, M., 315 Grewendorf, 153, 176 Groat, E., 243 Guardiano, C., 26–7, 31, 32, 34, 35, 44, 45, 47, 155 Guasti, M.-T., 322 Guilfoyle, E., 157, 263, 299 Gutman, E., 143 Haeberli, E., 67, 169, 184, 269 Haegeman, L., 5, 90, 127, 193, 194 Haider, H., 162, 176, 185 Haitian, 8, 22 Hakulinen, A., 200, 204, 211 Hale, K., 5, 19, 180 Hall, R. A., 163 Halle, M., 39–40, 169, 217 Harley, H., 69, 263, 293, 294, 295, 296 Harris, M., 304, 308–9 Hartmann, J., 186, 187 Haspelmath, M., 5, 21, 26 Hausa, 228 Hauser, M., 33 Hawkins, J., 27–8, 29–30, 39, 51–2 Hazout, I., 180 Hebrew, 3, 11, 62, 102, 104, 113, 134–5, 142, 143, 153, 201, 227–8 Biblical, 300 Helasvuo, M.-L., 202, 219, 222, 226 Hendrick, R., 292 Hermon, G., 153

361 Heycock, C., 167, 184, 278 Hindi, 113, 127 Hinterho¨lzl, R., 95, 176 Hoekstra, T., 180, 240 Holmberg, A., 5, 8, 10, 13, 14, 18, 19, 23, 26, 31, 34, 47, 49, 55, 56, 59–63, 70, 71, 72, 73, 74, 75, 79–80, 81, 85, 86, 87, 88, 90, 93, 94, 95, 97, 99, 101, 102, 104, 107–8, 109, 110, 113, 114, 116–17, 120, 121, 122, 123, 124, 125, 127, 128, 129, 130, 131, 132, 133, 135, 136, 138, 154, 155, 156, 159, 161, 162, 163, 168–9, 171, 179, 183, 189, 191, 192, 196, 199, 200, 204, 211, 212, 215, 218, 221, 222, 223, 226, 228, 229, 233, 260, 260, 261, 263, 264, 282, 287, 288, 290, 297, 308, 316 Hornstein, N., 58, 104, 139 Horvarth, J., 181 Hrafnbjargarson, G., 191, 192 Hro´arsdo´ttir, T., 110 Huang, J., 8, 42–3, 56, 62, 78, 83, 88, 128, 153, 278, 284 Hulk, A., 231, 304 Hung, H., 299 Hungarian, 248 I-head, 13, 34, 59, 60, 62, 154, 156 see also T-head Icelandic, 11, 19, 89, 102, 106–8, 109, 110–11, 112, 113, 115, 121, 122, 124, 127, 128, 153, 154, 155, 157, 158, 159, 167, 168–9, 170, 173–4, 175, 176, 177–80, 187, 189–94, 195–6, 201, 208, 218, 221, 227–8, 281, 285–90, 297, 302 Ihsane, T., 163 impersonal construction, 3, 12, 70, 201, 202, 212, 215, 228 impoverishment post-syntactic, 81 pre-syntactic, 48–9, 80–2, 84, 85, 87, 112, 155, 308, 309–10, 311, 319, 322 Inclusiveness Condition, 83–4 incorporation, 41, 64–8, 69, 70, 75, 76, 89, 97, 98, 99, 111, 117, 120, 122, 152, 201, 211, 217, 218, 222, 319 noun incorporation, 295 see also clitics Indic, 11, 93, 101, 113 see also Assamese; Marathi Indo-Aryan, see Indic Indo-European, 5, 21, 38, 44, 197 infinitives, 20, 37 see also null subject

362

Index

inflection, see agreement, inflection; tense, inflection Inversion, Complex, 315, 316, 318, 320, 321, 322, 324–5, 326, 327 Free see Inversion, Subject Stylistic, 307–8, 311, 317, 318, 325, 326 Subject, 4, 16, 17, 18, 19, 20, 21–2, 34, 72, 73, 231, 232–3, 237, 238–62, 307, 313, 322 Subject-clitic, 311–25 Irish, 5, 19, 198, 236, 291 Modern, 294–5 Old, 293–4, 295 Italian, 3, 4, 6, 7, 11, 12, 16, 17, 22, 49, 59, 62, 70–1, 71–3, 74, 78, 82, 88, 91, 92, 94, 95, 96, 99, 100, 104, 105, 112, 116, 118, 120, 121, 123, 154, 155, 175, 200, 201, 208, 223, 228, 231, 232, 234, 236, 237–40, 241, 242–3, 244, 246, 247–8, 249–54, 255, 257, 258, 260, 261, 264, 265, 267, 285, 290, 303–4, 311, 318, 319 Italian dialects, North, 304–7, 314, 321–5 Bellunese, 323 Donceto, 324 Veronese, 323 Vicentino, 323

Keyser, S., 180 King, G., 293 Kiparsky, P., 9, 206 Kiss, K., 211, 234 Koeneman, O., 186, 187, 264, 267, 269 Korean, 8, 83, 106, 122, 255 Kornfilt, J., 74, 132 Kosmeijer, W., 264 Koster, J., 153 Kratzer, A., 189 Kriyol, 8, 22, 121, 122, 124 Kroch, A., 170, 174, 281–2 Kuroda, Y., 10, 83

Jacobs, J., 176 Jackendoff, R., 55, 58, 234 Jaeggli, O., 155, 304 Jaggar, P., 228 Jamaican, 8, 22 Japanese, 8, 10, 13, 40, 42, 74, 78, 83, 122, 130, 161 Jayaseelan, K., 9, 122–3 Jelinek, E., 185 Jespersen, O., 3, 6, 281, 282 Johansen, L., 6 Johns, C., 120 Johnson, K., 177, 277, 285, 287 Jonas, D., 184, 186, 189, 289 Jo´nsson, J., 168, 191 Jouitteau, M., 165, 198, 300 Julien, M., 44

Ladusaw, W., 296 Laitinen, L., 200, 203, 204, 205, 206, 208, 209, 212, 219, 222, 226 Lambrecht, K., 232, 242, 326 Landau, I., 58, 98, 104, 115, 126, 134, 135, 136, 137, 138, 139, 150, 151, 217, 233, 249 Classification of verb types, 142 language faculty, 29, 33 Larson, R., 82 Lasnik, H., 76, 270 Latin Classical, 2, 5, 37–8, 40, 44, 274 Vulgar, 37–8 Lechner, W., 165, 269 Legibility Condition, 83 Levy, Y., 104, 127, 200, 204, 222 Lexicon, 32, 268, 301 Li, C., 29, 176, 198 Lightfoot, D., 172, 280 Lindseth, M., 132 Lobeck, A., 282 locality, 33, 52, 137 locative inversion, 167, 233, 238, 240, 248 Lo¨flund, J., 202 Logical Form, 17, 169, 189, 192, 203, 253, 275, 318 Logophoricity, 136 Lohndal, T., 19, 20, 193 Longobardi, G., 26–7, 31, 32, 34, 35, 44, 46, 47, 54, 155, 163, 293 Lyngfelt, B., 138 Lyons, C., 163

Kanerva, J., 205 Kato, M., 73, 248 Karttunen, L., 200, 204 Kashmiri, 114 Kauffman, S., 34 Kayne, R., 30, 37, 40, 41, 42, 44, 64, 68, 78, 94, 163, 254, 260, 272, 303, 305, 307–8, 311, 312, 314, 317, 318, 325, 327

Macedonian-Bulgarian, 68 Mac Eoin, G., 300 Mackenzie, D., 79 Macronesian languages, 109 Malagasy, 165, 198, 299 Malayalam, 40 Maling, J., 168, 191, 194, 206 Manninen, S., 202, 226, 227

Index Manzini, R., 13, 36, 58, 73, 87, 232, 305, 306–7, 319, 321, 322 MaOP, see Maximize Online Processing Maori, 296 Mapudungun, 123 Marantz, A., 63, 169, 217 Marathi, 11, 62, 74, 91, 101–2, 103, 113, 125, 126, 127, 128–9, 131, 132, 134, 135, 139, 140, 147, 152, 201, 227–8 Martin, R., 58 Massam, D., 69, 109, 165, 198, 295–6, 295–6 Masullo, P., 112 Mathieu, E., 192, 282, 290 Matushansky, O., 67, 98, 165, 269 Mauritian, 8, 22 Mavrogiorgos, M., 63 maximal category, 63, 67, 71 Maximize Online Processing, 52 May, R., 82 Mayan, 69, 296 Mayr, C., 193 McCloskey, J., 5, 19, 198, 291, 294 McShane, M., 154 Menuzzi, S., 232 Merge, 32 Meso-American languages, 109 MiD, see Minimize Domains MiF, see Minimize Forms minimal category, 63–5, 67 minimalism, see Minimalist Program Minimalist Program, 32, 156, 164, 232, 233, 267 Minimize Domains, 28–9, 52 Minimize Forms, 52 Mobbs, I., 29, 51–2 Modesto, M., 78, 104, 136, 139, 145, 146, 148, 154 Mohawk, 41, 123 Mohr, S., 159, 162, 167, 176, 179, 186, 187 Moltmann, F., 92, 106, 203–4, 219, 229 Montalbetti, M., 85 Moro, A., 180 morphology agglutinating, 10, 44, 74 fusional, 9–10, 44 see also agreement, inflection; tense inflection Morpurgo-Davies, A., 44 Mosse´, F., 278 Move, 33, 66–7, 76, 325 movement, 43 A, object, 40, 68, 69, 295 subject, see subject, extraction; subject, lowering; subject, raising; Inversion, Subject

363 VP, 40, 42, 68, 69, 74, 109, 254, 261, 295, 296, 297–8, 298–300 vP, 74, 173, 176, 177, 179, 184, 188, 192, 195, 287, 288, 299, 300 A0 (A-bar), 20, 300 VP/vP fronting, 267, 276, 277, 278, 325 wh, 3, 16, 18, 20, 22, 30, 33, 42, 169, 256, 315 see also complementiser-trace effects auxiliary, 68 covert, 17, 267 copy theory of, 58, 59, 66, 86 head, 43, 46, 49, 63–70, 87, 98, 165, 217, 269, 292, 301, 314 N-to-n, 43 Aux-to-C, 72, 240, 258, 262, 274, 276, 283, 318, 322 V-to-C, 293, 295, 296, 318 see also word order, V2 V-to-T, 23, 30, 42, 43, 69, 74, 98, 99, 156, 161, 163, 165, 173, 176, 178, 183, 184, 193, 196, 197, 238, 241, 263–6, 267, 269–70, 278–83, 285–8, 291, 296, 299, 300–1 V-to-v, 40, 43, 65, 74, 284 remnant, 40, 64, 69, 74, 163, 234, 254, 261, 287, 288, 299, 318 scrambling, 236, 254–5, 257, 260, 261, 289 see also Move; features, movement Mulder, R., 180, 240 Mu¨ller, G., 48, 80–2, 83–4, 85, 86, 87, 112, 154, 155, 179, 180, 192, 194, 268, 290, 306, 308, 310–11, 316, 318, 327 Munaro, N., 316 Muysken, P., 64 Nash, L., 13, 58 Nayudu, A., 10, 62, 63, 73, 74, 75, 82, 90, 104, 113, 114, 117, 125, 138, 145, 148, 212, 218, 222, 223, 228, 308 Neeleman, A., 9–10, 62, 79, 83, 161, 170, 185, 186, 187, 256, 264, 267, 299 Negra˜o, E., 73, 139 Nelson, D., 202, 226, 227 Newmeyer, F., 21, 23–32, 28–30, 35, 45, 51–2, 55, 166, 199 Newton, G., 293 Nicolis, M., 8, 22, 116, 121, 156, 197, 232 Nikkanne, U., 94, 102, 104, 109, 116, 128, 204, 211, 212 Niuean, 69, 165, 198, 295–6, 295–7, 299, 301 Niyogi, P., 53 N-head, 295 No Tampering Condition, 52 Norwegian, 106, 109, 167, 288 NSP, see null subject, parameter

364

Index

null object, 9, 12, 223–5 indefinite, 223 null subject, 2–5, 194, 226, 238, 263, 264, 266, 286, 287, 288, 301, 303, 308, 316, 321, 324, 325, 327 analyses, deletion, 59–70, 161, 163, 165, 170, 194, 195, 261, 327 pro as Dmin/max, 87 see also incorporation I-subject, 13, 59, 85, 87, 161, 162, 165, 193–4 pro-subject, 13–14, 58, 59–62 arbitrary, see null subject, indefinite, exclusive definite, 3, 4, 5, 9, 12, 16, 21, 22, 88, 89, 103, 105, 116, 125, 133, 211, 213, 215, 222 see also impersonal construction expletive, 5, 14, 21, 22, 23, 27, 70, 88, 107, 116, 128, 133, 152, 155, 166, 167, 170, 184, 198, 200, 228, 231, 243, 260, 290, 308 see also null subject, languages; “expletive null-subject” generic, see null subject, indefinite indefinite, 3, 11, 80, 89, 92, 105, 106, 117, 124, 128, 129, 152, 203, 207–16, 217–24, 227, 228, 229–30 inclusive, 92–3, 128, 130, 203, 204, 208, 229 exclusive, 92, 130, 203 languages, 1, 3–5, 14, 17, 19, 20, 88, 153, 154, 218, 231, 232, 243, 264–5, 266, 288 “consistent”, 6–8, 12, 27, 58–9, 62, 74, 75, 82, 85–7, 88, 89, 90–4, 94–5, 97–101, 103, 105, 114, 115, 117, 118, 119, 122, 125, 126, 127–8, 129, 130, 132, 133, 134, 153, 154, 156, 157, 160, 161, 163, 165, 193–4, 195, 201, 228, 229, 285, 288, 290, 306, 307, 311, 318, 326 “discourse pro-drop”, 8–10, 12, 27, 42, 43, 49, 74, 78, 82, 85–7, 106, 122–3, 130, 161 “expletive null-subject”, 8, 12, 112, 128, 153–7, 170, 196, 197, 199, 307 With null quasi-argumental subjects (Type I), 153, 154, 157, 159 With overt quasi-argumental subjects (Type II), 153, 154, 157, 159 “Full”, see null subject, languages, “consistent” “non-null-subject”, 4, 5, 17, 20, 86, 88, 89, 106–15, 126, 127, 128, 154, 163, 223, 303–4, 306, 325, 327

overt pronouns in, 3, 7, 16, 93, 116 “partial”, 5, 6, 10–12, 18, 19, 27, 50, 73, 74, 75, 78, 86, 87, 88, 89, 90–4, 94–5, 101–4, 105, 106–7, 115, 116, 117, 118–19, 122, 125, 126, 127–8, 131, 132, 133–4, 134–52, 153, 154, 160, 163, 228–9, 308, 311, 325, 326 Control in, 141 Summary of properties of, 132 “radical pro-drop”, see null subject, languages, “discourse pro-drop” “semi-pro-drop”, see null subject, languages, “expletive null-subject” typology of, 3, 5–13, 18–19, 22–3, 264 see also null subject, parameter, clustering effects non-referential, see null subject, expletive parameter, 5, 8, 14, 22, 24, 26, 27, 29–30, 31, 34–5, 41, 44, 48, 58, 86, 88, 107, 119, 131, 231, 232, 248, 260, 261, 264, 300, 303–4 clustering effects, 7, 14–23, 31, 35, 44, 55 formulations of, 8, 14, 17–18, 31, 34, 116, 264–5 referential, 22 see also null subject, definite; null subject, indefinite see also pro; pronouns, weak; impersonal construction Numeration, 174, 176, 180, 188, 267–8, 269, 292, 296 Nunes, J., 58, 66, 71, 84, 98, 233 observational adequacy, 15 O’Neil, J., 58 Optimality Theory, 249 Optionality, 159, 169, 183–4, 188, 192, 198 Ordo´n˜ez, F., 13, 58, 73, 244, 245, 250, 251–2, 253, 254 Oscan, 38 Otsuka, Y., 296 Ouali, H., 284, 288, 300, 301 Ouhalla, J., 198, 300 ¨ ztu¨rk, B., 6, 74 O P & P, see parameters, theory of Paduan, 322, 323, 324 Paesani, K., 299 Palmer, L., 37 Pancheva, R., 77 Papiamentu, 8, 22 parameters, Agglutination, 44 Case-Depencency of Agreement, 31

Index clustering effects, 14–23, 15–16, 18–19, 20–4, 27, 30, 35, 39, 282 see also null subject, parameter, clustering effects Direction of Agreement, 31, 194 expression of, 16 Head, 25, 30, 36, 39–40 High Analyticity, 42–4 Hierarchies of, 46–50, 86, 121–4, 155 microparameters, 30, 31, 35–9, 41–4, 47–50, 56, 86 macroparameters, 31, 35, 36–7, 38–9, 41–4, 45, 47–50, 51, 57, 87 markedness, 34, 35 of parameter systems, 39–41, 46, 57 Negative Concord, 30 Nominal Mapping, 9, 14, 37 Polysynthesis, 31, 36, 41–2, 44, 122 Serial Verbs, 31 Subject Side, 31 theory of, 15–16, 23–7, 30–57, 86, 117, 233, 260–2 triggers of, 16, 31 see also Borer–Chomsky Conjecture Parry, M., 320 Pashto, 78–9 passive, 12, 202, 205, 215, 225–7, 228 Paul, I., 198, 297, 298, 299 Pearson, M., 298 Performance-Grammar Correspondence Hypothesis, 28, 30, 52 Performative Hypothesis, 98 Perlmutter, D., 3–4, 16–17, 62, 81, 87, 303 Pesetsky, D., 61, 84, 120, 178, 216, 273 PF, see Phonological Form phase, 40, 43, 52, 54, 60, 63, 73, 220, 222, 235, 301–2, 303, 319, 322 phase-head, 64, 65, 66, 70, 72, 73 Phase Impenetrability Condition, 43, 52, 60, 151, 194 Phonological Form, 58, 61, 67, 76, 84, 161, 163, 169, 170, 178, 183, 189, 192, 194, 232, 233, 234, 235–6, 240, 249, 253, 255, 256, 262, 305, 312 Pianesi, F., 237, 247 Piattelli-Palmarini, M., 45 Piedmontese, 320 “pied-piping”, 40, 46, 67, 68, 75, 77, 165, 174, 175, 176, 178, 179, 180, 181, 183, 184, 188, 189, 198 Pinto, M., 101, 238, 239, 240, 242 Pittner, K., 185 Platzack, C., 13, 49, 58, 89, 99, 107–8, 153, 154, 159, 171, 179, 183, 196, 281, 287, 290

365 Poletto, C., 37, 64, 305, 306, 312, 314, 318, 319, 320, 321–3, 324, 325 Pollock, J-Y., 13, 58, 64, 67, 231, 254, 264, 269, 270, 275, 278, 279, 281, 307–8, 311, 312, 313, 314, 316, 317, 318, 319, 325, 327 Polynesian languages, 69, 263, 296, 297 Ponelis, F., 170 Poole, G., 54 Portuguese, 101 Brazilian, 11, 18, 19, 62, 73, 75, 78, 88, 91, 92, 93, 94, 101, 102, 103, 104, 105, 112, 113, 119, 125, 127, 128–9, 130, 131, 132, 133, 134, 135–40, 147, 151, 152, 174, 201, 227–8 European, 11, 73, 92, 100, 101, 112, 118, 119, 127, 130, 201, 228, 232, 234, 237–8, 239–41, 242, 243, 244–5, 246, 247–8, 249, 255, 257–61, 261 Postal, P., 167 poverty of the stimulus, 15, 22, 26, 53 see also explanatory adequacy primary linguistic data, 16, 17, 22, 25, 26, 27, 45, 51, 53 see also poverty of the stimulus Principles and Parameters Theory, see parameters, theory of pro, 13–14, 18, 20, 58–9, 62, 70–5, 79–80, 82, 117, 119, 135, 154, 160, 162, 170, 214, 232, 303, 306, 308, 311, 324, 327 proarb, 120, 201, 202 proEXPþA, 154, 155, 156, 159, 160, 179, 181, 196, 323 proEXPA, 154, 155, 156, 157, 159, 160, 161–2, 176, 181–2, 183, 184, 187, 193, 195 proEXP, 154, 156, 157, 158, 160, 162, 164, 167, 168, 169, 170–1, 177, 178, 196, 197, 199 proREF, 154, 155, 156 see also null subject, analyses, pro-subject PRO, 58, 87, 133, 134, 135, 136, 137, 142, 143, 150, 151, 170, 179, 205, 214, 226 pro-drop parameter, see null subject, parameter pronouns, 3, 9, 61–2, 71 clitic, see clitics definite, 7, 11, 214, 218 see also null subject, definite expletive, expletives; null subject, expletive generic, see pronouns, indefinite; null subject, indefinite impersonal, 179, 201, 228 indefinite, 108, 130, 200, 203, 204, 208, 209, 225

366

Index

pronouns (cont.) generic, 201–4, 219, 223, 228 arbitrary, 201–2, 228, 316 specific, 201–2 see also null subject, indefinite non-referential, see expletives null, see null object; null subject referential, 4, 8 see also null subject, referential; null subject, definite; null subject, indefinite; pronouns, definite; pronouns, indefinite strong, 64, 71, 72, 118, 307, 317 weak, 14, 62, 64, 70, 72, 73, 87, 308, 312, 313, 316, 317, 319 quantifier, floated, 72 universal, 106 Pullum, G., 272, 274, 314, 323 Puskas, G., 163 Pyatt, E., 293, 294 Rackowski, A., 69, 295, 296 Ramchand, G., 9, 43, 296 Raposo, E., 78, 234, 248 Rebusci, G., 132 Reime, H., 206 Reinhart, T., 233, 234, 252, 255, 256 Renzi, L., 7, 306, 321, 322, 323 Repetti, L., 306, 321, 322, 323–4 Rezac, M., 180 Rhaeto-Romansch, 5, 18, 37 Richards, M., 68, 74, 157, 158, 167, 169, 170, 175, 176, 179, 180, 181, 182, 186, 187, 188, 189, 194, 198, 267, 287, 289 Rivero, M-L., 228 Rizzi, L., 5, 8, 9, 13, 14, 16, 17, 18, 19–20, 21, 22–3, 25–6, 27, 29, 31, 34, 35, 56, 58, 59–60, 61, 63, 71, 72, 75, 78, 85, 88, 89, 94, 107, 108, 110, 117, 119, 120, 127, 154, 155, 156, 157, 158, 193, 194, 197, 208, 219, 223, 231, 237, 246, 263, 303, 304, 305, 312, 314, 315, 317, 319, 320, 322 Roberge, Y., 304, 322, 326 Roberts, I., 2, 5, 8, 9, 10, 11, 13, 14, 16, 19, 22, 23, 26, 30, 32, 34, 35, 39, 40, 41, 43, 45–6, 47, 48, 50, 54, 55, 63, 64, 65, 67, 68, 69, 73, 74, 75, 86, 87, 89, 94, 97, 98, 99, 108, 109, 111, 112, 114, 117, 120, 121, 122, 123, 130, 132, 133, 154, 155, 159, 161, 162, 163, 164, 165, 170, 172, 173, 174, 176, 177, 178, 184, 193, 194, 196, 197, 201, 217, 218, 220, 222, 261, 263, 264, 265, 268, 269, 276, 277,

278, 279, 280, 281, 282, 283–4, 285, 288, 293, 300, 303, 304, 306, 308, 312, 313–14, 315, 316, 317, 318–19, 320, 322, 324, 327 Rodrigues, C., 127, 133, 136, 139, 146 Ro¨gnvaldsson, E., 110 Rohrbacher, B., 264 Romance, 5, 7, 14, 18, 37–8, 41, 63, 64, 65, 68, 69, 74, 93, 99, 100, 130, 132, 159, 160, 165, 179, 193, 194, 197, 198, 201, 228, 231, 232, 235, 238, 265–6, 268, 269, 270, 278, 280, 283–4, 284–5, 286, 288–9, 291, 300–1, 318 Italo-, 31, 36–8 see also French; Italian; Rhaeto-Romansch; Romanian; Spanish Romanian, 159, 256 Rosenbaum, P., 87 Rosengren, I., 167, 176, 185, 186, 187, 189, 289 Ross, J. R., 4, 98, 277, 278 Roussou, A., 2, 34, 40, 45–6, 52, 54, 58, 87, 280 Rouveret, A., 13, 58, 292 rules, language-particular, 24–5, 27 Russel, B., 82 Russian, 11, 113 Sabel, J., 163, 180 Saeed, J., 132 Safir, K., 155 Saito, M., 9, 10, 43, 48, 62, 78, 83, 85, 87 Samek-Lodovici, V., 7, 95, 99, 126, 234, 261 Sanskrit, 44 Sapir, E., 44 Saramaccan, 8, 22 Savoia, L., 13, 36, 58, 73, 87, 232, 305, 306–7, 319, 321, 322 Scandinavian, 24, 106–14, 225 Insular, 23–4 Mainland, 17, 19, 20, 22, 23–4, 49, 69, 89, 106, 111, 112, 167, 168, 175, 195, 196, 222, 264, 267, 284, 285, 290, 295, 301 see also Swedish Schlegel, A., 44 Schleicher, A., 44 Schwartz, B., 158, 266, 287 Seediq, 198 Seely, D., 157, 231 Segal, G., 82 Select, 32 Selkirk, E., 189 Sells, P., 180 Semitic, 300 Serbo-Croatian, 3

Index Sheehan, M., 5, 10, 11, 13, 14, 16, 18, 62, 63, 73, 74, 75, 82, 88, 90, 100, 101, 102, 104, 109, 112, 113, 114, 117, 120, 121, 125, 127, 138, 154, 161, 193, 194, 198, 211, 212, 218, 222, 223, 228, 247, 255, 260, 264, 285, 288, 308 Sheppard, M. M., 228 Shlonsky, U., 20, 21, 104, 110, 133, 154, 193, 246, 300 Shore, S., 202, 226 Sigurðsson, H., 54, 98, 103, 104, 106, 109, 110, 128, 160, 179, 189, 194, 208, 219, 221, 222 Sindhi, 113 Slavic, 5, 123, 154, 166, 198, 228 Smallwood, C., 69, 198, 295, 296 Sobin, N., 153 Somali, 132 Sorace, A., 116, 167, 184 Spanish, 3, 4, 7, 16, 22, 62, 73, 74, 78, 91, 101, 109, 112, 121, 126, 127, 129, 154, 231, 233, 234, 236, 237–8, 241, 242, 243–7, 249–54, 257, 258, 258, 260, 262, 264, 266, 267, 290, 318 Spell-Out, 9, 267 Spencer, A., 274 Sportiche, D., 251, 303, 304, 312, 314, 315–17, 318, 324, 325 Starke, M., 14, 62, 64, 66, 70–1, 72, 73, 87, 94, 99, 118, 163, 312–13, 317, 324 Steele, S., 277 Stepanovic´, S., 233 Strong Minimalist Thesis, 164, 255 Strong Uniformity Thesis, 54 stylistic fronting, 23, 107, 109, 111, 168, 169, 179, 189, 191–3, 196, 281–2, 290 subject agreement, see agreement, subject canonical position, 13, 20, 34, 58, 72, 73–4, 85 control, see control, subject extraction, 20, 34 lowering, 74 null, see null subject oblique, see subject, “quirky” case “quirky” case, 23, 107 raising, 49, 156, 163, 191, 193, 195, 198, 299 Subset Principle, 41 Sun˜er, M., 73, 241 SUT, see Strong Uniformity Thesis Svenonius, P., 9, 54, 109, 157, 166, 179, 194, 220, 231 Swahili, 68

367 Swedish, 88, 90, 91, 93, 105, 106, 109, 115, 167, 170, 183, 196, 201, 266 Fenno-Swedish, 20, 114 Szabolsci, A., 163 Szendro˝i, K., 9–10, 62, 83, 161, 170, 249 Tagalog, 170, 198, 199, 300 Tallerman, M., 5, 19, 293, 300 Tamburelli, M., 155 Tamil, 122 Taraldsen, T., 117, 162, 232, 264, 303 Taylan, E., 74 Telugu, 229 tense head, see T-head inflection, 42, 263, 286, 299, 300 Poor, 267, 268, 286, 288, 296, 301 Rich, 69, 263, 266, 267, 268, 269, 280, 286, 288, 291, 296, 299, 301 Thai, 8 that-trace effects, see complementiser-trace effects T-head, 14, 21, 23, 34, 49, 59–60, 61, 62–3, 67, 68–9, 70, 72, 74–5, 78, 82, 85, 86, 87, 89, 94–5, 97, 98, 103–4, 104–5, 106, 108–9, 110, 111–12, 115–16, 117, 119, 121, 123, 125, 130, 132, 133–4, 150, 152, 161, 162–3, 164–5, 172, 173–4, 175, 177, 178, 180, 181–2, 183, 184, 189, 191–4, 195, 196–7, 198, 199, 206, 211, 216–17, 217–22, 222–3, 226, 229, 230, 246, 247, 260, 263, 264–6, 267–8, 268–72, 273–4, 275–8, 280, 288, 290, 291, 296–300, 301–2, 304, 306, 312, 313–14, 315–16, 317, 318, 321, 322, 323, 325, 326–7 Third factor, 51, 52, 56, 86, 122 Thompson, S., 176, 198 Thra´insson, H., 110, 154, 160, 167, 175, 184, 189, 264, 267, 285–6, 287, 288, 289 Timberlake, A., 206 Tomioka, S., 9, 62, 83 Tongan, 296 Topic, 95, 134, 211, 232, 243, 244, 245, 246, 257, 259, 260, 261, 298, 299 Aboutness-shift, 96, 97, 98, 99, 100, 101, 103, 105, 134 A-Topic, see Topic, Aboutness-shift Topic-drop, 127 Topic-prominent languages, 198, 199 Torrego, E., 61, 120, 216, 243, 244, 245 traces, 17, 58 see also complementiser-trace effects Travis, L., 69, 154, 165, 266, 295, 296, 299

368

Index

Trentino, 305 Montesover Trentino, 323 Trevin˜o, E., 244, 245 Trosterud, T., 205, 224, 225 Tuller, L., 132 Turkish, 6, 74–5, 91, 127 typology, 14, 17, 21, 22–3, 24–30, 36–44, 68, 264, 267 see also null subject, languages, typology of; parameters, clustering effects; null subject, parameter, clustering effects; drift, typological UCOOL, see Universal Constraint on the Order of Operations Ukrainian, 153 Umbrian, 38 Universal, 68 Universal Constraint on the Order of Operations, 179 Universal Grammar, 15, 16, 17, 25, 26, 31, 32, 33, 34, 45, 47, 51, 237, 269 see also parameters, theory of Uralic, 274 Uriagereka, J., 45 Vainikka, A., 99, 104, 127, 200, 204, 205, 206, 222 Val d’Ayas, 320 Valdoˆtain, 311, 319–21, 325, 326, 327 Vallader, 162 Vallduvi, E., 244 van Kemenade, A., 269, 279 van der Koot, H., 176, 185, 186, 187, 267, 299 van Gelderen, E., 170 Vanelli, L., 7, 306, 321, 322, 323 Vance, B., 196, 291 Vangsnes, ., 189 Venetic, 38 Veneto dialects, 305, 306, 311, 321–5 Loreo, 321 Venetian, 323 Vercelli, D., 45 Vicente, L., 67, 269 Vietnamese, 8 Vikner, S., 156, 157, 158, 159, 167, 168, 173, 179, 264, 265, 266, 268, 280, 285–6, 287 Vilkuna, M., 209 Vinet, M.-T., 299, 322, 326 V-head, 165, 264–5, 266, 267–8, 268–72, 275, 278, 280, 291, 295, 296, 301, 314 v-head, 40, 60, 65, 66, 69, 70, 74, 78, 86, 150, 152, 178, 195, 211, 220–1, 223, 229, 265, 268, 269, 271–2, 277–8, 280–1, 283–4, 301, 314, 319

Walbiri, 3 Wanner, D., 68 Warner, A., 280, 281 Wasow, T., 268–9 Weerman, F., 186 Welsh, Middle, 280 Modern, 291, 299–300 colloquial, 5 literary, 19, 299 Wexler, K., 17 Wiltschko, M., 64, 94 Williams, A., 170, 205 Williams, E., 272, 275 Willis, D., 19, 280, 293, 300 word order, 33 consistent, 28, 36, 39–40 see also word order, “head-final”; word order, “head-initial” “head-final”, 28, 36, 40 “head-initial”, 28, 36, 40 inconsistent, 28, 36, 40 OV, 24, 28, 36, 37, 38, 40, 44, 68, 74–5, 78, 79, 126, 145, 146, 149, 287 see also word order, SOV SOV, 19, 38, 74, 103, 134, 148 SVO, 36, 237, 245, 248, 299 VO, 24, 28, 36, 38, 40, 68, 79, 126, 145, 146, 287 see also word order, VOS; word order, VSO; word order, SVO VOS, 69, 74, 249–57, 258, 295, 296, 297, 298, 299, 300 VSO, 19, 20, 31, 69, 236, 238, 244, 246, 252, 253, 258, 295, 296, 298, 299 V2, 30, 33, 64, 107, 110, 114–15, 155, 157, 158, 160, 171, 172, 173, 177, 192, 196, 245, 246, 260, 263, 265, 266, 268–9, 270, 279–80, 284, 285, 287, 292, 293, 297, 298, 302, 313–14, 315–16, 319, 321, 323, 325–6 Wurmbrand, S., 183, 189, 190, 289 Yiddish, 24, 153, 154, 155, 158, 167, 173, 287, 288 Yoon, J., 153 Zribi-Hertz, A., 303, 311, 312, 314, 315, 316, 326–7 Zubizarreta, M.-L., 74, 112, 241, 244, 245, 246, 250, 251, 252, 253, 254, 256 Zwart, C. J.-W., 146, 186, 246, 266, 323 Zwicky, A., 272, 274, 314, 323