A deletion analysis of null subjects - University of Cambridge

A Deletion Analysis of Null Subjects: French as a Case Study1

Ian Roberts

Downing College, University of Cambridge

[email protected]

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,

2006), Manzini & Savoia (2005), Nash & Rouveret (1997), Ordoñez (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 corollary, the idea that the canonical subject position is occupied by pro is 1 The research reported here was carried out under the auspices of the project “Null Subjects and the Structure of Parametric Theory” funded by the Arts and Humanities Research Council of Great Britain (Grant No. APN14458). I’d like to thank the other members of the 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 paper 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 and Cambridge, and at the Encontro Lingua Falada e Escrita V, Maceió, Brazil. I’d like to thank the participants at those meetings, especially Claire Blanche-Benveniste, João Costa, Eric Haeberli, Ad Neeleman, Liliane Haegeman and Ur Shlonsky, for their comments and questions. All errors are my own.

1

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-subjects” approach to null subjects (the idea

being that I, as the functional head bearing subject agreement, has the subject role).

In the first part of this paper 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). In the second part of the

paper, I will apply the analysis to pro in French; this will lead to a revision and

updating of the analysis of French subject enclitics proposed in Sportiche (1999), as

well as a rough characterisation of some aspects of register variation in French. On

both of these points, my proposals largely converge with those in Zribi-Hertz (1994).

1. Holmberg (2005)

Holmberg (2005) shows that the widely accepted pro-licensing analysis of null

subjects put forward in Rizzi (1986a:518-523) is incompatible with the feature-

valuing system put forward in Chomsky (1995, 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

(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 φ-features of T, indicated in (2) as [3pl]. We can think of

identification as a process whereby the values of the pro’s features are assigned.

As Holmberg (2005:536-7) 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 (1995, 2000, 2001) recent work. According to this approach,

formal features such as φ-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 φ-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 φ-features,

and as such are fully interpretable and able to value the φ-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 φ-features of

I [i.e. T, IGR], as those features are themselves inherently unspecified” (2005:537).

It is useful to restate 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

(3) In a well-formed Agree relation of which α and β are the terms, where α’s

feature matrix contains [Atti:__] and β’s contains [Atti: valj], for some feature

F = [Atti: (val{..,j,..})], copy valj into __ in α’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 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).

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 φ-features.

Whichever one it is will be able to value those of the other one. Accordingly,

Holmberg considers the following two hypotheses:

(4) Hypothesis A: in null-subject languages, the φ-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 φ-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.

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 φ-features are valued by T is Nominative and DP whose φ-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 §2 (see Note 13).

4

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 φ-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, sitä:

(5) a. Puhun englantia.

speak-1Sg English

“I speak English”

b. Sitä meni nyt hullusti.

EXPL went now wrong

“Now things went wrong.”

The expletive sitä does not cooccur with referential null subjects:

(6) a. *Sitä puhun englantia.

EXPL speak-1sg English

b. Oletteko (*sitä) käyneet Pariisissa?

have-2pl-Q EXPL visited Paris?

Holmberg concludes that Hypothesis B is right: pro occupies SpecTP.3 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

3 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.

5

pronounced”. Clearly, one way to see this 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 φ-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 φ-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

(2005:553-4) mentions Brazilian Portuguese, Marathi and Hebrew; see also

Holmberg, Johns, Nayudu & Sheehan (this volume)). The third type is found in many

East Asian languages, including notably Chinese, and is discussed in Huang (1984,

1989) and Tomioka (2003), as well as by Neeleman & Szendrői (forthcoming).

Leaving aside the East Asian type,4 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 φP, a

deficient pronoun that receives the ability to refer to an individual or group from I

containing D” (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, Johns, Nayudu & Sheehan (this

volume). Now, however, it is time to introduce the technical ideas which will

motivate and restrict the environments of pronoun-deletion which give rise to null

subjects.

4 I will briefly return to the East Asian kind of null subject in §2.5 below, where I look at the nature of “rich” agreement in relation to null subjects.

6

2. Defective goals, clitics and weak pronouns

2.1 Clitics and incorporation

In Roberts (2006a) 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 (2006a) 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,

which depart only slightly from the standard conception in Chomsky (1995):

(7) a. The label L of category α is minimal iff α dominates no category β

whose label is distinct from α’s.

b. The label L of category β is maximal iff there is no immediately

dominating category α whose label is non-distinct from β’s.

These definitions allow 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):

7

(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 (1999:168-169) 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.5 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 (2006a) adopts Déchaîne & Wiltschko (2002:428-31)’s

terminology in labelling Romance clitics as φPs, rather than DPs. Romance clitics

thus differ from the strong complement pronouns of a language like English in being

5 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 subject-clitic inversion involving verb-movement to C in favour of remnant TP-movement; these questions are discussed in §3.2, see particular Note 35.

8

φmin/max, rather than Dmin/max. I will say more about the third element in Cardinaletti &

Starke’s typology, the weak pronouns, below.6

Since the label of (active, transitive) v* contains φ-features, in fact, unvalued versions

of the very φ-features that make up the clitic, the clitic’s label is not distinct from

v*’s. More precisely, the clitic’s features, essentially φ-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

[iφ] v*min le Root/Vmin v*min

voit [iV, uφ]

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 φmin/max, it allows for cliticisation

of the Romance kind. Where object pronouns are Ds, as in English, cliticisation (to

v*) is not possible (Roberts (2006a, §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.7

6 Implicit in this is a rejection of Déchaîne & Wiltschko’s (2000:421-426) proposal that English 3rd-person pronouns are φPs while 1st- and 2nd-person pronouns are DPs. See Roberts (2006a, Note 28) for discussion. 7 One objection to incorporating φmin into v*min as shown in (9) is that the operation violates the Extension Condition (all operations must extend the root, Chomsky (2000:136-8), 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 (2005), 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

9

What is the trigger for cliticisation? Roberts (2006a) 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:__] D[Pers:a, Num:b], [Case:__]

b. Outcome of Agree:

v*[Pers:a, Num:b] 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 Note 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),8 and, as defined in (3), values v*’s unvalued features.

Schematically, then, we have the situation in (12):

(being an A-position, see Chomsky (2005:16)); 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 (9) is quite licit, as in this structure, the goal [iφ] is incorporated into the probe and the probe c-commands the copy of this feature bundle contained in its sister VP. 8 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)).

10

(12) a. Trigger for Agree:

v*[[Pers:__], [Num:__] φ[Pers:a, Num:b]

b. Outcome of Agree:

v*[Pers:a, Num:b] (φ[Pers:a, Num:b])

Here v’s [Pers:a, Num:b] features are valued as an automatic consequence of Agree of

φmin 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 constitute an identical copy of the copied feature bundle.9 But, precisely in the

case of incorporation, this is what happens. For this reason we see the PF effect of

movement, with the φ-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 9 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.

11

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).10 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 (13’):11

(13’) 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 (2006a, §3.5.1, and

Chapter Four). 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 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

10 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 (2006a), Vicente (2005), 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. The formulation in the text suffices for present purposes, however. 11 Thanks to Eric Haeberli for drawing my attention to this matter.

12

which show adverbal 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, at least in the case of the Romance

languages, it is fairly clear that the modern clitic system developed after the change

from OV to VO order (see Wanner (1987)).

The second prediction concerns verb- and auxiliary-movement. Roberts (2006a,

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 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 (14):12

(14) 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 (14) 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 (2006a, Chapter 4) argues that

12 AbsP in (14b) 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.

13

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). Now, given

the proposal in Biberauer & Roberts (this volume) 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. This

prediction certainly seems to be borne out in Niuean, in which tense/aspect

information is carried by sentence-initial particles, which Massam (2000:101)

concludes represents “a portmanteau Complementiser/Tense element”. In the terms

described above, we can simply take these elements to be realisations of T-features in

C. Both Tongan and Māori appear to pattern in the same way (Churchward (1953),

Chung & Ladusaw (2003), cited in Chung (2005); see also Otsuka (2005) on

Tongan)). This line of argument is developed and documented more fully in Roberts

(2006a, Chapter 4).

Here we have seen the basic mechanics of cliticisation/incorporation, according to

Roberts (2006a). The most important point for what follows is the incompatibility of

incorporation into a probe bearing an EPP feature, stated in (13). In the light of this,

let is 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 clitic-incorporation

yields a notion of defective goal, which derives directly from the general

characterisation of the precondition for incorporation in (10):

(15) A goal G is defective iff G’s formal features are a proper subset of those of

G’s Probe P.

Since clitics are φ-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 (15) is a relative rather than an absolute notion; in

14

general a category is defective if is probed by a category of a similar type. As argued

in Roberts (2006a, 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

(1999:175-176) 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” (Cardinaletti & Starke (1999:175)). These points are illustrated by

the following Italian examples:

(16) 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.”

c. pro/*lui è molto costoso.

It is very expensive

d. Lui/*pro è veramente bello.

He (over there) is really nice.

15

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:

(17) Gianni ha telefonato quando pro/*lui è 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 proinoun. 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, §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:

16

(18) 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, …”

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.”

(18a) shows that overt weak pronouns cannot be dislocated, so presumably the same

applies to pro. (18b) 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, Ch 3)). Finally, dislocated elements are not

allowed in complementiser-deletion contexts like (18c) (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.13

Furthermore, pro cannot appear in the “freely inverted” subject position. Cardinaletti

(1997:36-37) summarises two arguments that have been given for this. First, Burzio

13 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 φmin/max. For an alternative proposal regarding the feature content of egli, which is also compatible with the proposals made here, see Cardinaletti (2004:149-150).

17

(1986) showed that overt preverbal pronominal subjects are not allowed in

presentational sentences in Italian:

(19) a. *Io ci sono alla festa.

I there am at-the party

b. Ci sono io alla festa.

There am I at-the party

“I am at the party.”

Null subjects are not allowed here:

(20) *Ci sono alla festa.

There I-am at the party

If pro can only occur in preverbal subject position, then the ungrammaticality of (20)

is assimilated to that of (19a).

Second, Rizzi (1987) showed that only preverbal subjects can license a floated

quantifier:

(21) 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

d. *Sono andati tutti via i bambini.

Are gone all away the children

“All the children have gone away.”

18

e. Sono andati tutti via.

Are gone all away

“They have all gone away.”

In (21e), pro must be preverbal, since a postverbal subject cannot cooccur with a

floated quantifier, as the contrast between (21b) and (21d) shows.

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:14

(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.”

14 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.

19

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 φmin/max (in this I differ from Holmberg

(2005) – see §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: hence, it is

a Dmin/max). Thus, pro 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. Suñer (2002) argues essentially the same for Spanish,

as does Ordoñez (2006); Costa (2006) argues the same for European Portuguese, and

Adragão & Costa (2006) argue this for child European Portuguese (Brazilian

Portuguese appears to be a partial null-subject language; see the papers in Kato &

Negrão (2000), Holmberg, Johns, Nayudu & Sheehan (this volume) and the

references given there). Given (13), this is also consistent with the conclusion in the

previous section that pro is not a clitic.

Sheehan (2006) 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. See

also Sheehan & Roberts (this volume).

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, Johns, Nayudu and Sheehan (this volume)). In her sketch of

the grammar of Turkish, Kornfilt (2003, IV:305) says “[b]ecause of the richly

20

differentiated agreement system, subjects of both main and embedded clauses … can

(and preferably do) remain unexpressed when interpreted as personal pronouns” (see

also Kornfilt (1997:132); cf. Öztürk (2001) for a different view). Here is a simple

example:

(24) Ben ev- e 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

“I came home. I did some reading. I watched TV.” (Öztürk (2001:241))

Assuming that this is a genuine case of an Italian-type “consistent” null subject in

Holmberg’s (2005) terminology (and see Öztürk (2001) for a dissenting 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+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 Old English and the modern West Germanic languages. The analysis

can presumably extended to OV languages like Turkish. What this means is that T has

an EPP feature in Turkish, one which is associated with φ-features on T which probe

the subject.15 Turkish merely has the additional extra pied-piping requirement. Most

important for our purposes, pro can be the goal; it can be inside a larger pied-piped

category which satisfies the EPP.

15 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.

21

Furthermore, Holmberg’s (2005) argument for a deleted pronoun in null-subject

sentences in Finnish depends on the idea that pro is able to satisfy the EPP (in

SpecTP, as Holmberg (2005:542-543) argues). This can be seen from contrasts like

that in (5), repeated here:

(5) a. Puhun englantia.

speak-1Sg English

“I speak English”

b. Sitä meni nyt hullusti.

EXPL 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.16 Following Holmberg

(2005:555), I take it that languages of this type have D-feature in T.17 Also following

16 Cf. also the following remark by Apollonius Dyscolus (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. 17 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

22

Holmberg, I take this D-feature to be correlated with “rich” agreement (I will consider

the nature of rich agreement in more detail in §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, φ 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

§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 §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 §2.5 on this). In §2.1, we suggested that the reason for

(26) may be connected to the nature of chain reduction in the sense of Nunes (2004);

see also Note 9 for further discussion. To the extent that pronouns are often defective

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.

their peculiarities; see Holmberg, Johns, Nayudu & Sheehan (this volume). In Roberts (2006b) I speculate as to how pro may have lost its D-feature in the recent history of Brazilian Portuguese.

23

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

EPP feature, 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 rather than a

full-fledged wh-DP. Null operators appear in the following contexts:

(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 -- ]]].

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

24

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.

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 pied-piping 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,18 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 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 (Italian)

b. la personne que/*qui Bill a vue (French)

the person that/who Bill has seen

A further prediction stemming from (26) concerns v. If consistent, agreement-licensed

null subjects are only found where T has φ-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 φ-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). 18 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 (1995:263)).

25

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.19 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

Note 17 and Holmberg, Johns, Nayudu & Sheehan (this volume)). 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 agreement pattern in tenses formed from the present

verb stem and an ergative-absolutive one in tenses formed from the past verb stem

(MacKenzie (2003:255-256)). 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 maņa 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 (1984:535-536))

19 These might be cases of topic deletion (in the sense of Huang (1984), Raposo (1986), Modesto (2000)), which it would be natural to account for in terms of the notion of defective goal. However, if Chomsky (2005) 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 Note 22 for a further brief comment on “radical” pro-drop.

26

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?

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 (depending on its person). 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 Müller’s (2005) notion of presyntactic impoverishment as a way of

accounting for the relation of “rich” verbal agreement inflection to null subjects. I will

show how Müller’s idea can be integrated into the proposals being made here. I will

argue that this entails, 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

27

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” (Müller (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 Müller 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, Müller (2005:5) gives the

following two impoverishment rules for German verbal inflection:

(31) a. [±1] ø/[-2,-pl,+past] __

b. [±1] ø/[-2,+pl] __

These rules delete the 1st-person feature ([+1] 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.

Müller (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 α if α has

been subjected (perhaps vacuously) to a φ-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 analysis, too). That is, pro in

28

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 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 (2005:548-550) and Holmberg, Johns, Nayudu and Sheehan (this volume)).

Following Russell (1905), I take the notion of definiteness to involve existence and

uniqueness.20 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

φ-feature specifications:

(33) If a category α has D[def], then all α’s φ-features are specified.

As we have seen, impoverishment removes certain φ-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 its features have been subject to impoverishment.21 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 Müller’s pro generalisation from the postulate about

the interaction of D[def] with φ-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 Müller, between rich agreement and the licensing

of consistent null subjects.

20 May (1985:8) analyses the as a generalised quantifier as follows: (i) the(X, Y) = 1 iff X = X ∩ Y = {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 the denotation of VP (Larson & Segal (1995:320)). 21 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.

29

In these terms, then, consistent null-subject languages have non-impoverished φ-

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 φ-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 φ-features on T and hence pattern with English

rather than with consistent null-subject languages in this regard; here, however, other

deletion processes are at work, such as those identifying null topics in the sense of

Huang (1984) (but see Note 19 above), as well as possibly N’-deletion as in Tomioka

(2003) and/or context-free zero-realisation of morphologically transparent pronouns

as proposed in Neeleman & Szendrői (2005).22

How does all this bear on the question of the precise nature of pro’s non-overtness?

Müller (2005:7-8) gives three conceptual arguments against post-syntactic lexical 22 Neeleman & Szendrői (2005) treat fully specified nominals as KPs (since they inherently contain a syntactic position for Case, according to Neeleman & Szendrői) 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 +pronoun, -anaphor, 3rd person, Singular, Masculine, Accusative]. The general “radical pro-drop” rule is the context-free zero-realisation rule (i): (i) [KP +pronoun, -anaphor] Ø 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 (i). 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. On the other hand, as Neeleman & Szendrői (2005:16) point out, fusional pronouns may be compatible with consistent null subjects (as determined by rich agreement, in what we can take to be the sense described by Müller’s pro generalisation) thanks to “context-sensitive zero-realization” of the general kind in (ii): (ii) [KP +pronoun, -anaphor, φi ] Ø/ ___ φi. Neeleman & Szendrői take “φi” to denote some set of person and number features; the coindexation is intended to indicate agreement with the same features of KP. Given the text proposal, we can take it to be a non-impoverished set of φ-features combined with a D feature.

30

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 &

Müller (2005), Müller 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 component). In this

connection, Müller 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. Müller concludes that “a morphological

component of grammar that meets minimalist requirements should be pre-syntactic

rather than post-syntactic” (Müller (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 Müller 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

undergo 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.

31

Therefore, I follow Müller’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 I take to be at the point of transfer to PF (but

see Note 23). 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.

I conclude 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 §2.1, and, plausibly, to the general operation of

copy-deletion. Whether this is a PF operation or internal to core syntax is a further

question. Since copies need to be available at LF, presumably it does not feed LF, and

therefore must take place at Spell Out at the earliest; this is a standard view and I see

no reason to dissent from it here.23

23 It is, however, worth pointing out that pro’s features are properly included in those of the T which probes it. Therefore no LF-relevant information is lost by the deletion process; T’s φ- 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, 2006) 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 (2006)). 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: 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 §1 and the arguments from Burzio, Rizzi and Cardinaletti in §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. I will not pursue this intriguing hypothesis further here, however.

32

This concludes the general characterisation of pro: it is a weak pronoun, 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 φ-features

have undergone (pre-syntactic) impoverishment; this establishes a connection with

“rich” agreement, since non-impoverished φ-features will tend to be realised by

distinct vocabulary items (although a certain amount of accidental homophony and

null realisation may exist). This, in the current theoretical context, is the nature of the

“pure pronominal empty category”: it is a pronominal goal deleted under identity by

pronominal probe.

In the final part of this paper, I apply this approach 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 the considerations we have raised in this section.

3. Null subjects in French

3.1 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 (1989), Taraldsen (1978)). The

following contrasts with Italian motivate this ((34b) is a grammatical imperative, but

ungrammatical as a declarative):

(34) a. Parla italiano.

b. *Parle italien.

“(He/she) speaks Italian.”

(35) a. Chi hai detto che -- ha scritto questo libro?

b. *Qui as-tu dit qu’ – a écrit ce livre?

Who have-2sg said that – has written this book

33

(36) a. E’ arrivato Gianni.

b. *Est arrivé Jean.

Has arrived John.

“John has arrived.”

(37) a. Hanno telefonato molti studenti.

b. *Ont téléphoné beaucoup d’étudiants.

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 φ (and, given our analysis,

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)). Thus, we would have a null subject in an example like (38):

(38) 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 is usually 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 in the subject-clitic paradigms in Northern Italian

dialects but not in French, and pronominal paradigms do not normally show such gaps

34

(this is related to the point made at the end of the previous section that the incidence

of pro cannot be purely accidental). Second, in many (but not all) Northern 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:

(39) a. Nessuno gl’ha detto nulla. (Florentine)

Noone SCL.has said nothing

b. *Personne il n’a rien dit. (French)

Noone he not.has said anything

“Noone said anything.”

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:

(40) a. No te ghe l’hai dit. (Trentino)

Not you to-him it.have said

b. Un tu gliel’ha detto. (Florentine)

Not you to-him-it.have said

“You have not said it to him.”

Fourth, under coordination, the clitics must be repeated in both conjuncts in Northern

Italian dialects, while this is not the case in French:

(41) a. Il chante et danse. (French)

He sings and dances

b. La canta e *(la) balla (Trentino)

she sings and she dances

35

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.

Nevertheless, with the possible exception of those found in certain Veneto dialects,

they contrast in their distribution with French subject pronouns along the lines just

illustrated. For these reasons, 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. According to Rizzi and Brandi & Cordin (see also Manzini & Savoia (2005,

I: 117ff.)), subject clitics in many Northern Italian dialects are a form of subject

agreement, arguably a realisation of T’s φ- or D-features, and presumably

contributing thereby to the licensing of null subjects.24,25

Another case where it has been argued that French allows at least expletive null

subjects is in Stylistic Inversion (Styl-Inv), as analysed by Kayne & Pollock (1978)

and Pollock (1986):

(42) a. A qui a parlé ton ami?

To whom has spoken your friend?

“Who did your friend speak to?”

b. l’homme à qui a parlé 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 24 Cardinaletti & Repetti (2003) dissent from this general view. They argue that subject clitics, at least in certain Veneto dialects, are subject pronouns. 25 I will not speculate here on how this kind of analysis of Northern Italian subject clitics would interact with our account of Müller’s pro-generalisation discussed in the previous section.

36

position. In their more recent treatment, however, Kayne & Pollock (2005) argue that

the postverbal 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 (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 (42).

They give three pieces of evidence for this. First, in certain non-colloquial registers

3rd-person strong pronouns seem to appear as both preverbal and postverbal subjects,

but not 1st or 2nd-person ones, which must co-occur with an overt subject clitic:

(43) a. LUI a mangé/Qu’a mangé LUI?

HE has eaten/what has eaten HE?

“HE has eaten/what has HE eaten?”

b. *TOI as mange/*Qu’as mangé TOI?

YOU have eaten/what have eaten YOU?

“YOU have eaten/what have you eaten?”

c. TOI tu as mangé.

YOU you have eaten

“YOU have eaten”

Kayne & Pollock suggest that there is a 3rd-person null subject in (43a) in the position

occupied by tu in (43c). (43b) is ungrammatical because 2nd-person null subjects are

not allowed in French.26

Second, indefinite subjects are not good in Styl-Inv:

26 In more colloquial registers, a weak subject pronoun (il) obligatorily appears in (43a). Kayne & Pollock observe that registers which require the 3rd-person subject clitic here do not allow Styl-Inv at all (these include Quebec French).

37

(44) *Quel gâteau a mangé quelqu’un?

Which cake has eaten somebody?

If there is a weak subject pronoun in Styl-Inv, this can be assimilated to cases like

(39b), 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 postverbal in

Styl-Inv:

(45) a. *Il, il mangera ce gâteau.

He, he will-eat this cake

b. *Quand a parlé il?

When has spoken he?

Again, these reduce to a single fact if il is “dislocated” in (45b) 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 left dislocation and Styl-Inv

constructions are quite convincing. In the light of the arguments and analyses

presented in the preceding sections, 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, Johns, Nayudu & Sheehan (this volume).

Let us very briefly consider how French complies with Müller’s pro-generalisation.

Are there cases of system-defining syncretism in the French verbal-agreement

paradigm? Harris (1988:224) gives the following summary of French verbal

conjugation:

38

(46) Present indicative by conjugation type:

donner “to give” voir “to see” rompre “to break” finir “to finish”

donn-e voi-s romp-s fini-s

donn-es voi-s romp-s fini-s

donn-e voi-t romp-t fini-t

donn-ons voy-ons romp-ons fin-iss-ons

donn-ez voy-ez romp-ez fin-iss-ez

donn-ent voi-ent romp-ent fin-iss-ent

(47) Other tenses (all conjugations):27

Imperfect Pres. Subjunctive Future Conditional

donn-ais donn-e donne-r-ai donne-r-ais

donn-ais donn-es donne-r-as donne-r-ais

donn-ait donn-e donne-r-a donne-r-ait

donn-ions donn-ions donne-r-ons donne-r-ions

donn-iez donn-iez donne-r-ez donne-r-iez

donn-aient donn-ent donne-r-ont 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 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

27 There are two further tenses which are not part of spoken French, the preterit (or passé 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.

39

be deleted in non-liaison contexts by the very general process of final-obstruent

deletion (see Anderson (1982), Dell (1985) for details). The same is arguably true of

the final <t> of the 3pl and the final <z> 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 Müller (2005):28

(48) a. /onz/ > [ +pl, +1 ]

b. /ez/ > [ +pl, -1, +2 ]

c. /ət/ > [ +pl, -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”.29 What this means is that we have syncretism

between 3rd singular and 3rd plural in the present tense.30 It is possible that we have

this syncretism in verbs of all conjugations in the imperfect and conditional tenses;

however, we can take the “imperfect” ending (common to both tenses) as /ε ~ 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 plural as /ət/, with a regular phonological rule

deleting /ə/ after /ε/. In that case, the singular forms are as in (49):

(49) a. Ø > [-Pl, +Imperfect, αPersi, -αPersj] (i ≠ j)

b. /t/ > [-Pl, +Imperfect, -1, -2]

c. /e/ > [-Pl, +Future, +1]

d. /a/ > [-Pl, +Future, -1]

28 These are somewhat approximate phonological forms: /onz/ for example is realised with nasal vowel /õ/ and no nasal consonant. This results from regular phonological processes, however (see in particular Dell (1985)). 29 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. 30 This does not apply to 1st-conjugation verbs, as shown in the paradigm for donner in (46), or for the subclass of 4th-conjugation verbs illustrated by ouvrir (“to open”), which have 3sg present ouvre, 3pl ouvrent.

40

(It is unclear whether the homophony of the 2sg and 3sg future endings really

represents a system-defining syncretism in Müller’s sense). In all tenses except the

future, there is syncretism between 1st and 2nd person. This latter point creates a

difficulty for Müller’s feature system for persons, since, by positing the features [±1]

and [±2], the prediction is made that 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 α-notation in (49a), intended to indicate that the positive value of one of the

person features, [+1] or [+2], must be chosen, along with the negative value of the

other one. Leaving aside the difficulty of formulating this rule given the feature

system Müller adopts (which does not in itself pose a problem for the general idea of

system-defining syncretism), it is clear that there is system-wide syncretism here. So

it seems clear that French conforms to Müller’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 null-subject language, given Kayne & Pollock’s

observations reported above).31

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

Müller’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 in French. I now turn to this.

31 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 (43a), 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 the concluding section below for more on register variation.

41

3.2 Null subjects and subject-clitic inversion

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 language, and that

the subject clitic apparently enclitic to C in these contexts is in fact a realisation of the

φ-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 (50):

(50) 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), Pollock (2006), see Note 35).

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):

(51) a. ?Has, by the way, John seen Mary?

b. ?*Have, by the way, you seen Mary?

In French, interpolation is impossible here:

(52) **As, à propos, tu vu Marie? (=(51b))

In general, non-clitic subjects are impossible in the position occupied by the clitic in

(50):

42

(53) *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:

(54) a. Il ou elle connait bien le problème.

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?

Further, as can in fact be seen in (54b), subject-clitic inversion with a 3sg clitic is

associated with a specific phonological operation, here insertion of an epenthetic /t/.32

Cardinaletti & Starke (1999:167) observe that subject clitics in enclitic position

cannot be omitted in the second conjunct of a coordinate structure:

(55) a. Il aime les choux, mais – ne les mange que cuits? 32 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 §3.1, 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 à la poupée. (/ižutalapupe/) They play with the doll 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- à la poupée. 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 §3.1. Pollock (2006:627-630) 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.

43

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 (2006a),

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 φ-features are really features of C, as proposed in

Chomsky (2005), 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 (2006a,

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 (2005) suggestion that T’s φ-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-

44

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-être (“perhaps”)), C simply does not

pass its features to T. C’s φ-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 φ-set of residual V2 C; we can think of them as

realisations of C[+Q], or whatever feature best characterises residual V2 C. We can

clearly capture the presence of epenthetic /t/ in this way (see Notes 29 and 32 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 §3.1, 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 (51-55) 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 (51), we can assume that, since there are no φ-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 Agree relation holds between C’s φ-set and the

subject, and indeed the subject may be attracted to SpecCP, giving rise to complex

inversion. The grammatical version of (51) is thus:

(56) Jean a-t-il vu Marie?

John has-3sg seen Mary?

“Has John seen Mary?”

45

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

(57b)), but not readily between the auxiliary and the participle in examples like the

following:

(57) a. Les électeurs sont ??(quand le vote a eu lieu) allés à la pêche.

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 électeurs sont ??(quand le vote a eu lieu) allés à la

pêche?

In which town the voters are 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.33

33 The difficulty of combining subject extraction with an apparent enclitic subject pronoun might be seen as a problem for this analysis: (i) *Qui a-t-il mangé de la soupe? Who has-3sg eaten of the soup However, Sportiche (1999:215) points out that examples of this type are attested, citing Grevisse (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.

46

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)):

(58) pro A-t-il vu Marie?

has-3sg seen Mary?

“Has he seen Mary?”

The null subject occupies the specifier of 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 Müller’s feature system; these are

realisation of φ-features on C rather than T, hence there is a further contextual

restriction here that is not specified in (59)):34

(59) a. [+1, -2, -pl] > ø

b. [-1, +2, -pl] > /ty/

c. [-1, -2, -pl, -fem] > /til/

d. [-1, -2, -pl, +fem] > /tεl/

e. [+1, -2, +pl] > /nuz/

f. [-1, +2, +pl] > /vuz/

g. [-1, -2, +pl, -fem] > /tilz/

h. [-1, -2, +pl, +fem] > /tεlz/

34 There are two further forms which need to be considered here: the generic element on, which surfaces here as /tõ/, 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)). 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 (1994:141-143) for a proposal and references. Thanks to Claire Blanche-Benveniste for drawing my attention to these forms.

47

(The final /z/ in the plural forms only surfaces in careful liaison contexts, e.g. ont-ils à

faire cela? /õtilzafεrsla/ “Do they have to do that?”). Given these forms, we expect C

to be able to have a D-feature, by the reasoning given in §2.5 above, and therefore to

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:

(60) a. Si Jeani était venu, ili aurait décidé.

If John was come, he would-have decided.

“If John had come, he would have decided.”

b. *Ili aurait, si Jeani était venu, décidé

he would-have, if John were come, decided

c. *Aurait-ili, si Jeani était venu, décidé

would-have-he, if John were come, decided

As Sportiche points out, (60b) is a Principle C effect, since il binds Jean. The effect is

also found in (60c); 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:

(61) *Où il est-il allé?

Where he is-3sg gone?

This relates to the distribution of weak, as opposed to strong pronouns. It appears

from (61) 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 il must delete in this context.

48

Second, only 3rd-person strong pronouns are allowed:

(62) a. Quel livre eux ont-t-ils apporté?

Which book they have-3pl brought?

“Which book have they brought?”

b. *Quel livre toi as-tu apporté?

Which book you did-2sg bring?

This contrast parallels that in (43) 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 (50)). 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 Note 13 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 (43a) above, but this does not fully undermine their case for a null subject in Styl-

Inv contexts. Clearly, there is in any case a contrast between 3rd person and other

persons, in that non-3rd-person strong pronouns cannot appear in this position, as

(62b) shows.35

35 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-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 head-movement analysis of verb-movement into the C-system do not in fact apply, given the assumptions about morphology adopted in §2.5. 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 prévoyez (“you foresee it”), the clitic would have to attach to the prefix. Kayne goes on to suggest that

49

Valdôtain has cases like (63), discussed in Roberts (1993) and Pollock (2006:632-

636), which appear to be instances of the same string as in (61):

(63) a. Cen que dz’i dzo fe?

What that I-have I done?

“What have I done?”

b. Dze medzo-dzò an pomma?

I eat I an apple

“Do I eat an apple?”

Similar examples, from various Provençal varieties, are mentioned by Poletto

(2000:54-55). One possibility is to treat the enclitic as a realisation of C’s φ-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. This in turn implies, given (13), 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 clitic+verb 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 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 verb-movement 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) a. Tu [XP le [YP as [ZP fait ]]] (movement of ZP) 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 Note 32). 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 (2006a), we do not need to draw the conclusions drawn by Kayne and adopted by (Poletto and) Pollock. Although I have been assuming a version of the LCA (cf. the discussion of OV languages in §§2.3 and 2.4), I do not assume that all morphology is syntactic affixation, and therefore that the LCA necessarily applies at the 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 (2006a), 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 (2006a; §3.4). Thus we are not forced to a remnant-movement analysis of verb-movement into the C-system (see Roberts (2006a, Chapter 4) on how full and residual V2 should be handled).

50

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 (63) with an expletive

pronoun. However, examples such as the following suggest that we do find expletive

proclitics in the construction analogous to (63) in Valdôtain:

(64) a. Cenque l’y est-ë ba-lé?

What it-there is-it over-there

“What is there over there?”

(“Qu’y a-t-il là-bas?” Chenal (1986:389))

b. Que l’est-ë que te va trové?

What it-is-it that you will find?

“What will you find?” (Chenal (1986:388))

Furthermore, Valdôtain has complex inversion, where the preverbal subject cooccurs

with a subject clitic:

(65) a. Ton ommo l’est-ë dza tornà?

Your man he is-he already come-back

“Has your man already come back?” (Chenal (1986:350))

b. Pierre l’at- ë mandà euna lettra a Caterine?

Pierre he-has-he sent a letter to Catherine?

“Has Pierre sent a letter to Catherine?” (Chenal (1986:373))

Finally, Valdôtain 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 (1993):36

36 (66a) is an example of another phenomenon of interest in Valdôtain (also found in some Piedmontese varieties; Parry (1994)), which Roberts (1993:329) calls “OCL-for-SCL”. Here it seems that there is just one morphological “slot” for a proclitic. Where there is more than one proclitic, 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.

51

(66) a. Gnunc l’a viu-me.

No-one he-has seen me

“Noone saw me.” (Roberts (1993:330))

b. Ki y et vinyà?

Who he is come?

“Who has come?” (Roberts (1993:328))

For all these reasons, it seems best to regard the apparent proclitic subject in (63) and

(64) as a realisation of φ-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 φ-features twice over, although the interrogative conjugation may be seen

as the syncretic realisation of the Q-morpheme and φ-features, while the “proclitic” in

these cases is the realisation purely of φ-features. A possibly 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 φ-features.

The idea that the apparent enclitic subject pronoun in contexts of subject-clitic and

complex inversion is really a manifestation of the φ- 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 (2006a,b) for a different perspective). This idea is developed in Roberts

(2006a, Chapter 3). What I have tried to show here is that this analysis accounts best

for the facts of French subject-clitic and complex inversion, largely following

No-one me-saw “No-one saw me.” In other varieties, e.g. Châtillon, this is not possible, and objects must be enclitic to the participle in the presence of a subject clitic.

52

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.

3.3 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 (43a) and stylistic inversion; otherwise it 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 français très évolué (FTE, or “very evolved French”) subject

proclitics are to be analysed as realising φ-features of T rather as in certain Northern

Italian dialects and Valdôtain, as we saw at the end of the last section. This is shown

by the fact that they cooccur with non-referentially quantified subjects, as in the

attested examples in (67):

(67) 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, à Toulon.

No-one he does anything at Toulon

“No-one does anything in Toulon.”

(Zribi-Hertz’ (19a, e), p. 137); (67b) from P. Mille Barnavuax et

quelques femmes, 1908)

Further evidence for the clitic status of these pronouns comes from the fact that they

cannot be dropped under coordination:

53

(68) 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 mangé 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’ (21), p. 138)

As Zribi-Hertz points out, if il realises T’s φ-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:

(69) (i) “high” registers:

allowing (43a) and stylistic inversion: (very) partial null-subject system in

TP, fully null-subject in CP;

(ii) registers not allowing (43a) and stylistic inversion: non-null-subject in

TP, consistently null-subject in CP;

(iii) colloquial registers in which complex inversion and subject-clitic

inversion are lacking: fully non-null-subject systems;

(iv) the variety/varieties seen in (67) and (68) in which subject proclitics are

to be analysed as realising φ-features of T rather as in some Northern

Italian dialects and Valdôtain: fully null-subject in TP.

The variety in (69i) is formal literary French, or perhaps français standard moderne

(FSM) in the terminology of Zribi-Hertz (1994:136), which she describes as “the

productive standard language, without archaisms, i.e. well-formed constructions of

54

Classical French which nowadays are no longer productive.”37 This variety also

makes use of the cases of careful, optional liaison discussed in §3.1. The variety in

(69ii) may correspond to Quebec French (Kayne & Pollock (2005:13)), as well as

slightly less formal varieties than (69i) spoken elsewhere. Both (69iii) and (69iv)

represent the colloquial spoken French often referred to as français avancé. More

precisely, we can perhaps identify (69iii) with Zribi-Hertz’ (1994:137) français parlé

courant (FPC, or “common spoken French”). Zribi-Hertz points out that français

avancé lacks enclitic subject pronouns, since it lacks subject-clitic and complex

inversion. Finally, variety (69iv) is Zribi-Hertz’ FEC, as already mentioned.

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 §3.1 above. It seems likely that this

kind of optional liaison is unavailable in français avancé, and possibly in variety

(69ii). Other syntactic features may correlate with these distinctions: français avancé

lacks the preverbal negative element ne (as can in fact be observed in (67b)), and may

have systematic wh-in-situ in root clauses, for example.

The approach to null subjects outlined in §2 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 Müller’s pro-generalisation; we saw

also that the latter is derivable from the general approach.

4. Conclusion

We began this paper by observing that the typology of empty categories, including

empty pronouns, that was developed in GB theory cannot be 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-

37 “la langue standard productive, c’est-à-dire débarrassée des ses archaïsmes, i.e. des tournures bien formées du français classique devenues improductives aujourd’hui” (Zribi-Hertz (1994:136)). My translation.

55

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 (2006a). 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 φ-features have undergone (pre-syntactic) impoverishment; this

establishes the connection between null subjects and “rich” agreement, since non-

impoverished φ-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 Müller (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-subjects”

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. Whether or not that

view turns 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 (2006a). 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.

56

References

Adams, M. (1987a) “From Old French to the Theory of Pro-Drop,” Natural Language

and Linguistic Theory 5:1-32.

Adams, M. (1987b) Old French, Null Subjects and Verb-Second Phenomena. PhD

Dissertation, UCLA.

Adragão & J. Costa (2006) “On the status of preverbal subjects in null subject

languages: evidence from acquisition,” ms. Universidade Nova de Lisboa.

Alexiadou, A. & E. Anagnostopoulou (1998) “Parametrizing Agr: word order, verb-

movement and EPP-checking.” Natural Language and Linguistic Theory 16: 491-539.

Anderson, S. (1982) “The analysis of French schwa, or how to get something for

nothing,” Language 58:534-573.

Armstrong, L. (1962) The phonetics of French. A practical handbook, London: Bells

& Son

Baker C.L. (1970) “Notes on the description of English questions: the role of an

abstract question morpheme,” Foundations of Language 6: 197-219.

Barbosa, P. (1995) Null Subjects. PhD Dissertation, MIT.

Barbosa, P. (2006) “Two kinds of subject pro,” talk given at the Workshop on Null-

Subject Languages, Cambridge.

Barbosa, P., E. Duarte & M. Kato (2005) “Null Subjects in European and Brazilian

Portuguese,” ms. Universities of Minho, Rio de Janeiro and Campinas.

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:

John Benjamins, pp. 47-131.

Biberauer, T. & I. Roberts (2005). Changing EPP-parameters in the history of

English: accounting for variation and change. English Language and Linguistics 9: 5

– 46.

Biberauer, T. & I. Roberts (this volume). “Subjects, Tense and Verb-movement in

Germanic and Romance.”

Boeckx, C. & N. Hornstein (2003) “Reply to ‘Control is not Movement’” Linguistic

Inquiry 34: 269-280.

Boeckx, C. and Hornstein, N., 2004. “Movement under control.” Linguistic Inquiry

35, 431-452.

57

Borer, H. (1986) “I-Subjects,” Linguistic Inquiry 17: 375-416.

Borer, H. (1989) “Anaphoric Agr” in O. Jaeggli & K. Safir (eds) The Null Subject

Parameter. Dordrecht: Kluwer, pp. 69-109.

Brandi, L. & P. Cordin (1989) “Two Italian Dialects and the Null Subject Parameter,”

in O. Jaeggli & K. Safir (eds) The Null Subject Parameter, Dordrecht: Kluwer, pp.

111-142.

Burzio, L. (1986) Italian Syntax: A Government-Binding Approach. Dordrecht:

Kluwer.

Cardinaletti, A. (1997) “Subjects and clause structure,” in L. Haegeman (ed) The New

Comparative Syntax. London: Longman.

Cardinaletti, A. (2004) “Towards a cartography of subject positions.” In L. Rizzi (ed)

The Structure of IP and CP: The Cartography of Syntactic Structures, Volume 2.

Oxford: Oxford University Press, pp. 115-165.

Cardinaletti, A. & L. Repetti (2006a) “Proclitic vs Enclitic Pronouns in Northern

Italian Dialects,” talk given at the 1st Cambridge Italian Dialect Syntax Meeting.

Cardinaletti, A. & L. Repetti (2006b) “Subject Clitics in Interrogative Sentences:

Phonology, Syntax and Microvariation,” ms. SUNY Stony Brook/University of

Venice.

Cardinaletti, A. & 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: de Gruyter, pp. 145-235.

Carnie, A., S. A. Dooley & H. Harley (2005) Verb First. Amsterdam: John

Benjamins.

Chenal, A. (1986) Le franco-provençal valdôtain. Aosta: Musumeci Editeur.

Chomsky, N. (1973) “Conditions on Transformations,” in S. Anderson & P Kiparsky

(eds) A Festschrift for Morris Halle. New York: Holt, Reinhart & Winston, pp. 232-

286.

Chomsky, N. (1981) Lectures on Government and Binding. Dordrecht: Foris.

Chomsky, N. (1993) “A Minimalist Program for Linguistic Theory,” in K. Hale &

S.J. Keyser (eds) The View from Building 20: Essays in Linguistics in Honor of

Sylvain Bromberger. Cambridge, Mass.: MIT Press, pp. 1-52.

Chomsky, N. (1995) The Minimalist Program. Cambridge, Mass.: MIT Press.

58

Chomsky, N. (2000) “Minimalist Inquiries: The Framework,” in R. Martin, D.

Michaels & J. Uriagereka (eds) Step by Step: Essays on Minimalist Syntax in Honor

of Howard Lasnik. Cambridge, Mass.: MIT Press, pp. 89-156.

Chomsky, N. (2001) “Derivation by Phase,” in M. Kenstowicz (ed) Ken Hale: A Life

in Language. Cambridge, Mass.: MIT Press, pp. 1-52.

Chomsky, N. (2005) “On Phases,” ms. MIT.

Chomsky, N. & H. Lasnik (1977) “Filters and Control,” Linguistic Inquiry 8:425-304.

Chung, S. (2005) “What Fronts? On the VP Raising Account of Verb-Initial Order.”

in A. Carnie, H. Harley & S. Dooley (eds) Verb-First, Amsterdam: John Benjamins,

pp. 9-29.

Chung, S. & W. Ladusaw (2003) Restriction and Saturation. Cambridge, Mass.: MIT

Press.

Churchward, C. (1953) Tongan Grammar. Oxford.

Cinque, G. (1988) “On si-constructions and the theory of arb,” Linguistic Inquiry

19:521-81.

Culicover, P. & R. Jackendoff (2001) “Control is not movement,” Linguistic Inquiry

32: 493-512.

Cyrino, S., E. Duarte & M. Kato (2000) “Visible Subjects and Invisible Clitics in

Brazilian Portuguese,” in M. Kato & E. Negrão (eds) The Null Subject Parameter in

Brazilian Portuguese, Frankfurt: Vervuert-IberoAmericana, pp. 55-74.

Davies, W. & S. Dubinsky (2004) The Grammar of Raising and Control. Oxford:

Blackwell.

Déchaîne, R. & M. Wiltschko (2002) “Decomposing Pronouns,” Linguistic Inquiry

33: 409-442.

Dell, F. (1985) Les règles et les sons. Paris: Hermann.

Duarte, E. (1995) A Perda do Princípio “Evite pronome” no Português Brasileiro.

PhD Dissertation, Unicamp.

Figueiredo-Silva, C. (2000) “Main and embedded null subjects in Brazilian

Portuguese,” in M. Kato & E. Negrão (eds) The Null Subject Parameter in Brazilian

Portuguese, Frankfurt: Vervuert-IberoAmericana, pp. 127-146.

Galves, C. (1987) “A sintaxe do português brasileiro” Ensaios Lingüísticos 13:31-50.

Grévisse, M. (1980) Le bon usage. Paris: Glembloux.

Harris, M. (1978) The Evolution of French Syntax: A Comparative Approach.

London: Longman.

59

Harris, M. (1988) “French,” in M. Harris & N. Vincent (eds) The Romance

Languages. London: Routledge, pp. 209-245.

Hawkins, J. (1979) Definiteness and indefiniteness: A study in reference and

grammatiacality prediction. London: Croom Helm.

Holmberg, A. (2005) “Is There a Little Pro? Evidence from Finnish,” Linguistic

Inquiry 36: 533-564.

Holmberg, A., C. Johns, A. Nayudu and M. Sheehan (this volume) “Partial Null-

Subject Languages.”

Hornstein, N. (1999) “Movement and Control,” Linguistic Inquiry 30:69-96.

Hornstein, N. (2003) “On control.” In Hendrick, R. (ed), Minimalist Syntax

Blackwell, Oxford, pp. 6-81

Householder, F. (1981) Syntax of Apollonius Dyscolus. Amsterdam: John Benjamins.

Huang, C-T. J. (1984) “On the Distribution and Reference of Empty Pronouns,”

Linguistic Inquiry 15: 531-574.

Huang, C-T. J. (1989) “Pro-drop in Chinese: A generalized control theory,” in Jaeggli

& K. Safir (eds) The Null Subject Parameter, Dordrecht: Kluwer, pp. 185-214.

Hulk, A. (1986) “Subject clitics and the pro-drop parameter,” in P. Coopmanns, Y.

Bordelois & B. Dotson-Smith (eds) Going Romance: Formal Parameters of

Generative Grammar II. Dordrecht: Foris, pp. 107-120.

Hulk, A. (1991) “Les pronoms clitiques sujet et la théorie linguistique,” Actes du

XVIIIe congrès international de linguistique et de philologie romanes. Tübingen:

Niemeyer, oo. 504-514.

Jaeggli, O. (1982) Topics in Romance Syntax, Dordrecht: Foris.

Kato, M. (1999) “Strong Pronouns, Weak Pronominals and the Null Subject

Parameter,” Probus 11:1-37.

Kato, M. & E. Negrão (eds) The Null Subject Parameter in Brazilian Portuguese,

Frankfurt: Vervuert-IberoAmericana.

Kayne, R. (1972) “Subject inversion in French interrogatives,” in J. Casagrande & B.

Saciuk (eds) Generative Studies in Romance Linguistics. Rowley, Mass: Newbury

House, pp. 70-126.

Kayne, R. (1975) French Syntax: The Transformational Cycle. Cambridge, Mass.:

MIT Press.

Kayne, R. (1983) “Chains, categories external to S, and French complex inversion,”

Natural Language and Linguistic Theory 1: 107-139 (reprinted in Kayne (1984)).

60

Kayne, R. (1991) “Romance Clitics, Verb Movement and PRO,” Linguistic Inquiry

22: 647-686. (Reprinted in Kayne (2000:60-97)).

Kayne, R. (1994) The Antisymmetry of Syntax, Cambridge, Mass.: MIT Press.

Kayne, R. (2000) Parameters and Universals, Oxford: Oxford University Press.

Kayne, R. (2005) Movement and Silence. Oxford: Oxford University Press.

Kayne, R. & J.-Y. Pollock (1978) “Stylistic Inversion, Successive Cyclicity and Move

NP in French,” Linguistic Inquiry 9: 595-621.

Kayne, R. & J.-Y. Pollock (2001) “New Thoughts on Stylistic Inversion,” in A. Hulk

& J.-Y. Pollock (eds) Subject Inversion in Romance and the Theory of Universal

Grammar, Oxford: Oxford University Press, pp. 107-162. (Reprinted in Kayne

(2005:3-49)).

Kiparsky, P. (1973) “’Elsewhere’ in Phonology,” in S. Anderson & P Kiparsky (eds)

A Festschrift for Morris Halle. New York: Holt, Reinhart & Winston, pp. 93-106.

Kornfilt, J. (1997) Turkish. London: Routledge.

Kornfilt, J. (2003) “Turkish,” in William J. Frawley (ed) International Encyclopedia

of Linguistics (2nd edition). Oxford: Oxford University Press, Volume IV, pp. 301-

307.

Lambrecht, K. (1981) Topic, Antitopic and Verb Agreement in Non-Standard French.

Amsterdam: John Benjamins.

Landau, I. (2000) Elements of Control: Structure and Meaning in Infinitival

Constructions. Dordrecht: Kluwer.

Landau, I. (2003) “Movement out of control,” Linguistic Inquiry 34: 471-498.

Landau, I. (2004) “The scale of finiteness and the calculus of control.” Natural

Language and Linguistic Theory 22, 811-877.

Larson, R. & G. Segal (1995) Knowledge of Meaning. Cambridge, Mass.: MIT Press.

Mackenzie, D. (2003) “Pashto,” in William J. Frawley (ed) International

Encyclopedia of Linguistics (2nd edition). Oxford: Oxford University Press, Volume

III, pp. 251-257.

Manzini, M.-R. & L. Savoia (2005) I dialetti italiani e romanci, Alessandria: Edizioni

dell’Orso (3 volumes).

Manzini, M.-R. & A. Roussou (2000) “A minimalist theory of A-movement and

control,” Lingua 110: 409-447.

Marantz, A. (2001) “Words,” WCCFL XX handout, University of Southern

California.

61

Marantz, A. (2006) “Phases and Words,” ms. New York University.

Martin, R. (1996) A Minimalist Theory of PRO and Control. University of

Connecticut PhD Dissertation.

Martin, R. (2001) “Null Case and the distribution of PRO.” Linguistic Inquiry 32,

141-166.

Massam, D. (2000) “VSO and VOS: Aspects of Niuean Word Order,” Carnie, A. & E.

Guilfoyle (2000) The Syntax of Verb-Initial Languages. Oxford: Oxford University

Press, pp. 97-116. Reprinted here as Volume V, Chapter Nine.

Massam, D. & C. Smallwood (1997) “Essential Features of Predication in Niuean and

English” in K. Kusumoto (ed) Proceedings of NELS 27, GLSA, University of

Massachusetts, Amherst. pp. 236-272.

Matushansky, O. (2006) “Head Movement in Linguistic Theory,” Linguistic Inquiry

37: 69-110.

Mavrogiorgos, M. (2006) “Morphology is at the Edge,” ms. University of Cambridge.

May, R. (1985) Logical Form. Cambridge, Mass.: MIT Press.

Modesto, M. (2000) “Null Subjects without “rich” Agreement,” in M. Kato & E.

Negrão (eds) The Null Subject Parameter in Brazilian Portuguese, Frankfurt:

Vervuert-IberoAmericana, pp. 147-174.

Montalbetti, M. (1984) After Binding. MIT PhD Dissertation.

Müller (2005) “Pro-drop and impoverishment,” ms. University of Leipzig.

Munaro, N. & J.-Y Pollock (2005) “Qu-est-ce-que (qu)-est-ce-que? A Case Study in

Comparative Romance Interrogative Syntax,” in G. Cinque & R. kayne (eds) The

Oxford Handbook of Comparative Syntax. Oxford; Oxford University Press, pp. 542-

606.

Muysken, P. (1982) “Parametrizing the notion ‘head’” Journal of LinguisticResearch

2: 57-75.

Nash, L. & A. Rouveret (1997) “Proxy Categories in Phrase Structure Theory”, in K.

Kusumoto (ed) Proceedings of NELS 27, pp. 287-304.

Neeleman, A. & K. Szendrői (2005) “Case Morphology and Radical Prodrop,” ms.

University College London and Utrecht Institute of Linguistics.

Neeleman, A. & K. Szendrői (forthcoming) “Radical pro-drop and the morphology of

pronouns,” in M.T. Biberauer (ed) The Limits of Syntactic Variation. Amsterdam:

John Benjamins.

62

Negrão & A. Müller (1996) “As Mudanças no Sistema Pronominal do Português do

Brasil: Substitução ou Especialização de formas?” DELTA 12:125-152.

Negrão, E. & E. Viotti (2000) “Brazilian Portuguese as a Discourse-Oriented

Language,” in M. Kato & E. Negrão (eds) The Null Subject Parameter in Brazilian

Portuguese, Frankfurt: Vervuert-IberoAmericana, pp. 105-126.

Nunes, J. (1995) The Copy Theory of Movement and Linearization of Chains in the

Minimalist Program. University of Maryland, College Park, PhD Dissertation.

Nunes, J. (2004) Linearization of Chains and Sideward Movement. MIT Press.

O’Neil, J. (1997) Means of Control: Deriving the Properties of PRO in the Minimalist

Program. Harvard University PhD Dissertation.

Ordoñez, P. (1997) Word order and clause structure in Spanish and other Romance

languages. CUNY Graduate Center PhD Dissertation.

Ordoñez, P. (2006) “The order of subjects in Spanish and Catalan,” talk given at the

Encontro Lingua Falada e Escrita V, Federal University of Maceió.

Otsuka, Y. (2005) “Two derivations of VSO: a comparative study of Niuean and

Tongan,” in A. Carnie, S. A. Dooley & H. Harley (2005) Verb First. Amsterdam:

John Benjamins, pp. 65-90.

Öztürk, B. (2001) “Turkish as a non-pro-drop language,” in E. E. Taylan (ed) The

Verb in Turkish. Amsterdam: John Benjamins, pp. 239-260.

Parry, M. (1994) “Posizione dei clitici complemento nelle costruzioni verbali

perifrastiche del piemontese,” in G.P. Clivio & C. Pich (eds) At dël (X) Rëscontr

antërnassional dë Studi an sla Lenga e la Literatura piemontèisa 8: 247-259.

Perlmutter, D. (1971) Deep and Surface Constraints in Syntax. New York: Holt,

Rinehart & Winston.

Pesetsky, D. (1989) “Language particular processes and the Earliness Principle,” ms.

MIT.

Platzack, C. (2004) Agreement and the Person Phrase hypothesis. Working Papers in

Scandinavian Syntax, 73: 83-112.

Poletto, C. (2000) The Higher Functional Field in the Northern Italian Dialects,

Oxford: Oxford University Press.

Poletto, C. & 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, pp. 251-296.

63

Pollock, J.-Y. (1986) “Sur la syntaxe de EN et le parameter du sujet nul.” In M. Ronat

& D. Couquaux (eds) La grammaire modulaire. Paris: Les editions de minuit, 211-

246.

Pollock, J.Y. (1989) “Verb movement, Universal Grammar and the structure of IP”

Linguistic Inquiry 20:365-424.

Pollock, J.Y. (1997) Langage et Cognition: Introduction au Programme Minimaliste

de la Grammaire Générative. Paris: Presses Universitaires de France.

Pollock, J.Y. (2006) “Subject-Clitic Inversion, Complex Inversion and Stylistic

Inversion in French,” in M. Everaert & H. van Riemsdijk (eds) Syncom; The Syntax

Companion, Oxford, Blackwell, Vol V.

Rackowski, A. & L. Travis (2000) “V-initial Languages: X or XP Movement and

Adverbial Placement,” in Carnie, A. & E. Guilfoyle (2000) The Syntax of Verb-Initial

Languages. Oxford: Oxford University Press, pp. 117-142.

Raposo, E. (1986) “On the Null Object in European Portuguese,” in O. Jaeggli & C.

Silva-Corvalán (eds) Studies in Romance Linguistics. Dordrecht: Foris, pp. 373-390.

Renzi, L. & L. Vanelli (1983) “I pronomi soggetto in alcune varietà romanze” in Studi

in onore di G.B. Pellegrini, Padua.

Richards, M. & M.T. Biberauer (2005) “Explaining EXPL”, in M. den Dikken & C.

Tortora (eds) The function of function words and functional categories. Amsterdam:

Benjamins, pp. 115-154.

Rizzi, L. (1982) Issues in Italian Syntax. Dordrecht: Foris.

Rizzi, L. (1986a) “Null Objects in Italian and the Theory of pro,” Linguistic Inquiry

17: 501-557.

Rizzi, L. (1986b) “On the Status of Subject Clitics in Romance,” in O. Jaeggli & C.

Silva-Corvalán (eds) Studies in Romance Linguistics. Dordrecht: Foris, pp. 391-420.

Rizzi, L. (1987) “Three Remarks on Null Subjects,” talk given at the Workshop on

Dialectology, GLOW X, University of Venice.

Rizzi, L. & I. Roberts (1989) “Complex Inversion in French,” Probus 1, 1-30.

Roberge, Y. (1986) The Syntactic Recoverability of Null Arguments. University of

British Columbia PhD Dissertation.

Roberge, Y. & M-T. Vinet (1989) La variation dialectale en grammaire universelle.

Montréal: Presses Universitaires de Montréal/Editions de l’Université de Sherbrooke.

Roberts, I. (1993a) "The Nature of Subject Clitics in Franco-Provençal Valdôtain," in

A. Belletti (ed) Dialects of Italy, Turin: Rosenberg & Sellier, pp. 319 - 353.

64

Roberts, I. (1993b) Verbs and Diachronic Syntax: A Comparative History of English

and French, Dordrecht: Kluwer.

Roberts, I. (2006a) “Clitics, head movement and defective goals,” ms. University of

Cambridge.

Roberts, I. (2006b) “Two Ways to Lose Null Subjects,” talk given at the Encontro

Lingua Falada e Escrita V, Federal University of Maceió.

Rosenbaum, P. (1967) The Grammar of English Predicate Complement

Constructions. Cambridge, Mass. :MIT Press.

Russell, B. (1905) “On Denoting,” Mind 14: 479-493.

Di Sciullo, A-M. (1990) “On the Properties of Clitics,” in Binding in Romance,

Carleton University, Ottawa: Canadian Linguistics Association.

Sheehan, M. (2006) The EPP and Null Subjects in Romance. University of

Newcastle-Upon-Tyne PhD Dissertation.

Shlonsky, U. (1989) “The hierarchical representation of subject-verb agreement.”

Shlonsky, U. (forthcoming) “Partial null subjects in Hebrew,” ms. University of

Geneva.

Sportiche, D. (1983) Structural Invariance and Symmetry in Syntax. MIT PhD

Dissertation.

Sportiche, D. (1999) “Subject Clitics in French and Romance, Complex Inversion and

Clitic Doubling”, in K. Johnson & I. Roberts (eds) Beyond Principles and

Parameters: Essays in Memory of Osvaldo Jaeggli, Dordrecht: Kluwer, pp. 189-222.

Suñer, M. (2002) “The Lexical Preverbal Subject in a Romance Null-Subject

Language: Where Art Thou?” Nuñez-Cedeño, R., L. Lopez & R. Cameron (eds) A

Romance Perspective on Language Knowledge and Use. Amsterdam: John

Benjamins.

Taraldsen, T. (1978) “On the NIC, vacuous application and the that-trace filter,” ms.

MIT.

Taylan, E. E. (ed) (2001) The Verb in Turkish. Amsterdam: John Benjamins.

Tomioka, S. (2003) “The semantics of Japanese null pronouns and its cross-linguistic

implications,” in K. Schwabe & S. Winkler (eds) The interfaces: Deriving and

interpreting omitted structures. Benjamins, pp. 321-40.

Vance, B. (1989) Null Subjects and Syntactic Change in Medieval French, PhD

Dissertation, Cornell University.

65

Vance, B. (1997) Syntactic Change in Medieval French: Verb Second and Null

Subjects, Dodrecht: Kluwer.

Wanner, D. (1987) The Development of Romance Clitic Pronouns from Latin to Old

Romance. Berlin: Mouton de Gruyter.

Vanelli, Renzi & Benincà (1985) “Typologie des pronoms sujets dans les langues

romanes,” Actes du XIIe Congrès des Linguistique et Philologie Romanes. Aix-en-

Provence.

Vicente, L. (2005) “Towards a unified theory of movement: an argument from

Spanish predicate clefts,” in M. Salzmann & L. Vicente (eds) Leiden Papers in

Linguistics 2:43-67.

Zubizarrets, M.-L. (1998) Word Order, Prosody and Focus. Cambridge, Mass.MIT

Press.

Zribi-Hertz (1994) “La syntaxe des clitiques nominatifs en français standard et en

français avancé,” in G. Kleiber & G. Roques (eds) Travaux de linguistique et de

philology. Strasbourg-Nancy: Klincksieck, pp. 131-147.

Zwicky, A. & G. Pullum (1983). Cliticisation vs. inflection: English n’t.” Language

59: 502 – 513.

66