A unified account of object clitics and referential null objects in French

A Unified Account of Object Clitics andReferential Null Objects in French

Theres Gruter

Abstract. It has traditionally been assumed that French is a non-null-object language ona par with English. Yet the analysis of adult French corpora has shown the consistentoccurrence of referential null objects in speech and writing. These constructions,although clearly marked, put into question a major premise of syntactic analyses ofobject-clitic constructions—namely, that sentences referring to a specific, referentialobject but lacking both a clitic and a postverbal object are necessarily ungrammatical.The goal of this paper is to present a revised analysis of object-clitic constructions thatis capable of integrating referential null objects. It is proposed that the zero morphemeconstitutes the default realization of the accusative clitic head and is inserted if thishead is underspecified for Case. The analysis extends naturally to account for predicatele, as well as to errors observed in child language development, in particular theoveruse of masculine singular le as well as object (-clitic) omission. It is proposed thatthe syntactic representation underlying clitic drop in child and adult French is identical,yet child and adult French differ with regard to the reason for the underspecification ofthe clitic head.

1. Introduction

At least since Kayne�s French Syntax (1975), object-clitic constructions,illustrated in (1a), have been one of the most extensively discussedgrammatical properties of French, and of the Romance languages moregenerally. Within this discussion and the analyses proposed it is typicallytaken as a premise that French is a non-null-object language (see, e.g., Huang1984, Raposo 1986). In other words, a sentence referring to a specific,referential object but lacking both a clitic and a postverbal object DP, as in(1b), is traditionally considered ungrammatical.1

I wish to thank Kyle Johnson and Yves Roberge, whose comments on an earlier version of thispaper, which constituted part of my Ph.D. dissertation (Gruter 2006), were instrumental. I wouldalso like to thank the three anonymous Syntax reviewers, whose suggestions on the present versionhave been extremely helpful. This research was supported in part by a postdoctoral fellowshipfrom the Swiss National Science Foundation (SNSF), for which I am grateful. Any remainingerrors are my own.

1 Generic, or nonspecific, null objects, an example of which is given in (i), are not discussedhere. Unlike referential null objects, generic null objects occur fairly freely in most languages.(Note that the English translation of (i) is acceptable.) It thus seems that a different analysis isrequired for null objects in generic constructions, as proposed, for example, by Cummins &Roberge (2004, 2005), to which I refer the reader for further detail.

(i) Les ecrivains attirent Ø sexuellement.�Writers attract Ø sexually.� (M. Duras, reported in Lambrecht & Lemoine 1996:286)

� 2009 The AuthorJournal compilation � 2009 Blackwell Publishing Ltd

Syntax 12:3, September 2009, 215–241 DOI: 10.1111/j.1467-9612.2009.00128.x

(1) a. Luc le/la/les met sur la table.Luc cl-(masc-sg/fem-sg/pl) put-3sg on the table�Luc is putting him(it)/her/them on the table.�

b. */?Luc met sur la table.Luc put-3sg on the table�Luc is putting (it) on the table.�

Recent evidence suggests, however, that this is not entirely true. A growingbody of corpus-based findings from both written and spoken French illustratethat referential null objects do occur (Fonagy 1985; Lambrecht & Lemoine1996, 2005; Larjavaara 2000; Noailly 1997). Yet syntactic analyses of thisphenomenon have been all but absent, with the notable exception of work byCummins & Roberge (2004, 2005), which I will discuss in more detail later.A few examples of attested ‘‘clitic-drop’’ constructions (to use the terminologyof Cummins & Roberge) are given in (2)–(4).

(2) A: J�ai un truc pour toi si ca t�interesse.�I have something for you if you�re interested.�

B: C�est quoi?�What is it?�

A: Je crois que t�aimes bien, toi, ce genre de truc. J �ai trouve hier.�I think that you like this sort of thing. I found Ø yesterday.�

(spoken interaction; Lambrecht & Lemoine 1996:297)

(3) Un jour, je me disais, je mettrais une petite annonce dans Le Provencal:[…] Mais je renvoyais toujours a plus tard.�One day, I told myself, I would put a classified ad in Le Provencal:[…] But I kept putting Ø off until later.� (literary text; Larjavaara 2000:63)

(4) …Et la tete qu�il fait le jour ou on rapporte au logis un store decored�une photo de Marilyn. S �il deteste vraiment, on le case dans la sallede bain.�…And the look on his face the day you bring home a blind decoratedwith a photo of Marilyn [Monroe]. If he really hates Ø, you stick it inthe bathroom.� (Cosmopolitan, August 1996, p. 118; reported in

Noailly 1997:100, trans. Cummins & Roberge 2005:52)

(Note that the English translations of these sentences are strongly unaccept-able.)In all of these examples, using a clitic instead of a null object would also

have been acceptable. Indeed, Cummins & Roberge (2005:52) initiallyobserve that null objects of this type appear ‘‘in the same contexts as nullobjects recovered by a clitic and [their] interpretation is identical.’’ However,both Lambrecht & Lemoine (1996:296–304) and Larjavaara (2000:63–76), as

216 Theres Gruter


well as Cummins & Roberge (2005) later on, point out that certain factorsfavor the occurrence of referential null objects. These include third-personreference, nonhuman reference, the cooccurrence of a pronominalized dativeargument, and reference to a proposition or process. Lambrecht & Lemoineobserve furthermore that null objects appear in cases where an object cliticcould not be substituted, as in (5).

(5) A: Je vais avoir trente ans.I will have thirty years�I will be thirty.�

B: J �ai deja eu, moi.I have already had me�I already am.� (spoken interaction; Lambrecht & Lemoine

1996:299, 2005:40)

Lambrecht & Lemoine (2005:40) argue that ‘‘in this example, the personalpronoun (#Je les ai deja eus, moi �I�ve already had them�) would beinappropriate because it would lend to the complement a specific referentialstatus incompatible with the indefinite quantified antecedent NP.’’ Exampleslike this lead them to conclude that a speaker can have recourse to null objectsas ‘‘a kind of default solution’’ (p. 40) in cases where ‘‘the grammar of Frenchoffers no convenient alternative’’ (p. 39).Thus there appears to be converging evidence that referential null objects

are indeed attested in French, although their exact distribution is only poorlyunderstood at this point. Although earlier studies proposed that referential nullobjects were confined to a closed (albeit large) class of transitive verbs(Fonagy 1985, Tuller 2000), more recent work assumes the null instantiationof referential direct objects to be a fully productive, though stronglystigmatized, option in contemporary French, constrained by pragmatic,discursive, and stylistic factors (Cummins & Roberge 2005, Lambrecht &Lemoine 2005). This observation puts into question whether French is a non-null-object language on a par with English—an assumption that previousaccounts of object-clitic constructions have taken as a premise. It also impliesthat syntactic accounts of object-clitic constructions within which sentenceslike (1b) and (2)–(5) cannot be the result of a convergent derivation are notdescriptively adequate. The goal of this paper is therefore to present asyntactic analysis capable of accounting for both clitic (e.g., (1a)) and clitic-drop (e.g., (1b), (2)–(5)) constructions in a unified manner. To this end, I drawextensively on the insights of Cummins & Roberge (2004, 2005) regardingreferential null objects, on the one hand, and on Sportiche�s (1996, 1998)seminal analysis of object-clitic constructions, on the other. In section 2,I begin with a brief review of Sportiche�s proposal. In section 3, I develop anadaptation of his analysis within a more recent minimalist framework(Chomsky 2000, 2001), combined with the view of morphology and thelexicon proposed by Halle & Marantz (1993, 1994; Distributed Morphology).

A Unified Account of Object Clitics and Referential Null Objects in French 217


In section 4, I illustrate how such an analysis extends rather naturally to clitic-drop constructions, capturing the same intuitions as Cummins & Roberge(2005) proposed for this construction specifically, as well as the discourse-dependence of referential null objects observed in previous work. In section 5,I point out a welcome side effect of the proposed analysis regarding predicatele. Finally, in section 6, I address clitic drop and the overuse of the masculinesingular clitic le in child French, suggesting that the syntactic analysisproposed in this paper allows for a coherent analysis of both of these errortypes in French child language.

2. A Brief Overview of Sportiche�s (1996) Analysis

Sportiche�s (1996) analysis presents a reconciliation of two previouslycompeting views of Romance object-clitic constructions. Movementapproaches (e.g., Kayne 1975) held that the clitic was base-generated in thecanonical object position as the complement of V0, followed by successivecyclic movement (or a movement transformation, in the terminology of thetime) to its preverbal surface position. Base-generation approaches (Strozer1976, Rivas 1977) argued for base-generating the clitic in its actual surfaceposition, with later analyses in this vein positing an empty category in thecomplement of V0 in (nondoubled) clitic constructions to satisfy theta and caserequirements (Borer 1984; Jaeggli 1982, 1986). Sportiche (1996) adopts theinsight of these latter proposals that the internal argument of object-cliticconstructions may be an empty category, while the clitic itself constitutes anagreement marker of some sort, although the exact status of clitics remainssomewhat open in all of these accounts.2 At the same time, Sportiche pointsout a number of properties that remain uncaptured in base-generationapproaches, yet are accounted for straightforwardly by movement theories.These include the well-known participle-agreement phenomenon, illustrated in(6), whereby a participle can show gender and number agreement only with apreceding argument or clitic, as well as the locality restrictions on cliticplacement, which abide by the Specified Subject Condition (SSC), a conditionthat also constrains phrasal movement (Kayne 1975, Burzio 1986).

(6) a. Jean a peint(*e) la porte.Jean has painted(*fem) the door�Jean painted the door.�

2 Jaeggli (1982:54 n. 10), for example, argues that clitics are not agreement markers, based onthe observation that they have ‘‘a more independent syntactic status’’ (but see Auger 1994, 1995for arguments that the morphological and the syntactic status of clitics are independent dimen-sions). For Borer (1984:41) clitics are affixes that are the output of an inflectional rule applyinglate in the derivation. Roberge (1990:167–174), who shares the basic assumptions of Jaeggli andBorer, discusses agreement markers versus clitics and comes to the conclusion that the two mustbe distinct. Suner (1988:393), on the other hand, claims explicitly that (Spanish) clitics aremanifestations of object agreement. For Sportiche�s (1996) position on this issue, see below.

218 Theres Gruter


b. La portei que Jean a peint(e) ti.the door that Jean has painted(fem)�The door that John painted.�

c. Jean l� a peint(e) e.John it has painted(fem)�John painted it.� (Sportiche 1996:227)

The SSC states that phrasal movement cannot cross a specified subjectposition. As illustrated in (7), accusative-clitic placement appears to besensitive to precisely this condition: it cannot reach the main clause over thesubject of the embedded clause. If clitic placement is the result of movement,this restriction is explained straightforwardly.

(7) a. Jean laisse le chat manger la souri.Jean lets the cat eat the mouse�Jean lets the cat eat the mouse.�

b. Jean le laisse la manger.Jean it (= cat) lets it (= mouse) eat�Jean lets it eat it.�

c. Jean laisse le chat la manger.d. *Jean la laisse le chat manger.

Jean it lets the chat eat

Drawing on arguments from both base-generation and movement analyses,Sportiche (1996) proposes that clitics are base-generated as heads of their ownprojections (called ‘‘Clitic Voices,’’ in the case of direct object clitics:‘‘AccP’’) that select as their specifier an accusative DP, thus triggeringmovement of the object DP to this position by LF. Depending on language-specific settings for three ‘‘[c]litic construction parameters’’ (Sportiche1996:237), the head of AccP may be lexically realized or null, movementto its specifier may be overt or covert, and the object DP may belexically realized or null (pro). This entails the partial phrase structure shownin (8).



(Partial) representation of clitic constructions, illustrating(8) Sportiche’s (1996) analysis

head movement

XP movement (covert)

AccP

AgroP

Agro0

VP

V DP

Acc0

CL

XP movement (overt or covert)

pro

Sportiche leaves open the exact order of projections, yet he argues that AccPmust be located very high in the clause. But note that, as indicated in (8), theclitic is not assumed to remain in AccP in the surface structure. The verb raisesthrough the Acc head, at which point the clitic incorporates into the verb. Thecomplex Acc+V head then moves up to an even higher functional projection(for further details, see Sportiche 1996:240).This analysis accounts for both movement properties such as participle

agreement and locality restrictions, as well as clitic-doubling constructions, asobserved, for example, in some varieties of Spanish (see, e.g., Jaeggli 1982).The latter can now simply be treated as constructions with an overt object DP(and movement delayed until LF). French differs from clitic-doublinglanguages only in that the object DP must be realized as pro in the presenceof an overt clitic head. Participle agreement, as in movement-only accounts,can be treated as the result of a spec-head agreement relation at anintermediate landing site of the object (DP or pro), identified in (8) asSpec,AgrOP (but see Friedemann & Siloni 1997 for arguments that participleagreement is not checked in AgrOP). The locality restrictions (SSC) alsofollow straightforwardly, given that movement of pro is phrasal movementin Sportiche�s account.As regards the status of the clitic itself, Sportiche (1996), following Jaeggli

(1982, 1986) and Borer (1984), assumes that pro generated in the complementof V is the internal argument of the verb. Clitics, then, have no argumentalstatus (see Gruter 2008 for additional arguments in favor of the clitic�s

220 Theres Gruter


nonargumental status). Instead, they are treated ‘‘as complex agreementmorphemes’’ (Sportiche 1996:237). To what extent they align with moregeneral instances of object agreement, however, remains somewhat open (seealso note 2). Sportiche is quick to follow the statement cited above with afootnote starting with ‘‘[b]ut not identical to agreement morphemes’’ (p. 270n. 18). This is reiterated at a later point in the paper: ‘‘[f]undamentally they[agreement and clitics] are identical, both being heads agreeing with phrasalspecifiers. They also differ’’ (p. 265). For example, Sportiche points out thatagreement morphemes typically impose no particular interpretive require-ments on the agreeing DP, although the clitic in this case is obligatorily linkedto specificity (see below). Moreover, agreement triggered by a DP may appearon more than one element in the clause, whereas clitics are typically unique(see Sportiche 1996:265, as well as Roberge 1990:167–174 for furtherdiscussion). In sum, although the exact nature of clitics as ‘‘complexagreement morphemes’’ is not entirely resolved, I take the minimallynecessary assumption of Sportiche�s framework to be that clitics arenonargumental inflectional material of some sort.Importantly, the clitic is seen as the head of a functional projection (the

Clitic Voice) whose function it is to license a certain property [+F] in an XP inits specifier. As Sportiche (1996:264) remarks, ‘‘[i]n this respect, clitics arelike [+wh] Cs, which license wh-phrases, [+neg] heads, which license negativequantifiers and polarity items, and [+focus] heads, which license focalizeditems.’’ In the case of clitics, the relevant property [+F] is assumed to be thatof specificity (p. 264), thus triggering the movement of a [+specific] DP to thespecifier of the Clitic Voice (by LF) in order for its specificity feature to belicensed (or checked). Whether specificity is exactly the right property,however, is not entirely clear. Sportiche himself admits that he is onlymoderately confident with regard to specificity being the right choice offeature (Sportiche 1998:13 n. 4). In later work, he suggests that ‘‘the propertylicensed by the Accusative projection is in fact the property that definitedeterminers denote, call it [+definite], whatever this may be’’ (Sportiche1999:701).3 For the present purpose, I will continue to assume that therelevant feature is [+specific], with the caveat that better understanding of thesemantics of specificity and definiteness may well lead to a revision withregard to the nature of the relevant feature [+F] in clitic constructions. In sum,then, clitics in Sportiche�s account are treated as a special type of agreement,more specifically as ‘‘specificity licenser[s]’’ (Sportiche 1996:268).Sportiche�s proposal is conceived predominantly within a government-and-

binding framework (Chomsky 1981, 1982).4 In the following section, I willattempt to reformulate his account within a more recent minimalist framework

3 But see Suner 1988:397 for arguments that in accusative-clitic-doubling constructions inPorteno Spanish, the relevant feature should be [+specific] rather than [+definite].

4 Although Sportiche�s paper was only published in 1996 (the year of publication is variablygiven as 1996 or 1995), it was written in late 1991 or early 1992 (Sportiche 1998:10), andcirculated as a manuscript for several years.



(Chomsky 2000, 2001), including insights from work in DistributedMorphology (Halle & Marantz 1993, 1994).

3. A Minimalist Adaptation of Sportiche�s Analysis

Following one of the key assumptions of minimalism, I will assume thatsyntactic computation is driven strictly by feature checking. Thus the first taskin formulating a minimalist version of Sportiche�s proposal will be todetermine which features are involved, and in which manner, in the derivationof clitic constructions. Adopting a key assumption of Distributed Morphology(Halle & Marantz 1993, 1994), I assume that these features lack phonologicalcontent during the syntactic computation.5 It is only at the level ofMorphological Structure (MS), adopting a view of the grammar as shown in(9), that bundles of morphosyntactic and semantic features are assignedphonological content in a process of Vocabulary Insertion.

Model of the grammar (cf. Halle & Marantz 1993:114) (9)

Lexicon

MS (Morphological Structure)

PF

LF

This process consists of matching the feature bundles on a terminal headagainst an ordered list of underspecified Vocabulary Items, which make up theVocabulary Entry for the terminal node in question. This is illustrated in (10).The vocabulary item that constitutes the largest proper subset of the featuresrepresented on the terminal node is chosen for insertion.

(10) Vocabulary entry for a terminal node Xvocab. item a. [F1, F2] M PA (= a set of phonological features)vocab. item b. [F1] M PBvocab. item c. M PC

Thus the second major task here will be to specify the Vocabulary Entry forthe terminal node in question, the clitic head (Cl0).I will begin with the first task, that is, determining the checking relations

involved in clitic constructions. Given that no relevant modifications are

5 This assumption is shared by other ‘‘separationist’’ approaches to morphology, which holdthat there is a logical separation between the representation of morphosyntactic features and theirphonological realization(s) (e.g., Anderson 1992, Beard 1995). Any such approach to morphologyis, in principle, compatible with the proposal outlined here.

222 Theres Gruter


required within a minimalist framework, I will continue to adopt the phrasestructure proposed by Sportiche (1996). A very general aspect to be consideredin determining the relevant checking relations is the nature of, and motivationfor, agreement and movement relations between constituents. Recentminimalism (following Chomsky 2000, 2001) differs from earlier work inthat agreement configurations for the purpose of feature checking are nolonger restricted to the specifier-head relation, as assumed, for example, inSportiche (1996). Instead, the ideal agreement configuration is the relationAgree, which may hold between two constituents a, a probe, and b, a goal,within a certain domain (see, e.g., Chomsky 2000:122). In other words,movement of the lower term (the goal) to the specifier of the higher term (theprobe) is no longer necessary to obtain an appropriate agreement relationship.Obtaining an appropriate agreement relationship is precisely what is at the

heart of Sportiche�s (1996:236) Clitic Criterion, reproduced here in (11).

(11) Clitic Criterioni. A clitic must be in a Spec-head relationship with a [+F] XP at LF.ii. A [+F] XP must be in a Spec-head relationship with a clitic at LF.

Given the possibility of Agree in a minimalist framework, the Clitic Criterioncan be reformulated as in (12).

(12) Clitic Criterion—minimalist versioni. A clitic must be in an Agree relation with a [+F] XP.ii. A [+F] XP must be in an Agree relation with a clitic.

The crucial consequence of (12) is that the direct object—the [+F] XP—is nolonger required to move to the specifier of the Clitic Voice for the purpose ofagreement. I therefore assume, in departure from Sportiche (1996), thatpro—and any other [+F] XP—in the complement of V establishes anagreement relationship with the clitic head through Agree at a distance, that is,without movement.Eliminating movement of pro raises the question how the movement

properties of clitic constructions observed above, in particular their abidanceby the Specified Subject Condition, are accounted for within the presentaccount. I expect that within a phase-based framework the effect described bythe SSC can be derived from restrictions on Agree that follow from cyclicSpell-Out, more specifically, the Phase Impenetrability Condition (PIC;Chomsky 2000:108, 2001:14). The general claim of the PIC is that elementswithin the domain of a strong phase are inaccessible to operations outside thatphase. The proposed Agree relation between the clitic and pro in thecomplement of V will almost certainly be barred by the PIC in sentences like(7d), as the clitic is located outside the lower CP whereas pro is likely toremain within its domain. With the exact formulation of the PIC still evolving,however, a more precise elaboration of this argument will have to be left for



future work. It is important to note, however, that even if it turns out that, forindependent reasons, pro does have to move, this would not affect the Agreerelation in (12) and thus would not directly affect the present account.A further consequence of the possibility of Agree at a distance is that the

Clitic Criterion in (12) can be assumed to hold throughout the derivation, ratherthan at LF only. Sportiche (1996) confined this requirement to LF to allow forcovert movement of XP to the specifier of the clitic phrase, an analysis heassumed for clitic-doubling constructions. In the current framework, theClitic Criterion can be fulfilled in these constructions by Agree at a distancebetween the clitic and the doubled phrase in the complement of V. This Agreerelation may hold at any point in the derivation. I therefore assume thatthe Clitic Criterion in (12) applies throughout the derivation, and holds ofSpell-Out.6

The next point to be addressed is the nature of the features involved inthis Agree relation. In Sportiche 1996, the driving feature is what he initiallylabels [+F] and later identifies as [+specific] (but see the caveat above).Within a checking framework, this can be operationalized as follows: theclitic head (Cl0) is the probe bearing an unvalued specificity feature, whichmust be valued (or checked) through an Agree relation with a goal bearing avalued specificity feature. This goal can either be a lexical direct objectbearing a [+specific] feature in D, or it can be pro, which I assume to belexically specified for [+specific], given that ‘‘pronouns are quintessentiallyspecific DPs’’ (Sportiche 1996:264).7 Thus an Agree relation between Cl0

and the direct object is established, and the specificity feature on Cl0 isvalued.8 (See (14)–(18) for a step-by-step illustration of the proposedderivation.)The direct object will also bear valued (interpretable) /-features, of which

those for number and gender are relevant for the present purpose.9 This is

6 The availability of covert movement also played a role in Sportiche�s (1996) analysis of theoptionality of past participle agreement in French. He suggested that in constructions lackingparticiple agreement, movement of pro to Spec,ClP (through Spec,AgrOP) is delayed until LF.Within the present framework, Agree between pro and Cl0 need not trigger participle agreement,given that there are no strict relativized minimality restrictions on Agree. If participle agreement ispresent, an additional Agree relation must be assumed, the conditions for which are subject tofurther investigation, to the same extent that the conditions for delaying movement until LFremained open in Sportiche�s original account.

7 Specifying pro [+specific] in the lexicon may appear to be in conflict with Rizzi�s (1986)proarb, an empty pronominal with an arbitrary (i.e., generic) interpretation. Note, however, that thecontexts for which Rizzi proposed proarb correspond to the generic-null-object contexts discussedby Cummins & Roberge (2005; see note 1). I assume, with Cummins & Roberge (2005:55–59),that the null element in these contexts is not pro but rather an empty bare noun.

8 This is analogous to the analysis proposed by Bruening & Rackowski (2001:74) for the Defmorpheme in Wampanoag (Eastern Algonquian), which the authors hypothesize to correspond toSportiche�s Clitic Voice.

9 Potentially, person is also relevant. Given the diachronic and synchronic differences betweenfirst- and second-person versus third-person clitics (e.g., Kayne 1975:101, Uriagereka 1995:79),the present discussion is restricted to third-person contexts. Yet it seems that, if need be, the persondimension could be integrated into the present analysis without serious consequences.

224 Theres Gruter


obvious for lexical objects, on the general assumption that number and genderare intrinsic lexical features of N (Chomsky 1995). It is somewhat less obviousfor pro, as it is clear that pro cannot have lexically specified /-features (Rizzi1986, Bennis 2006). I assume that referential pro acquires its semantic contentfrom its discourse antecedent, as has been proposed for null argumentsin languages without rich agreement (see, e.g., Hoji 1998 for Japanese).I propose that /-features are included in the specifications that pro acquiresfrom its discourse antecedent. In consequence, when the direct object (lexicalor pro) enters into the Agree relation with Cl0, Cl0 will also acquire theobject�s /-features, as an Agree relation between two terms generally entailsthe copying of all features from one term to the other (i.e., not just the one[s]that triggered the checking relation). Thus as a result of the Agree relationbetween Cl0 and the direct object, Cl0 not only values its specificity feature,but it also acquires the object�s /-features.Furthermore, like any nominal, the direct object (lexical or pro) needs to

check structural Case. Expressed in minimalist terminology, the object bearsan uninterpretable Case feature. This Case feature must be checked by arelevant Case ‘‘assigner,’’ which I assume to be at or within vP. (The exactlocus of accusative-Case checking within this domain is irrelevant for thepresent purpose.) Thus before entering the Agree relation with the clitic headdiscussed above (i.e., before Cl0 is merged), the direct object establishes achecking relation with the accusative-Case assigner. As a result, the objectnow bears a valued [acc] Case feature. According to Chomsky (1995:278),Case features are always uninterpretable, even when checked. Thereforethey need to be eliminated before the derivation reaches the interfaces.Chomsky (1995:280) suggests that as soon as Case features are checked, theyare erased, that is, they disappear entirely from the representation. I willassume that this is true in the general case, namely that of overt DPs. Inthe case of pro, however, I propose that the valued Case feature is noterased immediately, assuming a requirement for Case to be expressed overtlybefore it can be erased, call it the ‘‘Case-expression requirement,’’ statedin (13).10

(13) The Case-expression requirementA valued Case feature must be associated with a non-null elementbefore it can be erased.

A lexical object and object pro will therefore differ in their featurespecification after Case checking: the lexical object no longer has a Casefeature, whereas pro continues to bear [acc].

10 Wexler (2002) also suggests that the Case feature on pro in object-clitic constructions maynot be erased immediately on checking, despite its being uninterpretable: ‘‘perhaps the case featurehangs around at least through ClP derivation, being part of same �phase.�’’



With this in mind, let us reconsider the Agree relation between the directobject and the clitic head. By the time this relation is established, assumingthat Cl0 is merged higher than the accusative Case assigner, the object willhave checked its Case feature. In the case of lexical objects, this means that theCase feature is no longer present at that point, therefore nothing else needs tobe said. In the case of object pro, however, the [acc] Case feature is stillpresent. Thus [acc] will be copied onto Cl0 along with the object�s /-features.Note, therefore, that just in the case when the object is silent (pro), the clitichead will acquire a Case feature, which will be expressed overtly on the clitichead, and erased from the derivation at that point, a procedure reminiscent ofBorer�s (1984) view of clitics as spell-out of Case features. The proposedderivations for both clitic and lexical object constructions are illustrated stepby step in (14)–(18).

Step 1: Merge pro or lexical object (la petite fille ‘the little girl’). (14)

DPV

VP

pro/la petite fille[+specific][number][gender][uCase]

(Prefixed u- denotes an unvalued feature.)

Step 2: Merge AgrO0 (or whatever the accusative (15)

Case assigner may be) and check Case.

pro/la petite fille

DPV

VPAgrO

0

AgrOP

[+specific][number][gender][ACC]

(16) Step 3: If Case-expression requirement (see above) is fulfilled, delete [Case].fi [Acc] is deleted on lexical object but not on pro.

226 Theres Gruter


Step 4: Merge Cl0. (17)

a. object = pro

[uspecific]

DP

VP

C1P

pro [+specific]

V

[number] [gender] [ACC]

C10

b. object = lexical DP

[uspecific] VP

C10

C1P

la petite fille

DP V

[+specific][number][gender]

Step 5: Check specificity. (18)

a. object = pro b. object = lexical DP

[+specific] [+specific] [number] [number] [gender] DP V DP V

VP VP C10 C10

C1P C1P

[gender] [ACC]

pro la petite fille [+specific] [+specific] [number] [number] [gender] [gender] [ACC]

By the time the derivation exits the narrow syntax and enters themorphological component (MS), the clitic head will have the followingfeatures: (i) [+specific], (ii) /-features (number, gender), and (iii), iff theobject is silent, the Case feature [acc]. With this established, we can now turnto the second task, the specification of the Vocabulary Entry for Cl0. Theproposed representation is given in (19).

(19) Vocabulary Entry for Cl0

a. [+specific], [acc], [pl] M /les/b. [+specific], [acc], [fem] M /la/c. [+specific], [acc] M /le/d. [+specific] M Ø



Note that (19) mandates that unless the terminal node bears the Case feature[acc], the chosen item will always be item (d), the zero morpheme. Asdiscussed above, Cl0 bears [acc] if and only if the direct object is silent. Thus(19) correctly derives the fact that whenever the direct object is overtlyrealized, the clitic is silent.11 If the feature bundle on Cl0 includes [acc], it willbe overtly realized (items a–c), with the choice of items depending on thenature of /-features.I assume that these features aremonovalent (following, for example, Harley&

Ritter 2002). Thus what is traditionally described as the ‘‘masculine singular’’clitic (item c, le), is characterized in (19) by the absence of both number andgender features. This underspecification of le has several desirable side effects,which I discuss in more detail below. As regards items a and b, which arespecified for an equal number of features, intrinsic ordering is necessary tocapture the fact that in the case of an object specified for [pl] and [fem], only itema can be chosen, as illustrated in (20).

(20) A: Est-ce que tu connais ces filles?q you know these girls�Do you know these girls?�

B. Oui, je les/*la connais.yes, I cl know�Yes, I know them/*her.�

Although this ordering could simply be stipulated (Halle & Marantz1993:127), this is probably not necessary, as there appears to be convergingevidence that gender depends in some sense on number. This is expressed byGreenberg�s (1963) Universal 32 (‘‘Whenever a verb agrees with a nominalsubject or object in gender, it also agrees in number’’) and captured morerecently in Noyer�s (1992, 1997) Universal Feature Hierarchy, which includesthe statement ‘‘number features > gender features.’’ Under the view thatmorphological features are ordered in structural geometries (e.g., Bonet 1995,Harley & Ritter 2002), gender could be more deeply embedded in the featurestructure than number. Whatever the best solution will turn out to be, itappears that a relation of number > gender can be derived independently, andthus the ordering of items a and b in (19) will follow automatically.In sum, the Vocabulary Entry in (19), together with the agreement and

checking relations discussed in the first part of this section, correctly derive the

11 Additional assumptions are necessary to account for clitic-doubling languages, a discussionof which is beyond the scope of this paper. Note, however, that to account for Kayne�s Gener-alization—‘‘[a]n object NP may be doubled by a clitic only if the NP is preceded by a preposition’’(Jaeggli 1982:20)—the general solution has been to attribute independent Case-assigning prop-erties to this preposition, so that the Case assigned by the verb (i.e., [acc]) remains available forthe clitic. Thus even in clitic-doubling constructions, it appears possible for the clitic to spell out[acc]. Given this possibility, an integration of clitic-doubling constructions into the present ac-count should not encounter any fundamental difficulties.

228 Theres Gruter


descriptive facts of third-person accusative object-clitic constructions inFrench. In the next section, I show how clitic-drop constructions can beintegrated into this account rather straightforwardly.

4. Integrating Referential Null Objects

Clitic-drop constructions of the type illustrated in (2)–(5) above are typicallynot mentioned in the context of syntactic accounts of clitic constructions,including Sportiche 1996. To the best of my knowledge, the only currentlyavailable syntactic account of this construction is the one proposed byCummins & Roberge (2004, 2005). I will therefore begin with a brief reviewof Cummins & Roberge�s analysis and then illustrate how it combines with theanalysis of clitic constructions proposed in the previous section to yield aunified account of clitic and clitic-drop constructions in French.Cummins & Roberge establish a typology of null objects in French,

including both referential and generic cases (2005:61). This typology containstwo types of (nondeictic) referential null objects (NOs): (i) ‘‘definite NOsrecovered by a clitic,’’ and (ii) ‘‘definite NOs not recovered by a clitic.’’ Thefirst type corresponds to the familiar clitic construction discussed in theprevious section and illustrated in (1a), with the authors adopting a syntacticanalysis along the lines of Sportiche 1996. The second type corresponds to thereferential null objects illustrated in (2)–(5). Under Cummins & Roberge�sanalysis, these constructions differ only minimally from the familiar cliticconstructions ‘‘except that there is no overt clitic’’ (2005:52). This perspectivetheoretically allows for two alternatives: either the clitic is syntactically absent,that is, there is no Cl0 head; or the clitic is phonologically absent, that is, Cl0 ispresent but realized by a zero morpheme. The former would align theseconstructions with object drop in classic null-argument languages such asChinese, where referential null objects have traditionally been analysed as nullvariables bound by an empty topic (Huang 1984). Cummins & Roberge argueagainst this option, based on the claim that null objects in French are notsubject to the distributional restrictions that one would expect under anoperator-variable analysis, citing examples like (21), in which the overtmaterial in CP would be expected to block a null object, if this null objectwere a variable bound by an operator in CP. This example (althoughconstructed, i.e., not drawn from a corpus) appears as acceptable as thosewithout material in Spec,CP (Cummins & Roberge 2005:53).

(21) (in a video store)Si on prenait Tigre et Dragon? Qui a vu Ø?�How about Crouching Tiger, Hidden Dragon? Who has seen Ø?�

(Cummins & Roberge 2005:53)

Instead, Cummins & Roberge explicitly advocate the second option. Theypropose that ‘‘(a) clitics are analyzed as morphological markers on the verb;



(b) they are represented by features throughout the derivation and spelled outat PF; and (c) in the clitic-drop construction no morphophonologicalrealization is assigned to the features’’ (2005:53).This analysis is fully compatible with the one I proposed for (overt) clitic

constructions in the previous section. In particular, recall that the VocabularyEntry proposed for Cl0 (see (19)) contains as the least specified, or default,item the zero morpheme. Thus Cummins & Roberge�s analysis of referentialnull objects can be integrated directly with the analysis of clitic constructionsproposed here: referential null objects are cases in which Cl0 is spelled out bythe default zero morpheme. More precisely, I propose that in the case ofreferential null objects, Cl0 remains underspecified for Case (i.e., it lacks[Acc]), leading to the selection of item d, the null morpheme, from theVocabulary Entry for Cl0 shown in (19).The question raised by this proposal is how the derivation underlying an

utterance with a referential null object and a null clitic (as in (21)) can beconvergent, assuming convergence to require the valuation of all unvaluedinterpretable features and the elimination of all uninterpretable features (see,e.g., Chomsky 2000:95). In the derivation of clitic constructions as proposedhere, what is thus required for convergence is (a) the valuation of the specificityfeature on Cl0, and (b) the checking of the Case feature on pro. If either of thesefail, the derivation will crash and thus, strictly speaking, should not be able to bepronounced. Note that (a), the valuation of the specificity (or ‘‘[+F]’’) feature onCl0, remains entirely unaffected by the presence or absence of a Case feature onpro. Pro being intrinsically marked [+specific] (see above) will value thespecificity feature on Cl0 through andAgree relation regardless of Case features.More needs to be said about (b), that is, the putative absence of a Case

feature in clitic-drop constructions, which is necessary for the selection of thezero morpheme (item d). In the current proposal, the assumption is that Caseon pro is checked within AgrOP, yet not erased immediately, due to the Case-expression requirement (see (13)). Consequently, [acc] is copied onto Cl0

along with all other features of pro as part of the Agree relation between proand Cl0. In the case of clitic-drop constructions, Case checking can beassumed to proceed as usual, with accusative Case checked within AgrOP,eliminating the uninterpretable Case feature on pro. The additional assumptionnecessary to integrate clitic-drop constructions is that the Case-expressionrequirement is violated here—that is, Case on pro is erased on checking andwill therefore not be copied onto Cl0.The status of the Case-expression requirement must thus be seen as that of a

violable constraint. This appears natural, as it is obviously not a universalrequirement: languages allowing null arguments in the absence of relatedmorphological marking (e.g., Japanese) clearly do not abide by it. Moreover,deriving referential null objects in French through the violation of a constraintcorrectly captures their marked status in the language.12 The question that

12 I am grateful to an anonymous reviewer for pointing this out.

230 Theres Gruter


remains to be addressed regards the conditions under which the Case-expression requirement may or may not be violated.The present account assumes that the Case-expression requirement is

violated in exactly those instances where a null referential object appears. Aconverging observation in the literature on null referential objects in adultFrench has been that the null instantiation typically marks the highest degreeof salience, discourse cohesion or topicality. For example, Noailly (1997:107)remarks that ‘‘[l]a representation Ø signale le degre de la plus forte cohesiondiscursive [the null representation marks the strongest degree of discoursecohesion].’’ In a similar vein, Lambrecht & Lemoine (2005:44) observe that‘‘[f]or null-instantiation to be appropriate, it is not sufficient that the addresseebe able to establish the link between the null form and the intended referent.The pragmatic status of the referent in the discourse must be salient enough forits occurrence in the proposition to be considered highly predictable for thehearer at the time of utterance.’’ In sum, it seems that a null referential objectappears most frequently and is most acceptable in a context where there can beabsolutely no doubt as to the identity of its referent. Returning to the Case-expression requirement and its violability, we can thus observe that thisrequirement appears to be violable in particular by an object pro whosereferent has a very high degree of discourse salience or topicality. Note that theassumption within the present proposal has been that pro receives its semanticcontent from its discourse antecedent, similar to proposals for null argumentinterpretation in languages like Japanese. Given this close interpretive relationbetween pro and its discourse antecedent, I propose that in addition to thesemantic content of the antecedent, pro will also bear information regardingthe pragmatic or discourse salience of its referent, which I will express througha pragmatic feature [predictable]. Thus we may say that the Case-expressionrequirement is violable if and only if pro is marked [predictable].In light of this proposal, one may wonder whether a relationship between

the expression of Case and the predictability or discourse salience of anargument can be observed beyond the account presented here. I would like tosuggest that it can. De Hoop & Malchukov (2007:1637), for example, claimthat ‘‘case marking in general attributes to the interpretation and helps to avoidambiguity.’’ From a typological perspective, it has long been argued that oneof the core functions of case marking is discriminatory—that is, todisambiguate the syntactic, semantic, and pragmatic roles of participants(see, e.g., Comrie 1989, Whaley 1997). The need to avoid ambiguity appearsto be of least importance if the referent of a potentially ambiguous argument isextremely salient in the discourse. In other words, if the function of casemarking is to avoid ambiguity with regard to the referent of the argument itmarks, case marking is particularly unimportant if this referent is entirelyunambiguous in a given discourse context. The claim that the Case-expressionrequirement can be violated under the condition that the referent of pro ishighly salient or predictable may thus reflect a functional requirement of casemarking more generally.



In sum, I have argued that referential null objects in adult French can beintegrated into the account of object cliticization proposed in this paper byassuming that in constructions with a referential null object, Cl0 does not beara Case feature, which leads to the selection of the underspecified zeromorpheme in the morphological component of the grammar. The absence of aCase feature on Cl0 is attributed to a violation of the Case-expressionrequirement, which is permissible if and only if pro is marked [predictable],that is, when its referent is highly salient in the discourse context. Not onlydoes this proposal allow for a unified account of object clitics and referentialnull objects, but it also captures Lambrecht & Lemoine�s (pretheoretical)observation that the null option represents ‘‘a kind of default solution’’(2005:40). Referential null objects, or under the present assumptions moreprecisely null clitics, can now be seen as an instance of a ‘‘retreat to thegeneral case’’ (Halle & Marantz 1994:278), a scenario that typically arises incases of ‘‘impoverishment’’ within the morphological component (MS), thatis, the deletion of one or more syntactic features on a terminal node, resultingin the insertion of a less-specified Vocabulary Item (Halle & Marantz1994:278, see also Bonet 1991).

5. A Welcome Side Effect: The Analysis of Predicate le

The analysis proposed here, relying on underspecification and its morpho-logical effects, makes the clear prediction that in cases of underspecification,less-specified items will be chosen over more richly specified items, not viceversa. In other words, we expect to find item d (the zero morpheme) chosenover items a–c (les, la, le), not vice versa. We may also expect item c (le) to beselected over items a and b (les, la) in cases of underspecified number andgender, and not vice versa. Indeed, this latter scenario is precisely whatappears to be underlying predicate le, that is, constructions where the clitic‘‘pronominalizes’’ a nonnominal phrase, as illustrated in (22). (Note that thenonnominal phrases in (22) are obligatorily dislocated; they do not constituteinstances of clitic doubling.)

(22) Louis l�a ete en colere/a plaindre/ professeur/Louis it-has been in a rage/to pity/ professor/fidele a ses amis/ adore de ses enfants/ trahi par ses amis.faithful to his friends adored by his children/ betrayed by his friends

(Sportiche 1995:294)

As Sportiche (1995:294) remarks, ‘‘[t]he clitic seems to be able to stand for aPP, an infinitival CP, an NP, an AP or an Adjectival participal phrase, or aVerbal participal phrase.’’ In this function, the clitic always takes the form ofthe masculine singular le.The analysis proposed here extends directly to predicate le. Assuming that

only DPs bear interpretable /-features, it is predicted that items a and b, which

232 Theres Gruter


are specified for number and gender respectively, will never be chosen in caseswhere the ‘‘pronominalized’’ XP is not a DP. One might wonder, however,why it is item c, rather than d, which is chosen in these contexts. That is, onemight question the presence of an accusative Case feature in these cases.However, Sportiche (1995) points out some curious restrictions on thedistribution of predicate le, which lead him to conclude that (i) predicatele always pronominalizes a CP, and (ii) it ‘‘could only pronominalizeaccusative marked CPs’’ (Sportiche 1995:317; for supporting arguments andfurther detail, I refer the reader to the original source). The vocabulary entryfor Cl0 proposed in (19) captures precisely this observation: item c—le—canbe inserted only if the predicate it agrees with bears an [acc] Case feature.13

The facts presented by predicate le thus lend independent support to the lexicalrepresentation proposed in (19).

6. Further Support: Developmental Data

The acquisition of object-clitic constructions in the Romance languages hasbeen a field of extensive research and debate, at least since Eve Clark�sobservation that ‘‘object clitic pronouns are a fairly late acquisition’’ in childFrench (Clark 1985:714), a phenomenon confirmed by numerous subsequentstudies (e.g., Hamann, Rizzi & Frauenfelder 1996; Muller, Crysmann &Kaiser 1996). These studies have found that French-speaking children who arethree and four years old continue to use object clitics in contexts where theymight be expected in the adult language at rates considerably lower than thosefound in the speech of French adults. The reasons for this delay continue to bedebated (for a recent review of developmental accounts, see Gruter 2006). Atleast two facts, however, are clear: (i) children make errors involving theomission of an object clitic (23), and (ii) when the clitic is produced, errors ofgender and number mismatch are found (24).

(23) (situation: Barbie brushing Smurfette)Q: Que fait Barbie a Schtroumpfette?

�What is Barbie doing to Smurfette?�R: Elle brosse.

�She is brushing.� (JUL, 3;1; Van der Velde 2003)

(24) Exp: Que fait Barbie a Schtroumpfette?�What is Barbie doing to Smurfette?�

Ch: Elle le nettoie.she cl-masc-sg clean-3sg.�She is cleaning him.� (Cle, 4;1; Jakubowicz & Nash, to appear)

13 How this predicate acquires or values this Case feature is not entirely clear. As Sportiche(1995:323 n. 12) points out, ‘‘[t]his would mean that the verb be may assign accusative.’’



Interestingly, gender and number mismatches appear to be almost entirelyunidirectional—that is, they consist of the substitution of (masculine singular)le for (feminine) la or (plural) les. Jakubowicz & Nash (to appear), forexample, report gender errors on about 10% of all (singular) accusative cliticsproduced by three- and four-year-olds (3A: 11.1%, 4A: 9.8%), and note thatthese errors typically consist of the use of the masculine instead of thefeminine clitic. Chillier Zesiger and colleagues (2003) observe similar rates ofgender-related errors for all age groups (4;0–6;0), and point out the sameasymmetry in the direction of the error. Chillier Zesiger et al.�s elicitation taskalso included plural contexts, which enabled them to investigate potentialmismatches in terms of number marking as well. They report an average of7.3% number errors on plural targets across all age groups. Again, the errorappears to be unidirectional, with almost no number errors on singular targets(< 1%). The majority of errors on plural targets consisted of the substitution ofmasculine singular le for plural les.This error pattern is entirely consistent with the analysis proposed here. The

(near) absence of more richly specified items (la, les) in contexts requiring aless specified item (le) is precisely what is expected. Thus the asymmetryobserved in the error patterns with regard to gender and number mismatcheson object clitics in children�s speech lends further support to the underspe-cification approach pursued here. Children�s utterances with a mismatchedle can now be seen as the result of the underspecification of the clitic withregard to gender and number. What remains to be addressed is the reason forthis (variable) underspecification in child language. I will return to thisquestion shortly.Within the model proposed here, the analysis of mismatch errors extends

naturally to the other observed error type, namely object (clitic) omission, as in(23). These utterances can now be seen as the result of even furtherunderspecification, leading to the selection of the least specified item, the zeromorpheme, for the clitic head. In other words, the claim is that the syntacticrepresentation underlying the child utterance in (23) is identical to that ofclitic-drop constructions observed in adult corpora (e.g., (2)–(4) above). Inparticular, in both cases, it is assumed that Cl0 is underspecified for Case,leading to the selection of the default zero morpheme.However, a closer look at both the distribution and frequency of clitic drop

in child and adult corpora suggests that the reason for the underspecificationof Cl0 may not be the same in child and adult French. The currently availableevidence suggests that clitic drop is more frequent in child French than in adultFrench (see Pirvulescu 2006 for a quantitative comparison of clitic drop inchild and child-directed adult speech, showing a substantial difference inomission rates). Moreover, clitic drop in child French often appears in contextswhere it is judged illicit by adult speakers of the language. For example, mostadult speakers of French will judge (23) illicit. It is interesting to note,however, that there is very little agreement among native speakers with regardto the acceptance of clitic drop. In a study by Pirvulescu & Roberge (2005),

234 Theres Gruter


four adult judges agreed on the grammaticality of only 16% of all instances ofclitic drop found in a child corpus. In sum, these facts suggest that clitic dropin child French is a more wide-spread phenomenon than in the adult language.The question is why.One possibility is that children can violate the Case-expression requirement

more easily than adults. More specifically, it may be the case that they assignthe feature [predictable] more liberally, as a result of an immatureunderstanding of the distinction between speaker and hearer knowledge.Schaeffer (2000) proposed that children under the age of three lack apragmatic principle termed ‘‘Concept of non-shared knowledge,’’ which statesthat ‘‘[s]peaker and hearer knowledge are always independent’’ (Schaeffer2000:90). If this principle is not (consistently) respected, the distribution of thepragmatic feature [predictable] is expected to be less constrained. Anexplanation along these lines would view children�s clitic drop as essentiallyan overgeneralization of clitic drop in the adult language, due to an immaturityof the child�s pragmatic system. However, Schaeffer (2000:100) observed thatby about age three, the Concept of non-shared knowledge appears to beoperative. By contrast, illicit clitic drop in the speech of French-speakingchildren has been shown to persist beyond the age of three (e.g., ChillierZesiger et al. 2003; Perez-Leroux, Pirvulescu & Roberge 2006; van der Velde2003). This explanation also predicts that French-speaking children shouldaccept interpretations requiring a referential null object more liberally thanadults. Recent findings from truth-value judgment experiments have notconfirmed this prediction (Gruter 2006, 2007). Finally, an explanation alongthese lines would have nothing to say about the other error type observed inchild language—namely, the overuse of masculine singular le discussedabove.I would like to suggest instead that the more frequent underspecification of

Cl0 in French child language—and the consequent choice of le or the zeromorpheme—is due to a factor that is not primarily linguistic: namely,processing capacity limitations, specifically in the domain of working memory(see also Prevost 2006 for the suggestion that clitic drop in child L2 French isdue to processing difficulties). If it is true, as has been assumed in the morerecent, phase-based minimalist literature, that working memory plays a role inthe syntactic derivation itself (see, e.g., Chomsky 2001:11, 2005:16), weexpect diminished working-memory capacity to have direct effects on thesyntactic computation. In particular, constructions involving long-distancedependencies are expected to be affected, as they involve the retention offeatures in ‘‘active memory’’ (Chomsky 2001:11) across phase levels. Notethat in the syntactic analysis of object-clitic constructions adopted here, thereis a clear dependency relation between the clitic (Cl0) and the empty category(pro) in the complement of V, and thus a requirement for the relevant featuresof pro to remain accessible until Cl0 is merged. It is also clear that these twoelements are merged at a considerable distance: after the merge of pro, themerge of Cl0 occurs only after a number of intermediate computational



operations, and after at least one phase boundary (vP). Suppose, then, thatunder circumstances where working memory capacities are constrained, forexample in young children (see, e.g., Vance, Stackhouse & Wells 2005), notall features on pro remain accessible at the point where the clitic is merged.Consequently, these features will not be copied onto Cl0, which will then enterthe morphological component underspecified. If only /-features are lost in thecourse of the derivation, Cl0 will be marked for specificity and Case, and itemc, le, will be inserted. If the Case feature does not survive until the merge ofCl0, however, item d, the zero morpheme, will be selected. Thus theunderspecification account presented here is able to provide a unifiedexplanation for the two error types most commonly observed in cliticconstructions in French child language.This proposal raises serious questions with regard to the reality of the

syntactic derivation in real-time production that I cannot pursue here. (Formore detail and discussion, see Gruter 2006). Yet it also makes clear andtestable empirical predictions. First, if the syntactic distance between pro andCl0 is relevant and susceptible to memory constraints, the prediction is thatclitic drop should be elevated in clitic climbing contexts, where a clitic relatedto the argument of a lower predicate appears in a position preceding the higherpredicate, as illustrated in (25).

(25) a. Caillou la fait rire.Caillou cl makes laugh�Caillou makes her laugh.�

b. *Caillou fait la rire.

In these constructions, the distance between the clitic and its associated emptycategory is presumably even greater than in clauses with a single predicate,including in particular an additional clause boundary. In consequence, therewill be more ground for ‘‘forgetting,’’ because the relevant features will haveto be held in working memory over a greater distance. I am not aware of anyempirical evidence from child or adult French bearing on this prediction.It thus remains for future empirical studies to investigate.14 A secondprediction of an account that holds working-memory limitations responsiblefor clitic drop is that the amount of clitic drop and an independent measure ofthe speaker�s working memory capacity should be in negative correlation. Thisprediction has recently been tested with child second-language learners ofFrench, with findings (Gruter, Erdos & Genesee 2008) indeed revealingnegative correlations between the rate of object-clitic drop in elicited

14 But Bottari, Cipriani & Chilosi (2000) present evidence from the spontaneous speech of adysphasic Italian child showing that object clitics were omitted significantly more often in cliticclimbing contexts. It is also interesting to note in this context that in experiments on bindingeffects of object clitics, French-speaking children have shown particular difficulty with the correctinterpretation of clitics in clitic climbing contexts (Hamann, Kowalski & Philip 1997; Hamann2002).

236 Theres Gruter


production and scores on at least some subtests of the Working Memory TestBattery for Children (Pickering & Gathercole 2001).In sum, the account of object-clitic drop in adult French proposed in this

paper, together with the independently established assumption that childrenpossess more limited working memory capacity, is able to capture importantfacts about French child language, namely the overuse of the masculinesingular clitic le, as well as the higher incidence of clitic drop. In particular,I have proposed that the syntactic representation underlying clitic drop inchild French is identical to that of clitic-drop constructions observed inadult corpora—that is, in both cases, Cl0 is underspecified for Case, whichleads to the selection of the default zero morpheme. Children and adultsdiffer, however, with regard to the factors that lead to this underspecifi-cation. Whereas for adults, underspecification of Cl0 is the result ofviolating the Case-expression requirement, which is legitimate under theright discourse conditions, the underspecification of Cl0 in child language isargued to be the result of independent limitations in the domain of workingmemory.

7. Conclusion

The goal of this paper was to present a syntactic analysis capable ofaccounting for both clitic constructions and their closely related counterpartswithout an overt clitic (clitic drop), as observed in several recent corpora ofadult French. To this end, I have drawn extensively on Sportiche (1996) forclitic constructions and on Cummins & Roberge (2005) for clitic drop, toarrive at a unified analysis of these two constructions. The analysis reliescrucially on the underspecification of the clitic head, which leads to theinsertion of less-specified vocabulary items in the morphological componentof the grammar in the spirit of Distributed Morphology. Data from predicate lewere found to lend independent support to this underspecification approach.Moreover, errors found in child French, in particular the overuse of themasculine singular clitic le and the higher incidence of clitic drop, were shownto be directly compatible with the present account of adult French, under theadditional assumption that children�s syntactic derivations are affected by theirmore limited working memory capacities. The syntactic analysis proposed inthis paper should thus be able to serve as a unified theoretical framework forfuture empirical work on clitic and clitic-drop constructions in both adult andchild French, data that ultimately must be accounted for by any syntacticanalysis of French.

References

Anderson, S. 1992. A-morphous morphology. Cambridge: Cambridge University Press.Auger, J. 1994. Pronominal clitics in Quebec colloquial French: A morphologicalanalysis. Doctoral dissertation, University of Pennsylvania.



Auger, J. 1995. Les clitiques pronominaux en francais parle informel: Une approchemorphologique [Pronominal clitics in informal spoken French: A morphologicalapproach]. Revue quebecoise de linguistique 24: 21–60.

Beard, R. 1995. Lexeme-morpheme base morphology: A general theory of inflectionand word formation. Albany, NY: SUNY Press.

Bennis, H. 2006. Agreement, pro, and imperatives. In Arguments and agreement, ed.P. Ackema, P. Brandt, M. Schoorlemmer & F. Weerman, 101–127. Oxford: OxfordUniversity Press.

Bonet, E. 1991. Morphology after syntax: Pronominal clitics in Romance. Doctoraldissertation, MIT.

Bonet, E. 1995. Feature structure of Romance clitics. Natural Language & LinguisticTheory 13: 607–647.

Borer, H. 1984. Parametric syntax: Case studies in Semitic and Romance languages.Dordrecht: Foris.

Bottari, P., P. Cipriani & A. M. Chilosi. 2000. Dissociations in the acquisition of cliticpronouns by dysphasic children: A case study from Italian. In The acquisition ofscrambling and cliticization, ed. S. M. Powers & C. Hamann, 237–277. Dordrecht:Kluwer.

Bruening, B. & A. Rackowski. 2001. Configurationality and object shift in Algon-quian. In Proceedings of WSCLA 5, ed. S. Gessner, S. Oh & K. Shidara, 71–84.Vancouver: UBCWPL.

Burzio, L. 1986. Italian syntax: A government-binding approach. Dordrecht:Reidel.

Chillier Zesiger, L., M. Arabatzi, L. Baranzini, S. Cronel-Ohayon, J. Franck, U.Frauenfelder, C. Hamann, L. Rizzi & P. Zesiger. 2003. The acquisition and masteryof French pronouns in normal children. Ms., University of Geneva.

Chomsky, N. 1981. Lectures on government and binding. Dordrecht: Foris.Chomsky, N. 1982. Some concepts and consequences of the theory of government andbinding. Cambridge, MA: MIT Press.

Chomsky, N. 1995. The Minimalist Program. Cambridge, MA: MIT Press.Chomsky, N. 2000. Minimalist inquiries: The framework. In Step by step: Essayson minimalist syntax in honor of Howard Lasnik, ed. R. Martin, D. Michaels &J. Uriagereka, 89–155. Cambridge, MA: MIT Press.

Chomsky, N. 2001. Derivation by phase. In Ken Hale: A life in language, ed.M. Kenstowicz, 1–52. Cambridge, MA: MIT Press.

Chomsky, N. 2005. Three factors in language design. Linguistic Inquiry 36: 1–22.Clark, E. 1985. The acquisition of Romance with special reference to French. In Thecrosslinguistic study of language acquisition, ed. D. I. Slobin, 688–782. Hillsdale,NJ: Erlbaum.

Comrie, B. 1989. Language universals and linguistic typology: Syntax and morphology.Chicago: University of Chicago Press.

Cummins, S. & Y. Roberge. 2004. Null objects in French and English. In Contem-porary approaches to Romance linguistics, ed. J. Auger, C. Clements & B. Vance,121–138. Amsterdam: John Benjamins.

Cummins, S. & Y. Roberge 2005. A modular account of null objects in French. Syntax8: 44–64.

De Hoop, H. & A. Malchukov 2007. On fluid differential case marking: A bidirectionalOT approach. Lingua 117: 1636–1656.

Fonagy, I. 1985. J�aime, je connais: Verbes transitifs a object latent [I like, I know:Transitive verbs with latent objects]. Revue Romane 20: 3–35.

Friedemann, M.-A. & T. Siloni 1997. Agr-object is not agr-participle. The LinguisticReview 14: 69–96.

238 Theres Gruter


Greenberg, J. H. 1963. Some universals of grammar with particular reference to theorder of meaningful elements. In Universals of language, ed. J. H. Greenberg, 73–113. Cambridge, MA: MIT Press.

Gruter, T. 2006. Object clitics and null objects in the acquisition of French. Doctoraldissertation, McGill University. http://www.arts.mcgill.ca/programs/linguistics/mcgwpl/theses/gruter_thesis.pdf.

Gruter, T. 2007. Investigating object drop in child French and English: A truth valuejudgment task. In Proceedings of the 2nd conference on Generative Approachesto Language Acquisition North America (GALANA), ed. A. Belikova, L. Meroni &M. Umeda, 102–113. Somerville, MA: Cascadilla Proceedings Project. http://www.lingref.com, document #1551.

Gruter, T. 2008. When learners know more than linguists: (French) direct object cliticsare not objects. Probus 20: 211–234.

Gruter, T., C. Erdos & F. Genesee 2008. Working memory and the underlying repre-sentation of missing objects in the development of French. Paper presented at the XIInternational Congress for the Study of Child Language (IASCL), Edinburgh,Scotland, August.

Halle, M. & A. Marantz. 1993. Distributed Morphology and the pieces of inflection. InThe view from Building 20: Linguistic essays in honor of Sylvain Bromberger, ed.K. Hale & S. J. Keyser, 111–176. Cambridge, MA: MIT Press.

Halle, M. & A. Marantz. 1994. Some key features of Distributed Morphology.In MITWPL 21: Papers on phonology and morphology, ed. A. Carnie & H. Harley,with T. Bures, 275–288. Cambridge, MA: MITWPL.

Hamann, C. 2002. From syntax to discourse: Pronominal clitics, null subjects, andinfinitives in child language. Dordrecht: Kluwer.

Hamann, C., O. Kowalski & W. Philip. 1997. The French ‘‘delay of principle B’’ effect.In Proceedings of the 21st BUCLD, ed. E. Hughes, M. Hughes & A. Greenhill, 205–219. Somerville, MA: Cascadilla Press.

Hamann, C., L. Rizzi & U. Frauenfelder. 1996. On the acquisition of subject and objectclitics in French. In Generative perspectives on language acquisition, ed. H. Clahsen,309–334. Amsterdam: John Benjamins.

Harley, H. & E. Ritter 2002. Person and number in pronouns: A feature-geometricanalysis. Language 78: 482–526.

Hoji, H. 1998. Null object and sloppy identity in Japanese. Linguistic Inquiry 29: 127–152.

Huang, C.-T. J. 1984. On the distribution and reference of empty pronouns. LinguisticInquiry 15: 531–574.

Jaeggli, O. 1982. Topics in Romance syntax. Dordrecht: Foris.Jaeggli, O. 1986. Three issues in the theory of clitics: Case, doubled NPs, andextraction. In The syntax of pronominal clitics (Syntax and Semantics 19), ed.H. Borer, 15–42. San Diego, CA: Academic Press.

Jakubowicz, C. & L. Nash. To appear. Why accusative clitics are avoided in normal andimpaired language development. In Essays in syntax, morphology, and phonologyin SLI, ed. C. Jakubowicz, L. Nash & K. Wexler. Cambridge, MA: MIT Press.

Kayne, R. 1975.French syntax: The transformational cycle. Cambridge,MA:MITPress.Lambrecht, K. & K. Lemoine. 1996. Vers une grammaire des complements zeroen francais parle [Toward a grammar of null complements in spoken French].In Travaux linguistiques du CERLICO 9: Absence de marques et representationde l�absence, ed. J. Chuquet & M. Fryd, 279–306. Rennes, France: PressesUniversitaires de Rennes.

Lambrecht, K. & K. Lemoine. 2005. Definite null objects in (spoken) French: Aconstruction-grammar account. In Grammatical constructions: Back to the roots, ed.M. Fried & H. C. Boas, 13–55. Amsterdam: John Benjamins.



Larjavaara, M. 2000. Presence ou absence de l�objet: Limites du possible en francaiscontemporain [Presence or absence of the object: Limits of the possible incontemporary French]. Helsinki: Academia Scientiarum Fennica. http://ethesis.helsinki.fi/julkaisut/hum/romaa/vk/larjavaara/presence.pdf.

Muller, N., B. Crysmann & G. Kaiser 1996. Interactions between the acquisition ofFrench object drop and the development of the C-system. Language Acquisition 5:35–63.

Noailly, M. 1997. Les mysteres de la transitivite invisible [The mysteries of invisibletransitivity]. Langages 127: 96–109.

Noyer, R. 1992. Features, positions, and affixes in autonomous morphological struc-ture. Doctoral dissertation, MIT.

Noyer, R. 1997. Features, positions, and affixes in autonomous morphological struc-ture. New York: Garland.

Perez-Leroux, A. T., M. Pirvulescu & Y. Roberge. 2006. Early object omission in childFrench and English. In New perspectives on Romance linguistics, ed. J.-P. Montreuil& C. Nishida, 213–228. Amsterdam: John Benjamins.

Pickering, S. & S. Gathercole 2001. Working Memory Test Battery for Children.London: Psychological Corporation Europe.

Pirvulescu, M. 2006. Theoretical implications of object clitic omission in earlyFrench: Spontaneous vs. elicited production. Catalan Journal of Linguistics 5:221–236.

Pirvulescu, M. & Y. Roberge. 2005. Licit and illicit null objects in L1 French. InTheoretical and experimental approaches to Romance linguistics: Selected papersfrom the 34th Linguistic Symposium on Romance Languages (LSRL), ed. R. S. Gess& E. J. Rubin, 197–212. Amsterdam: John Benjamins.

Prevost, P. 2006. The phenomenon of object omission in child L2 French. Bilingualism:Language and Cognition 9: 263–280.

Raposo, E. 1986. On the null object in European Portuguese. In Studies in Romancelinguistics, ed. O. Jaeggli & C. Silva-Corvalan, 373–390. Dordrecht: Foris.

Rivas, A. 1977. A theory of clitics. Doctoral dissertation, MIT.Rizzi, L. 1986. Null objects in Italian and the theory of pro. Linguistic Inquiry 17: 501–557.

Roberge, Y. 1990. The syntactic recoverability of null arguments. Montreal andKingston, ON: McGill–Queen�s University Press.

Schaeffer, J. 2000. The acquisition of direct object scrambling and clitic placement:Syntax and pragmatics. Amsterdam: John Benjamins.

Sportiche, D. 1995. Predicate clitics and clause structure. In Small clauses (Syntax andSemantics 28), ed. A. Cardinaletti & M. T. Guasti, 287–324. San Diego, CA:Academic Press.

Sportiche, D. 1996. Clitic constructions. In Phrase structure and the lexicon, ed.J. Rooryck & L. Zaring, 213–276. Dordrecht: Kluwer.

Sportiche, D. 1998. Partitions and atoms of clause structure. London: Routledge.Sportiche, D. 1999. Pronominal clitic dependencies. In Clitics in the languages ofEurope, ed. H. van Riemsdijk, 679–708. Berlin: Mouton de Gruyter.

Strozer, J. 1976. Clitics in Spanish. Doctoral dissertation, UCLA.Suner, M. 1988. The role of agreement in clitic-doubled constructions. Natural Lan-guage & Linguistic Theory 6: 391–434.

Tuller, L. 2000. Aspects de la morphosyntaxe du francais des sourds [Aspects of themorphosyntax of the French of deaf persons]. Recherches Linguistiques deVincennes 29: 143–156.

Uriagereka, J. 1995. Aspects of the syntax of clitic placement in western Romance.Linguistic Inquiry 26: 79–123.

240 Theres Gruter


Vance, M., J. Stackhouse & B. Wells 2005. Speech-production skills in children aged3–7 years. International Journal of Language and Communication Disorders 40:29–48.

Van der Velde, M. 2003. Determinants et pronoms en neerlandais et en francais:Syntaxe et acquisition [Determiners and pronouns in Dutch and in French: Syntaxand acquisition]. Doctoral dissertation, Universite Paris 8. http://umr7023.free.fr/Downloads/MvdVeldeTh_somm.html.

Wexler, K. 2002. The theory of clitics, a parameter of object agreement, and theextension of the unique checking constraint to massive cross-linguistic differences inobject clitic omission: Evidence from normal development and SLI. Handout ofpaper presented at XXVIII Incontro di Grammatica Generativa, Lecce, Italy, March.

Whaley, L. 1997. Introduction to typology: The unity and diversity of language.Thousand Oaks, CA: Sage.

Theres GruterStanford University

Department of PsychologyCenter for Infant Studies

Jordan Hall (Building 420)Stanford, CA 94307-2130

USA

[email protected]



A unified account of object clitics and referential null objects in French

Documents