Clitic Climbing from Finite Clauses and Long Head Movement

Clitic Climbing from Finite Clauses and Long Head Movement Arhonto Terzi University of Ottawa A number of accounts of clitic climbing have regarded it a phenomenon wrrelating exclusively with infinitives. This paper shows that the above correlation should not be taken to argue that restructuring configurations must be monoclausal. By presenting instances of clitic climbing from finite subordinates, it also demonstrates that proposals which associate clitic climbing with lack of Tense (or Tense raising) are not satisfactory either. I claim that while clitic climbing is an instance of head movement, thns subject to the HMCIECP, it is also constrained by the requirement that the specifier positions of the heads tbrough which the clitic moves be coindexed. This requirement I attribute to the fact that clitic climbing is followed by an XP movement at LF, which proceeds via the respective specifier positions. The possibility of such coindexing in finite contexts makes clitic climbing available in Romance languages with Balkan clausal structwe, but is nonexistent in the rest of the Romance, where control and raising is restricted to infinitivais. This paper explores a number of issues related to the phenomenon of clitic climbing (henceforth CC), illustrated below with a typical example from Spanish:' I will not be concemed here with the initial step of cliticization, namely, the relation of the clitic in (i) with the XP argument it is associated with in (ii). (i) a. Maria 10 ley6. María 3sgAcc rad-PAST-3sg b. Maria ley6 el libro. María read-PAST3sg the book Catalan Working Papers in Linguistics (CatWPL) 3.2 (1994): 97-122 Universitat Autbnoma de Barcelona (1) a. Quiero leerlo. wantlsg read-3sgAcc b. Lo quiero leer. 3sgAcc wantlsg read 'I want to read it.' In the traditional literature CC has been associated with infinitival complement clauses. Here, I focus on this correlation and investigate the factors that underly it. By introducing data from languages that employ finite clauses in standard restructuring configurations, I show that CC does not depend crucially on the presence of infinitival subordinates. Rather, it is the coreference of matrix and embedded subjects (manifested in subject control or raising contexts) that is of critica1 importance. Consequently, the cooccurrence of CC and infinitives is only a result of the fact that raising and control configurations are associated exclusively with infinitives in the majority of the well-known languages with CC. The data discussed in this paper provides new evicience against which various approaches to CC are evaluated. From those, I will consider two types distinguished mainly along the following lines: first, analyses which claim that restructuring configurations involve a VP subordinate clause and therefore make direct predictions for the issue I arn addressing and, second, those which argue for a more elaborate clausal structure of restructuring complements but do not provide a clear answer to the question of why CC out of finite clauses is not attested.2 Thus, I will avoid debates over whether 10 is a base-generated element or has moved from the position occupied by el libro 'the book'. I consider an approach dong the iines oí' Sportiche (1992) to provide a complete amunt of the above issues and aiso to offer the possibility of disassociating CC from the process that takes place within the VP and gives rise to (i). 2 The tem restructuring is used here as a mnemonical device rather than as a commitment to Rizzi's (1982) approach to CC. I provide evidence against the first type and follow anaiyses that consider restructuring complements to be full IPS (or perhaps CPs). Subsequently, I argue that the reason why CC is not commonly obsewed in the presence of finite subordinates is because in the majority of relevant languages finite clauses do not involve obligatorily coreferent subjects (namely, they are not control or raising contexts). I propose that in order for CC to be possible, the specifier positions of the phrases crossed by the moved clitic have to be coindexed and that coindexing cannot hold if some specifier hosts a contentful operator (such as a wh-operator in [Spec, CP]). Likewise, coindexing of specifiers cannot hold when matrix and embedded subjects are disjoint in reference (as in Romance subjunctives) but is a possibility in the Balkan type of subjunctives which can be control configurations. 1. Clitic Climbing and the Structure of the Embedded Clause Clitic climbing has attracted a considerable amount of research mainly within the Romance literature (Luján (1980), Rizzi (1982), Burzio (1986), Kayne (1989, 1991), Rivero (1992), Roberts (1992, 1993) to mention some recent studies). While these accounts investigate to different degrees the syntactic structure of the complement clause, none has explicitly addressed the significance of the fact that all attested cases of CC are associated with infinitivais. In other words, is CC out of finite subordinates to be expected and if not, what exactly precludes this possibility? Nevertheless, a number of approaches to the clausal structure of restructuring configurations make clear prodictions with respect to the above question: restructuring configurations are monoclausal thus CC out of finite clauses is excluded as the latter involve more syntactic structure than a bare VP (Zagona (1982), Picallo (1990), Moore (1992)). Picailo (1990), for instance, argues that restructuring complexes are monosentential in both D-structure and Sstructure. Had poder ('may') in (2) taken an IP complement, CC would be rendered impossible according to her analysis, since VP would constitute a barrier for movement of the clitic from its lower position. (2) a. La Joana pot [vp enviar-li cartes]] the Joana may3sg send-3sgDat letters b. La Joana [IIln lii pot [w enviar-e. cartes]] the Joana 3sgDat may3sg send letters 'Joana may send letters to himlher.' (Catalan) In the same vein, Moore (1992), adopting the distinction between T and T' heads proposed by Li (1990), argues that Infl (as a T'-head) prevents the clitic (also a T'-head) from headgoverning its lower position. Therefore, as in Picallo (1990), it is crucial for him that complements of restructuring verbs be VPs, as indicated in (3b), or else CC is rendered impossible. (3) a. Curro quiere conocerte. Curro wants know-2sgAcc b. Curro te, quiere [vp conocer ei]. Curro 2sgAcc wants know 'Curro wants to know you.' On the basis of independent considerations, however, other studies of the syntactic structure of infinitival clauses indicate that it is rather implausible that they are bare VPs. Raposo (1987) and Kayne (1991), for instance, propose that infinitives are associated with the inflectional head Infn, which hosts the infinitival suffix. Kayne (1991) argues, more specifically, for the existence of another functional head immediately above Infn. This is a nonlexical head dominating To and its existence is independently supported by the presence of clitics which attach to it. Furthermore, according to Chomsky and Lasnik (1991), To is present in infinitival clauses and allows PRO to check nul1 Case in its specifier posj.tion.3 See Martin (1993) for evidence supporting the presence of a functionai projection dominating Tense in English control infinitivais. Finally, AgrO, according to some proposals, is also indispensable for accusative Case checking.4 Therefore, by concentrating on standard Romance languages, accounts which view restructuring configurations as monoclausal because of the lack of overt morphological inflection on the infinitival verb are only superficially justified. Nevertheless, if we take clitics to undergo head movement from the relevant argument position, as in (2) and (3), such accounts are the only ones to offer a direct explanation for the absence of CC from finite subordinates. By presenting evidence hs to why infinitives involve a more elaboraie clausal structure than a bare VP we are leaving the issue open again. Approaches such as Kayne's (1989) differ in various respects from the approaches just described. Kayne considers CC to be an instance of head movement, subject to antecedent government and attributes the ungrammaticality of sentences like (4b) to the blocking effect of the lexical complementizer. (4) a. Non so se farli. not knowlsg if do3plAcc 'I don't know whether to do them.' b. *Nonlii so se fare ei. not 3plAcc know-lsg if do His account touches on the issue of CC from finite clauses less directly than the previous approaches, yet, it does not seem to exclude it. As already mentioned, Kayne (1991) argues that Romance infinitivais involve the functional head Infn which hosts the infinitival suffix and the nonlexical To to which clitics attach. It is, therefore, expected that infinitival Infl should not be significantly different from finite Infl in its interaction with CC. Furthermore, if Kayne To these arguments we should perhaps add the Extended Projection Principle which according to Chomsky (1992) reduces to morphological properties of Tense. (1991) is right in considering Romance clitics to left-adjoin to the embedded Agr, the embedded Infl (or, more precisely, the different functional heads which Infl is currently taken to consist of) should not interfere with movement of clitics, since clitics do not move past it. Nevertheless, CC over finite clauses is not possible as the following example illustrates (from (Kayne 1989)). (5) *Gianni li vuole che (Maria) veda. Gianni 3plAcc wants that Maria see-SUBJSsg Kayne attributes the status of (5) to an ECP violation induced by che ('that'). But notice that his explanation cannot extend to comparable sentences in Spanish which are equally ungrammatical. (6) *Juan 10s quiere que (Maria) vea. Juan 3plAcc wants that Maria see-SUBJ3sg This is because CC over que is possible in Spanish infinitives as (7) shows. On the other hand, attributing the well-formedness of (7) to the fact that que occupies [Spec, CP] (as in Kayne (1991)) would leave (6) unexplained. (7) Juan 10s tiene que ver. Juan 3plAcc has that see 'Juan has to see them.' Concluding, it appears that Kayne's account does not have a clear way to account for contrasts such as (6) and (7), except by resorting to the nonfiniteness of the latter structure, and thus utilizing an option that does not follow naturally from the rest of his analysis. This is perhaps why other accounts which, just like Kayne's, do not have an internal mechanism able to bar CC from finite clauses, have to stipulate at some point that only infinitival Agr (or Tense) is in some way associated to the presence of CC. Roberts (1992), for instance, relates CC to the ability of infinitival Agr to properly govern the clitic trace in CO. In the following sections I will show that such stipulations are not necessary, as CC from finite subordinates is indeed possible. More precisely, I will offer evidence that the occurrence of an IP node does not act as a minimality blocker for CC, thus providing independent support for those analyses which do not view restructuring complements as bare VPs. Evidently, the area to turn to are languages which have clitics but in which potential CC contexts involve finite subordinate clauses. 2. Balkan Subjunctive Subordinates and Clitic Climbing Such are the languages of the Balkans. Structures like (8), often referred to as subjunctives in the Balkan literature, are control configurations in Greek and Romanian, as argued in Terzi (1991, 1992). Similar claims for Greek are found in Iatridou (1988), Felix (1989) and Hornstein and Varlokosta ( 1992). (8) a. Maria insearca PRO sZ scrie. b. I Maria prospathise PRO na grapsi. the Maria tried PRT writes 'Maria tried to write.' (Romanian) (Greek) The clausal structure of Balkan subjunctives appears in (9). Rivero (1988), Tsimpli (1990), and Terzi (1992) argue that the particle that precedes the finite verb in (8) or (9) is not a complementizer, but the inflectional head MO that heads the maximal projection Mood Phrase (MP). The subordinate clause is headed by a nonlexical CO and, according to Temi (1991, 1992) the specifier position of Md Phrase is occupied by PRO. Object clitics are available in the Balkan languages. Furthermore, Greek and Romanian clitics are not second-position clitics as in the northern BalkanlSlavic languages, but rather, they precede the finite verb, as shown in (10), as is standardly the case among Romance finite structures. (10) a. Maria insearca i o scrie. b. I Maria prospathise na to grapsi. the Maria tried PRT 3sgAcc writes 'Maria tried to write it.' (Romanian) (Greek) Furthermore, CC is attested in Romanian, in some of the rare instances where true infinitives are still enmuntered. (11) 11 pot minca. 3sgAcc can-lsg eat 'I can eat it.' Nevertheless, clitic climbing out of the complement clause ln (10) is impossible, as (12) shows, despite the absence of a lexical CO in the embedded clause. (12) a.*Maria o insearca s; scrie. b. *I Maria to prospathise na grapsi. the Maria 3sgAcc tried PRT writes (Romanian) (Greek) The ungrammaticality of (12) is well-accounted for by the two types of approaches we have reviewed. For those accounts that propose that restructuring verbs take a VP complement the mere presence of IP rules out CC in the Balkan languages. For accounts such as Kayne's and Roberts' the presence of a finite IP is rather irrelevant, the minimality blocker is presumably MO. To conclude this section, I have to admit that the data from Greek and Romanian are not particularly illuminating with respect to the possibility of CC over finite subordinates. This is because in the Balkan languages, potential restructuring wntexts are always associated with the so-called subjunctive particle which itself blocks CC (in ways we will elaborate in following sections) with the result that the role of the finite Infl per se in the process of CC cannot be isolated. 3. Salentino Subjunctive Subordinates and CIitic Climbing The crucial evidence for the issue we are addressing comes from Salentino, a Southem Italian dialect which also lacks infinitives to a large extend. Calabrese (1991) points out that complements of control verbs in this dialect share the clausal structure of their Balkan counterparts, that is, they fall under the representation in (9). As illustrated in (13), sentential complements of subject control and object control verbs are introduced by a particle comparable to that encountered in the Balkan languages, with the embedded verb to the right of the particle being inflected for subject agreement and tense. (13) a. Karlu voli ku vveni krai. Karlu wants PRT comes tomorrow 'Karlu wants (himlher) to come tomorrow.' b. Karlu e cunvintu Maryuku vveni. Karlu has persuaded Maryu PRT comes 'Karlu persuaded Maryu to come.' the specific details of his analysis. There it is proposed that it is the embedded I, rather than the clitic itself, which moves to the matrix I in restructuring contexts and the result of this movement is that matrix and embedded Agr end up being coindexed. This idea is consistent with subsequent proposals (Kayne (1991)) according to which Romance infinitives involve a nonlexical inflectional head (the highermost functional head of the subordinate clause) with which V is not obliged to merge and to which clitics attach.10 (23) V... Cl+T ... [ldn V+Infn [vp [v e]]] It is presumably by virtue of this functional head being empty that the correct surface order clitic - matrix verb - infinitive is obtained after the embedded I moves to matrix I in the standard Romance languages. The above process does not carry over to languagea like Salentino, however, since the highermost functional head (for concreteness, let us iake it to be Agr) is lexical, and the verb has to raise to it. Therefore, movement of embedded Agr to matrix Agr -of the type that Kayne proposes to capture clitic climbingwould carry dong the verb in finite configumtions, thus, give rise to the unacceptable surface order clitic-embedded verb-restructuring verb. In the following, I will sketch out an analysis which captures the coreference of matrix and embedded subject manifested by restructuring configurations but avoids the technical problems that Kayne's account faces when it comes to finite restructuring complements. I would like to attribute the coreference of matrix and embedded subjects, which correlates with all instances of CC, to a condition that requires the specifier positions of the functional l0 We will overlook here the discrepancy between Kayne (1989) and Kayne (1991) with respect to the nature of the higher functional head (i.e., Agre or To) since it is of minor irnportance for our central concems. projections which clitic movement crosses to be coindexed. Considering clitics to move via CO, the relevant specifiers are [Spec, AgrS] of the embedded clause, [Spec, CP] of the embedded clause and [Spec, AgrS] of the matrix clause. More specifically, I propose that in order for CC to be possible, the following two conditions must be met:" a) non L-related heads may not intervene in the chain formed by CC (a fact that is captured in one way or another by all accounts which consider CC to involve head movement and to be subject to the ECP, and which was discussed in the previous section), and b) coindexing of the specifier positions in the path of CC must be established (or, more precisel y, no contraindexing may be possible). Both conditions are met in a typical case of CC such as (I), repeated below. The specifier positions of the heads through which the clitic is found are coindexed. Moreover, CO is empty and the clitic can move through it. (25) Lo quiero leer. 3sgAcc want-lsg read 'I want to read it.' Considering que ('that') to occupy [Spec, CP], as in Kayne (1991), we see why CC is also possible in instances as the following: (26) Lo tengo que hacer. 3sgAcc have-lsg that do 'I have to do it.' l1 In this respect my analysis is partially reminiscent of Unagereka's (1988) in that CC is taken to be accounted for not on the basis of the ECP alone but also by means of some other independent condition. However, while for Unagereka this condition has to do with the event matrix of the relevant structure I consider it to be related to the obligatorily coreferent subjects involved in restructuring coníigurations. The clitic is able to move via the empty CO position and coindexing of specifiers presumably holds although [Spec, CP] is occupied by que. Que does not seem to prohibit coindexing of specifiers because it has no semantic content. Now contrast (26) with the ungrammatical (27), where just as in (26), [Spec, CP] is occupied by que and CO is empty. (27) *?No te SC qu6 decir. not 2sgDat know-lsg what say Obviously there is no clear reason why (27) is ill-formed if qué('what') is in [Spec, CP] and clitics move via head movement, as proposed in Kayne (1989). The contrast of (26) vs. (27) follows from the second of the conditions I propose, namely, from the fact that coindexing of specifiers cannot take place. I take this to be related to the fact that [Spec, CP] in (27) does not host a complementizer devoid of semantic content, as in (26), but a wh-operator. Although ungrammatical, (27) is slightly better than the sentence that follows. (28) *No 10 SC si hacer. not 3sgAcc know-lsg if do The sharp ungrammaticality of (28) is due to the fact that neither of the conditions I proposed is met. Clitics cannot move via CO, as it is occupied by si ('if') and no specifier coindexing can take place because [Spec, CP], is occupied by a nonlexical wh-operator.12 1.2 The slightiy different status of the following two ungrammatical sentences from Italian pattems that of (27) (28) from Spanish and can be analyzed dong the same lines. (i) a. sem on ti saprei che dire not 2sgDat would-know-lsg that say b. *Non li so se fare. not 3plAcc know-lsg if do To sum up, I argued that irrespectively of the structure of the embedded clause (finite or nonfinite) matrix and embedded subjects of restructuring configurations must be coindexed in order for CC to take place. I proposed, although somewhat more tentatively, that this amounts to a condition on the coindexing of the specifier positions of the projections through which CC proceeds and appears not to obtain if a specifier is occupied by a contentful operator.13 An explanation for the reasons behind Spec-coindexing may be available if one views CC as an instance of Xo movement followed by XP movement at LF. This idea is actually similar to Notice that simply considering CC an instance of Xo movement is notable to explain the status of (ia). On the other hand, (ib) remains unexplained if the XP movement approach to CC is adopted. Thus, in light of wntrasts such as the one above, it appears that a condition on coindexing of the specifier positions present of restructuring configurations is necessary, irrespectively of the type of movement one adopts for CC. l3 An apparent wunterexample for the Spec-to-Spec coindexing proposal is the following instance of CC from earlier and literary French (from Kayne (1989)). where windexing of the impersonal matrix subject and PRO is presumably not possible, yet CC may take place. (i) Il le faut faire. It 2sgAcc is-necessary do In view of structures such as (ii) it may be suggested that the dative subject has moved to the position of the impersonal at LF in (i) and rendered Spec-to-Spec coindexing possible (which suggests that the proposed Speo to-Spec windexing actually takes place at LF). (ii) I1 me faut ces livres. It lsgDat is-necessary these books Notice, that the same process seems not to be able to save the Spanish example in (iii). This can be attributed to the fact that hacer falta 'to be necessary' (contrary to falloir) agrees with a postverbal subject (iv), and thus the matrix subject has to be windexed with it instead. (iii) *Me 10 hace falta leer. lsgDat 3sgAcc is-necessary read (iv) Me hacen falta dos libros. 1sgDat are-necessary two books Sportiche's proposal concerning the initial step of cliticization. Recall that Sportiche proposes that in (b) of the run-off-the mill pairs as the following: a) Quiero comprar un libro ('I want to buy a book') b) Lo quiero comprar ('I want to buy it'), a covert NP has moved to the specifier position of H (the functional head that dominates the clitic) in satisfaction of his Clitic Criterion. CC (a less important issue for his proposals) is then considered an instance of XP movement, without explaining why the Clitic Criterion does not have to hold any more. By extending his initial proposa1 to CC we offer an answer to this question capturing at the same time the requirement on coindexed specifiers for CC. Furthermore, see fn. 13 for evidence as to why the XP movement that amounts to Spec-coindexing takes place at LF. 6. Romance vs. Balkan Subjunctives and Clitic Climbing. According to my analysis, the contrast in (6) and (7) from Spanish, repeated below as (29) and (30), which is left essentially unexplained by a head movement approach to CC, is now easier to understand. (29) Juan 10s tiene que ver. (30) *Juan 10s quiere que (Maria) vea. Assuming que ('that') to occupy [Spec, CP], the clitic can presumably move via CO in both cases. Furthermore, que is not the wh-operator that we encountered in (26) and therefore coindexing of [Spec, CP] with [Spec, Agr] should be possible. Finally, (1) has demonstrated that querer ('want') is a restructuring verb. Nevertheless, (30) is sharply ungrammatical. In light of the Salentino facts, it is not clear whether the contrast of (29) vs. (30) should be attributed to the Tensed status of the latter structure (as in Rivero (1992) where the [+Tensel CO of finite clauses was held responsible for the impossibility of CC). A comparable type of CO would presumably have to be postulated for the Salentino example in (19), yet it does not exclude CC (19b). Moreover, Salentino subjunctive subordinates appear to demonstrate fewer Tense dependencies than their standard Romance counterparts -i.e., they are not subject to the usual Tense dependencies of Romance subjunctivesand thus justify the postulation of a [+Tensel CO. (31) Vulia ku vveni. want-PAST-lsg PRT come-PRES3sg (32) a. *Quise que venga ahora. want-PASTlsg that come-PRES-SUBJ3sg now b. Quise que viniera ahora. want-PAST-lsg that come-PAST-SUBJ-3sg now (Salentino) (Spanish) Furthermore, it is undesirable to associate CC with the Tense properties of the infinitival subordinate clauses that demonstrate the phenomenon (see, Roberts (1993), for instance, and his idea of Tense raising as related to CC). In recent work on English infinitivals, Martin (1993) argues for the different Tense properties of control vs. raising infinitives, that is, while the former are associated with [+Tensel (correlating with the requirement for PRO to check its nul1 Case) the latter are not. Providing that Martin is on the right track and that his claims can be transposed to Romance, it is not clear how precisely CC depends on Tense, since it is a phenomenon manifested in the context of both types of infinitival complements. According to the analysis I have suggested, the status of (30) follows from the fact that coindexing of specifiers cannot take place. This time, however, it is not [Spec, CP] that prevents coindexing but rather the fact that the nul1 subjects occupying the two [Spec, AgrS] are disjoint in reference (a typical case of subject obviation in Romance languages, as in Picallo (1985)). In other words, the ungrammaticality of (31) follows from the same reasons that exclude CC from an object control configuration.14 l4 A reviewer asks what the difference between a restructuring verb like 'want' is, and verbs that do not allow CC. Here, I would have to stipulate that only restructuring verbs are able to govern the clitic trace in the intermediate C position. Additional support for the above proposals is provided by (33), one of those rare instances where CO can be empty in Spanish subjunctives. (33) *Lo espero comas. 3sgAcc hope-lsg eat-SUBJ-2sg It is not clear how the ill-formedness of (33) can be explained by evoking an ECP violation alone.15 The status of (33) clearly follows from my analysis, since matrix and embedded subjects cannot be coindexed. On the other hand, in a comparable structure in Salentino, shown in (19b) and repeated in (34). the subject of the so-called subjunctive complement clause is coreferent with the matrix subject when the particle ku is absent (or else, the embedded subject of (34) is PRO) and thus coindexing of specifiers is possible. (34) Lu voggyu kkattu. 3sgAcc want-lsg buy-lsg 'I want to buy it. Put differently, while Salentino subjunctives rnay be involved in control configurations (35), this is never the case in standard Romance, where control and raising (and, subsequently, CC) is confined to infinitives (36).16 (35) Maryui voli Iu proil*, kkatta. Maryu wants 3sgAcc buybsg (36) a. Manoi quiere que 10 pro,/*, compre. Mario wants that 3sgAcc buy-SUBJ3sg 15 We will follow Kayne (1989) in attributing the status of hidden causatives to those instances of object control configurations where CC seems to be possible in Spanish. l6 Unless one adopts Rizzi's (1982) proposals that complements of subjunctives in Italian (and presumably in Spanish) involve I-to-C movement, and consider CC an instance of head movement. (36) b. Marioi quiere PROipj comprarlo. Mario wants buy-3sgAcc 7. Conclusions This paper has investigated the role of the Infl components in restructuring configurations. I offered empirical and theoretical support for the claim that cooccurrence of CC and infinitival subordinates cannot be attributed to the VP status of restructuring complements. I relied on approaches that consider CC an instance of head movement and elaborated on the clausal structure of restructuring configurations and the interaction of the different types of functional heads with CC. Subsequently, I showed that what is crucial for CC and holds across both finite and infinitival restructuring contexts is that the subject positions of matrix and embedded clause be coreferent and I proposed to capture this fact by a condition on coindexing of the specifier positions crossed in the process of CC. I suggested that coindexing follows from the fact that although CC is an instance of head movement, it is followed by an XP movement at LF, that proceeds via the specifier positions. I have shown that it is control (more precisely coreference of matrix and embedded subjects as manifested in subject control, and also in raising contexts) which is crucial for CC, while the type of clausal structure (infinitival or not) may differ from language to language. The Balkan type of subjunctive clauses can be control configurations, and thus, providing no intervening functional heads are present, are able to manifest CC. On the other hand, the possibility of coreferent subjects is never available in standard Romance subjunctives, and consequently CC is nota possibility in their presence.17 l7 This should not be taken to associate CC with those restructuring verbs that have the option of selecting for a nonfinite clause (as in Quicoli (1976)). Rizzi (1982) has offered evidence from Italian which demonstrates that While writing this paper I was supported by SSHRCC Grant 410-91-0178. I would like to thank Maria Teresa Guasti, Richard Kayne, Maria Luisa Rivero, and Ken Safir for helpful suggestions. My deepest thanks to Andrea Calabrese who discussed Salentino with me extensively and provided most of the data presented here. References Baker, C. M. (1989) 'Object Sharing and Projections in Serial Constructions', Linguistic Znquiry 20, 513-553. Burzio, L. (1986) Ztalian Syntax, Reidel, Dordrecht. Calabrese, A. (1991) The Sentential Complementation in Salentino: a Study of a Language without Infinitival Clauses', to appear in A. Belletti (ed) Syntactic Theory and Dialects of Ztalian, Rosenberg, Turin. Chomsky, N. and H. Lasnik. (1991) 'Principles and Parameters Theory', to appear in J. Jacobs, A. von Stechow, W. Sternefeld and T. Vennemann (eds) Syntax: An Znternational Handbook of Contemporary Research, Walter de Gmyter, Berlin. Chomsky, N. (1992) 'A Minimalist Program for Linguistic Theory', rns. MIT. Felix, S. (1989) 'Finite Infinitives in Modern Greek', in C. Bhatt, E. Lobel, and C. Schmidt (4s) Syntactic Phrase Structure Phenomena, 1 13-13 1. Varlokosta S. and N. Homstein. (1992) 'Control in Modem Greek', Proceedings of NELS 23. Iatridou, S. (1988) 'On Nominative Case Assignement and a Few Related Things', ms. MIT (to appear in MZWPL). Kayne, R. (1989) 'Null Subjects and Clitic Climbing' in O. Jaeggli and K. Safir (eds) The Null Szlbject Parameter, Kluwer Academic Publishers, 239-262. Kayne, R. (1991) 'Romance Clitics, Verb Movement, and PRO', Linguistic Znquiry 22,647686. this is not the case. From our point of view, the requirement on CC is coindexation of the relevant specifier positions, while the complementarity issue of (36) should be treated separately. Kayne, R. (1993) 'The Antisymmetry of Syntax', ms. CUNY (to appear in Linguistic Inquiry). Li, Y. (1990) 'Xo Binding and Verb Incorporation', Linguistic Inquiry 21,399-426. Luján, M. (1980) 'Clitic Promotion and Mood in Spanish Verbal Complements', Linguistics Martin, R. (1993) 'On the Distribution and Case Features of PRO', paper presented at the LSA Winter Meeting. Moore, J. (1992) 'Romance Cliticization and Relativized Minimality', paper presented at the LSA Winter Meeting. I Ouhalla, J. (1988) The Syntm of Head Movement: A Study of Berber. PhD. University College, London. Picallo, C. (1985) Opaque Domains, PhD. CUNY Graduate Center. I Picailo, C. (1990) 'Modal Verbs in Catalant, Natural Language and Linguistic Theory 8, 281 I 3 12. Quicoli, C. (1976) 'Conditions on Clitic-Movement in Portuguese', Linguistic Anulysis 2, 199223. Raposo, E. (1987) 'Romance Infinitival Clauses and Case Theory' in C. Neidle and R. NuñezCedeño (eds) Studies in Romance Languages, 237-249, Foris, Dordrecht. Rivero, M. L. (1988) The Structure of the Clause and V-movement in the Languages of the Balkans', ms. University of Ottawa (revised version to appear in Natural Language and Linguistic Theory) Rivero, M. L. (1992) 'Clitic and NP climbing in Old Spanish' in H. Campos and F. MartínezGil (eds) Current Studies in Spanish Linguistics, Georgetown University Press, Washington, DC, 241-282. Rizzi, L. (1982) Zssues in Italian Syntm, Foris, Dordrecht. Roberts, 1. (1992) Two Types of Head Movement in Romance', ms. University of Waies. Roberts, I. (1993) 'Clitic Dependencies and A-dependencies', GLOW Newsletter. Sportiche, D. (1992) 'Ciitic Constructions', ms. UCLA. Terzi, A. ( 1991) 'Governed PRO and Finiteness', Proceedings of ESCOL 8.