Clitic Climbing and Null Subject Languages

Abstract The aim of this paper is to derive Clitic Climbing from restructuring together with the Null Subject property. Data are drawn mainly from Catalan. I propose a biclausal analysis for restructuring constructions in which clitic climbing (like any clitic-Argument dependencies) is analysed as an A-dependency, assuming that clitics are AGR-heads which check features in the Agreement-Case system. Clitic climbing would be then a «long distance» A-dependency, which is made possible by assuming that restructuring is always raising, and that raising in NSLs is a transparent structure for A-dependencies, due to the nature of Nominative Case checking in these languages. To this end, a reformulation of locality conditions for checking (Agree, Chomsky 1998) is needed, which has some independent plausibility. Key words: verbal agreement, object clitics, Case, subject raising, clitic climbing. 0. Introduction Clitic climbing is a intricate phenomenon in the syntax of Romance Languages: it is clearly linked to «restructuring» (whose exact nature has always been a matter of debate1) but it also appears to be linked to the Null Subject status of the language. It is furthermore subject to a number of wellformedness conditions in interaction with other side effects of restructuring (auxiliary assignment, participle agreement, SE-passive). This paper is an attempt at reassessing the overall picture of clitic climbing and restructuring in the light of both empirical and theoretical consider- *. The present work has been supported by grant BFF2000-0403-C02-01 from the Spanish Government and by grant 2001/SGR/00150 from the Generalitat de Catalunya. I have also benefited from comments at the Grup de Gramàtica Teòrica’s seminar. I want to specially thank the invaluable comments by the volume editor Gemma Rigau and two anonymous reviewers, who had to face many obscure and imprecise points throughout the first draft. 1. For different views on restructuring and clitic climbing, see Rizzi (1982), Zubizarreta (1982), Aissen & Perlmutter (1983); Manzini (1983), Hernanz & Rigau (1984), Burzio (1986), Guéron & Hoekstra (1988), Kayne (1989), Martineau (1989), Llinàs Grau (1990-91), Luján (1993), Treviño (1993); Bok-Bennema & Kampers-Mahne (1994), Roberts (1997), Gonçalves (1997a, b), Cinque (2000). For GPHG or HPHG accounts, see Monachesi (1998, 1999). Catalan Journal of Linguistics 1, 2002 225-255 Clitic Climbing and Null Subject Languages* Jaume Solà Universitat Autònoma de Barcelona. Departament de Filologia Catalana 08193 Bellaterra (Barcelona). Spain [email protected] Cat.Jour.Ling. 1 001-259 26/2/03 15:54 Página 225 ations. On the empirical side, most of the data are drawn from Catalan, which is close enough to Spanish and Italian in most respects. The connection between null subjects and clitic climbing has been challenged. Cinque (2000) adduces 17th century French and Kru languages as cases of clitic climbing without null subjects. The present proposal relies on certain assumptions on the nature of I(nflection) in NSLs, together with a bare IP analysis for the restructured infinitival clause. If indeed clitic climbing is possible in non null-subject languages, the present analysis may still be valid provided these languages feature a more reduced type of infinitival with restructuring. For the moment, let us proceed by assuming that clitic climbing is a property tied to null subject in modern Romance, postponing the discussion of possible counterevidence to section 4.1. In order to provide an account for Romance clitic climbing, one has to address the following issues: 1) Which verbs are restructuring? In section 1we will adduce some clitic climbing verbs from Catalan that are not standardly considered in restructuring accounts. 2) What does restructuring consist in? In section 2 we will argue for the view that restructuring involves a biclausal structure, and more specifically one of raising. 3) What does clitic climbing consist in? In section 3 we argue that clitic climbing is ‘long’ DP-movement out of the embedded clause, which crucially can only take place in a Null subject language. 1. Restructuring verbs As a first approximation, the class of restructuring verbs in Catalan can be described as consisting of the following classes (which I exemplify with clitic climbing and without):2 a) Auxiliary verb:va ‘past perfect’: (1) Ho van trencar. / Van trencar-ho. it-PAST-3pl to-break PAST-3pl to-break-it ‘They broke it.’ b) Epistemic modals: poder ‘may’, haver de ‘must’, deure ‘must’. (2) Ho deuen tenir. / Deuen tenir-ho. it-must-3pl to-have must-3pl to-have-it ‘They probably have it.’ 226 CJL 1, 2002 Jaume Solà 2. In this paper I put aside Romance causative verbs (Catalan fer ‘make’ and deixar ‘let’), which involve a different construction, in which clitic climbing is not sensitive to la Null Subject status of the language (thus French does not allow clitic climbing with restructuring, but it does with causative constructions). Cat.Jour.Ling. 1 001-259 26/2/03 15:54 Página 226 c) Root modals: haver de ‘have to’, poder ‘can’, voler ‘want to’, gosar ‘dare to’, tractar de ‘try to’ mirar de ‘try to’, provar de ‘try to’, intentar ‘try to’, saber ‘know how’, aconseguir ‘manage to’, arribar a ‘manage to’. (3) El vull veure. / Vull veure’l. him-want-sg to-see want-sg to-see-him ‘I want to see him.’ d) Aspectuals: soler ‘usually’, començar a ‘start V-ing’, continuar +GER ‘keep V-ing’, anar +GER ‘go on V-ing’, deixar de ‘stop V-ing’, tornar a ‘be back V-ing’; estar +GER ‘be V-ing’, acabar de ‘have just V-ed’ acabar de ‘finish V-ing’; anar a‘be about to’: (4) El torno a llegir. / Torno a llegir-lo. it-return-1sg to to-read return-1sg to to-read-it ‘I am reading it again.’ e) Some verbs of motion: anar ‘go’; venir ‘come’: (5) L’aniré a visitar. / Aniré a visitar-lo. him-go-FUT-1sg to to-visit go-FUT-1sg to to-visit-him ‘I will go and visit him.’ The claim has been made that some or all of these verbs are functional verbs that do not have a full clausal complement: they are rather functional morphemes belonging somewhere in the functional hierarchy of the clause. Picallo (1990) makes this proposal for epistemic modals, and assigns root modals a non full-verb status. Cinque (1999a, 1999b, 2000) proposes that modal and aspectual verbs are functional morphemes corresponding to a variety of functional categories which are arranged in a universal hierarchy present in each clause. Within Cinque’s a proposal, restructuring (hence clitic climbing) with modal and aspectual verbs can be derived almost trivially: these verbs involve no biclausal structure and therefore clitic climbing turns out to be a clause-bound operation (just like clitic placement in any simple clause). Cinque (2000) in fact makes the stronger claim that the class of restructuring verbs is coextensive with the class of functional verbs. Restructuring is therefore ‘obligatory’ (a functional verb can not be lexical).3If independently motivated, this account appears to provide the simplest solution to restructuring. Not all restructuring verbs, though, are equally likely to be «only functional». Auxiliaries are uncontroversially functional. Epistemic modals are good candidates to be functional verbs: they do not impose selectional restrictions on their Clitic Climbing and Null Subject Languages CJL 1, 2002 227 3. Of course, there might be a limited number of cases of lexical ambiguity between functional and lexical status, but not a systematic alternation. Cat.Jour.Ling. 1 001-259 26/2/03 15:54 Página 227 subject; they must precede auxiliaries (see Picallo 1990) and they seem to be paraphrasable with some modal adverb. Next, aspectual verbs do not impose selections on their subject either; and at least some of them have an adverbial paraphrase too: (6) a. Torna a ploure. - Plou una altra vegada. returns to to-rain rains an other time ‘It’s raining again.’ b. Acaba d’ arribar. - Ha arribat ara mateix. finishes of to-arrive has arrived now self ‘S/he’s just arrived.’ Root modals do not satisfy the transparency criterion for subjects. It is not clear that they assign a standard theta role to their subject, but at least they seem to attribute to it some relational property (capability, obligation or volition). They do not lend themselves to adverbial paraphrase. Finally, we have verbs of motion such as the ones in (7) from Catalan, whose meaning does not easily lend itself to an aspectual or modal interpretation. (7) a. Ho ha anat a arreglar. it-has gone to to-fix ‘S/he has gone to fix it.’ b. T’he vingut a veure. you-have-1sg come to to-see ‘I have come to see you.’ It is true that motion verbs meaning ‘go’ or ’come’ often become grammaticalized as modal or aspectual verbs (with some aspectual or modal interpretation, typically ‘future’). But the more basic meaning of these verbs (which allows restructuring) is clearly one of motion to a (not mentioned) space location in order to do something. Verbs meaning ‘go’ with a more specific path interpretation (‘go up’, go down’, etc.) are also restructuring:4 (8) a. L’he pujat a veure. him-have-1sg gone-up to to-see ‘I have gone up to see him.’ 228 CJL 1, 2002 Jaume Solà 4. In Catalan, venir ‘come’ is restructuring not only in its allative construal, but also, more marginally, in its ablative construal: (i) ?Ara m’hi vinc d’inscriure. now me-tere-come-1sg from-to-register ‘I’m just back from getting registered for it.’ Cat.Jour.Ling. 1 001-259 26/2/03 15:54 Página 228 b. L’he baixat a buscar. him-have-1sg gone-down to to-fetch ‘I have gone down to fetch him.’ c. Hi entraré a parlar. there-go-in-1sg to to-talk ‘I’ll go in to talk to him.’ d. El va sortir a veure. him-PAST-3sg go-out to to-see ‘S/he went out to see him.’ e. El passaré a saludar. him go-by-1sg to to-greet ‘I’ll go by to greet him.’ Beyond verbs of motion, there are some other apparently lexical verbs that allow clitic climbing. One of them is aprendre ‘learn’. (9) Ho he après a fer. it-have-1sg earned to to-do ‘I have learned to do it.’ Most strikingly, Catalan allows for clitic climbing with a number of inherently reflexive verbs. These verbs are compatible only with a restricted (basically locative) set of climbing clitics (in the following examples, climbed clitics are in italics, while inherent clitics are in normal case): (10) a. Se n’hi va a viure. / *Se n’ho va a veure. SE-EN-there-go-3sg to to-live SE-EN-it-go-3sg to to-see ‘S/he’s going there to live there.’ ‘S/he’s going there to see it.’ b. No s’hi atreveix a anar. / *No se n’atreveix a parlar. not SE-there-dares to to-go not SE-of-it-dares to to-speak ‘S/he doesn’t dare to go there.’ ‘S/he doesn’t dare to do talk about it.’ c. S’hi ha proposat anar. / *Se l’ha proposat llegir. SE-there-has determinedto-go SE-it-has determined to-read ‘S/he has determined to go there.’ ‘S/he has determined to read it.’ d. (?)S’hi disposava a anar. / *Se’n disposen a venir SE-there-get-ready-PAST-3sg to to-go SE-EN-get-ready to to-come molts. many. ‘S/he was getting ready to go there.’‘Many are getting ready to come.’ e. (?)S’hi ha descuidat d’anar. / ??Se l’ha descuidat de comprar. SE-there-has forgotten of-to-go SE-it-has forgotten of to-buy ‘S/he forgot to go there.’ ‘S/he forgot to buy it.’ Clitic Climbing and Null Subject Languages CJL 1, 2002 229 Cat.Jour.Ling. 1 001-259 26/2/03 15:54 Página 229 Clitic climbing is also possible with some psych-verbs, which show similar restrictions: (11) a. (?)M’hi agradaria anar. / *Me l’agradaria llegir. me-there-would-like to-go me-it-would-like to-read ‘I’d like to go there.’ ‘I’d like to read it.’ b.?M’hi interessaria ser present. / *Me l’interessaria llegir. me-there-would-interest to-be present me-it-would-interest to-read ‘I’d be interested in going there.’ ‘I’d be interested in reading it.’ Even if a modal-aspectual flavor could be argued for in most cases, it should be assessed whether functional verbs are to be expanded to such a generous extent, including apparently lexical meanings like ‘learn’, and allowing for the possibility of inherent clitics. In the following section we will provide tests to decide between alternative structural options for the analysis of restructuring. Probably the optimal solution is a unitary account for all restructuring cases, extending from the most functional-like cases to the most lexical-like ones.5 2. Restructuring Ideally, an account of restructuring should satisfy the following desiderata: (12) a. Restructured constructions should minimally differ from their non-restructured counterparts, to the extent that (at least some) restructuring constructions have a systematic relation to a non-restructured counterpart. 230 CJL 1, 2002 Jaume Solà 5. A further issue to consider is whether (apparent) object or dative control verbs can be restructuring. One candidate is ensenyar ‘teach’ (and its counterpart in other Romance languages), a dative control verb: (i) M’hi ha ensenyat a anar sol. / M’ho ha ensenyat a fer. me-there-has taught to to-go alone me-it-has taught to to-do ‘S/he taught me how to go by myself.’ ‘S/he taught me how to do it.’ Cinque (2000) argues that this is a case of hidden causative. Indeed, there are several verbs that display causative behaviour when clitic climbing occurs. Verbs like forçar/obligar ‘force’ or ajudar ‘help’, which basically are object control verbs, show the typical object/dative alternation of causative verbs (depending on the transitivity of the infinitive) with clitic climbing: (ii) a. ?L’en han obligat a sortir. / ?Li ho han obligat a pagar. ACC-EN-have-3pl forced to come-out DAT-it-have-3pl forced to pay ‘They forced him to come out of there.’ ‘They forced him to pay it.’ b. ?L’en han ajudat a sortir. / ?Li ho han ajudat a pagar. ACC-EN-have-3pl helped to come-out DAT-it-have-3pl helped to pay ‘They helped him to come out of there.’ ‘They helped him to pay it.’ The verb ensenyar, instead, always displays dative in clitic climbing, as shown in (iii): (iii) Li n’ha ensenyat a parlar bé. / Li ho ha ensenyat a fer. DAT-EN-has taught to speak well DAT-it-has taught to do ‘S/he taught him to praise it.’ ‘S/he taught him to do it.’ Cat.Jour.Ling. 1 001-259 26/2/03 15:54 Página 230 b. Restructuring should make transparency effects possible (perhaps not obligatory). Consider first (13). It has been traditionally assumed that while (13.a) is an instance of restructuring (and clitic climbing), (13.b,c,d,e) (where clitic climbing is impossible due to several opacity factors) are not: (13) a. El vull veure. (Vull veure’l.) him-wantg-1sg to-see want-1sg to-see-him ‘I want to see him.’ INFINITIVAL NEGATION: b. *El vull no veure més. / Vull no veure’l més. him-want-1sg not to-see more want-1sg not to-see-him more ‘I want not to see him anymore.’ FOCUS FRONTING OF THE INFINITIVE: c. *VEURE, el vull! / VEURE’L, vull! see him-want-1sg see-him want-1sg ‘It’s seeing him that I want.’ DISLOCATION OF THE INFINITIVE: d. *Veure, ara no el vull. / Veure’l, ara no vull. see now not him-want-1sg see-him now not want-1sg ‘As for seeing him, I don’t want right now.’ CLEFTING OF THE INFINITIVE: e. *És veure, que el volia. / És veure’l, que volia. is to-see that him-wanted-1sg is to-see-him that wanted-1sg ‘It’s to see him that I wanted.’ It is plausible that all instances of voler ‘want’ in (13) are to be ascribed to the same lexical entry. Therefore, differences between the restructured and the nonrestructured options should be minimal. Cinque (2000) makes the most radical claim in this connection: there is only one structure with restructuring verbs, the «restructured» one, which in fact simply involves a monoclausal structure with the «restructuring» verb in its functional position. A verb like voler (and, mutatis mutandis, any other restructuring verb) is inserted as the head of some Modal (or Aspectual) phrase, whose complement is the next functional category in sentence structure. In Cinque’s view, the fact that clitic climbing (and other transparency effects) are blocked in cases (13.b,c,d,e) is not due to absence of restructuring, but to some locality problem which occurs even with restructuring. Specifically, focus fronting, dislocation and clefting would involve Null Complement Anaphora (and not a trace) as the resumptive element, which would break the chain conClitic Climbing and Null Subject Languages CJL 1, 2002 231 Cat.Jour.Ling. 1 001-259 26/2/03 15:54 Página 231 nection of the climbed clitic to its base position in the infinitival (forcing control in the displaced infinitival). Let us call Cinque’s hypothesis the OnlyFunctional Hypothesis. There are two conceivable alternatives to Cinque’s proposal. One is that restructuring configurations basically derive from full biclausal structures, with a control infinitival (perhaps also from raising structures, if the restructuring verb is arguably raising). This is the view in Rizzi, 1982, Kayne 1989b, Bok-Bennema & KampersMahne 1984, Roberts 1987. According to this view, in essence what triggers restructuring is not the basic structure but some feature specification that somehow allows (or forces) some special transformations (generally head incorporation). Let us call this hypothesis the Full Structure Hypothesis. The second alternative is that restructured clauses are always somehow reduced from the base, but are nevertheless biclausal: just like raising verbs are optionally complemented by a partial structure (IP instead of CP), restructuring verbs would (optionally) take a reduced clausal complement. Let us call this hypothesis the Partial Structure Hypothesis. Within the Full Structure and the Partial Structure hypotheses, restructuring may in principle be conceived as obligatory (restructuring always applies with restructuring verbs) or optional (restructuring optionally applies with restructuring verbs). Let us proceed to check these alternatives. There is one test that distinguishes the Full Structure Hypothesis from the other two. In Control Structures, the Full Structure Hypothesis predicts control to hold of restructured constructions (assuming that transformations cannot «destroy» control). As observed by Picallo (1990), the subject of a restructuring verb behaves like an unaccusative subject (allowing en-cliticization) just in case the infinitival verb is unaccusative: (14) a. N’hi volen anar alguns. / Volen anar-n’hi alguns. EN-there-want-3pl to-go some want-3pl to-go-EN-there some ‘There are some that want to go there.’ b. *En volen protestar alguns. EN-want-3pl protest some ‘There are some that want to protest.’ The possibility of en-cliticization is lost when the opacity inducing factors occur: (15) *N’hi voldrien no anar alguns. / *Voldrien no EN-there-wanted-3pl not to-go some wanted-3pl not anar-n’hi alguns. to-go-EN-there some ‘There are some that would like not to go there.’ 232 CJL 1, 2002 Jaume Solà Cat.Jour.Ling. 1 001-259 26/2/03 15:54 Página 232 Obviously, the facts in (14) are incompatible with control, as can be seen from the behavior of control non-restructuring verbs:6 (16) *Lamenten (no) anar-n’hi alguns. regret-3pl (not) to-go-en-there some ‘There are some that regret not to go.’ The facts in (14) are compatible with both the Only Functional Hypothesis and a version of the Partial Structure Hypothesis that postulates IP structure (hence raising) for the restructured infinitive. There is another test that shows essentially the same result. In restructuring constructions (unlike in non-restructuring ones), subject inversion (both free and indefinite) can be shown to place the inverted subject inside the infinitive: (17) a. Volen venir els cosins a dinar. want-3pl to-come the cousins to to-lunch ‘Our cousins want to come and have lunch.’ b. *Es proposen venir els cosins a dinar. SE-determine-3pl to-come the cousins to to-lunch ‘Our cousins are determined to come and have lunch.’ c. Venen els cosins a dinar. come-3pl the cousins to to-lunch ‘Our cousins are coming to have lunch.’ (18) a. Han provat d’entrar lladres a la casa. have-3pl tried of-to-enter thieves to the house ‘Thieves have tried to get into the house.’ b. *S’han decidit a entrar lladres a la casa. SE-have-3pl decided to to-enter thieves to the house ‘Some thieves have made their mind to get into the house.’ c. Han entrat lladres a la casa. have-3pl entered thieves to the house ‘Thieves got into the house.’ (19) a. No ho ha gosat tocar ningú amb les mans. not it-has dared to-touch nobody with the hands ‘Nobody dared to touch it with their hands.’ Clitic Climbing and Null Subject Languages CJL 1, 2002 233 6. These facts (and the subject inversion facts below) constitute strong counterevidence against any proposal that postulates control in restructured infinitives: Kayne (1989b), Bok-Benema & KampersMahne (1994), Roberts (1997). Cat.Jour.Ling. 1 001-259 26/2/03 15:54 Página 233 AGROhead, while other clitics (dative and oblique) stand for other AGR-heads. We will assume that checking for clitic AGRs is generally pure agree, without move: (32) [AGRP [AGRº cl ] ... DP ...] Let us call them clitic AGR-heads.14 Within such a view, consider clitic climbing. If we want to analyze clitic climbing as a direct AGR-argument dependency, two problems arise: a) Why should the clitic AGR be generated in (or moved to) the upper clause (instead of in the infinitive), considering that it is associated with the infinitival argument positions? b) Are «climbed» clitic AGRs local enough to their associated arguments? We address these questions in turn. 3.1. Clitic climbing as attraction Clitic climbing should satisfy a requirement which has been largely accepted within the Minimalist Program, namely that movement should be driven by (functional) features that need to be checked against matching features in a lower constituent. In this connection, it is to be remarked that the «landing site» for clitic climbing is always exactly the clitic position corresponding to the restructuring verb: either a proclitic of the finite restructuring verb or an enclitic of the infinitival or imperative restructuring verb. This is highly suggestive of movement to a specific target. Let us examine some facts that convincingly show that clitic climbing is sensitive to the «attracting» clitic position. If clitic climbing was head movement with no attracting or preexisting target, the expectation is that any clitic or clitic cluster originating in the restructured infinitive should be able to climb to the main clause (just as, say, any phrase is able to scramble in the right configuration). This is indeed the case with many restructuring verbs. However, we have seen that in Catalan there is an significative number of restructuring verbs which have «clitics of their own», namely inherent clitics (anar-se’n go-SE-EN ‘leave’, disposar-se ‘get ready to’, atrevir-se ‘dare’ descuidar-se ‘forget’). In these cases, clitic climbing is possible, but in a limited way: only certain clitics can climb (in the following examples, I indicate climbed clitics in italics): 240 CJL 1, 2002 Jaume Solà Further, the patterns of variation for clitic doubling (either with any NP or only with pronouns) parallel those of AGRS-doubling: while Italian has AGRSdoubling with any NP, some well-known Northern Italian dialects have AGRS-doubling only with pronouns. 14. An assumption also compatible with our account would be that clitics are voices in Sportiche’s (1992) sense (Roberts 1997 adopts this option). Then clitic associates undergo A-movement to their Spec,AGR (this being either clause-bound or climbing movement), subsequently moving to their Voice Phrase in a clause-bound fashion. The essential point is that clitic climbing is A-movement out of the infinitive. In other words, clitic climbing is cyclic, whereas clitic placement is not (Luján 1993:259). Cat.Jour.Ling. 1 001-259 26/2/03 15:54 Página 240 (33) a. Se n’hi va a viure. / *Se n’ho va a veure. SE-EN-there-go-3sg to to-live SE-EN-it-go-3sg to to-see ‘S/he’s going there to live there.’ ‘S/he’s going there to see it.’ b. S’hi disposava a anar. / *S’ho disposava SE-there-get-ready-PAST-3sg to to-go SE-it-get-readyPAST-3sg afer. to to-do ‘S/he was getting ready to go there.’ ‘S/he was getting ready to do it.’ c. No s’hi atreveix a anar. / *No s’ho atreveix a fer. not SE-there-dares to to-go not SE-it-dares to to-do ‘S/he doesn’t dare to go there.’ ‘S/he doesn’t dare to do it.’ d. S’hi ha descuidat d’anar. / **Se s’ha descuidat d’afaitar. SE-there-has forgotten of-to-go SE-SE-has forgotten of to-shave ‘S/he forgot to go there.’ ‘S/he forgot to shave himself.’ Morphological incompatibilities are expected to arise in such cases, stemming from the clitic cluster pattern of possible forms. This would be the case with the ungrammatical sentence in (33d), where the climbed reflexive clitic is incompatible with the inherent reflexive clitic of the restructuring verb (descuidar-se): Catalan clitic clusters do not allow two reflexive clitics. But the incompatibilities observable in such cases are far from obvious within a merely morphological account. One consistent type of incompatibility is that between the inherent reflexive clitic from the restructuring verb and the climbed accusative clitic. This incompatibility is not morphological in any obvious sense, since reflexives are generally compatible with 3rd person Accusative clitics. Suppose we try to relate these incompatibilities to Case Theory. In (33a,b,c) we can see that Accusative clitic climbing in not compatible with the inherent reflexive of a restructuring verb. This can be taken to derive from the fact that (a subset of) inherent reflexives absorb Accusative Case. Take the case of disposar-se ‘get ready’. This verb coexists with a transitive counterpart (disposar ‘to arrange’), which suggests that its inherent reflexive somehow has absorbed Accusative. In the other cases (anar-se’n ‘go’, atrevir-se ‘dare’), there is transitive alternation, but we can still assume that the clitic is an indication of lack of Accusative. This Case account does not cover all the incompatibilities that arise with these verbs: in fact, only the clitic hi ‘there’ can generally climb with reflexive restructuring verbs.15 Let us tentatively assume, nevertheless, that clitic climbing is restricted by the Case (and possibly other) properties of the upper clause clitic configuration. Clitic Climbing and Null Subject Languages CJL 1, 2002 241 15. Consider: (i) No s’hi atreveix a anar. not SE-there-dares to to-go ‘S/he does not dare to go.’ Cat.Jour.Ling. 1 001-259 26/2/03 15:54 Página 241 We then propose that clitic climbing involves Case checking at the upper clause: it would not consist in mere displacement of constituents that have already been checked for case in the lower clause. Rather, we assume that in clitic climbing constructions, there is no object Case checking in the lower clause. Let us reconsider our suggestion that clitics are AGR-heads (clitic AGRheads). In the cases above, a clitic AGR position in the main restructuring clause contains clitics associated to both main and infinitival arguments or Cases. This suggests that clitic AGRs are not necessarily tied to every clause structure, but rather are projected when possible and necessary. Now the question is: what makes clitic AGR-projections not necessary in the infinitive and possible in main clause. 3.2. Null subjects and clitic climbing The connection between clitic climbing and NSLs has been claimed for Romance languages. Kayne (1989) proposes that what crucially characterizes Null-Subject languages is I(nflection)’s ability to l-mark VP. An l-marked VP loses barrierhood and the clitic can escape the VP as a necessary first step for clitic climbing. This proposal captures the Null-Subject status of clitic climbing in a simple way, but we cannot adopt it as it stands because we have discarded a head-movement account of the clitic «displacement». Assuming that restructuring occurs in non-NSLs such as French (perhaps in all languages), and assuming further that raising is also present in restructuring in the other languages, we must conclude that raising per se can not amount to 242 CJL 1, 2002 Jaume Solà (ii) No s’hi atreveix a parlar. not SE-there-dares to to-talk ‘S/he does not dare to talk to him.’ (iii) *No s’ho atreveix a fer. not SE-there-dares to to-do ‘S/he does not dare to do it.’ (iv) *No se n’atreveix a sortir. not SE-EN-dares to to-go-out ‘S/he does not dare to get out.’ (v) *No se li atreveix a dir res. not SE-DAT-dares to to-say nothing ‘S/he does not dare to tell her/him anything.’ Rigau (1994) argues that inherent reflexives generally absorb Partitive Case. Indeed, Partitive Case is not possible with reflexive restructuring verbs, whether in unaccusative or in transitive constructions: (vi) *No se n’atreveixen a venir (d’admiradors). not SE-EN-dare to to-come (of fans) ‘No fan dares come’ (vii)*No se n’atreveix a comprar (de condons).’ not SE-EN-dares to buy (of condoms) ‘S/he does not dare to buy condoms.’ Cat.Jour.Ling. 1 001-259 26/2/03 15:54 Página 242 clitic climbing. Optimally, the simple interaction between the Null Subject «property» and raising should give Clitic Climbing as a result. Let us try. There has been, during decades, a minority line of research that has contended that the Null Subject property of the Romance kind is to be defined by a single core property which simultaneously derives null subjects and free subject inversion (see Rosselló, 1986, Solà 1992, Picallo 1998, and Rosselló 2000, that makes a critical review of other recent proposals). According to this view, free subject inversion is a misnomer for the parametric option of licensing Nominative Case in its basic (theta) position, without resorting to NP movement to (or expletive insertion in) Spec,IP. In a recent proposal, Rosselló (2000) convincingly argues that I (T) checks its associate DP inside the vP or VP directly (by mere agree, in Chomsky’s 1998 terms),16 with no EPP feature requiring raising of the DP or merge of expletive.17 Assuming this line of reasoning, we should now ask what the implications of such a theory should be for raising. Raising (and ECM) infinitives have been generally characterized as involving a defective I (Chomsky 1988 ties this property to the absence of CP). Defective I (Idef) contains an EPP feature but it cannot check (erase) a Case feature.18 Now, it becomes apparent that the hypothesis sketched above for NSLs must inescapably address the issue of defective I: if NSLs generally lack an EPP feature in I, and this extends to Idef, Idef will not project a subject specifier. In fact, Idef will neither be able to check Case features (due to its deficiency in phi-features) nor trigger DP movement. Assuming that Idef has some features to check, its role will solely be to ensure that some DP is still active for checking when the derivation reaches I’. In other words, Idef needs an associate, and makes sure that this associate will not be Case marked when the derivation reaches I’. Let us put all the pieces together. Restructuring involves a raising configuration, headed by Idef. This defective I lacks an EPP feature in NSLs. This means that no subject position will be projected in the specifier of Idef. This in turn implies that NSLs will lack a potential minimality barrier for A-movement, which will be present in non-NSLs. Is all this amenable to an account of clitic climbing?19 Clitic Climbing and Null Subject Languages CJL 1, 2002 243 16. Rosselló shows that a standard account in not only unnecessary, but even impossible for Catalan, in view of the never-raising behavior of subjects of certain existential verbs (quedar ‘be left’, faltar ‘be lacking’). 17. This kind of approach challenges the EPP principle, perhaps redefining it as a parameter: roughly, some nominal features must be strongly present in IP, either as «rich» AGR in the I head or, if AGR is poor, as a DP in Spec,I. Under the view that subjects are checked in their base position, preverbal subjects must be analysed as dislocated or base generated DPs that bind a resumptive pro in theta position. 18. The EPP feature has emerged into the theory due to the inescapable need to justify DP movement to, or expletive merge in, Spec,I, in the infinitive of an ECM construction, in the face of no possible Case account. 19. If restructuring (as raising) triggers clitic climbing, typical raising verbs (counterparts of seem or turn out) should also allow clitic climbing. This is so for Italian. For Spanish and Catalan, these verbs do not rallow clitic climbing. There should be some structural property that distinguishes Catalan and Spanish raising verbs from restructuring verbs: perhaps they involve a more complex infinitival structure. Cat.Jour.Ling. 1 001-259 26/2/03 15:54 Página 243 Indeed, we have claimed that clitic climbing configurations involve a «long distance» dependency between an upper clitic position and its associated argument positions in the infinitive. Consider the following abstract configuration for a simple clause (where AGRSand AGROare descriptive labels for any categories responsible for Nominative and Accusative, respectively, while EA and IA are the external and the internal argument DPs, respectively): (34) AGRSAGRO[vPEA IA] AGRSshould check EA and AGROshould check IA, giving a crossing paths pattern. This configuration raises a problem for minimality, which has been solved by ensuring equidistance between the two DPs competing for AGRO(see Chomsky 1985:187,356). Chomsky’s (1995:Ch4, 1998) solution is to assimilate AGROto v, which both projects the EA and checks the IA, thus becoming a hinge point for locality. With AGRSassimilated to T, we have: (35) T [ IA [ EA v[VP ... tIA ]]] In (35), EA is accessible to T, given appropriate definitions on equidistance. Now, consider what a clitic climbing configuration would be for a sentence like El vull veure it-want-1sg to-se ‘I want to see it’, according to our previous assumptions: (36) AGRSAGRO[IP Idef [vPEA IA]] If Idef is inactive, the local relations between the two AGRs and the two DPs are essentially the same as in (34): crossing paths. But for this case locality cannot be accommodated as in (35), because AGROis in a different clause. In order to pursue an account of clitic climbing in terms of ‘long distance AGRO’, AGRO must be conceived as independent from v. Suppose then that we want to essentially preserve the patterns in (34) and in (36) for simple clauses and clitic climbing configurations, respectively. We need some substantial elaboration to ensure that locality is respected. I want to pursue one of the essential insights in Chomsky (1995:ch.4): vis the key-stone for enabling Accusative Case checking and projecting the external argument (so deriving Burzio’s Generalization). This connection provides a local configuration that minimizes the «crossing» pattern for Nominative-Accusative checking. But instead of assimilating vto AGRO, I propose to assimilate it to another kind of AGR. Consider the participial agreement visible in Italian compound tenses (see Kayne 1989a), which is always agreement with the internal argument: it agrees with unaccusative subjects and with transitive (clitic) objects. It is an adjectivallike agreement (only Number and Gender features). Therefore it is not likely to 244 CJL 1, 2002 Jaume Solà Cat.Jour.Ling. 1 001-259 26/2/03 15:54 Página 244 check Accusative Case, if Nominative and Accusative checking is a property of categories with a full set of phi-features (Person, Number and Gender).20 Suppose we assimilate vto participial agreement, in an abstract sense generalizable to all sentences (even if in Italian only compound tenses show its overt manifestation). This means that vcan perform a task similar to that in Chomsky (1995,1998): checking the internal Argument and, possibly, «attracting» it to its specifier. But now we propose that no Accusative Case is involved in this checking, for the above-mentioned reason that this kind of agreement is [Number, Gender] only, no [Person] being involved. We assume, with Chomsky, that only a full set of phi-features can check (structural) Case. Let us assume that transitive v(i.e., vprojecting an external argument EA) has an EPP feature that brings the internal argument to its specifier. The result will be the configuration in (37), where the IA has raised and the EA is projected (for transitive verbs). (37) [vP IA [vPEA v[VP ... tIA ]]] Now, even if, with Chomsky, we assimilate AGRSto T, AGRO(as well as other clitic AGRs) is to be kept an independent category. It is essential that AGROis projected (or activated) only when and where necessary: in clitic climbing contexts it need not be projected in the embedded clause (although it can, if clitic climbing is indeed optional). Now the configurations for a simple clause and for clitic climbing, after IA raising to Spec,vhas taken place, will be (38) and (39) respectively: (38) T AGRO[vP IA [vPEA v[VP ... tIA ]]] (39) T AGRO[VP V [ Tdef [vP IA [vPEA v[VP ... tIA ]]]] If Tdef in (39) does not project a specifier, the locality conditions are now the same in both constructions: both T and AGROcan access both IA and EA, assuming that specifiers of the same projection (vP) are equidistant For simplicity, we have omitted main clause v, which does not project a specifier either, assuming that restructuring verbs, as raising verbs, do not have an EA, hence vdoes not have an EPP feature. One problem remains to be solved: as its stands, our proposal predicts that either T or AGROcan access either IA and EA. We must ensure that AGROchecks IA while T checks EA. To this effect, we narrow the definition of equidistance from (40) to (41): Clitic Climbing and Null Subject Languages CJL 1, 2002 245 20. One plausible job for this agreement is checking of Partitive Case, assuming that [Gender, Number] is a sufficient phi-set for Partitive checking. It appears that Partitive is not a «self-standing» Case: its needs to be supplemented by a structural (Nominative/Accusative) Case, as assumed in the text that follows (see Belletti 1988). See, however, Rosselló (2000:123,124) for an interesting piece of evidence against this view. Cat.Jour.Ling. 1 001-259 26/2/03 15:54 Página 245 (40) Terms of the same minimal domain are equidistant (Chomsky 1985, 1988). (41) Two terms of the same minimal domain are equidistant if one of them is inactive. In configurations (38) and (39), (40) makes IA and EA equidistant to both T and AGRO. (41), instead, predicts that IA is closer to AGROat the moment AGRO is merged. It is the only option for AGRO, which will check it under agree (crucially, not under move), rendering it inactive. At the moment T is merged, however, EA is equidistant from IA, which will not intervene in the T-EA checking operation by agree (with no move either, we assume for NSLs). The optionality of clitic climbing within restructuring is accounted for by assuming that the structure in (42.a) is a legitimate free alternative to (42.b) (= (39)): (42) a. T [ Tdef AGRO[vP IA ([vPEA) v[VP ... tIA ]]]] b. T AGRO[T def [vP IA ([vPEA) v[VP ... tIA ]]]] Since AGROprojects when possible and necessary, it can be perfectly projected in the embedded infinitive, giving a non-climbing configuration. Now, for non pro-drop languages like French (or English), Tdef projects a specifier due to its EPP feature, so that only (43.a) (non-climbing), but not (43.b) (climbing), is possible: (43) a. T [ EA Tdef AGRO[vP IA [vPtEA v[VP ... tIA ]]]] b. T AGRO[Spec Tdef [vP IA [vPEA v[VP ... tIA ]]]] In (43.a) AGROagrees with IP and inactivates it; Tdef subsequently agrees with EA, which moves its specifier (where it will be accessed by the higher T). In (43.b), Tdef cannot access EA. If then Tdef (incorrectly) accesses IA, EA will be left inaccessible to both AGROand T, due to the intervention of Spec,Tdef, leading the derivation to crash. 4. Further extensions and discussion 4.1.On the relevance of null subject parameter The present analysis relies on the idea that in (Romance) null-subject languages T (hence Tdef) does not have an EPP feature; therefore Tdef does not create a minimality barrier for long distance checking between main clause AGROand its associate internal argument (IA) in the infinitive. This predicts that languages being like Catalan except fort the null-subject property will lack clitic climbing in restructuring constructions. This does not predict, however, that clitic climbing is confined to null-subject languages: clitic climbing should also be possible with a more impoverished infinitival structure lacking the TP projection, since this too would provide the relevant transparency effects by eliminating the Tdef specifier. The Romance 246 CJL 1, 2002 Jaume Solà Cat.Jour.Ling. 1 001-259 26/2/03 15:54 Página 246 causative construction (which allows clitic climbing in both French and null-subject Romance languages) is likely to instantiate this possibility. It is beyond the scope of this paper to assess whether 17th century French or Kru languages (which have been claimed to allow clitic climbing in the absence of null subjects, as we mentioned in the introduction) can be accommodated in this way. 4.2. On the status of blocking factors A further issue that has not been addressed thus far is why negation and other factors block restructuring (hence clitic climbing). Let us pursue the idea that blocking factors somehow force a richer structure than a bare TP. Negation in Romance has been claimed to be a functional category above TP (Zanuttini 1991). As for the other factors (focusing, extraposition over an adverb, dislocation), let us assume that a bare TP is not suited for such operations (while CP is). Now, we can assume with Chomsky (1998) that Tdef can be selected only by V. Therefore, any clause structure larger than a bare TP will not contain a Tdef; and restructuring and clitic climbing will be impossible. For the blocking status of negation, a head movement account has often been adduced (Kayne 1989, Bok-Bennema & Kampers-Mahne 1994, Roberts 1997). Such an account seems unavailable here. It should be remarked, in addition, that a preposition often intervenes between the restructuring verb and the infinitive. This preposition is likely to be lexical (see Amadas Simon 1999:5.3) and may be a problem for head incorporation analyses. 4.3. On the status of v Let us shift attention to a central idea in our proposal, namely that vis responsible for checking and attracting an internal argument, but without checking its Case (due to its incomplete set of phi-features). We have suggested that v-agreement is visible in past-participial agreement in languages like Italian. Past participial agreement in Italian is triggered by both (clitic) objects and unaccusative subjects. If this is to be captured by the present proposal, vshould be present no only in transitive (including unergative) structures but also in unaccusative structures, much like Chomsky’s (1998) generalized v(vs. transitive v*). There are several pieces of evidence in favor of the structure [IA [EA vVP]] postulated here. If this structure generalizes to all kinds of verbs, we have the following three representations: (44) a. [ IA [EA v[VP ... tIA ]]] (transitive structures) b. [ [EA v[VP ... (IA) ]]] (unergative structures) c. [ v[VP ... IA ]] (unaccusative structures) For the unergative configuration (44.b), the postulated internal argument would probably incorporate to V (ultimately to v). Clitic Climbing and Null Subject Languages CJL 1, 2002 247 Cat.Jour.Ling. 1 001-259 26/2/03 15:54 Página 247 One piece of evidence in favor of (44) is that it mirrors the shape and behavior of past participles: «passive» past participles (when EA is absorbed: hence unergatives do not have a «passive» past participle) and «active» past participles in compound tenses, where EA, if present, remains accessible. (45) also corresponds to the infinitive complement of causative verbs in Romance, an ECM construction where Accusative Case is assigned exactly to the DP which is most accessible in each of the structures in (45): IA with transitives and unaccusatives, and EA with unergatives: (45) a. El faig arreglar (a en Joan/per en Joan). it-make-1sg to-repair (to the John/by the John) ‘I have it repaired (by John).’ b. El faig caure. it-make-1sg to-fall ‘I make it fall.’ c. El faig treballar. him-make-1sg to-work ‘I make him work.’ A further piece of evidence concerns word order. In transitive structures IA moves to the left of EA. It must be actual move, not just agree, for the locality theory advanced here to work. Now, is there any evidence for this movement? The answer is clearly yes. Assuming that the verb ([vV v]) moves past both arguments, we predict the order V IA EA for sentences with subject inversion (where the EA stays in its base position). That is what we standardly find in Catalan and Italian (even if cooccurrence of both overt object and overt subject is somehow marginal in some cases): (46) a. Només menja patates en Joan. / *Només menja en Joan patates. only eats potatoes the John only eats the John potatoes b. Pagarà la factura la companyia. / *Pagarà la companyia la factura. will-pay the bill the company will-pay the company the bill It is not accurate to claim that these facts point to a VP final position for the subject: inverted subjects unmarkedly precede VP adjuncts and also datives and governed locatives:21 (47) a. Ho ha escrit en Joan amb l’ordinador /a casa seva /aquesta tarda. it-has written the John with the-computer at home his this afternoon ‘This was written by John with the computer/at his home/this afternoon.’ 248 CJL 1, 2002 Jaume Solà 21. In (47) the inverted subject is not interpreted as (restricted) focus: rather the whole VP is interpreted as the comment. (47.b), for example is an adequate reply to «What’s this letter you found?». We cannot then support the view that subject inversion is focalization (Belletti 1999). Cat.Jour.Ling. 1 001-259 26/2/03 15:54 Página 248 b. Aquesta carta, la hi va enviar l’avi a l’àvia. This letter it-he-PAST to-send the-grandfather to the-grandmother ‘This letter was sent by grandfather to grandmother.’ c. Els diners, ja els ha posat la Maria al banc. the money already it-has put the Mary in-the bank ‘The money was already but in the bank by Mary.’ We conclude, then, that the functional structure we propose has some independent theoretical and empirical motivation 4.4. On the prevalence of raising over ECM One of the questions that have been left unanswered is why raising is the general option for restructuring verbs, and not ECM. If we postulate that the crucial property of restructuring is a reduced infinitive (Tdef), the prediction should be that either raising or ECM should be possible depending on the properties of the restructuring verb. For epistemic restructuring verbs, raising is a natural option, since they do not seem to select any argument other than the clausal complement itself. But for deontic modal verbs like voler, or motion verbs like anar, raising is not expected. Rather, ECM seems to be predicted, if the restructuring verb is able to assign Accusative (which in the present proposal comes for free provided locality conditions are met). Let us consider what the ECM option would look like for a NSL in the present account (suppose we have a transitive verb in the infinitival): (48) a. T AGRO[vPEA1v[VP V[ TP Tdef AGRO[vPIA [EA2v ]] ]]]] In this ECM configuration, the main vprojects an external argument (EA1), hence has an EPP feature, which must attract an accessible DP. Since IA in the embedded clause has been checked by the infinitival AGRO, the upper vcan attract EA2, giving the standard ECM results. In other words, ECM is predicted to be possible and to work essentially as in non NSLs. Clitic climbing with ECM is predicted to be impossible, since both the main and the embedded AGROmust be present for all DPs to be checked. Now, since ECM is not attested with Romance restructuring verbs, we must conclude that the DP argument of these verbs is never an EA: then vlacks an EPPfeature and cannot attract the DP to be «exceptionally case-marked». The lack of an EA is easy to argue for with restructuring motion verbs, which are plausibly unaccusative. As for deontic verbs we might suggest that their DP argument is an Experiencer, and that Experiencers tend not to be EAs in Romance. Since control is also impossible with a reduced infinitive (Tdef cannot check Null Case), we are left with the raising option, however problematic it is to assume that a control structure may turn into a raising structure whenever the restructuring verb shifts from selecting a CP to selecting a TP. If this is a real Clitic Climbing and Null Subject Languages CJL 1, 2002 249 Cat.Jour.Ling. 1 001-259 26/2/03 15:54 Página 249