Clitic Climbing in Causative Constructions

Clitic Climbing in Causative Constructions Xavier Villalba Universitat de Barcelona In this paper it is argued that the distribution of Clitic Climbing in Catalan causative Constructions may receive a principled account if two independently motivated processes are assumed: Det Incorporation, conceived as a proper means to check Case features, and Verb Incorporation, a process at stake in Causative Constructions. The interaction of both head movement processes with general principies of Universal Grammar not only explains the bulk of empirical data, but also gives a proper answer to the theoretical problems that optional Clitic Climbing poses for Chomsky's (1992) Last Resort F'rinciple. The distribution of (third person) pronominal clitics in Catalan causative constructions (hereafter: CC) raises many puzzling questions that have not received a principled and comprehensive analysis. Consider the well-known contrast between externa1 and interna1 arguments of the embedded verb with respect to Clitic Climbing (hereafter: ClCl): (1) a. *Faré dormir-10s junts, els nens. make-FUT sleep-them(ACC) together the children b. Els faré dormir junts, els nens. them(ACC) make-FUT sleep together the children The children, I will make them sleep together.' (2) a. Els regals, faré posar-10s junts a la Maria. the gifts make-FUT put-them together to the o Mary Catalan Working Papers in Linguislics (CafWPL) 3.2 (1994): 123-152 Universitat Autbnoma de Barcelona (2) b. Els regals, els far6 posar junts a la Maria. the gifts them make-FUT put together to the Mary The gifts, I will make Mary put them together.' Whereas the causee clitic must climb, the object one may climb or not. It is tempting to account for this contrast in terms of an internal vs. external argument distinction. However, such a distinction is helpless if unaccusative verbs are taken into account: *Va fer venir-10s aviat, els metges. PAST make wme-them(ACC) soon the doctors Els va fer venir aviat, els metges. them(ACC) PAST make come soon the doctors The doctors, he made them come soon.' Que va fer venir-ne molts, de metges? that PAST make come-of.it many of doctors Que en va fer venir molts, de metges? that of.it PAST make come many of doctors 'Did he m&e many doctors come?' The clitic related to the internal argument of the unaccusative verb venir ('come') is expected to behave as the object clitic in (2) with respect to CICI: it should optionally climb. However, such an expectation is only partially fulfilled: when accusative, it must climb, just like causee clitics do; nevertheless, when partitive, it may climb or not, just like interna1 argument clitics do. An explanation on the basis of the internal vs. external nature of the argument must thus be revisited. Furthermore, dative clitics pose another challenge because clitics related to dative internal arguments cannot climb, as the interpretative contrast between (Sa') and (Sa") shows: (5) a. Li far6 enviar-[lzi] cartes. himlher-DAT make-FLIT send-them-DAT ietters a'. *'I will make them send letters to himlher.' a". I will make himlher send them letters.' The contrast between accusative and dative internal argument clitics is again an unexpected result when compared with the wntrasts shown above. After considering all these examples, it is extremely difficult to make any empirical generalization on the range of CIC1 in Catalan CC. Moreover, the preceding examples raise many theoretical questions as well. Firstly, let us take into account the optionality of CICI, as it is shown in (2). Chomsky's (1992) Last Resort Principle (LRP) states that operations (i.e., 'move a') take place just in order to make the derivation converge, in other words, if an operation need not take place, it cannot. Nevertheless, the existence of a phenomenon like optional CIC1 prima facie poses a problem for the LRP. Another point is the relation between cliticization and Case. We must ask ourselves whether the different case of the clitic has any role in the distribution of clitics in Catalan CC, as seems to be the case. If the answer is affinnative, we must go further and investigate the place for cliticization in a theory of Feature Checking like that of Chomsky (1992). In this paper it will be suggested that an anaiysis of CC and cliticization crucially involving head-movement (Verb Incorporation and Deteminer Inwrporation respectively) can account for the preceding contrasts. Moreover, it will be shown that such an analysis can also provide us with an answer for the two theoretical problems: the optionality of CIC1 and the relation between cliticization and Case. The organization of this paper is as follows. In the first section I will outline the analysis of cliticization I will assume throughout. Aftenvard, I will briefly work out the issues concerning CC. In the third section I will deal with the distribution of clitics in CC. Finally, the main conclusions will be addressed. 1. Cliticization as Det Incorporation In this paper it will be assumed, following Uriagereka (1989, 1992), Laenzlinger (1990) and Roca (1992), which pursued seminal unpublished work by E. Torrego, that cliticization is in fact Determiner Incorporation (hereafter: DI). This is not surprising if looked at from a diachronic perspective: both determinen and clitics have the same origin, Latin demonstratives. From a synchronic point of view, the above-mentioned authors have shown that there is evidence enough to assume that both deteminers and clitics are Dets heading their own DP projection. Consequently, the representation I will assume for clitics is the following: 1 Given this analysis, cliticization is an instance of head movement. We thus expect its distribution to be constrained by general principies constraining movement (the ECP, if looked at from a standard approach, or a condition on chain formation, if looked at from a minimalist one). This paper is not the place to argue for the advantages of the 'Det Hypothesis' over the 'Affix Hypothesis'. Nevertheless, I will briefly show in the next paragraph how the former might subsume the basic intuitions of the latter, especially with respect to Case. LaenzJinger (1990) and Roca (1992). following Uriagereka (199'2). explicitiy assume the Det to select a nul1 pro as a complement. Another noteworthy option would be to consider clitics as intransitive Dets, in the way Abney (1987) treats pronouns. Anyway, the decision between one of these analysis is not relevant for the discussion. So then I arn not making a choice here. 1 .I. A Minimalist Approach to Cliticization Borer (1984), the classical approach to the 'Affix Hypothesis' of cliticization, arrives at the following main conclusions: (a) clitics are generated as features of the head of their phrase -i.e., they are the 'spell-out' of the Case properties of the head; (b) the argument position is filled by a lexical or nul1 NP depending on whether a Case-assigning device is available or not; (c) the clitic governs the argument NP; and (d) the clitic and the argument NP must be coindexed. Both (c) and (d) straightfonvardly follow from the 'Det hypothesis': the moved Det must govern its trace and obviously they must have the same index. Conclusion (b) refers to the question of 'clitic doubling'. As Roca (1992) has argued, the 'Det Hypothesis' can account for this phenomenon without further stipulations (even though the version he adopts cannot): the doubled DP appears on the Spec of the DP headed by the clitic, the complement position being occupied by a pro. The status of this doubled DP is far from clear, but clarifying it would exceed the scope of this paper, so I am focusing my attention on question (a). The intuition that clitics are the 'spell-out' of the Case properties of heads, even though productive, has not received a clear technical implementation in the literature. The approach ta feature checking developed in Chomsky (1992) seems to provide us with a good solution. Nevertheless, some technical background is needed. The next paragraph will be devoted to sketching the basis of this approach. Aftenvard, we will dea1 with the relationship between cliticization and Case. 1.2. On Feature Checking Chomsky's (1992) approach to the role played by morphological features in syntax is based on two main points: (a version of) the Strong Lexicalist Hypothesis (SLH) and a theory of feature checking. The SLH assumes, in the spirit of Lexical Phonology, that lexical elements are inserted in syntax with all their morphological features. Verbs will thus be drawn from the lexicon fully inflected for Tense and Agreement (V-features). The same holds for DPs with respect to Case and +-features (N-features). However, if feature assignment devices and, more important, the conditions regulating their application are dispensed with, how is the correct matching of features to be achieved? It is obvious that a theory of feature checking is needed. Such a theory should give an answer to two basic questions: why should features be checked? and how are features checked? Answering the first question is quite straightfonvard if the minimalist approach to linguistic theory is adopted. Chomsky (19926) claims that "the linguistic expressions are the optima1 realizations of the interface conditions, where 'optimality' is determined by the economy conditions of UG". From this point of view, it seems quite plausible to regard morphological features as elements that may be dispensed with at LF, because they play no role at this interface level. Therefore, if a feature is not checked before LFinterpretation applies, the input to this interface level will not be optimal, making the derivation crash. The question of how features are checked may be divided into two aspects: the formal checking device and the structural conditions regulating its application. Regarding the first aspect, it is worth noting that while answering the question of why checking was necessary a sketch of a formal checking device has implicitly been suggested. If features must disappear before LFinterpretation applies, it is logical to regard checking as an elision process: features are checked, if they correctly match, they are eliminated; if they don't, they remain and the derivation crashes at LF. The structural conditions regulating checking are quite straightfonvard: the (element canying the) feature must be in a local relation with an appropriate head, i.e., a head canying the same feature. Technically speaking, a feature must enter into the checking domain of a head carrying the same feature, where the 'checking domain' of a head a is the minimal set of nodes (the complement excluded) contained in MAX (a) that are distinct from a and do not contzin a (see Chomsky (1992)).2Namely, the checking domain of a head includes its specifier and anything According to Chomsky (1992: 15) "the category a contains $ if some segment of a dominates $" adjoined (adjunction being allowed to its maximal projection, its specifier, or its head). A feature may be checked in any of these positions. 1.3. Case Checking According to the SLH, argument DPs enter the syntactic component with a Case feature.3 So then, they must enter into the checking domain of a Case-bearing head to have this Case feature checked. Three logical options exist: raising to the specifier, adjunction to the specifier, and adjunction to the head.4The former -raising to the specifieris, according to Chomsky & Lasnik (1991) and Chomsky (1992), the core realization of structural Case checking. Given the highly articulated structure of the clause, argument DPs, which are generated VP-intemally, move to the specifier of their respective agreement heads (Agr-S and Agr-O). There they enter into the checking domain of V, which has previously adjoined to the Agr head to check its inflectional features (see Chomsky (1992) for a detailed analysis). The second option -adjunction to the specifierseems to be restricted to a single case: sentences with a there-DP pair of the kind: (7) There is a unicorn in the garden. The DP a unicorn bears nominative case, but it cannot check it in situ (but see Belletti (1987) and Lasnik (1993) for a different view). Note that the presence of the expletive there makes raising to the specifier of Agr-S (or Tense) impossible. The only possibility is to adjoin the DP It is a matter of discussion whether Case is a property ofDetls or of N's (see Chomsky (1992:n. 35)). I will nevertheless make the assumption -rather plausible if we look at a language like German, where the Det is iníiected for Casethat it is Det rather than N that bears N-features (i.e., Case and $-features). Indeed, a fourth logical possibility exists: adjunction to the maximal projection. Much research on this possibility, which has not been taken into amunt in the literature, is needed. This phenomenon provides us with a clear-cut test to set up the c-commanding relations within V2*. Moreover, given our hypothesis that the causee asymmetrically c-commands the internal arguments of V2, a prediction can be made: a quantifier in the causee position would share its index with a pronoun within the object position, but not conversely. The prediction is borne out, as the LF representations of the following sentences show: (10) a. El Pere ha fet convidar tothom a la seva mare. the Peter has made invite everybody to the his mother 'Peter has made his mother invite everybody.' a'. *[for every x, x a person, Peter made x's mother invite x] a". [for every x, x a person, Peter made y's mother invite x] b. El Pere ha fet convidar la seva mare a tothom. the Peter has made invite the his mother to everybody 'Peter has made everybody invite his mother.' b'. [for every x, x a person, Peter made x invite x's mother] Furthemore, wh-phrases show the same behavior: (1 1) a. Qui vas fer convidar a la seva mare? Who PAST make invite to the his mother 'Who(m) did you make his mother invite?' a'. *[for which x, x a person, you made x's mother invite x] a". [for which x, x a person, you made y's mother invite x] b. A qui vas fer convidar la seva mare? to whom PAST make invite the his mother 'Whom did you make invite his mother?' b'. [for which x, x a person, you made x invite x's mother] Whereas a quantifier or a wh-phrase in object position yields a WCO effect when binding a pronoun within the causee position at LF, the sentence becomes perfect in the converse configuration. It must be concluded thus that the causee asymmetrically c-commands the intemal arguments of V2. There is thus enough evidence to assume our initial hypothesis: the causee is the X-bar subject of V2? 2.2. Causative Verbs Select a Verbal Projection 2.2.1. Lack of Tense. There exist many empirical arguments supporting the hypothesis that causative verbs select a verbal projection as a complement. Lack of embedded tense is the most striking one.8 Consider the following sentences: (12) a. Ahir ti van ordenar atacar el campament avui. yesterday him-DAT PAST ordered attack the carnp toda^ 'Yesterday they ordered him to attack the camp today.' The conclusions arrived at in this paragraph go against Li (1990), who claims that the causee and the intemal arguments of V2 mutually c-command in languages like Catalan. Moreover, a corollary of this conclusion is that dative causees are not PPs but DPs. In other words, the element a ('to') is not a true preposition, but an inserted Case-assigner. In Branchadell (1992) it is convincingly argued that the same conclusion holds for al1 instances of lexical and non-lexical datives. However, for the differences between dative causees and other dative DPs see Villalha (1993). Obviously, there exist instances of CC involving a CP complement in many languages. Consider: (i) El Pere va fer que la Maria compr6s el diari. the Peter PAST make that the Mary buy(SUi3JCT) the newspaper 'Peter made Mary buy the newspaper.' Nevertheless, sentences like that in (i) lack the bulk of characteristics making CC interesting and seem to be quite parallel to other instancm of selected tensed CPs. NegP -ex. (21).llI arn not entering into details, however. For our purposes, the evidence presented so far is enough to motivate VI in Catalan CC. 2.3.2. Theoretical Motivation and Consequences. We have just seen that an analysis of Catalan CC involving VI was tenable on empirical grounds. However, something has to be said on its theoretical motivations and consequences. One theoretical motivation of VI might follow from the structure of CC assumed so far. We have seen that the causative complement had no Tns projection at all. However, as different authors have suggested (see Gutron & Hoekstra (1988). Zagona (1988)), events are to be licensed by Tns. In that case, the event argument of the embedded VP would have no chance of being licensed in situ.12 The only way for the embedded event to be licensed is VI: once V2 incorporated, the embedded event is accessible to the main Tns and can hence be licensed. This analysis straightfonvardly follows from the constituent structure of CC proposed in (8), so I will assume it to be correct. For other arguments, see Guasti (1992), Li (1990) and Villalba (1993). What about the consequences that VI has for UG? While discussing Noun Incorporation, Baker (1988) raises the question of how complex heads behave with respect to Case. Simple heads have their Case features listed in the lexicon, but something new has to be assumed for heads formed in syntax. It seems quite plausible to assume, following Di Sciullo & Williams (1987), that Case (and other relevant) features of the simple verbs involved in VI percolate up to the complex one. Nevertheless, an immediate question arises: is feature percolation somehow l1 After V2 incorporates to V1 the latter may excorporate, at least in some Catalan dialects (the same is true for Italian and French), allowing some adverbs, íloating quantifiers and the negative particle to appear between both verbs. See Guasti (1992) and Villalba (1993). l2 It would be a matter of discussion whether CC involve two or just one event. In Guasti (1992) and Villalba (1993). it has been argued on the basis of non-temporal adverbs scope ambiguities that V2 has its own event argument. See also Alsina (1992) and Ritter &Rosen (1993). restricted or is it rather a feature addition process? Baker (1988) gives ovenvhelming evidence against feature addition. Furthermore, he proposes the following principle constraining feature percolation: (22) Case Frame Preservation Principle A complex X of category A in a given language can have at most the maximal Case assigning properties allowed to a morphologically simple item of category A in that language. (Baker 1988: 122) Putting its conceptual adequacy aside, the Case Frame Preservation Principle (hereafter: CFPP) seems to be highly desirable on empirical grounds. It provides us with an explanation of the following ungrammatical sentences:l3 (23) a. *Vaig fer comprar un cotxe als nens al Pere. PAST make buy a car to-the children to-the Peter 'I made Peter buy the children a car.' b. *Vaig fer comprar un cotxe el Pere. PAST make buy a car(ACC) the Peter(ACC) 'Mary made Peter buy a car.' Even though, both fer ('make') and comprar ('to buy') can each legitimate an accusative and a dative DP, the resultant complex verb cannot legitimate neither two lexical datives nor two accusatives. In other words, no Case feature addition seems to be allowed at all. The CFPP thus constrains feature percolation in complex verb formation: complex verbs would bear as many case features as a simple one. This will have crucial consequences for our analysis of dative clitics (see paragraph 3.2.) l3 It is worth noting that some speakers find sentences like (23a) acceptable, however. See Villalba (1993) for a possible explanation. 3. On Clitic Climbing in Causative Constructions Let us make a summary of the conclusions we have arrived at so far: (a) cliticization may receive a proper analysis as an instance of DI; (b) DI is a proper means to check Case features; (c) CC have the constituent structure in (8); and (d) CC involve complex verb formation -¡.e., VIunder the constraints posed by the CFPP. Now we have the basic skills to return to the contrasts sketched at the begiming of this paper. 3.1. Non-Dative Clitics 3.1.1. Causee Clitics. Let us consider the behavior of causees with respect to CICI: (24) a. *Fart? dormir-10s junts, els nens. make-FUT sleep-them(ACC) together the children b. Els fart? dormir junts, els nens. them(ACC) make-FUT sleep together the children The children, I will make them sleep together.' The Det head of the causee cannot incorporate to V2. As it has been shown in paragraph 2.1, the causee asymmetrically c-commands the internal arguments of V2. It must also be the case that it asymmetrically c-commands V2 as well. As a result, the causee Det incorporating to V2 from the causee position would create an ill-formed chain (see footnote 10): the head would not c-command its trace making the derivation crash.14 The independently motivated constituent structure of CC can easily explain why causees must climb.15 3.1.2. Interna1 Argurnent Clitics: Transitive Verbs. Consider again the following contrast: (25) a. Els regals, faré posar-10s junts a la Maria. the gifts make-FüT put-them together to the Mary b. Els regals, els faré posar junts a la Maria. the gifts them make-FüT put together to the Mary 'The gifts, I will make Mary put them together.' Let us assume that the (simplified) constituent structure underlying both sentences is: l4 It might be argued that there exists a proper representation for (24a): once VI applies, incorporating both V2 and Det to Vl, the moved Det does c-command its tram. However, if, as Chomsky (1992:n. 20) suggests, chain formation is better understood derivationally rather than representationally, then the violation would still be operative, ruling out the sentence. 15 However, as F. Benucci @.c.) points out, there exist several non-standard Romance dialects where a causee clitic appears (sometimes as a copied clitic) attached to V2 (some examples can be found in Benucci 1990:ex. (15)). I have no explanation for these examples. Now let us suppose that DI precedes VI. Det must check its Case features with an appropriate head. Given the structure in (26), three possible landing-site heads arise: V2, V1, and a higher head. However, only the former, V2, is available. The Det incorporating either to V1 or to a higher head would yield a violation of economy principles -namely, the condition that the links of a chain must be minimal (see Chomsky (1992:21))-, because a more accessible landing site -i.e. V2would be skipped. The Det must thus incorporate to V2 in order to check its Case features. Once incorporated, the Det must remain adjoined to V2, because its moving up would violate economy principles. On the one hand, incorporation to a head higher than V1 would still form a non-minimal chain, making the derivation crash. On the other hand, incorporation to V1 would now form a minimal chain. However the Det has already checked its Case features, so it has no features to check with VI. This movement would thus violate the LW, because it is not necessary to make the derivation converge. As there is no chance for DI to apply, VI will apply incorporating the complex formed by V2 plus the Det to VI. Once there, the things do not change for the Det: even if DI were still allowed to apply, the Det could still not incorporate to VI, because of the LW. Furthermore, it could not incorporate to a higher head either, because a non-minimal chain would result. The resulting complex verb will show the following structure: (27) [vi Vl [v2 V2+Detll The Det must remain adjoined to V2, a correct result for sentences like (25a). We have just seen how the non-climbed version is the only possibility whenever DI precedes VI. Let us see what happens when it is VI that precedes DI, instead. As it has been shown in paragraph 2.3, V2 incorporates to VI, forming a complex verb. Moreover, it was also suggested, following Di Sciullo & Williams (1987), that the features of each verb percolate up to the complex one (under the constraint of the CFPP). This has crucial consequences for our analysis: V1 and V2 stop bearing any feature. So then, once VI has applied, neither V1 nor V2 can further count as landing sites for the Det moving to check its Case features: the movement would not be motivated by feature checking, yielding a LRP violation. The only way for the Det to check its Case features is thus incorporation to the complex verb.16 Note that this movement satisfies economy principles. It satisfies the LRP because it is triggered by Case checking. Furthermore, the resulting chain is minimal because feature percolation plausibly affects indices as well, so the potential intervening heads -i.e. V1, V2, and the trace of V2do not count as heads distinct from the complex verb but rather as segments of it. The resultant complex verb would have a structure like the following: The Det must adjoin to the complex verb, a good result for sentences like (25b). Let us make a summary of the analysis just suggested. We have seen how the optionality of ClCl showed in (25) is not optional at all, rather it can be derived from the interaction of two independent processes: DI -an independently motivated mechanism to check Case featuresand VI -an operation at stake in CC. If DI precedes VI, the non-climbed version necessarily results -i.e. (25a). If VI precedes DI, clitics must climb -i.e. (25b). Before closing this paragraph, let us make a brief comment on the theory-interna1 aspects of this approach. Leaving aside its empirical plausibility, this approach to optional CIC1 seems to be plausible on theoretical grounds as well. On the one hand, because of its being the result of the interaction of two independent processes -i.e. VI and DI-, no additional machinery nor stipulation is needed to account for the data. On the other hand, it gives a principled solution to the puzzle that optional ClCl posed to the LRP: optional ClCl is perfectly compatible with a l6 It is neediess to say that incorporation to a higher head is not a option, as we have just seen before. Neither is incorporation to the trace of V2. Probably, traces bear no Case features as long as the whole chain (V2, t) percolates its features up. principle like the LRP since no optional movement is involved at all. The apparently optional behavior of CIC1 is mther the result of the order in which DI and VI apply. 3.1.3. Interna1 Argument Clitics: Unaccusative Verbs. Let us now consider the distribution of clitics heading the interna1 DP argument of an unaccusative verb. (29) a. *Va fer venir-10s aviat, els metges. PAST make come-them(ACC) soon the doctors b. Els va fer venir aviat, els metges. them(ACC) PAST make come soon the doctors 'The doctors, he made them come soon.' (30) a. Que va fer venir-ne molts, de metges? that PAST make come-of.it many of doctors b. Que en va fer venir molts, de metges? that of.it PAST make come many of doctors 'Did he make many doctors come?' The behavior of internal arguments of unaccusative verbs is prima facie surprising: if the Det is accusative, it must climb, whereas if partitive, it may optionally climb. From what we have seen in previous paragmphs, the expected distribution is that of (30): the Det heading ai internal DP argument may climb or not depending on whether VI applies before or after DI. Namely, the sentences in (30) would receive the same analysis that those in (25). But what is this not so in (29) as well? The answer is quite straightforward for our approach to cliticization as a Case checking device. Let us consider (29a). It seems to be the case that the accusative Det has incorporated to V2 in order to check its [+acc] Case feature. However, this poses an unsolvable problem: on the one hand, venir ('to come') obviously has no [+acc] feature to be checked with; on the other, the Det cannot move up, as we have seen in paragraph 3.1.2. We have then an [+acc] feature (that of the Det) that cannot be checked, making the derivation crash. In other words, sentence (29a) cannot receive a convergent derivation. Hence, the contrast between accusative and partitive clitics with unaccusative verbs follows from simple assumptions on the structure of UG, namely, the requirement that features must be checked. 3.2. Dative Clitics Let us finally take into account dative clitics: (31) a. Li far6 enviar-[lzi] cartes. himlher-DAT make-FUT send-them-DAT letters a'. *'I will make them send letters to himlher.' a". I will make himlher send them letters.' The non-climbed clitic must be understood as the indirect object of V2, whereas the climbed one can only be understood as the causee. Let us see how this distribution can be accounted for under our analysis. We have two dative DPs -the indirect object and the causeeand two heads -V1 and V2bearing a [+dat] feature each.17 According to this, a proper derivation for (31) would be as follows. DI precedes VI. Consequently, the head of the indirect object DP must incorporate to V2 exactly as the direct object had to (see paragraph 3.1.2 for a complete derivation). On the other hand, the head of the causee cannot incorporate to V2, as it has been shown in paragraph 3.1.1 (see also footnotes 10 and 14). Therefore, the causee Det incorporates to VI, where it checks its [+dat] feature. Afterward, VI applies yielding the following complex verb structure corresponding to the correct interpretation (3 la"): l7 On the nature of the case bome by the causee and the Case properties of V1 see Villalba (1993). To sum up, if DI precedes VI, the correct interpretation results. On the one hand, economy principies force the indirect object Det to incorporate to V2. On the other hand, the independently motivated hypothesis that the causee is the X-bar subject of V2 can straightfonvardly account for the obligatoriness of causee Det climbing. What happens with (31at)? Let us see why such an interpretation cannot be obtained. It has just been shown that the dative causee Det cannot appear on V2 regardless of its case. However, no principled reason has been suggested to prevent a Det heading a dative DP internal argument from climbing. In fact, a derivation like that of non-dative climbed Dets -i.e., VI precedes DIprima facie seems to be possible. Nevertheless, the crucial point is that the parallelism with non-dative Dets leaves off being operative at this point. Given the analysis of VI sketched in paragraph 2.3, if VI precedes DI, the resultant complex verb will inherit the Case features of both V1 and V2 under the conditions posed by the CFPP. This had no relevant consequences for non-dative climbed Dets, because the incorporated Dets borne a different Case feature each: the CFPP allowed the complex verb to inherit enough Case features to license the non-dative internal argument and the causee -i.e. one accusative and one dative feature. However, when two dative DPs come up, the things radicaliy change. The CFPP still allows the complex verb to inherit one accusative and one dative feature, but now these features are not enough to license both the indirect object and the causee. In other words, even though V1 and V2 bear a [+dat] feature each, the resultant complex verb will bear just one [+dat] feature, othenvise a violation of the CFPP would result. That amounts to saying that once VI applies one of the two dative Dets will not be able to check its Case feature, making the derivation crash at LF (see paragraph 2.1). Therefore, (31a') could not receive a convergent derivation, a good result for our anaiysis. Moreover, such a Case-based approach to the contrast in (31) can easily be tested. It allows us to make the following prediction: if we had a dative interna1 argument but no causee, the former would climb without any problem. The prediction is borne out, as CC involving unaccusative embedded verbs show: (33) a. [alzi] far6 arribar aquelles cartes. them-DAT make-FUT arrive those letters 'I will mai1 them those letters.' b. Aquell &tel li va fer venir gana. that cocktail himlher-DAT PAST make come appetite That cocktail whetted his appetite.' The Det heading the dative interna1 argument of the unaccusative verb may climb because only one dative DP is to be licensed by the complex verb.18 It seems thus that the existence of sentences like (33) gives us additional support for the Case-based approach to the distribution of dative clitics in Catalan CC. 4. Conclusions Our approach to the distribution of clitics in CC has rested on very simple grounds. Firstly, the constituent structure of CC gave us an answer to the obligatoriness of ClCl with externa1 arguments. Secondly, optional ClCl was derived from the interaction of two independent processes -DI and VIand general principies of UG. Depending on the order in which DI and VI take place, the clitic climbs or not. This optionality does not extend to interna1 dative clitics, since the complex verb can only license a [+dat] feature because of the CFPP. Consequently, for a CC with two dative Dets to converge DI must precede VI, following the correct distribution of dative Dets. Finally, such an analysis has proved to correctly predict the distribution of Dets heading dative internal arguments in CC with unaccusative verbs. 18 The same explaaation holds for internal dative clitics in the so-called jaire-par construction. See Kayne (1975) and Burzio (1986) for the relevant examples.