Case, Incorporation, and Economy : an Approach to Causative Constructions

-. . CASE, INCORPORATION, AND ECONOMY: AN APPROACH TO CAUSATIVE CONSTRUCTIONS Xavier Villalba Universitat Autbnoma de Barcelona In this paper it is argued that the causee behaves as a subject because it is a subject. It is shown that many different phenomena related to Binding, Control and Predication Theory may be accounted for under an analysis of Causative Constructions taking the causee as the subject of a Vmax selected by the causative verb. Moreover, the proposed constituent structure interacts with both Case Theory and head movemeut, yielding a principled explanation of aoss-linguistic variation of Causative Constructions with respect to Case and word order. Finally, it is claimed that a last resort de, formulated as 'Discharge an unsatisfied case' and constrained by independentiy motivated economy principies, is responsible for the assignment of dative case to the causee in Catalan and many other languages. Since the initial works in the seventies -Bordelois (1974) or Kayne (1975) for example-, many linguists have focused their attention on Causative Constructions (henceforth CC): Rouveret & Vergnaud (l980), Zubizarreta (l985), Burzio (1986), and Baker (1988) constitute an incomplete but relevant sample of works within the GB framework. The conclusions that every author has anived at are extremely diverse in several points except in one: the causee is the structural subject of the causativized verb.' It can be said that this is an undiscussed point that underlies all works on CC, at least in the GB framework. Bordelois (1974) and subsequent work as Bordelois (1988) are an exception within the GB framework. She claims that the causee is in fact an indirect object of the causative verb that controls the subject of the embedded sentence. However, she adopts such an analysis just for the faire-infinitive construction and only when the embedded verb is not ergative/unaccusative. In the other cases (farre-par and faire-infinitive with ergativelunaccusative verbs), the analysis she adopts is the 'classical' one, which represents the causee as the subject of the causativized verb. The exception is then relative. Calalan Working Papers in Linguislics (1992): 345-389. Universitat Autbnoma de Barcelona Nevertheless, such an assumption has been called in doubt by Alsina (1991) from the LFG. He claims that no argument can be submitted to support it, and gives severa1 arguments that point at considering the causee as an intemal argument of the causativized verb. In this article I will give some piece of evidence that the causee must be represented as the higher argument of the embedded verbal projection, i.e., as a phrase-structure subject;Z I will do it in section 1. In section 2, a principled account of CC in Catalan will be provided; I will argue that the interaction of Case Theory, Incorporation and Economy suffices to explain the behavior of CC with respect to case and word order. Section 3 attempts to show how the analysis outlined in section 2 might explain cross-linguistical variation in CC making use of independently motivated principles. Section 4 is devoted to addressing some of the remaining problems. Finally, a summary of the main conclusions will be given. 1. The Causee as a Phrase-Structure Subject I .I. The Causee and Binding Theory (I): C-Commanding dornains Let us consider the following sentences: (1) a. Els professors faran inscriure'si al Jomi. the teachers will-make enrol himself tethe John The teachers will make John enrol himself.' 2 It might be argued that the discussion is not on the correct track. As Baker (1988) suggests. notions such as 'subject' or 'object' cannot be further considered as primitives of the theory. Rather they should be defined with respect to the different principes and modules of the Universal Grammar. From this point of view, the causee can no Ionger be considered a 'subject' in the broad sense: it may be seen as a subject with respect to Thematic or Binding Theory. but as an object with respect to Case Theory, for example. However, systematizing such a theory of Grammatical Functions is beyond the scope of this article. -. (1) b. *~ls professorsj faran inscriure'sj al Joan. the teachers will-make enrol themselves to-the John The teachers will make John enrol themselves.' An interesting fact follows from (1): the anaphor in object position must be bound by the causee in order to be licensed, that is, the presence of a closer binder -the causeeprevents the main subject from binding the anaphor. In tems of Chomsky (1986)'s Theory of Binding, it may be assumed that the minimal Complete Functional Complex containing a BT-compatible index for the anaphor in object position includes the causee (in fact, it is the causee which provides a BTcompatible index to the anaphor), but excludes the main subject. Obviously, that amounts to saying that the causee c-commands the other arguments of the embedded verb, what is far from being surprising, if we assume the causee to be a phrase-structure subject. The causee behaves then like subjects, which c-command all the interna1 arguments even in their verb internal position (see Koopman & Sportiche (1991)). Such a behavior is far from being a singularity of Romance languages. Grimshaw (1990) brings us evidence that this is so in Japanese. Japanese zibun 'self', only allows subjects as antecedents: (2) Taroo wa Hanako o zibun no kuruma kara oros-i-ta. Taro-TOP Hanako-ACC self-GEN car from come down-PAST Taro brought Hanako out of his/*her own car.' However, zibun allows a causee as antecedent: (3) Taroo wa Hanako o zibun no kuruma kara ori-sase-ta. Taro-TOP Hanako-ACC self-GEN car from come down-CAUS-PAST Taro made Hanako come out of hislher own car.' This contrast confirms the hypothesis in a sharp way: if zibun is, as it has been commonly accepted, a subject-oriented anaphora, then the causee must be a subject. Note moreover that, as Grimshaw (1990: 168-169) points out, such a contrast is a very strong argument for rejecting lexical approaches to causative constructions. It is needless to say that Japanese crucially differs from Catalan in allowing binding by any subject, not just the closest one. This seems to be true also for Italian. Guasti (p.c.) points out that the contrast in (1) does not hold in Italian: (4) a. Maria ha fatto accusare se stessoi a Giannii. Maria has made accuse himself to Gianni 'Maria has made Gianni accuse himself.' b. Mariai ha fatto accusare se stessai a Gianni. Maria has made accuse herself to Giami 'Maria has made Gianni accuse herself.' It must be concluded that se stessolse stessa 'himselflherself', like zibun, does not impose a closeness constraint on its antecedent.3 3 However, the facts are far from being clear. Benucci @.c.) claims that only (4b) is possible in his diaiect (Veneto). However, when the causee is realized as a clitic, both possibilities are perfect: (i) a. Maria le ha fatto accusare se stessolse stessa. the Mary to-himlher has made accuse himselflherself 'Mary has made himlher accuse himselflherself.' It might be argued that in this diaiect the causee is in fact a PP instead of an NP. As he has pointed out to me, the causee, if realized as a clitic, may have the same form that the clitic which appears with some PPs. A detailed study of CC in Italian dialects is beyond the scope of this paper. This conclusion is confirmed by other related phenomena. As Kayne (1975) and Bunio (1986) point out, idioms containing a possessive -(5)- and phrases expressing inalienable possession -(6)- have an anaphoric nature (the coindexation in (6c) is just an informal attempt to express the bound nature of the inalienable expression and '#' expresses that the sentence would be grammatical but with a pragmatically odd reading): (5) a. La Maria porta la seva creu amb dignitat. the Mary canies the her cross with dignity 'Mary puts up with her problem with dignity.' b. #La Maria porta la teva creu amb dignitat. the Mary canies the your cross with dignity 'Mary puts up with your problem with dignity.' c. #La Maria lii porta la sevai creu amb dignitat. the Mary to-himlher carries the hislher cross with dignity 'Mary puts up with hislher problem with dignity.' (6) a. La Maria va ficar els nassos a I'assumpte. the Mary PAST put the noses in the business 'Mary stuck her nose into the business.' b. *La Maria va ficar els teus nassos a l'assumpte. 'Mary stuck your nose into the business.' c. *La Maria lii va ficar elsi nassos a I'assumpte. the Mary to-him PAST put the noses in the business 'Mary stuck his nose into the business.' So then, a paradigm like (1) is expected to be repeated. The prediction is borne out: (7) a. La Maria va fer portar la seva¡ creu amb dignitat al J~ani. the Mary PAST make cany the his cross with dignity to-the John 'Mary made John put up with his problem with dignity.' (7) b. #La Maria, va fer portar la seva, creu amb dignitat al Joan. the Mary PAST make cany the her cross with dignity to the John 'Mary made John put up with her problem with dignity.' (8) a. La Maria elsi far& ficar elsi nassos a I'assumpte. the Mary to-them will-make put the noses in the business 'Mary will make them stick their nose into the business.' b. *La Mariaj els fd ficar elsj nassos a l'assumpte. the Mary to-them will-make put the noses in the business 'Mary will make them stick her nose into the business.' (7)-(8) confirm to us that the causee is an appropriate binder for an anaphoric element in object position, so it must occupy a higher position in the embedded predicate in order to c-command the object position. Note furthermore that examples like (8) are a gocd argument for treating the causee as a subject. As Zubizarreta (1985272) points out, "it is a lexical property of a certain class of verbs to allow an 'Art + Body Part' object to be referentially dependent on the verb's subject by virtue of directly binding the determiner to the external argument." In other words, if inalienable possessors are usually referentially bound by a subject, then, having (8) in mind, it can be concluded that the causee is a subject (at least with respect to Binding Theory). We arrive at a similar conclusion when other idioms are iaken into account. As Kayne (1975) and Burzio (1986) have suggested, 'non-passivizable' idioms like (fer) l'orni 'wash one's hands' must be bound by a subject because of their anaphoric nature. If this assumption is correct, it is expected, under the adopted hypothesis, to have one of these idioms licensed by the causee. The prediction is borne out again: (9) a. E1 Joan (es) fa l'orni. 'John washes his hands.' b. llo orni Cs fet (pel Joan). 'His hands are washed by John.' (10) a. La Maria fa& fer l'ornii al Joani. the Mary will-make do the-orni to-the John 'Mary will make John wash his hands.' b. *La Mariaj far& fer l'ornij al Joan. the Mary will-make do the-orni to-the John 'Mary will make John wash her hands.' 1.2. The Causee and Binding Theory (11): Asymmetrical C-Command As it has been shown in the previous paragraph, the causee clearly c-commands the other arguments of the embedded verb. However, this cannot be taken as a conclusive proof of its subjecthood, unless a stronger relation is met: asymmetrical c-command. In X-bar tems, a subject may be defined as the argument that asymmetrically c-commands the other arguments of a given head. So then, asymmetrical c-command is a step that the proposed hypothesis cannot skip. Quantifiers like cada 'every' or cap 'nonet, unlike referential elements, must c-command a pronoun in order to share its index: (1 1) a. Els alumnes que havien de visitar el museu Fhan visitat de pressa. The students that should have visited the museum have visited it in a hurry.' b. *Els alumnes que havien de visitar cada museu Fhan visitat de pressa. The students that should have visited every museum have visited it in a huny.' c. *Els alumnes que no volien visitar cap museu l'han visitat de pressa. The students that wanted to visit no museum have visited it in a huny.' So, if the causee asymmetrically c-commands the arguments within the embedded VP, then a quantifier of the cadalcap class in the causee position is expected to bound a pronoun in the object position but not conversely. The prediction is borne out: (12) a. Vaig fer castigar el seu professor a cada alumne. 'I made every student punish his teacher.' b. No vaig fer castigar el seu professor a cap alumne. 'I made no student punish his teacher.' (13) a. *Vaig fer castigar cada alumne al seu professor. 'I made his teacher punish every student.' b. *No vaig fer castigar cap alumne al seu professor. 'I made his teacher punish no student.' If the interna1 arguments of the embedded verb were able to c-command the position of the causee, sentences like those in (13) would be grammatical. However, the facts point to the opposite direction: at LF the trace left by the quantifier is locally bound by the pronoun yielding a Principle C violation. In other words, if sentences like (13) are instances of Strong Crossover configurations -and it seems to be the case-, then they give us evidence enough to conclude that the causee asymmetrically c-commands the interna1 arguments of the embedded verb.4 - 4 Sentences like these in (i) seem to pose a problem to the claim that the causee asymmetrically c-commands the object position. (i) a. Vaig fer castigar cada un dels alumnes al seu professor. 'I made their teacher punish each of the students.' b. No vaig fer castigar cap dels alumnes al seu professor. 'I made their teacher punish noone of the students.' If my claim is conect, the distributive phrase cannot c-command the causee, so no coreference relation is expected to hold. However, it is not clear that the distributive phrases in (i) are equivalent to the quautifiers in (13). Note that they enter into coreference relations even in contexts where no c-commanding relation is available: (ii) a. Els alumnes que havien de visitar cada un dels museus els han visitat de pressa. The students that should have visited each of the museums have visited them in a hurry.' b. Els alumnes que no volien visitar cap dels museus han acabat visitant-10s de pressa. The students that wanted to visit none of the museums have ended to visit them in a hurry! 1.3. The Causee and Control Theory I have hypothesized that the causee is a phrase-structure subject, at least with respect to Binding Theory. We may now look over Control Theory in order to test whether the hypothesis fits in with obligatory control. Let us firstly consider the following sentences, where an obligatory subject controller is required: (14) a. El Joan desitja de tenir un cotxe per sortir els caps de setmana. 'John wants to have a car to go out on weekends.' b. Carl Lewis intenta batre el rkord del m6n dels cent metres. 'Carl Lewis tries to beat the one hundred meters world record.' Under my hypothesis, if we causativize a verb requiring control by a subject, the causee will be able to control the PRO in the mostly embedded sentence. The data confirm this prediction: (15) a. La publicitat li va fer desitjar de tenir un cotxe per sortir els caps de setmana. 'Publicity made him want to have a car to go out on weekends.' b. El frach als 'trials' li va fer intentar batre el rkord del m6n dels cent metres. The failure at the trials made him try to beat the one hundred meters world record.' Even though the causee is realized as apro bound to a dative clitic, it is still a phrase structure subject. Its controller properties remain thus unchanged. In other words, control brings us more evidence that the causee is indeed a phlase-structure subject.5 It must thus be concluded that distributive phrases are indeed referential, namely, they need not c-command a pronoun in order to corefer with it. So then, sentences like those in (i) do not contradict the claim that the causee asymmetrically c-comrnands the object position. Note that the presence of an overt causee makes the sentences considerably worse: (i) publicitat va fer desitjar de tenir un cotxe per sortir els caps de setmana al Joan. the publicity PAST make want to have a car to go-out the ends of week to-the John 'Publicity made him want to have a car to go out on weekends.' 2.1.3. The Causee as the Subject of VmM. The constituent structure in (18) accounts for all the exarnples quoted in section 1. Let us see the most relevant ones, repeated here as (24)-(26): (24) a. Els professors faran inscriure'si al Joani. the teachers will-make enrol himself to-the John The teachers will make John enrol himself.' b. *Els professorsj faran inscriure'sj al Joan. the teachers will-make enrol themselves to-the John The teachers will make John enrol themselves.' (25) a. Vaig fer castigar el seu professor a cada alumne. 'I made every student punish his teacher.' b. No vaig fer castigar el seu professor a cap alumne. 'I made no student punish his teacher.' (26) a. *Vaig fer castigar cada alumne al seu professor. 'I made his teacher punish every student.' b. *No vaig fer castigar cap alumne al seu professor. 'I made his teacher punish no student.' First of all, the behavior of anaphors in object position -(24)- is not surprising. A governor (the causativized verb), a subject (the causee), and a BT-compatible index are provided within Vmax. Moreover, Vm" is a Complete Functional Complex (CFC) because all the functions required by its head are satisfied within it (in this case, Vmax has both a subject and an object). So then Vis the Minimal Goveming Category (MGC) for an element in object position. The facts confirm this: an anaphor in object position must be bound within Vmax. The same explanation holds for the other anaphoric expressions, i.e., idioms containing a possessive, constructions expressing inalienable possession, and non-passivizable idioms (see 1.1. for the relevant examples). Let us consider now (25)-(26). It is obvious that if quantifiers, because of their non-referential status, must c-command a pronoun in order to corefer with it, the constituent structure in (18) accounts for (25)-(26) straightfonvardly. Another piece of evidence for the correctness of the analysis is given to us by clitic climbing. The obligatory adjunction to the causative verb of the clitic related to the causee position is a well known characteristic of CC, but it has not received a straightfonvard account in the literature. Consider the relevant data: (27) a. El Joan la va fer venir. the John her PAST make come 'John made her come.' b. El Joan li va fer menjar patates. the John tuhimlher PAST make eat potatoes 'John made himlher eat potatoes.' c. *El Joan va fer venir-la the John PAST make come-her 'John made her come.' d. *E1 Joan va fer menjar-li patates. the John PAST make eat-to-himlher potatoes 'John made himlher eat potatoes.' Such a compulsory movement only holds for clitics related to the causee position. The clitics related to internal arguments of the embedded verb may appear adjoined either to the causative verb or to the embedded one: (28) a. El Joan va fer comprar-ne a la Maria. the John PAST make buy-of-it to the Mary 'John made Mary buy some of it.' (28) b. El Joan en va fer comprar a la Maria. the John of-it PAST make buy to the Mary 'John made Mary buy some of it.' If we assume, following Torrego (1988), that cliticization is in fact Det incorporation, then the contrast in (27) can be accounted for straightfonvardly. Having (18) in mind, it is clear that a Det in the position of the causee cannot incorporate to the embedded verb without violating the ECP: the trace would not be governed by the moved Det. The only possible landing site for such an incorporating Det must be outside Vmax, that is, the causative verb or a higher functional category (see Villalba (in progress) for a more detailed analysis of clitic climbing in CC). So then the obligatoriness of clitic climbing when a causee is involved may be derived from the ECP. Furthermore, the constituent structure in (18) explains why causees behave like subjects with respect to Control and Predication Theory: they are subjects. To sum up, many facts related to Binding, Control, Predication or clitic climbing have been accounted for under the analysis that takes the causee to be the subject of a Vselected by the causative verb. 2.1.4. Word Order. Since Stowell (1981), the informational content of phrase-structure rules has been replaced by a word order parameter, commonly formulated as 'head first' vs. 'head last'. That parameter has proved to be a fruitful approach to many of the typoiogical generalizations Greenberg's framework on linguistic universals arrived at. However, it has recently been claimed that another parameter determining the order of the external argument is needed: the Head-Subject Parameter (see Giorgi and Longobardi (1991) and Georgopoulos (1991), who calls it the 'Specifier Parameter'). Giorgi and Longobardi argue that many word order differences between languages can be accounted for by means of the Head-Subject Parameter. Specifically, they suggest that Romance vs. Germanic differences with respect to the internal structure of NPs are due to a different setting of the Head-Subject Parameter: rightward in Romance vs. leftward in Germanic. Let us consider now the following contrast: (29) a. El Joan va fer comprar un llibre a la Maria. the John PAST make buy a book to the Mary 'John made Mary buy a book.' b. *EI Joan va fer la Maria comprar un llibre. 'John made Mary buy a book.' (30) a. *John made buy a book (to) John. b. John made Mary buy a book. Leaving aside the aspects concerning case, which will be addressed in the next subsection, it might be asumed that the main difference between CC in Catalan and English has to do with the position of the causee: whereas it must precede the embedded VP in English, it must follow it in Catalan. How is such a difference to be explained? The HSP might bring us a good chance of dealing with the contrast in (29)-(30) by means of assuming a different setting of the HSP, which has proved to be independently motivated for NPs. So then, the causee (i.e., the subject of Vmax) falls under the HSP: it is generated to the left of VP in English but to the right in Catalan. However, I am not pursuing such an analysis here for two reasons. First of all, in recent work by Kayne (invited lecture at the GLOW Colloquium, 1992, Lisbon) and Uriagereka (class lectures, GISSL 1992, Girona) it is argued that word order is not a property of UG but a condition imposed by the interfaces with other cognitive systems, especially by the interface with Phonetic Form. That is, Phonetic Form imposes linearization on the output of UG. If this approach to word order is correct, pararneters dealing with word order, such as the HSP, may be dispensed with. Empirical evidence supporting so may be found in the following contrast (the examples in (31) are quoted from Alsina (1991)): (3 1) a. Farem creurelconfiar la Maria en l'atzar. (we) shall-make believelrely the Mary iníon the-chance 'We shall make Mary believe inlrely on chance.' b. *~arem creurelconfiar en l'atzar la Maria. (we) shall-make believelrely inlon the-chance the Mary 'We shall make Mary believe inlrely on chance.' (32) a. Farem creurelconfiar en I'atzar a la Maria. (we) shall-make believelrely iníon the-chance to rhe Mary 'We shall make Mary believe inlrely on chance.' b. *~arem creurelconfiar a la Maria en l'atzar. (we) shall-make believelrely to the Mary idon the-chance 'We shall make Mary believe inlrely on chance.' If we assume the causee to be subject to the HSP, which takes the rightward value in Catalan, it is expected to appear to the right of the interna1 arguments. The facts seem then to contradict such a prediction. However, note that, in spite of surfacing between the complex verb and the embedded object, the causee still c-commands the arguments within the embedded VP in both cases: (33) a. Farem creurelconfiar la Maria en si mateixa (we) shall-make believelrely the Mary inlon herself 'We shall make Mary believe inírely on herself.' (33) b. Farem creurelconfiar en si mateixa a la Maria. (we) shall make believelrely idon herself to the Mary 'We shall make Mary believe idrely on herself. In other words, it seems to be the case that the causee is not always generated to the right of the VP.10 Furthemore, this also seems to be true for subjects in main sentences: (34) Els nens van confiar tots en si mateixos. the children PAST rely all in themselves The children all rely in themselves.' If Sportiche (1988)'s analysis of floating quantifiers is correct, then it must be assumed that subjects may generate to the left of the VP in Catalan (see Sola (1991) for a proposal along these lines). To sum up, both theoretical and empirical evidence allows us to dispense with the HSP. But, how is word order in CC to be explained? I claim that the causee may freely generate either to the left or to the right of the VP. However, even though no parameter nor constraint regulates the position in which the causee may appear, the conditions on case assignment will rule out some structures. In other words, I am having the work usually done by parameters fixing word order done by independently motivated conditions on case assignment. As far as I can see this is a highly desirable result on conceptual grounds. Nevertheless, I am not going into this analysis in detail unti1 conditions on case assignment are introduced (see 2.2.2. below). So I will turn back to this issue later. 10 Guasti (p.c.) suggests that PP scrambling wouid explain the conirast without assuming free generation of the causee. However, it is difficuit to see how a scrambling-based analysis would account for the binding facts in (33). 2.2. Case Assignment and Verb Zncorporation The behavior of CC with respect to case is exemplified by the following sentences: (35) a. El Joan fa plorar la Maria. the John makes cry the Mary 'John makes Mary cry.' b. El Joan l'ha feta plorar. the John her-has made+fem cry 'John made her cry.' (36) a. El Joan fa menjar patates a la Maria. the John makes eat potatoes to the Mary 'John makes Mary eat potatoes.' b. El Joan li ha fet menjar patates. the John to-himlher has made eat potatoes 'John made himlher eat potatoes.' The causee is assigned accusative case (as cliticization and past participle agreement show) when the embedded verb selects one argument, whereas dative (as cliticization and lack of past participle agreement show), if more than one argument is selected. Three questions immediately arise: A. Why two different cases are involved? B. Why does the causee receive dative case? C. How are the number of cases and the number of arguments to be related? Before answering these questions, a further refinement of the analysis of CC is needed. 2.2.1. Verb Incorporation. Let us assume that the causative verb and the embedded verb form a complex verb (see Burzio (1986)). Let us further assume, following Guasti (1991,1992), that such a complex verb formation is in fact a subcase of Verb Incorporation. The formation of complex words in the syntax poses many interesting problems. For exarnple, if two case assigners become a new complex word, will their case assigning properties be added or not? In other words, how is syntax related to morphology? The nul1 hypothesis consists of applying to complex words the same principles that hold for simple words. That amounts to saying that cases cannot be added. Baker (1988) also adopts such a hypothesis and states the following principle: (37) Case Frame Preservation Principk (CFPP) A complex XO of a category A in a given language can have at most the maximal case assigning properties allowed to a morphological simple item of category A in that given language. As a result, no more than two cases, one accusative and one dative, are available in CC, because verbs in Catalan are able to assign only two cases (for the sake of simplicity I am leaving nominative case aside). The following sentence confirms such a prediction: (38) *~l Joan va fer comprar un cotxe als nens a la Maria. the John PAST make buy a car to-the children to the Mary 'John made Mary buy a car for the children.' The complex verb must license three NPs, but there is no way for a verb to assign three cases in Catalan. So then (38) is ruled out as a Case Filter violation (see 2.2.2).11 Note that this l1 The sentence in (38) sharply contrasts with that of (i), where the causee is not lexically realized. (i) El Joan li va fer comprar un cotxe als nens. the John to-himlher made buy a car to-the children 'John made himlher buy a car for the children.' The wellfomedness of (i) is unexpected. A solution based on Noun Incorporation will be provided in 3.1. constraint on complex words does not hold in languages that do not trigger Verb Incorporation (the Dutch example in (39b) is quoted from Koster (1987)): (39) a. John made Mary give him a book. b. Jan liet Peter Marie een boek geven. John let Peter Mary a book give 'John let Peter give a book to Mary.' Even though no verb is able to assign three accusative cases neither in English nor in Dutch, the sentences are perfectly grammatical: the embedded verb assigns accusative case to its complements and the causative verb to the causee (see section 4 below for a more detailed account of CC in Germanic languages). Verb Incorporation gives us a good answer to question C: the relation between the number of cases and the number of arguments follows from the CFPP, i.e., from the constraints that UG imposes on the formation of complex words. However, a crucial point is still to be explained: Why is Verb Incorporation necessary? Let us assume, following Larson (1988) and Li (1990), that verbs must be governed by Infl in order to assign case. If this assumption is correct, a principled account of Verb Incorporation is provided. The embedded verb cannot assign case, because it is not governed by Infl but by the causative verb. So then, the only way for it to license its complements is VI (note that the embedded verb is a closer governor that prevents case-marking of the complements by the causative verb). Once incorporated, Infl governs the complex verb, allowing it to assign case.12 2.2.2. Case Assignment. Questions A and B remain unanswered: l2 Guasti (1991, 1992) adopts a morphology-based anaiysis: it is the morphological subcategorization of the items that triggers Verb Incorporation. As far as I can see, her proposai is pedectiy compatible with the analysis just outiined. A. Why two different cases are involved? B. Why does the causee receive dative case? From a pre-theoretical point of view it is easy to see that causativization poses a case problem. Causativization increases the value of a predicate (the embedded verb), so the number of arguments becomes higher than the number of cases that such a predicate could assign. It is then necessary for the complex verb to have at least one case more than the embedded verb in order to avoid a Case Filter violation. I arn thus assuming the causative verb may assign both an accusative and a dative case. Such an assumption is far from being implausible if sentences like the following are taken into account: (40) a. El Joan (li) va fer els deures a la seva germana. the John (to-himlher) PAST do the homework to the his sister 'John did the homework for his sister.' b. El Joan (li) va fer un poema d'amor a la seva xicota. the John (to-himlher) PAST do a poem of-love to the his girlfriend 'John did a love poem for his girlfriend.' However, even though a dative case is available, it may not be assigned because there is no argument receiving also a theta role from the causative verb. In other words, if a dative case were assigned either to the the embedded object or the causee, it would be a violation of the Uniformity Condition because inherent case must be linked to a theta role and obviously this is not the case in CC. Nevertheless, a dative case is assigned to the causee in transitive CC. How is this paradox to be solved? We have two options: (a) dispensing with the Uniformity Condition or (b) assuming the dative case of causees not to be an inherent case. I am pursuing the latter. Let us consider the nature of the dative case that appears in CC (the accusative case poses no special problem, let us assume without justification that accusative case must be assigned under Secondly, the last resort rule poses a conceptual problem: condition (i) states that economy principles must be satisfied by the rule. But this is nonsense, for when the rule applies (DStructure), there is no way to evaluate how costly a derivation would be. The crucial point here is that economy principles are not conditions on rules but on derivations (see Chomsky (1992)). So condition (i) may also be dispensed with: (51) Last Resort Rule (Final Version) Discharge an unsatisfied case. The outputs of such a rule would be constrained by economy principles. The Earliness Principle makes the rule apply at D-Structure, preventing the embedded object from receiving a last resort case. Afterwards, the Last Resort Principle will evaluate whether the rule was necessary for the derivation to converge. In case it wasn't, the derivation would be tuled out. Let us see how this analysis accounts for the whole pattem of CC in Catalan. (52) a. El Joan fa bullir la llet. the John makes boil the milk 'John boils the milk.' b. El Joan fa sortir la Maria. the John makes go-out the Mary 'John makes Mary go out.' c. El Joan fa plorar la Maria. the John makes cry the Mary 'John makes Mary cry.' d. El Joan fa menjar patates a la Maria. the John makes eat potatoes to the Mary 'John makes Mary eat potatoes.' (52) e. *~l Joan fa comprar un cotxe als nens a la Maria the John makes buy a car tethe children to the Mary 'John makes Mary buy a car for the children.' f. *El Joan fa telefonar als nens a la Maria the John makes phone to-the children to the Mary 'John makes Mary phone the children.' (52a, c) behave identically: the only one argument receives accusative case under adjacency from the complex verb. The last resort rule cannot apply, othenvise a violation of the Last Resort Principle would follow. We have deait with (52d) before. If the causee were generated to the left of VP, it would receive accusative case, leaving the embedded object with no case because the application of the last resort rule would violate the Earliness Principle (the Case Filter would be satisfied at SStructure, instead of at D-Structure). If the causee is generated to the right, no problem arises: the embedded object receives accusative case from the complex verb and the causee dative by means of the last resort rule. (52e, f) are indirectly ruled out by the CFPP. We have seen that complex verbs must behave indenticaily to simple ones with respect to case. So then the case frame of the complex verbs in (52e, f) contains just one dative case, which is to be assigned either to the indirect object, leaving the causee with no case, or to the causee, rendering the indirect object with no case. In other words, the last resort rule has no unsatisfied case to borrow, both sentences being thus ruled out as Case Filter violations. Finally, two more sentences must be accounted for: (53) a. Farem creurelconfiar la Maria en l'atzar. (we) shail-make believelrely the Mary inlon the-chance 'We shall make Mary believe inlrely on chance.' b. Farem creurelconfiar en l'atzar a la Maria. (we) shail make believelrely inlon the-chance to the Mary 'We shall make Mary believe in I rely on chance. The difference here is that the causee is generated to the left of VP in (53a), but to the right in (53b). As we have seen above, accusative case is available in (53a), preventing the last resort rule to apply because of the Last Resort Principle. In (53b) instead, accusative case is not available (the adjacency requirement is not met), allowing the last resort rule to apply without violating any economy principle. 3. Cross-linguistical Variation in Causative Constructions Leaving aside the aspects related to morphological variation (see Guasti (1991, 1992)), causativization falls into two patterns cross-linguistically. In Type 1 causatives, the embedded object functions like the object of the whole sentence: it triggers object agreement -(54a)- and may raise when the sentence is passivized -(54b). The causee, however, cannot. On the other hand, in Type 2 causatives, it is the causee that functions like the object of the whole sentence, so then it triggers object agreement -(Sa)- and may raise when the sentence is passivized -(55b). (54) a. Anyani a-na-wa-meny-ets-a ana kwa buluzi. (Chichewa; from Baker (1988)) baboons SP-PAST-OP-hit-make-ASP children to lizard The baboons made the lizard hit the children.' b. Ana a-na-meny-ets-edw-a kwa buluzi ndi anyani. children SP-PAST-OP-hit-make-PASS-ASP to lizard by baboons The children were made to be hit by the lizard by the baboons.' (55) a. Mwa:limu@-wa-andik-ish-ize wa:na xati. (Chimwiini; from Baker (1988)) teacher SP-OP-write-make-ASP children letter The teacher made the children write a letter.' b. Wa:na wa-andik-ish-iz-a: xati na mwxlimu. children SP-write-make-ASPIPASS letter by-teacher The children were made to write a letter by the teacher.' 3.1. Type 1 Causatives 3.1.1. CC and Noun Incorporation. We have seen in section 2 how Type 1 causatives are to be analyzed: a last resort rule assigns an oblique case to the causee, which othenvise would not pass the Case Filter. However, not all languages having Type 1 causatives make use of this option. Let us consider the following contrast found in Southern Tiwa (example quoted from Baker (1988)): (56) a. I-'u'u-kur-'am-ban. lsS:2sO-baby-hold-CAUS-PAST 'I made you hold the baby.' b. *~'ude i-kur-'am-ban. baby IsS:2sO-hold-CAUS-PAST 'I made you hold the baby.' The only way to have a transitive CC in Southern Tiwa is by means of incorporating the embedded object to the complex verb. Baker suggests that NI is a licit means to satisfy the Case Filter (which he understands as an instance of Chomsky's (1986) Full Interpretation Principle). So it seems to be the case that even though something blocks case assignment in (56), the language has a way of avoiding a Case Filter violation without making use of a last resort rule, namely Noun Incorporation. Compare the sentences in (56) with the following Catalan ones: (57) a. El Joan li va fer comprar un cotxe als nens. the John to-himlher PAST make buy a car to-the children 'John made him/her buy a car for the children.' b. *El Joan fa comprar un cotxe als nens a la Maria. the John makes buy a car tethe children to the Mary 'John makes Mary buy a car for the children.' It is the presence of an overt causee what makes (57b) ungrammatical: the complex verb cannot assign case to the four arguments because of the CFPP, that bars out complex verbs assigning more cases than simple verbs could do. In (57a) the situation seems to be identical, but I claim that here the causee has incorporated to the complex verb, satisfying the Case Filter without receiving any case. So then the Catalan and the Southern Tiwa examples are similar: in both cases an argument incorporates to the complex verb in order to avoid a Case Filter violation.16 3.1.2. CC and Argument Demotion. But NI doesn't make an end of the ways languages having Type 1 causatives solve the case problem posed by transitive CC. Consider the following Malayalam example quoted from Baker (1988): (58) Amma kuttiyekk~ta aanaye swasnstam wiittil wecca QdI-icc-u. mother-NOM child-ACC by elephant-ACC self's house at pinch-make-PAST The mother made the child pinch the elephant at mother's/*child's house.' l6 Note furthemore that this analysis if wmbined with independently motivated economy principles may also explain why the clitic must always be unambiguously understood as the causee. Inwrporating the causee, we are dispensing with the last resort de, a wstiy (i.e., marked) way of assigning case. But, if instead of the causee, the indirect object incorporates. then the last resort rule is needed anyway in order to assign case to the causee, yielding a more wstly derivation that is ruled out by the Last Resort Rinciple. That is, the unambiguity of (57a) is a direct result of economy principles operating in UG. The reflexive swa- 'self', which always takes a subject as an antecedent, cannot be bound by the causee but by the main subject. This is unexpected under the analysis of CC developed in section 2. Larson (1988) proposes that a theta role assigned by a head may be assigned to an adjunct of that head. He calls such a mechanism Argument Demotion. I claim that Argument Demotion is at work in Malayalam CC, yielding the described binding properties. The causee, being realized as an adjunct, cannot bind the reflexive nor provide it with a BT-compatible index within V-x. The reflexive must thus be bound by the main subject. However, binding phenomena are not the crucial point here. What is really important is that Argument Demotion may be regarded as a way to solve the case problem posed by transitive CC: (59) Amma kuttiyekkwta annaye ~41-icc-u. mother-NOM child-ACC-by elephant-ACC pinch-make-PAST The mother made the child pinch the elephant.' Note that as (60) proves the complex verb is able to assign a dative case (it must be pointed out that inanimate direct objects receive nominative in Malayalam): (60) Acchan (ammayekkonta) kuttikka pu&kam kotupp-icc-u. father (mother-ACC-by) child-DAT book-NOM give-make-PAST The father make the mother give the book to the child.' The contrast between (59) and (60) may be accounted for if we assume that Malayalam lacks the last resort rule formulated in (51). Then, even though a dative case might be available in the case frame of the causative verb, it could not be discharged anyway. The only way for transitive CC to be grammatical is thus Argument Demotion. Such a mechanism is also found in languages having a last resort rule (the Turkish example in (61a) is quoted from Comrie (1981)):17 (61) a. Ali Hasan-a mektub-u miidiir tarafindan goster-t-ti. Ali Hasan-DAT letter-ACC director by show-make-PAST 'Ali made the director show the letter to Hasan.' b. Gianni fa scrivere una lettera a Mimada Maria. John makes write a letter to Mima by Maria 'John makes Maria write a letter to Mima.' 3.1.3. CC and the Limits of Markedness. Finally, let us take into account a possibility: What would happen to transitive CC in a language having none of the mechanisms just described? The answer is quite simple: such a language would not allow transitive CC. Berber is one of these languages (exarnples quoted from Baker (1988)): (62) a. Y-ss-jen Mohand arba. 3sS-CAUS-sleep Mohand boy 'Mohand made the boy sleep.' b. *Y-SS-wt wryaz aggzin i-wrba. 3sS-CAUS-hit man dog to-boy The man made the boy hit the dog.' c. *Y -SS-icr wryaz tacurt i-wrba. 3sS-CAUS-steal man ball to-boy The man made the boy steal the ball.' l7 Neither Catalan nor Spanish easily allow Argument Dernotion (i.e., the faire-par construction). I cannot see any reason distinguishing these languages from Italian in that point. A detailed andysis of Romance CC dealing with the kind of variation just pointed out is still to be done. Obviously, such au analysis is beyond the scopeof this paper. 3.2. Type 2 Causatives Consider the following sentences involving Type 2 causatives (both the Swahili example in (63a) and the Chichewa one in (63b) are quoted from Baker (1988)): (63) a. Musa a-li-m-pik-ish-a mke wake chakula. Musa SP-PAST-OP-cook-make-ASP wife his food 'Musa made his wife cook food.' b. Catherine a-na-mu-kolol-ets-a nwana wake chimanga. Catherine SP-PAST-OP-harvest-make-ASP child her com 'Catherine made his child harvest the com.' It has been suggested above that the main difference between Type 1 and Type 2 causatives had to do with case, namely with the ability of assigning two structural cases. The problem Type 1 causatives had with case was due to the fact that one of the cases the complex verb was able to assign was an inherent one. Type 2 causatives do not have this problem, because all the cases the complex verb is able to assign are structural (for the sake of concreteness, henceforth I will call ACCl the first accusative case and ACC2 the second). Obviously case assignment conditions applying in Type 1 causatives are supposed to hold in Type 2 causatives: ACCl will be assigned under adjacency at S-Structure. But what happens with ACC2? I will assume that ACC2 is assigned under Spec-Head Agreement in the Spec of the main VP, henceforth [Spec,VP] (see Johnson (1991)). The interaction of these two case assignment conditions with the free generation of the causee either to the left or to the right of the embedded VP brings us the correct results. If the causee is generated to the right of the VP, it will never be adjacent to the complex verb at S-Stmcture, so it will not receive ACCI. It must raise to [Spec,VP] in order to receive ACC2 by means of Spec-Head Agreement. The embedded object instead will be adjacent to the complex verb at S-Structure, receiving ACCI. Aftenvards the complex verb will raise to Infl, following the correct order. If the causee is generated to the left, it will be adjacent to and receive ACCl from the complex verb at S-Stmcture. The embedded object must then raise to [Spec,VP] in order to get ACC2, but this movement would be blocked by Relativized Minimality: the causee would be a typical potential governor in an A-Spec for the trace left by the embedded object.18 Thus, generating the causee to the left of the VP would always yield an ungrammatical result. 4. Some Remaining Problems In spite of being very appealing and fruitful, this analysis has to dea1 with many problems. The first one has to do with the binding properties distinguishing type 1 from type 2 causatives. It has been shown in 1.1 that the causee was able to license an anaphor in object position in Cataian and Italian (but see note 3), languages that have type 1 causatives. However, Li (1990) quotes the following example from Yupik, a language having type 1 causatives, where an anaphor in object pition cannot be bound by the causee but by the matrix subject: (64) Arna-m annga-ni tuqute-vlar-aa ing'u-mun. (Yupik) woman-ERG brother-REFL kill-CAUS-3sS13sO guy-DAT The woman made the guy kill her own/*his own brother.' Li (1990) argues that such a behavior is due to the fact that a relation of symmetrical ccommand holds between the causee and the object, yielding a binding violation in case they share the same index. (48) is a mistery for my analysis. l8 Such an analysis apparently would fai1 to explain why the embedded object may raise to [SpecJP] when the complex verb passivizes. However. it might be argued that a different kind of movement would be involved here. narnely. A'-movement (see Diesing (1990) and Koopman and Sportiche (1991). where it is argued that [SpecJP] may be either an A or an A'-position). So then, the causee, an A-specifier would prevent the embedded object from moving to an A-position ([Spec.VP]), but not to au A'-position ([SpecJP]). Another problem has to do with CC in Germanic. Let us consider the relevant English and Dutch examples (=(39)):19 (65) a. John made Mary give him a book. b. Jan liet Peter Marie een boek geven. John let Peter Mary a book give 'John let Peter give a book to Mary.' It has been suggested that the embedded verb was unable to assign case because it was not governed by Infl. This was assumed to trigger Verb Incorporation in CC (besides any morphological factor). But, even though no Verb Incorporation applies at all in (65), the internal arguments of the embedded verb receive case. Moreover, (65) poses a problem for the CFPP, because there exists no verb assigning three structural accusative cases. Obviously Verb Incorporation cannot be proposed for CC in Germanic languages, othenvise the CFPP would be violated. I claim that the solution to this puzzle should be based on reanalyisis. Baker (1988) argues that reanalysis may be seen as abstract incorporation, that is as a coindexing relation between two heads that, even though it satisfies the conditions on head movement, involves no movement at all (possibly for morphological reasons). If we assume the sentences in (65) to l9 Dutch also presents CC very close to the Romance ones: (i) Marie liet de taart proeven aan Peter. Mary let the pie taste to Peter 'Mary let Peter taste the pie.' However, as Koster (1987). from where all the Dutch examples are quoted, points out, it has a very limited range. In fact, it can only function with stative verbs of wgnition. Verbs of action are not allowed: (ii) *~ij lit de boerderij bezoeken aan haar ouders she let the fann visit to her parents 'She let her parents visit the fann.' See Koster (1987:302fO for a detailed description.