THE CASE OF SUBJECTS IN THE ROMANCE CAUSATIVE *

One characteristic of causative verbs in French, Spanish, and Italian is that the subject of the embedded verb appears to the right of its object (if it has one). A second distinguishing property is the Case that appears on the subject embedded under the causative verb. It is always either accusative or dative, depending on the transitivity of the embedded verb. An explanation for these facts need not assume rules of Verb Phrase (VP) preposing, the preposing of any projection of V, rules involving the internalization of an external argument, or that the embedded subject is an argument of the matrix verb. Rather the causative facts can be explained within recent hypotheses that subjects are base-generated within the maximal projection of V. Specifically, it is argued that the subject appears inside the embedded VP complement of the causative, and is Case-marked by the embedded verb. It is suggested that the embedded verb is endowed with an extra Case-marking ability, transmitted from the causative verb. This account explains the distinct patterns of Case-marking that appear on the embedded subject for transitive, unergative, and lexical dative verbs in both the "faire"-infinitive and the "faire par" constructions. (JL) *********************************************************************** Reproductions supplied by EDRS are the best that can be made from the original document. *********************************************************************** THE CASE OF SUBJECTS IN THE ROMANCE CAUSATIVE*


Introduction
One characteristic of the Romance causative verbs is that the subject of the embedded verb appears to the right of the embedded verb and its object (if it has one). This word-order fact is illustrated in (1) -(3) for French, Spanish, and Italian. In addition to the causative verb 'make', this construction may occur with the verb 'let' and the perception verbs 'see ', 'hear', 'watch' etc. (1) Fre:ich a. Jean a fait manger les pommes a Marie.
'Jean made Marie eat the apples.' A second distinguishing property of the Romance causative construction is the Case that appears on the subject embedded under the causative verb. It is always either accusative or dative, depending on the transitivity of the embedded verb. This can be seen most clearly when the subject is pronominal because clitic pronouns distinguish accusative from dative, as illustrated in (4) (6). If the embedded verb has an accusative Case-marked object, the embedded subject appears in dative Case. If, however, the embedded verb has no accusative Casemarked object, then the embedded subject appears in accusative Case. The purpose of this paper is to show that an explanation for these facts need not assume either rules of VP preposing, or the preposing of any projection of V (as in Baker, 1988a;Burzio, 1986;Kayne, 1975;and Rouveret & Vergnaud, 1980). It also need not assume rules involving the internalization of an external argument (Di S,-3ullo & Williams, 1987;Zubizarreta, 1985;1987), or the assumption that the embedded subject is an argument of the matrix verb (Bordelois, 1988). The causative facts can be explained straightforwardly within the recent hypothesis that subjects are base-generated within the maximal I will argue that the Romance causative, the so-called faire infinitive construction, has the phrase structure representation c:,iven in (7); that is, the causative verb takes a VP complement rather than a full CP or even TP complement.3 This fact, along with the subject under VP hypothesis, will explain the position and Case-marking of the embedded subject.4 (7) V'

Evidence for VP Complement of Causatives
It is not new to posit that the Romance causative takes a VP complement. This notion has been proposed in various forms in the works of Burzio (1986), Di Sciullo & Williams (1987), Kayne (1989), 82 Marantz (1985), Rosen (1989), and Zubizarreta (1985;1987), for example. Several pieces of evidence, both old and new, suggest that the complement to the causative is correctly identified as a VP. They include the lack of inflectional material in the clausal complement of the causative verb, facts concerning clitic climbing, and a distinction in the ability to take VP-external subjects.
2.1 Lac'. of Clausal Negatioe on the Embedded Verb. One source of evidence that the complement of the Romance causative is a VP rather than a full clausal complement is to show that the material that ordinarily goes in the functional projections dominating VP cannot exist in the complement to a causative. Finding a test is difficult, given that much of the inflectional material is absent in any infinitival complement, given that they have no independent tense and generally do not allow modals. A comparison of (8a) with the tensed complement (8b) in English illustrates this point. However, clausal negation can appear in an infinitival complement, as the grammaticality of (8c) indicates (8) a. *We believe John must/can to be intelligent. b. We believe (that) John must/can be intelligent. c.
We believe John not to be intelligent.
It is generally assumed that clausal negation forms a maximal projection falling within the functional inflectional categories (cf. Pollock, 1989, and subsequent work on clausal functional categories). Thus, clausal negation provides a test case for the VP nature of the complement to the Romance causative. The VP analysis of the complement to the Romance causative constructions leads to the prediction that negation will not appear in the the embedded clause. Indeed, this is the case, as the examples in (9) (11)  However, under certain circumstances, negation on the embedded phrase is allowed. For example, speakers will marginally accept the examples in (10) if no, the negative element is stressed. And in recent work, Reed (1990b) argues that negation can appear on the clause embedded under the French causative, as in (12a), from Reed (1990b). However, my informants claim that these are Quite marginal, and that it is not clear that they have the semantics of clausal negation. In particular, wherever negation can appear embedded under the causative, a VP adverbial can be substituted for the negation (12b).
(Through his incantations,) the sorcerer made him not feel well for days.' b. Le sorcier l'a fait totalement se sentir bien pendant des jours. 'The sorcerer made him totally feel well for days.' In addition, as Rochette (1988;p.c.) has pointed out, double negation is possible in infinitival TP or CP complements, the first being clausal negation appearing within the functional categories as usual, and the second being VP modification. All these facts taken together, it seems likely that the negation in the clause embedded under the causative is equivalent to the second of these, i.e. a VP adjoined adverbial modifier. Thus, we can maintain the generalization that clausal negation is impossible under the causative. This follows from the analysis of the causatives as taking a bare VP complement. Since clausal negation forms a functional projection outside of VP, it will never appear in the complement to the Romance causative. 84 2.2 Clitic Climbing. Another phenomenon related to the Romance causative construction is clitic climbing. In this construction, an object clitic appears on the matrix verb rather than on the embedded verb which selects it, as illustrated in (13)  Since clitic climbing is limited to this and similar constructions, it appears that a clitic may only climb out of a VP complement. Assuming, following Kayne (1989), that clitics appear on one of the inflectional heads (Tense, for example), the clitic will be forced to climb out of a VP complement into the inflectional material. That is, if the embedded clause has no inflectional material, then the clitic must appear in the matrix clause. Thus, clitic climbing out of the complement of a causative follows from the assumption that the causative takes a VP complement. Level and Individual Level Predicates. Finally, Kratzer (1988) and Diesing (1988) argue that a semantic classification of predicates first pointed out in Carlson (1977) has specific syntactic ramifications for the base position of subjects. The semantic classification distinguishes predicates as either stage level or individual level. A stage level predicate denotes either an action or a temporary property of the subject, whereas an individual level predicate denotes a relatively permanent property of the subject. Kratzer argues that the subject of a stage level predicate is base generated inside the VP, and subsequently may move out of the VP. In contrast, the subject of an individual level predicate must be outside the VP at all levels of the derivation. If the Romance causatives take a VP complement, then this leads to the prediction that they will be compatible only with stage level predicates; if it is true that there are no functional projections dominating the complement to the causative, then there will be no place to project the subject of an individual level predicate. Indeed, the Romance causatives can take stage level predicates as their complements, as the examples in (17)  In addition, one can show that (20) -(22) are not judged ungrammatical solely because of the pragmatics --that it seems difficult to make someone like or know something. The Spanish examples in (23) indicate that the same meaning can be conveyed with a causative verb and an individual level predicate in the subordinate clause, but only when the complement is in the form of a full tensed CP. The sentences in (23) are syntactically well-formed, because the subordinate clause is a tensed CP with an overt complementizer. In the full CP clause, there is a position in the specifier of one of the inflectional projections (outside the VP) for the subject to be generated. (The contrast in the English examples in footnote 6 provides more evidence against a pragmatic explanation. With the causative verb 'make' it is perfectly grammatical to embed an individual level predicate, but the same individual level predicate is bad under the causative use of 'have'.) Thus, an individual level predicate can only be projected within a full CP or TP clausal complement, and never within a VP complement. Notice that in the examples in (23), the embedded subject appears to the left of the embedded verb. I take this to be the Spec of TP position, as diagrammed in (23c). This suggests that in general any CP or TP complement will be compatible with an individual level predicate. Compare the French faire construction in (20) with (24), in which, once again, similar semantic informb lion is conveyed using the verb forcer, which clearly takes a full CP complement.

Stage
(24) a. Jean a force Marie a aimer la cuisine francaise.
'Jean forced Marie to like French cooking.' b. Jean a force Marie a connaitre le francais. 'Jean forced Marie to know French.'

TP Complement Verbs and the Stage Level/Individual Level
Distinction. The verb 'to let' and the perception verbs in Romance may enter into one of two constructions. One is the causative construction, which I will call the VP complement construction. In the other, the verb takes a full TP complement. The contrast between these two constructions is exemplified in the sentences in (25) and (26) for the perception verb and in (27a,b) for French laisser ('let'). The most obvious distinguishing characteristic is the position of the embedded subject argument. In the VP complement construction, the subject appears to the right of the embedded verb and its object. In contrast, the subject in the TP complement construction appears between the two verbs.  (26) and (27b) must contain at least one inflectional projection. In these cases, material that generally appears within the inflectional categories can appear in the complement, providing evidence that the complement in these sentences is not a bare VP. First, the position of the embedded subject between the two verbs is indication that this is not to be treated as the same as the VP complement construction.
Second, clitic climbing out of the embedded object position is impossible in this construction, as the examples in (28) (30) show.
Assuming that clitics move to the closest governing T position, the clitics in (28)   Third, Guasti (1989) has provided evidence for the existence of inflectional functional categories under the perception verbs, using the analysis of V-to-Agr-to-T raising of Pollock (1989), and the relative I osition of the embedded verb with respect to that of adverbials and negation. The structure of the inflectional system that Pollock proposes is given in (31). Under his analysis, T(ense) takes NegP as its complement; Neg takes AgrP; and Agr takes VP. When there is no negation, T directly takes AgrP as its complement. Each phrase, of course, has its own Spec position, which, for the sake of simplicity, is not included in the diagram here. Poii.,ck argues that in Romance the verb raises up to Agr and subsequently to T in order to receive its inflectional features. One can determine the position of the verb by the relative order of the verb and negation, and also the order of the verb and VP adjoined adverbials.
Assuming Pollock's structure for the inflectional categories, Guasti shows that the verb embedded under a perception verb undergoes V raising just like any verb in the language. In (32a), (33a) and (34a), the verb raises up around a VP adjoined adverbial at least as far as Agr. Notice that in Italian and Spanish ((32b) and (33b)), verb raising is obligatory, though it may be optional in French (34b). All the sentences in (32) (34)  Guasti further shows that in Italian and Spanish, the verb embedded under the perception verb must appear before the negative element (piii or mas 'more' in (35) -(36)). In French, the verb remains after negation, which is precisely what one would predict given Pollock's analysis of V raising in French. The French infinitival only raises as far as Agr, and not to T, unlike raising in Italian and Spanish infinitives which go all the way up to T. The data, again from Guasti (1989: 3-4), are given in (35)  endendu les enfants ne pleurer plus. b. J'ai endendu les enfants ne plus pleurer.
'I have heard the kids no cry more/no more cry.' Examples such as those in (32) -(37) indicate that the complement to the perception verbs can be realized as a TP. If there were just a VP complement, ,.o V raising could take place around an adverbial in the embedded clause, and there would be no place for the embedded negation. I propose, then, that the perception verbs and the verb laisser may take a TP complement. If the embedded subject of a stage level predicate is base-generated under the VP, it will move up to the Spec of TP position, just as in any full clausal construction. And, as will be shown below in Section 4, it will receive Case directly from the matrix verb, by exceptional Case marking.
This leads to the prediction that the perception verbs and laisser should also be able to take individual level predicates. Recall that the subject of an individual level predicate is base-generated outside the VP. In the VP complement construction of the causatives, the embedded complement could not be an individual level predicate. Given the current analysis, one would expect that individual level predicates would be incompatible with the VP complement construction of laisser and the perception verbs, but that the TP complement construction would accept individual level predicates. When informants were asked, the predicted pattern of grammaticality judgements resulted. Data on French laisser appear in (38) and (39). The TP complement construction is exemplified in (38a) and (39a), and the VP complement construction in (38b) and (39b), as one can tell from the position of the embedded subject. The individual level predicate embedded under laisser is consistently better in the TP construction than in the VP construction. Similar data are obtained in Spanish for the perception verbs. Again, the TP complement construction is exemplified in (40a) and (41a), and the VP complement construction in (40b) and (41b). In each case, the individual level predicate is significantly better in the TP construction than in the VP construction. Thus, I conclude that laisser and the perception verbs can optionally take a full TP complement, and that the embedded subject appears in the Spec of TP at S-structure, and indeed originates there if the embedded clause denotes an individual level predicate. And, as will be argued below in Section 4, the embedded subject is then Casemarked by the matrix verb in an ECM construction: The conclusion drawn from the previous section is that the Romance causative verb, faire/fare/hacer, takes a VP complement, with the subject of the embedded verb resting in its base position within the VP. Laisser and the perception verbs sometimes take a VP and sometimes take a TP complement. This section will establish that in the VP complement construction, the embedded subject is directly Case-marked by the embedded verb, but that the Case assigned to the subject is transmitted from the matrix causative verb. By contrast, in the TP complement construction, the matrix verb directly Case-marks the embedded subject in its Spec of TP position (ECM). I will presently sketch out exactly how the Case-marking applies.
First, the Romance causative verb is an accusative Case assigner. This is supported by the fact that the causative verb can take an accusative Case-marked NP complement: 'Gianni made a run.' An additional piece of evidence that these verbs have an accusative Case to assign comes from similar verbs that optionally enter into the causative construction. These verbs were introduced in section 2.4, where evidence was presented indicating that laisser and the perception verbs are exceptional Case markers (ECM); thus they must be Case-markers. In sentences like (46) (48), the perception verb takes a TP complement. In such sentences, the embedded subject receives accusative Case directly from the matrix verb in its S-structure position in the Spec of TP (cf. also Guasti (1989) for a similar proposal). Cliticization of the embedded subject out of the Spec of TP subject position shows that it always receives accusative Case.8 In these examples, the embedded object is cliticized to the embedded verb simply to ensure that we have the TP complement construction (recall that climbing of the object clitic is impossible out of an embedded TP). The existence of the ECM construction further indicates that these verbs are accusative Case assigners.9 The perception verbs also enter into the VP complement construction, just like the causative verbs (cf. the examples in (25) above). In addition, the Case-marking patterns in the VP complement construction of the perception verbs are identical to that of the causative verbs. Thus, I conclude that the causative and perceptior verbs are alike in being accusative Case assigners.
Returning to Case assignment in the causative construction, let us assume that accusative Case is assigned under adjacency, as suggested in Stowell (1981). In addition, it appears that the verb in Romance assigns its Case rightward. The adjacency requirement on Case assignment and directionality of Case assignment taken together mean that the causative verb can only assign its Case to the VP complement, the constituent directly adjacent and to the right of the causative verb. However, a VP neither requires Case, nor is it capable of bearing Case. Since there is no other constituent that is capable of bearing Case, the Case feature is transmitted from the embedded VP to its head, the embedded verb (as suggested, for example, in Rouveret & Vergnaud, 1980). This Case transmission process is illustrated in (49) with the dotted line. (49) faire Evp Iv reparer NP NP Case Let us say, further, that Case transmission is only possible to a head that is capable of assigning accusative Case. The result will be that Case 95 transmission cannot apply in either a CP or a TP complement. The head of CP (C) is not a Case assigner at all, and the head of TP (T) is not an accusative Case assigner. Therefore it is only possible to transmit Case to the category V.11 The result of Case assignment by the causative verb is that the embedded verb always has one extra Case to assign. Therefore, if the embedded verb has n Cases to assign ordinarily, it will functionally have n + 1 Cases to assign when embedded under the causative.
3.1 Case to the Embedded Subject. There is a generalization across languages in the Case-marking patterns within causative constructions. If a language allows double accusatives in general, then the embedded subject and the embedded object will both receive accusative Case; but if a language does not allow don' _-: accusative, then one VP-internal NP will receiv-accusative Case and the other will receive some other Case, like dative (Aissen, 1979;Baker, 1988a;Rosen, 1989). This generalization strongly suggests that one verb is responsible for assigning Case to all the arguments of the embedded verb, including the subject. As an example of a double object language, Kinyarwanda allows two accusative Case-marked bare NPs within one VP. Thus, for a triadic predicate like 'give', the two internal arguments both behave as if they have structural accusative Case (cf. Kimenyi, 1980;Baker, 1988a). In the causative construction, the embedded object and the embedded subject also appear as bare NPs, and both behave as if they have accusative Case in the same ways that the double object verb does.
In the examples (50) from Kimenyi (1980: 31, 164), one can see that the pattern of Case-marking appears to be the same in the double object construction of (50a) and the causative construction of (50b).12 Baker (1988a) shows that either of the bare NPs in sentences like (50a) and (50b) can become the subject of a passive, a property only of objects that receive accusative Case in active clauses.
(50) a. Umugabo y-a-haa-ye umugore igitabo. man he-past-give-asp woman book 'The man gave the woman the book.' b. Umugabo a-ra-som-eesh-a abaana ibitabo. man he-pres-read-cause-asp children book 'The man is making the children read the books.' Unlike Kinyarwanda, Romance verbs never assign two accusative Cases. If a verb has two structural Cases to assign, they are 96 always accusative and dative, as in (51a) in which the verb is a triadic predicate. And like the simple verbs with two internal arguments, in the causative construction, the Cases to the object and subject are also accusative and dative (51b). Other languages with similar Case patterns include Japanese, Turkish, and Malayalam (cf. Rosen (1989) for further discussion of the Case patterns in simple VPs and in the VPs embedded under the causatives). In the Romance causative construction, the Cases that are assigned within the embedded VP are exactly those structural Cases that a verb in the Romance languages generally assigns --accusative when the verb is transitive, and accusative and dative when the verb is ditransitive. So, the object and subject arguments of the embedded verb are Case-marked as if they were both object arguments; they are both Case-marked by the verb, even though one argument is clearly a subject (that is, an external argument projected outside the V`).
The conclusion one must draw from this discussion is that one verb simultaneously gives Case to both the object and the subject arguments inside the VP. The mechanism I propose for this is Case transmission. The causative verb has a Case to assign, but its VP complement cannot bear Case. Therefore, the Case is transmitted down from the VP to its head V. The V then Case-marks its arguments within its own maximal projection. The embedded subject happens to rest within the VP, and therefore will be Case-marked just like the other complements of the V.
If the VP embedded under the Romance causative has two NPs (the erb's object and its subject), then they will be assigned accusative and dative Case respectively. The Case assigning process is illustrated in (52). A verb like reparer 'repair' has one structural Case of its own to assign. Faire also has a Case to assign, but this Case feature is transferred down to reparer. Reparer will subsequently have two Cases to assign. The analysis assumes that the NP adjacent to the verb will receive accusative Case, and that the adjacency requirement does not hold for dative Case (cf. Stowell, 1981). Since the direct object is generated adjacent to the verb, within V', it receives accusative Case j 9 97 from the verb. This leaves dative Case to be assigned to the subject, which is generated outside the V'.13 (52)  For unergative and unaccusative predicates, the embedded verb has no Case of its own to assign. It acquires one Case from the matrix causative verb, resulting in one Case to assign. The sole argument receives accusative Case, as assigned by faire via the embedded verb. Implicit in this account is the assumption that accusative Case is always assigned first; thus if the verb has only one structural Case to assign, it will always be accusative. The result of the Case-assignment process is that the Case realized on the embedded subject is directly determined by the transitivity of the embedded verb. This prediction is borne out in the simple cases. In addition, as shown in Burzio (1986: 241, 243), and as pointed out in Rouveret and Vergnaud (1980), a clause with two datives is not accepted by most speakers. To the extent that they are accepted by speakers (some speakers accept (54) at least marginally, according to Burzio), the outermost dative phrase is unambiguously interpreted as the subject.
(54) Jean fait porter une lettre a Marie a Paul. 'Jean made Paul take a letter to Marie.' What happens, though, if the base-generated order of arguments is at odds with the adjacency requirements of Case 98 assignment? This occurs in verbs that are lexically marked as taking dative objects (inherent Case). The object is an internal argument, and therefore is base-generated within the V'. The subject is base-generated outside V'; it must receive accusative Case from the verb, but this configuration dc.es not meet the adjacency requirement on accusative Case assignment.: Case Acc A concrete example will help at this point. The verb telephoner is lexically marked as taking a dative object, as the sentence in (56a) shows. When embedded under a causative, we actually find that the lexically marked dative phrase appears at S-structure after the accusative subject. The order of the arguments of telephoner in (56b) is opposite of that expected by the base-generation of the internal argument inside V', and the external argument outside V' as diagrammed in (56c). How does one get the internal argument of the embedded verb to appear outside its external argument?

Pierre a fait [vp [v telephoner a Marie J Jean I
In (56b), the subject of telephoner is Jean, and the internal argument is Marie. The expected D-structure of for (56b) is given in (56c), with the internal argument projected inside the V' and the external argument projected outside V'. I suggest that the sentence in (56b) is derived from the D-structure given in (56c) by extraposing the internal argument PP. Jackendoff (1977: 75) describes instances in which argument PPs appear to the right of adverbial phrases (as in 'John gave the beans quickly to Bill'), and suggests that this is derived by a process of PP extraposition. In the VP complements to causatives, PP extraposition will be driven by Case Theory. Assignment of accusative Case requires string adjacency, and therefore the embedded subject must be adjacent to the verb in order to receive accusative Case 99 from it. This will force the PP to move out from between the verb and the subject.
In addition, the present theory predicts that the subject of an intransitive verb under the causative will appear in accusative rather than dative Case. As (56b) shows, this is correct. The verb telephoner has one inherent Case and no structural Cases to assign. Faire assigns Case to the embedded VP, and when the Case feature is taken over by telephoner, this verb has one structural Case to assign. The subject will then receive accusative Case.
33 Faire par and Case Assignment. One final piece of evidence for the current approach to Case assignment in the causatives comes from the faire par construction. In faire par: what seems to be the embedded subject appears after the preposition `by' (par in French, por in Spanish, and da in Italian). Unlike the faire infinitive construction, the embedded subject is always either realized in a by phrase, or is completely absent.14 Some examples of the faire par construction are provided in (57).

Italian
Giovanni ha fatto riparare la macchina (da Paolo). 'Giovanni made Paolo/someone repair the car.' Zubizarreta (1985;1987) has argued that in the faire par construction, the external argument of the embedded verb has been suppressed as in passivization. Because the external argument is suppressed, it cannot appear as an argument, but only as an adjunct. Because it is an adjunct, it is correctly predicted to be optional. Zubizarreta shows that the by phrase in the faire par construction does not behave like an argument. The evidence she brings to bear on the issue is based on the ability of the logical subject to bind the possessive pronoun sa. The examples in (58) show that by phrases in general cannot bind sa-it must be bound by an argument. In 2,2 100 the sentences in (59), one can see that faire infinitive contrasts with faire par in the ability of the embedded 'subject' to be the antecedent for sa. The contrast is fully expected within Zubizarreta's passivization account, because sa maison can only be referentially dependent on Jean if Jean is an argument. The par phrase of (59b) is not an argument, and therefore cannot bind the anaphor sa. However, in (59a) can does count as an argument, and therefore may bind sa, as expected. The fact that the by phrase behaves identically in the passive in (58b) and in the faire par in (59b) is good indication that they should be treated as the same phenomenon, and that it is not an argument in either case.
These examples are from Zubizarreta (1985: 270, 263). Given the evidence that the by phrase is not an argument, one can assume that the faire par construction is derived by a process of suppression, much like passivization, applying to the embedded verb. In traditional GB accounts, suppression of an external argument applies in the argument structure component, and takes away the accusative Case assigning abilities of the verb (Burzio, 1986); this forces the object to move to subject position in the syntactic component in order to obtain Case from T.15 Returning to faire par, suppose that the external argument of the embedded verb is suppressed, effectively taking away its Case. This means that with respect to Case assignment, the embedded verb will act like an intransitive verb. Except in this instance, it has an object argument. When faire assigns Case to its complement VP, and the Case feature percolates down to the head V, the embedded verb obtains one structural Case. This Case is assigned to the embedded verb's object. Thus, the verb whose external argument was suppressed has no Case of its own to assign, but it can transmit the Case of the matrix causative verb. Further, because the external argument of the embedded verb is suppressed, there is no subject inside the embedded VP. The logical 'subject' may only appear as an adjunct by phrase. Because the embedded verb obtains an extra Case to assign from the 1J 101 causative verb, the object of the embedded verb is expected to receive accusative Case even though the embedded verb has no Case of its own to assign. Further, there is no subject argument requiring Case. The expected Case pattern, then, is an accusative marked object, and an optional adjunct by phrase, as diagrammed in (60). This is exactly the Case-marking pattern found in the faire par construction.

Case Marking in the TP Complement Construction
In section 2.4 I argued that verbs like laisser and the perception verbs in Romance may enter into either a VP complement construction or an infinitival ECM construction. In the ECM construction, these verbs take an untensed TP complement. The embedded subject appears at S-structure in the Spec of the embedded TP, where it is string-adjacent to, and governed by the matrix verb. In such a configuration Exceptional Case Marking (ECM) is expected to apply; the matrix verb should Case mark the embedded subject. Thus, the Case-marking pattern should be quite different for the same verbs when they enter into the TP complement construction and when they enter into the VP complement construction. I will now show that this is correct.
An important argument for the ECM analysis of the Romance perception verbs is the Case that is realized on the embedded subject. In the ECM construction, an embedded pronominal subject may cliticize onto the matrix verb; in the pronominal form, one can actually tell whether the subject is accusative or dative. The TP complement is distinguishable from the VP complement construction by the Case realized on the embedded subject clitic. In the VP complement construction illustrated in (61) (62) The examples classified as the VP complement construction appear to be interchangeable with those classified as the TP complement construction. Thus, it is necessary to ensure that the examples just given are classified correctly. In fact, there is a correlation between the position of the object clitic and the Case that is realized on the embedded subject. Recall that clitic climbing may only apply out of the VP complement. If the object clitic climbs into the matrix clause, the embedded subject receives dative Case. This is illustrated in (65). The sentences in (65) represent the VP complement construction, in which the embedded subject receives Case from the embedded verb; because the embedded verb has an accusative Casemarked object, the subject must be dative. If, however, the object clitic remains within the embedded clause, the subject receives accusative Case, as illustrated in (66). The sentences in (66) represent the TP complement, in which the subject receives Case directly from the matrix verb; in this case, the embedded subject must be accusative even though there is an accusative Case-marked object. Further evidence for the ECM analysis comes from passivization and the interaction of passive and clitic placement. It is well-known that passivization can apply to an ECM verb, taking away the Case assigning ability of the matrix verb, and thus the Case to the embedded subject. This subject will then move to the matrix subject position, where it receives nominative Case from the matrix T, as in the English sentence (67).
(67) Johni was believed ti to be intelligent.
Passivization also applies to the Romance perception verbs in the ECM construction (with some lexical idiosyncrasies in the ability of a given verb to undergo passivization, as well as some differences across speakers in acceptability). This is illustrated in (68) (70), with (69a) from Burzio (1986: 300 3. The literature contains some differences concerning the actual inflectional functional category labels. Some argue that the top functional category is TP (for Tense Phrase), some use IP; and and Belletti (1988) analyses the inflectional categories with Agreement Phrase (AgrP) above TP. For purposes of this paper, the actual node labels are irrelevent. What is crucial here is that there be at least one functional head of S. I will continue to use TP as the highest projection, taking an AgrP, and an optional NegP in between.
4. Given the position of the embedded subject in causatives, it might be tempting to analyze free inversion in Romance, as in (i) (examples from Burzio, 1986), in the same manner. However, even though it is possible that the inverted subject position is actually its base position, there are crucial differences between the subject under a causative and the subject in free inversion. As Burzio (1986) has argued, an inverted subject forms a chain with the Spec of TP subject position, and receives nominative Case from T via this chain. In the causatives, I will argue that the VP-internal subject is Case-marked by V, and not by T. 'Giovanni has spoken.' 5. The Italian and Spanish equivalents of (16b) are grammatical, however. The matrix verb 'want' in Italian (volere) and Spanish (quiere) is a restructuring verb. The explanation for the fact that clitic climbing applies in restructuring is that these verbs optionally take a VP complement. When the complement is a VP, clitic climbing results. When the complement is larger than a VP, there is no clitic climbing (cf. Picallo, 1985;Rochette, 1989;Rosen, 1989). 6. A parallel distinction was pointed out in Ritter and Rosen (1991) for the English causatives have and make. Th're it was argued 108 that the complement of have is a bare VP, and the complement of make is headed by an inflectional functional projection. As illustrated in (i), Ritter and Rosen found a strong contrast in the ability of the causative verb to take an individual level predicate as its complement. 7. Some French speakers accept (20a), but clearly interpret it as referring to a change of state, rather than a permanent, ongoing property of Marie (thus, I have indicated it as marginally acceptable rather than totally out). Similarly, speakers report that (20b) can only mean that Marie is made to come to know about French, not that she is forced to know (how to speak) French. In like vein, (i) can only mean that Marie is forced to speak French at a particular moment, and not that she is forced to be able to speak French: (i) Jean fait parler le francais a Marie.
8. Unlike the object clitic in the TP complement, the subject must move up to the matrix T. The subject rests in Spec of TP (where it receives accusative Case from the matrix ECM verb); moving onto the embedded T would constitute unlicensed downward movement. Thus, the subject clitic must move up to the closest c-commanding T. 9. It is apparent that the causative verb cannot enter into the ECM construction that 'let' and the perception verbs can. That is, speakers will consistently reject sentences like (i).
'Jean made Pierre repair it.' It is not known exactly why this is, and an exploration of this matter is beyond the scope of this paper. However, suggestions have been made concerning differences in the lexical representation and selectional characteristics of the verbs (Rosen, 1989). In addition, in the Mexican, Columbian, Peruvian and Bolivian dialects of Spanish, the causative verb hacer does appear to enter into the TP complement construction. Speakers of these dialects will accept sentences like (ii), in which the embedded subject appears between the two verbs. See Treviiio (1990) for a full description of the data. (ii) Hice a Juan arreglarla. 'I made Juan repair it.' 109 10. Judgements of (47c) vary depending upon dialect. In the Mexican dialect, speakers will generally use only the accusative clitic (47b) for the embedded subject. However, in River Plate Spanish, speakers accept the dative clitic.
11. Case transmission should also be possible through PP, but this is irrelevant here. See, however,  for use of such a phenomenon of Case transmission in PP complements.
12. The parallel between double object predicates and causatives in a language like Kinyarwanda is relevant within an analysis of morphological causatives in which the causative verb takes a VP complement, just as in Romance. V-to-V raising subsequently takes place to satisfy the a ffixal requirements of the causative verb. For specifics of this verb incorporation, see Baker (1988a), and as applied directly within a VP complement account, see Rosen (1989).
13. As mentioned in note 9, there are some dialects of Spanish in which the causative verb may take a TP complement, and therefore will exceptionally Case-mark the embedded subject directly. In these dialects, the embedded subject appears between the two verbs, and clitic climbing does not take place (Trevino, 1990). This is illustrated in (i).
(i) Hice a Juan arreglarla. 'I made Juan repair it.' As pointed out by Reed (1990a;1990b), there al e also some nonstandard dialects of French in which the embedded object and subject may simultaneously appear in accusative Case, though this is only the case when the embedded subject is cliticized. An example from Reed (1990a) appears in (ii). a.
(ii) ...pour aller travailler. Mais...je peux pas la faire lacher l'ecole, elle est si juene... '...to go to work. But...I can't make her-acc drop out of school, she's so young...' Both Treviiio and Reed report that these respective constructions have an accompanying semantic distinction --they represent unambiguous direct causation, whereas the ordinary causative construction is ambiguous between a direct and an indirect causative reading. This semantic distinction is reminiscent of the distinction between direct/indirect causatives found in Japanese (cf., for example, Shibatani 1973). In Japanese, the distinction is between the -o (accusative) and the -ni (dative) causative. In the -o causative, the embedded subject appears in accusative Case, and the construction receives a direct reading. In contrast, in the -ni caustive, the subject appears in dative Case, and the construction receives an indirect reading.
There appears to be a correlation between accusative Case marking on the embedded subject and the direct causative reading. I would like to suggest that in Japanese, and in these dia:ects of French and Spanish, the causative verb is directly Case-marking the embedded subject, and the Case that it gives to the subject is accusative. When the causative verb directly Case-marks the embedded subject the semantics is one of direct influence on that subject, and the direct causative reading results. The exact syntactic mechanisms for this are right now unknown, but the correlation between the direct reading, and accusative Case or the position of the subject is clear.
14. Though given that the two constructions look identical other than the realization of the embedded subject, it is difficult to tell whether the optional subject stems from the faire par or the faire infinitive construction. It will become clear from the analysis that when the 'subject' does not appear on the surface, this must be the faire par construction.
15. But see Baker, Johnson, and Roberts (1989) for quite a different view of passivization. It is not crucial here that this process be viewed as passivization per se, but only as the suppression of an external argument, and the Case-absorption that follows.