A MODULAR APPROACH TO "PASSIVES" Osvaldo A. Jaeggli University of Southern California Within the tradition of generative grammar the properties of the construction commonly referred to as "the passive construction" have been accounted for by a single rule which typically preposes the direct object of an active -like structure, positions the subject of such a structure in an agentive prepositional phrase, and inserts the auxiliary and morphological trappings found in passive sentences. Within the framework of the Standard Theory, such a rule can be stated approximately as in (i): (cf. Chomsky(1957)). O. (1) X - NP1 - AUX - V - NP2 - Y X - NP2 - AUX +be -en - V - by +NP1 - Y We can call this approach to passive the "single -rule approach." More recently, an alternative approach has been suggested, where the properties of passive constructions do not follow from a rule as in (1), but rather are derived from the articulation of several subtheories within a tightly structured theory of grammar, in particular the theory of Abstract Case and the theory of A- roles. See Chomsky (1981) and references cited there. We will call this alternative approach, which will be explicated in much detail below (see section 2), the "modular approach ". The modular approach is succintly characterized in Farmer (1980:128) as follows: The task of accounting for the phenomena of, say, "passive," is viewed as a process of properly apportioning to the levels , of the grammar the various aspects of the passive [construction]. This view is incompatible with a theory that attributes to a single rule all the effects of [this construction]... Instead, we will construct a system whose subparts operate independently of each other. We will show in this paper that there is a class of constructions in Romance languages which necessitate a modular approach to passive. Without such an approach, important generalizations are missed, and it is impossible to give a truly explanatory analysis of the phenomena we are about to investigate. The analysis will assume the general theoretical framework of the Extended Standard Theory, taking into consideration recent developments such as the theory of Abstract Case and the theory of 9 -roles as put forth in Chomsky (1980a, 1981) and references cited there. 1. Modern Romance languages, with the exception of Rumanian, have a class of causative + infinitive constructions in which the subject of the infinitival surfaces as an agentive prepositional phrase. French, Spanish, and Italian examples are given in (2): (2) a. Le professeur a fait lire ce livre par tous les étudiants.) 'The professor had this book read by all the students.' - 41 - 42 b. El profesor hizo leer ese libro por todos los estudiantes. 'The professor had that book read by all the students.' c. Il professore ha fatto leggere quel libbro da tutti gli studenti. 'The professor had that book read by all the students.' This construction, which we will call 'the agentive causative,'2 has been analyzed in much detail in Kayne (1975) for French, Aíssen (1974) and Strozer (1976) for Spanish, and Radford (1977) for Italian. One of the many interesting features of agentive causatives is that they can be shown to be similar to passives in a number of significant ways. This similarity is not shared by infinitivals embedded under causatives which do not contain an agentive PP, i.e., the so- called 'a construction'. Kayne (1975: 234 -242) presents several arguments for French which clearly show that there is a systematic similarity between passives and agentive causatives. Strozer (1976:406 -412) notes that Kayne's arguments apply equally well to Spanish. (Radford (1978) mentions that Cinque (1975) argues the same point for Italian.) In this section we will present Kayne's arguments for French and Strozer's extensions to Spanish in order to assess the nature of this similarity.3 The first argument concerns the behavior of idiomatic expressions.4 It is well -known that certain idiomatic expressions retain their idiomaticity under the deforming influence of rules like passive, and others don't. The English expressions kick the bucket and take advantage of contrast in this respect, as shown below: (3) (4) a. Ron kicked the bucket. b. *The bucket was kicked by Ron. a. The rich took advantage of the poor once again. b. Advantage was taken of the poor by the rich once again. French and Spanish also exhibit ' passivizable' and 'non -passivizable' idioms. Consider the following examples: (5) a. Sa famille a cassé la croúte. lit.:His family has broken the crust 'His family had a snack.' b. *La croúte a été cassée par sa famille. The crust has been broken by his family (.6) a. Son fils te pretera assistance. 'His son will help you.' b. Assistance te sera pretee par son fils. 'Assistance will be given to you by his son.' 43 (7) a. Pedro metió la pata. lit.: Pedro stuck the foot 'Pedro stuck his foot in his mouth." b. *La pata fue metida por Pedro. The foot was stuck by Pedro (8) a. La Cruz Roja les prestó asistencia. 'The Red Cross helped them.' b. Asistencia les fue prestada por la Cruz Roja. 'Assistance was given to them by the Red Cross.' Exactly the same contrast appears in agentive causatives. Those idiomatic expressions which cannot be passivized, as in (5) and (7), cannot occur in agentive causatives; while those which can be passivized, as in (6) and (8), can also occur in agentive causatives. Cf.: (9) a. *I1 a fait casser la croate par sa famille. He b. has made to -break the crust by his family *Juan hizo meter la pata por Pedro. Juan made to -stick the foot by Pedro (10) a. Il te fera prgter assistance par son fils. 'He will make his son help you.' b. El gobierno les hizo prestar asistencia por la Cruz Roja. 'The government had the Red Cross help them.' Second, there is a restriction on passivization if the object is an 'inalienable possession,' i.e., if it is a part of the body understood as belonging inalienably to the subject. Cf: (11) a. Jean lèvera la main. 'Jean will raise his hand.' b. *La main sera levée par Jean.5 The hand will -be raised by Jean (12) a. Juan levantó la mano. 'Juan raised his hand.' b. *La mano fue levantada por Juan. The hand was raised by Juan 44 A similar distribution holds for overt possessives, as originally pointed out to Kayne by J. C. Milner: (13) a. Jean. apprendra son. rôle. 'Jean will learn his role.' b. *Sons rôle sera appris par Jean.. His role will -be learned by Jean (14) a. Juan. aprenderá su. papel. 'Juan will -learn his role.' b. *Su. papel sera aprendido por Juan.. His role will -be learned by Juan The same restrictions occur in agentive causatives. (15) a. Cf.: *Elle a fait lever la main par Jean. She has made to -raise the hand by Jean b. *Tu feras apprendre son rôle par Jean. You will -make to -learn his role by Jean (16) a. *Maria hizo levantar la mano por Pedro. Maria made to -raise the hand by Pedro b. *Tu haras aprender su papel por Juan. You will -make to -learn his role by Juan Third, sentences which contain reflexive clitics cannot be passivized. (17) a. Jean s'achétera ce jouet. "Jean will buy himself that toy.' b. *Ce jouet se sera acheté par Jean. That toy himself will -be bought by Jean (18) a. Juan se comprará ese juguete. 'Juan will buy himself that toy.' b. *Ese juguete se será comprado por Juan. That toy himself will -be bought by Juan The same restriction is operative in agentive causatives. Cf.: (19) a. *Nous ferons s'acheter ce jouet par Jean. We will -make himself -to -buy that toy by Jean Cf.: 45 b. *Nosotros haremos comprarse ese juguete por Juan. We will -make to- buy -himself that toy by Juan Fourth, certain superficial objects which are in some sense locative NPs may not undergo passivization or be embedded in agentive causatives. Cf.: (20) a. Jean quittera ma maison demain 'Jean will leave my house tomorrow.' b. *Ma maison sera quittée demain par Jean. My house will -be left tomorrow by Jean (21) a. Juan dejará mi casa marrana. 'Juan will leave my house tomorrow.' b. *Mi casa será dejada marrana por Juan. My house will -be left tomorrow by Juan (22) a. *Je ferai quitter ma maison par Jean demain. I will -make to -leave my house by Jean tomorrow b. *Yo haré dejar mi casa mañana por Juan. * will make to -leave my house tomorrow by Juan Finally, verbs which are compatible with passives in de (in French) can occur in a faire...de construction similar to the agentive causatives discussed above. Cf.: i _(23) Marie est haie de tout le monde. 'Mary is hated by everybody.' (24) Marie est arrivée á se faire hair de tout le monde.6 'Marie managed to get herself hated by everybody.' It should be noted that the ungrammatical sentences in (9), (15), (16), (19), and (22) become grammatical if the preposition par /por is In other words, only agentive causatives (i.e. substituted with a /a. the faire...par construction) share these properties with passives. Causatives with á/a behave in a completely different way. We have omitted examples of the a-construction for brevity's sake. The interested reader is invited to consider such cases which are exemplified in Kayne (1975:234 -239) and Strozer (1976:406 -412).7 The data presented above strongly suggests that there is a systematic An adequate analysis relation between agentive causatives and passives. In connection to should express this relation in its relevant aspects. this, Kayne (1975:242) writes: 46 The significant relationship between passives and the faire...par construction can be made precise only in the context of an analysis specifying the derivation of the latter. The most straightforward proposal would be to derive the faire...par construction through the embedding under faire of a sentence which has undergone the passive In this way (111) [ =Marie fera boire cette eau par transformation. cette eau être bu par son chien] would be derived from Marie fera son chien] ... (111) could be derived from such an intermediate structure ... if the auxiliary être were deleted. [ However, he ultimately concludes that "despite its initial plausibility, such an analysis would have serious defects ", Kayne (1975:243). What leads him to this conclusion? The central reason why such an analysis would have serious defects concerns the rule deleting the passive auxiliary être. This rule is a necessary component of such an analysis, since the intermediate structure mentioned above, e.g., Marie fera [cette eau âtre bu par son chien], is totally ungrammatical. Kayne (1975:243 -244) argues very convincingly that such a rule "is not well motivated and is subject to reservations on general grounds." We will not repeat his arguments here. We accept his conclusion as valid. Notice however, that such a rule is needed only under the "single -rule Under such an approach, every passive sentence contains approach" to passive. the passive auxiliary (Eng. be -en, Fr. être -é, Sp. ser -do, etc.), as can clearly be seen by looking at rule (1)8 Thus, a "single -rule approach" to passive fails to relate agentive causatives to passives in "the most straightforward way." On the other hand, if we assume a theory of passive which does not require that every passive sentence contain a passive auxiliary, but rather derives those cases in which an auxiliary does show up from independent considerations, we would then have a way of getting around the problem posed by the undesirable deletion rule mentioned above. In fact, within such a theory there would be The correct theory would explain why the no need for such a deletion rule. passive auxiliary is not permitted in those contexts. We argue below that a modular approach to passives provides just such a theory. In order to do so, however, we must look at that approach in greater detail, a task to which we turn in the next section. As mentioned in section 0, within a modular approach to passives the properties of such constructions do not follow from a single rule, as in (1), but rather are derived from the articulation of several subtheories of a The subtheories most centrally intightly structured theory of grammar. volved in such an account are the theory of Abstract Case and the theory In this section we will explicate the functioning of these of 9- roles. subtheories as they relate to the passive construction. An interesting property of modular accounts which is well -worth mentioning is that the theoretical principles which are used typically enter into explanatory accounts of a wide -range of data. That is, they are completely independently motivated; they are not 'construction- specific'. This consideration alone makes them preferable to construction- specific accounts such as the "single -rule approach" to passives. Needless to say, however, to present the wide range of facts which independently motivates such theories would fall way beyond the scope of this paper. Therefore, we will usually omit 2. 47 such presentation. (The reader is referred to Chomsky (1981) for a detailed discussion of these theories.) Instead, we will concentrate on the relevance of these principles to the passive construction. And even here we will limit our attention only to what is essential to our concerns, leaving much else aside. (For a detailed discussion of passive under a modular approach, see Chomsky (1981), chapter 2, section 7.) We will proceed as follows: we will first present the core principles of the theory of Abstract Case and the theory of 9- roles, and then we will show how such principles account for simple (tensed) passive constructions. What will emerge is a theory which analyzes the different components found in passive constructions, e.g., NP- Preposing, NP- Postposing, and the presence of the passive auxiliary, as being independent from each other. This is the theory promised in the last paragraph of the preceding section. In section 3 we will apply these methods of analysis to the problem of agentive causatives. The theory of Abstract Case regulates the appearance /non- appearance of lexical NPs in syntactic structures. Roughly speaking we can say that it specifies those structural configurations in which a lexically filled NP can occur. This is achieved through the device of Case assignment. We suppose, following Chomsky (1980a, 1981), that the fundamental properties of Case assignment are as in (25): (25) a. NP is nominative if governed by INFL.9 b. NP is objective if governed by V. c. NP is oblique if governed by P. d. NP is inherently Case - marked as determined by lexical properties of its [ -N] governor. Furthermore, let us suppose that there is a principle which embodies the requirement that every lexical NP be Case marked. This can be accomplished with a filter, as in (26): (26) *NP, where NP has a phonetic matrix but no Case (25) and (26) entail directly that lexical NPs can appear in the following positions: 1) the subject of a tensed sentence, 2) the direct object of a verb, 3) the object of a preposition, and 4) other objects of certain verbs which may be inherently Case - marked. Notice that the position subject -of an infinitive is conspicuously absent from this list. In fact, infinitival structures rarely have lexically filled subjects.10 When they do, some special mechanism will have to be invoked. This brief presentation of the theory of Abstract Case will suffice for the moment. We will return later to certain assumptions concerning this theory which are, in a sense, specific to passive constructions. The theory of 9-roles is concerned with traditional notions such as 'agent -of action,' 'goal -of- action,' 'experiencer,' 'theme,' etc. which enter into many different theories of semantic description. Following Chomsky (1981) we will assume that LE must be so designed that expressions such as John, the man in John saw the man are assigned thematic (or semantic) roles; that is, they are assigned the status of terms in a thematic relation. 48 Such expressions are called 'arguments,' as opposed to idiom chunks, expletive (or pleonastic, non -pronominal) it, or existential there, elements which assume no O-role. We will refer to a position in LF to which a 9 -role is assigned as a "9- position." A reasonable criterion of adequacy for LF so- conceived is (27): (27) The 0-Criterion Each argument bears one and only one 9 -role, and each 0-role is assigned to one and only one argument. The 9- Criterion requires that there be a one -to -one correlation between arguments and 9- roles. We will see that it has important consequences for NP- movement and other related matters. Although much more can be said about the theory of 9- roles, this much will suffice for our purposes.11 We can now sketch a modular approach to passive in detail. Before we consider how the theories of Abstract Case and 9 -roles enter into such an approach, it should be mentioned that the 'modularization' of passive was begun as early as Chomsky (1965). In fact, the first steps away from a rule as in (1) above were taken in Aspects of the Theory of Syntax, where it is argued that the passive rule should be decomposed into two parts: 1) NPPostposing, which vacates the subject position, and 2) NP- Preposing, which moved the direct object into the empty subject position, after NP- Postposing. We will assume that such a decomposition is essentially correct, but we will make the further assumption that NP- Postposing does not really exist at all. Rather, we assume, following Bresnan (1972), Hornstein (1975), Emonds (1976), and Dresher & Hornstein (1979), that agentive phrases are generated directly in their position by the base rules of the grammar. Consider in connection with this issue a sentence as in (28), mentioned in Dresher & Hornstein (1979): (28) John's portrait of Mary by Warhol. In (28), Warhol could not have been postposed from the NP subject position (i.e., the Spec -N position), since that position is filled with John's. Furthermore, John's could not have been preposed from some other position because all sources are already filled up. Such sentences demonstrate that at least in some cases NPs must be generated directly in w phrases. And if this is true in some cases, there is little motivation for positing another source. Notice that in Romance languages agentive phrases in nominals are always generated by the base. They could never be filled -in by NP- movement from determiner position, since NPs cannot appear in the determiner position of nominals. That is, a structure such as Warhol's portrait of Mary, which might serve as the source for the portrait of Mary by Warhol in English, is never allowed in Romance languages. Cf.: (29) a. *Juan retrato de Maria. Juan portrait of Maria b. *Jean portrait de Marie. Jean portrait of Maria 49 c. *Gianni ritratto di Maria. Gianni portrait of Maria In these cases, the rule of NP- Postposing would have to be made absolutely obligatory, and the only reason for positing it would be to maintain that phrases are always transformationally derived. The base -generation alternative seems clearly preferrable. We conclude, then, that there is phrases. Rather, agentive phrases no rule of NP- Postposing to create are generated independently by the base.12 Passive, then, consists of two operations: 1) NP- Preposing, and 2) These two operations must be considinsertion of the passive auxiliary. Note that the operation we have been ered independent from each other. calling NP- Preposing occurs equally well, with all its essential properties, in constructions which do not contain passive auxiliaries, namely Raising constructions, as in (30): (30) a. John seems [t to have left early] b. Juan parece [t haber salido temprano] c. Jean semble [t être parti tôt] d. Gianni sembra [t essere partito presto] Furthermore, (t marks the D- structure position of John /Juan /Jean /Gianni.) in some Romance languages it is possible to show that the insertion of the passive auxiliary does not require NP- Preposing. Consider the following Spanish and Italian examples: (31) a. Fue declarada la independencia por los revolucionarios. 'Independence was declared by the revolutionaries.' b. Fu arrestato Gianni. 'Gianni was arrested.' Italian is particularly telling in these cases. In Italian, ne- cliticization is possible only from object position, but not from pre -verbal subject position nor from the post - verbal position that results from adjunction of an inverted subject. Cf. Burzio (1981). Burzio observes the ne- cliticization is possible from a post -verbal passive subject, as in (32): (32) a. Ne erano riconosciuti molti. 'Many of them were recognized.' b. Ne sono limitati troppi. 'Too many of them were limited.' Thus, we cannot claim that passives with post -verbal subjects, as in (31), (At are the result of double movement: NP- Preposing + subject inversion. least, if inversion is always adjunction, as we assume.) We are led to conclude, rather, that no movement at all occured in (31). Thus, passive morphology does not require preposing. 50 Williams (1979) observes the same phenomenon in English. He points out that in some cases passive morphology appears but there is not evidence of movement. Cf.: (33) It was believed /held /reasoned that the conclusion was false. He observes that some of these verbs do not allow the "extraposed" S to appear in pre -verbal position (e.g. *That the conclusion was false was held). Thus, we cannot assume that in general there is preposing followed by extraposition in such cases. To summarize, we conclude that the phenomena typically associated with passive constructions, such as Preposing, NP- Postposing, and the presence of the passive auxiliary and passive morphology, should not be stipulated as formally dependent on each other. Rather, they are to be thought of as independent operations which need not co- occur. The obvious question that arises out of this conclusion is the following: why is it that these independent operations do (and must). co -occur in certain structures? That is, why are the following sentences ungrammatical? (34) a. b. *NP was hit John by Mary. *John hit Mary by Peter.13 (34a) shows that in some cases NP- Preposing is obligatory in the sense that if it does not apply the sentence is ungrammatical. (Contrast (34a) with (33), to which we return below.) .(34b) shows that in some cases an agentive a phrase is not allowed to appear in a sentence. What accounts for these requirements? As we shall see, the theories of Abstract Case and 9 -roles provide elegant answers to this question. In order to provide these answers, we must make explicit two crucial assumptions concerning Abstract Case and 9 -roles in structures of the type shown in (34). They are: (35) In a "passive" construction: a. [NP,S] does not receive a 9 -role. b. [NP,VP] does not receive objective Case. In other words, in a passive sentence, the [NP,S] position is non -thematic, and the direct object position is left with no Case. These assumptions are quite natural, given that 1) passive participles are not [ -N] elements, but rather [ +V] elements, and only [ -N] elements (i.e. true verbs and prepositions) are capable of assigning Case; and 2) the subject of a passive sentence is the subject of a VP headed by the verb be, a verb which does These assumptions, together with not assign a 9 -role to [NP,S] position. the principles of the theory of Case and the 9- Criterion, suffice to characterize the structures in (34) as ill- formed. Let's consider them one -by -one. (34a) is marked ungrammatical because it contains a lexical NP, John, The passive participle does not assign Case which does not receive Case. to [NP,VP] position. Thus, the Case filter (26) rules this sentence out. 51 Notice that if 'direct object' position is occupied by something other than an NP in a structure as in (34a), the Case filter will not rule the sentence out, since only NPs are required to have Case. This is what accounts for the grammaticality of Williams' example, sentence (33). This problem does not arise in (34b). This sentence does not contain a passive participle. Here the tensed verb can assign Case to [NP,VP] Something else must obviously be wrong with this sentence. position. Consideration of the A- structure of this sentence immediately reveals what the problem is. We can assume that John and Mary are both assigned appropriate e- roles -- perhaps 'agent' and 'theme,' respectively --as they are in the grammatical sentence John hit Mary. What about Peter? To answer this question we must consider what A -role is assigned by the preposition IT. As noted in footnote 13, this preposition is semantiIt can have a cally (at least) two -ways ambiguous -- perhaps more. locative sense, under which (34b) is perfectly grammatical, with the meaning of (roughly) "John hit Mary while he /she was next to Peter." The Assume then that by can assign the 9-role 'locative' to Peter. sentence will then meet the 9-Criterion, and all else being in order, (34b) is ungrammatical only if we it will be pronounced grammatical. interpret the by-phrase in an agentive sense. That interpretation assigns the 9-role 'agent' to Peter. But in that arises only when case, the 9-Criterion will rule the sentence out, since there will be a 9-role, namely 'agent,' which will be assigned to more than one argument, namely to both John and Peter. This is what rules out (34b).14 We conclude, then, that the interaction of the theory of Abstract Case and the theory of 9-roles explains why two independent processes, i.e., 1) the presence of passive morphology, and 2) "obligatory" NPPreposing, must co -occur in certain structures. To conclude this section I would like to return briefly to the Romance examples given in (31). In these sentences, NP- Preposing has not applied. If Case assignment in Spanish and Italian were identical to Case assignment in English, we would be at a loss to explain why these sentences are grammatical while (34a) is ungrammatical. That is, why doesn't the Case filter also rule out (31)? The answer involves no mystery. It is well -known that Spanish and Italian differ from English in that they can assign nominative Case to a post - verbal NP in certain structures, i.e., in tensed sentences. See Chomsky (1981), Burzio (1981) and Jaeggli (1981) for relevant discussion. This is what allows the sentences in (31) to escape the implacable effects of the Case filter. It should be noted that this is a very positive feature of modular It is necessary to determine, independently of analyses of passive. passive constructions, that post - verbal NPs in tensed sentences can get nominative Case in Spanish and Italian. Within a modular theory of passive, this is all that needs to be said. The results for this construction follow immediately, with no additional statements. This completes our presentation of a modular analysis of passive. Needless to say, much has been left rather vague. We do not claim that this section constitutes a complete presentation of this topic. Rather, 52 our intention is simply to provide an introduction to the principles which allow us to give an account of agentive causatives, and to exemplify the operation of these principles if only in a very sketchy manner. In the next section, we return to the problem of the analysis of agentive causatives. One of the central properties of agentive causatives, mentioned towards the end of the first section, is that the passive auxiliary and passive morphology are not allowed to occur. Consider the following ungrammatical sentences: 3. *Marie fera être bu cette eau par son chien. (36) a. Marie will -make to -be drunk that water by her dog *Maria hará ser tomada esa agua por su perro. b. Marie will -make to -be drunk that water by her dog (Identical facts hold in Italian). It is not hard to show that these facts follow directly from the theory of Case. Consider the structure associated with the sentences in (36): (37) NP 1 Vc [BE PassP NP2 BY NP3] (Vc stands for the causative verb faire /hacer; BE = Fr. être, Sp. ser; BY = Fr- par /Sp. por; and the symbol PassP means 'passive participle'.) All NPs in (37) are phonologically filled. Thus, by the theory of Case, they. all require Case. But NP2 does not get assigned Case, since passive participles do not assign structural Case. The Case filter then rules the sentences in (36) ungrammatical. In other words, these sentences are ungrammatical for the same reason that (34a) is ungrammatical in English. This provides a simple explanation for why passive morphology is not allowed in these contexts. Notice that NP- Preposing cannot be invoked to save these structures, as it is in tensed passives. Consider the result of applying NP- Preposing in these cases: (38) NP 1 Vc [NP2 BE PassP t2 BY NP3] In this structure, NP2 still has no way of getting Case. The principles given in (25) do not assign Case to the subject of an infinitive from within the infinitival clause. And Vc does not assign Case across a clause boundary. This last observatión is supported by the ungrammatical status of the following sentences: (39) a. *Jean a fait Marie manger la pomme. Jean had made Marie to -eat the apple b. *Juan hizo Maria comer la manzana. Juan has -made Maria to -eat the apple 53 (This is a well -known fact about the causative verbs discussed here.) Our theory predicts, then, that sentences with the structure given in (38) should also be ill- formed. And this prediction is borne out by the data. Cf.: (40) a. *Marie fera cette eau être bu par son chien. Marie will -make that water to -be drunk by her dog b. *Maria hard esta agua ser tomada por su perro. Maria will -make that water to -be drunk by her dog Thus, the theory of Abstract Case provides a uniform account of the ungrammaticalities in (38) -(40). The reader can verify for himself that such a uniform account is not readily available within non - modular theories. This analysis makes another interesting prediction. It predicts that if there were some way for an NP to be assigned Case in the following structure: (41) ...Vc [NP Vinf...] ... then passive morphology + NP- Preposing should give a grammatical output. That is, in such cases, sentences like the ones in (40) should be grammatical. Furthermore, a sentence like those in (39) should also be grammatical. This is so because this analysis establishes an explicit correlation between the ungrammaticalities of these sentences. This prediction finds startling confirmation when we consider the situation which obtains in Portuguese. Consider the following Portuguese sentence: (42) Una rajada de vento fiz o vaso cair da prateleira. A gust of wind made the vase to -fall from +the shelf 'A gust of wind made the vase fall from the shelf.' This sentence shows that Portuguese allows structures as in (41) above, i.e., the NP in (41) can get Case in Portuguese.15 With no further stipulations, we predict that passive morphology should be possible under causatives in Portuguese. And indeed it is. Cf.: (43) Fiz o livro ser lido por Carlos.16 I -made the book to -be read by Carlos 'I had the book read by Carlos.' This is very strong confirmation for our analysis which claims that the theory of Abstract Case plays a crucial role in determining in which contexts passive morphology will be allowed. Without the unifying generalization provided by such a theory, the data examined above would remain at best puzzling, if not totally mysterious. Our account extends naturally to cover a different set of facts also noticed in Kayne (1975:254 -255). Kayne observes that embedding 54 restrictions under faire are not limited to passive constructions. Another type of infinitival is also not allowed in such contexts. Consider the following sentences (the French examples are from Kayne (1975:254), the Spanish ones are my own): (44) a. *Son expression peinée fait sembler Jean souffrir. His expression pained makes to -seem Jean to- suffer b. *Son expression peinée fait paraître Jean être en colare. His expression pained makes to -seem Jean to -be in anger (45) a. *Su expresión triste hace parecer Juan sufrir. His expression sad makes to -seem Juan to- suffer N b. *La falta de compania hace soler Juan tomar. The lack of company makes to -be -wont Juan to -drink Here what cannot be embedded under the causative verb are sentences which contain Raising verbs. Sembler, paraître, parecer, and soler are all subject -to- subject raising verbs. I will assume that the D- structure of a sentence containing subject to subject raising verbs is approximately as in (46) : (46) [NPe] [ VR [ NP2 Vinf..']]' where VR is the raising verb NP- Preposing will apply to (46) to yield approximately the following structure. (47) NP2 [ VR [ t2 Vinf'..]] This is how sentences such as Jean semble souffrir /Juan parece sufrir 'John seems to suffer' are generated. Note that the NP2 position in (46) is not a Case position. Cf.: (48) a. *I1 semble Jean souffrir. it seems Jean to- suffer b. *Parece Juan sufrir. it -seems Juan sufrir Unless NP- Preposing applies to a structure as in (46), the Case filter will rule it out, accounting for the ungrammaticality of (48). Consider next the structure of the sentences in (44) and (45). (49) NP1 Vc E VR [NP2 Vinf...] Cf.: 55 Once again, NP2 is not within its own clause, Case from the VR which as was shown in (48). in a Case position. It cannot receive Case from since the clause is non -finite. It cannot receive governs it, since raising verbs do not assign Case, Thus, the Case filter rules such structures ungram- matical. -7 Notice once again that NP- Preposing does not save these structures from the effects of the Case filter. The result of applying NP- Preposing to (46) is roughly as in (50): (50) NP1 Vc [ NP2 VR,inf [ t2 Vínf " ']] But here, once gain, NP2 is not in a Case assigning position. That is, the Case filter also rules out (50), just as it rules out (38) and (49). In fact, sentences corresponding to the structure in (50) are all ungrammatical. Cf.: (51) a. *Son expression peinée fait Jean sembler souffrir. His expression pained makes Jean to -seem to- suffer b. *Su expresión triste hace Juan parecer sufrir. His expression sad makes Juan to -seem to- suffer This is precisely what our theory predicts.17 To conclude we would like to summarize the central results achieved Our main claim is that the theory of Abstract Case accounts for the impossibility of embedding infinitivals with passive morphology under faire /hacer.18 The same principles explain why NP- Preposing is not a saving device within these structures. In Portuguese, infinitivals with passive morphology can be embedded under fazer, contrasting with the situation which occurs in Spanish, French, and Italian. This correlates with a significant difference between Portuguese on the one hand, and Spanish, French, and Italian on the other, as concerns the possibilities of Case assignment within infinitivals. Roughly speaking we can say that Portuguese has an extra Case -assignment mechanism, not shared by the other languages. Given this mechanism, we predict the difference in behavior under causatives with no further stipulations. Finally, a partial account can be given for the impossibility of embedding raising constructions in the contexts under consideration. here. These results are made possible by our modular approach to passive constructions. A single -rule approach to passives would not achieve the same degree of explanatory adequacy. In particular, it would fail to account for the impossibility of embedding certain types of infinitivals under faire /hacer, a task elegantly accomplished by the modular approach with no ad hoc stipulations. These Romance facts, then, support the view of grammar outlined in Farmer (1980). Although our investigations have proceeded from completely different data bases, e.g. Japanese on the one hand, and Romance languages on the other, we have arrived at very similar conclusions concerning the general organizing principles of rules of grammar. 56 Indeed, one might say more. A careful comparison of several of the specific mechanisms proposed in Farmer (1980) to account for Japanese passives and the account of Romance passives given above reveals striking similarities. This is hardly surprising given modularity, since the principles of grammar utilized are not 'construction- specific,' as mentioned above. Unfortunately it would greatly exceed the scope of this paper to present a detailed analysis of these similarities. Nevertheless, we would like to point out that it is a tribute to modularity that such similar conclusions can be arrived at by examining such typologically distinct languages as Japanese and the Romance languages. We do not claim that the partial analysis of agentive causatives presented above constitutes the whole story concerning these constructions. Our goal was simply to show that these constructions support- and, indeed, require --a modular approach to passive. This much has been accomplished. We leave the many gaps which remain in this analysis open for future research. FOOTNOTES *This paper originated as a short note written in March of 1980. I want to thank Noam Chomsky for encouraging me to pursue this topic in more depth. I also want to thank the organizers of the Arizona Conference on Japanese Linguistics for allowing me to present at that occasion this material, which bears only very indirectly on the main topic of the conference. The questions and comments I received during the conference were very useful in helping me to clarify and deepen my ideas on this subject. 1The data in this paper will be drawn primarily from French, Spanish, Italian, and Portuguese. Since these languages are hardly exotic, we refrain from giving word -by -word glosses for grammatical sentences. Ungrammatical strings, on the other hand, will often be given only word by -word glosses, with no attempt made to translate them accurately. 2More precisely, we will generally use the term agentive causative to refer to the string of elements following the causative verb in sentences such as (2). This is slightly confusing, we admit, because in this sense an 'agentive causative' does not formally contain a causative verb. However, it is the string of elements following the causative verb that we are interested in. We need a term to refer to that part of the construction, and we have adopted this one since it is the standard one associated with these constructions in the literature. 3Most of the example sentences cited in this section will be taken from Kayne (1975) or Strozer (1976). I have added some additional Spanish sentences. 57 4The term idiomatic expression is used here in an extended sense to cover bcth a) clear cases of non -compositional idioms such as English kick the bucket, French casser la croûte, Spanish meter la pata, and b) cases like English take advantage of, French prêter assistance á 'to help,' and Spanish prestar asistencia a 'to help.' 5Sentences (lib), (12b), and (13a,b) are grammatical in an alienable interpretation, where the hand is understood as not being part of the subject. This is irrelevant for the point under discussion. 6Strozer (1976:411) gives the following Spanish examples which make the same point. (i) (ii) Maria es odiada de todo el mundo. 'Marie is hated by everybody.' Maria ha llegado a hacerse odiar de todo el mundo. 'Maria managed to get herself hated by everybody.' I find these sentences extremely literary in style. My dialect of Spanish does not easily accept de passives. For those dialects in which these sentences are common, they illustrate the same phenomenon as the French examples given in the text. 7Strozer (1976:395 -406) presents further differences (not mentioned by Kayne) between the a- construction and the por- construction. These concern the behavior of clitic pronouns --both reflexive and non- reflexive --in these constructions. Since we will have little to say about clitics in this paper, we leave these further differences aside. (1). 8Kayne's rule of passive is identical in all relevant respects to rule Cf. Kayne (1975:248). 9The relation of government is defined in Chomsky (1980a:24) as follows: is governed by ß if a is c- commanded by ß and no major category or major category or major category boundary appears between a and ß. S is an absolute barrier to government. This much is sufficient for our purposes. For a more detailed discussion, see Chomsky (1981, chapter 3, section 3.2.1). 10Were it not for certain marked constructions in English, namely those in which the subject of an infinitival is preceded by the prepositional complementizer for or certain verbs such as believe, etc., we could strengthen that statement to "infinitival structures never have lexically filled NP subjects." We will ignore these cases in English, since they fall outside the range of data we are considering. 11For more extensive discussion of 9 -roles and the theory of 9- roles, see Gruber (1965), Katz (1972), Jackendoff (1972), Borer (1979, 1980) and Chomsky (1981). 12If by- phrases are generated independently by the base, the reader may wonder what rules out ill- formed sentences such as (i) *John saw Bill by Mary. where the by- phrase is interpreted as an agentive phrase. (These structures are also ill- formed in Romance). We return to this question below. 58 13There is an irrelevant grammatical reading of (34c) in which the a phrase is a locative, not agentive, PP. We return to this reading below. 14There is an alternative, slightly more subtle, way of ruling out (34b) in its relevant (ungrammatical) sense. This alternative employs the same devices used in the account given in the text. Crucially, the 9- Criterion is also involved. However, this alternative account requires a more thorough presentation of the mechanism of "9 -role absorption." Since the issue is not immediately relevant, we prefer to omit presentation at this point --but see further ahead for relevant comments. Also, see Jaeggli (forthcoming) where the matter is discussed in much greater detail. 15We will not discuss the question of what makes these structures possible in Portuguese. For disSeveral alternatives come to mind. cussion, see Rouveret (1979), and Zubizarreta (1980). Nothing crucial to our discussion hinges on the exact formulation of the Case assignment mechanism in these sentences. 16Sentences (42) and (43) are taken from Radford (1978), where this phenomenon is noticed. 17We hasten to add that this may not be the complete story concerning the impossibility of embedding raising verbs under faire /hacer. Kayne (1975:255) notes further that raising verbs cannot be embedded under faire even when followed by a tensed clause. Cf.: (Kayne's judgments) (i) *Ce rapport fait sembler que la situation est très mauvaise. This report makes it seem that the situation is very bad.' *Le journal fait paraître qu'on va augmenter le prix du métro. 'The newspaper makes it appear that they are going to raise the price of the metro.' *L'aveu de Jean a fait s'avérer que Paul était innocent. 'Jean's confession made it turn out that Paul was innocent.' The Case filter cannot be invoked to rule these sentences ungrammatical. The reader can check that they do not contain NPs which lack Case. These facts remain unaccounted for within our analysis. It seems worthwhile pointing out, however, that the sentences in (i) above are not nearly as bad as the ungrammatical sentences in (51a), (48a) and (44). Indeed, sentences similar to those in (i) are sometimes considered quite acceptable by native speakers of French, I have found. And their Spanish translations do not sound particularly ungrammatical to me. Cf.: (ii) ?Este informe hace parecer que la situación está muy mal. ?El diario hizo parecer que iban a aumentar el precio del subte. 59 If they are ill -formed at all, they are without any doubt much less ungrammatical that their infinitival counterparts. The analysis presented here establishes such a distinction, insofar as it postulates some If principles which are violated only by one of the constructions. there turn out to be other principles which both constructions violate (perhaps weaker principles), we would still get a distinction in grammaticality, since one construction, e.g. the infinitival one, would violate two principles, while the other one would only violate one. 18This result is of some consequence for the question of the proper categorial specification of infinitival complements under causative faire/ hacer. I assume that all infinitival complements are clausal. The analysis presented here then explains on totally independent grounds why certain types of S's, e.g., those which contain the passive auxiliary and passive morphology, are not allowed in these structures. Thus, these facts do not force us to assume that the infinitival complements of faire /hacer are VP's (or VP's), as suggested for example in Zagona (ms.), Strozer (1976) and other studies. It has often been claimed that the lack of passive auxiliaries under causatives can be accounted for only if we adopt the position that the infinitival complements in these cases are non -sentential. Such a theory presumably provides an account of these facts because the passive auxiliary is supposed to be outside VP. (We find this latter claim rather questionable in itself. See Akmajian and Wasow (1975) for interesting discussion of this issue.) But even if such an account were to be correct, it can now be seen to be totally stipulative. There is no need to stipulate that causative verbs can only take VP's as infinitival complements. Within our theory, we can assume the unmarked situation; namely that these verbs, like many other verbs, subcategorize sentential complements. Note that faire /hacer must be subcategorized for sentential complements, anyhow, since tensed S's can appear in their complement structures. Cf.: (i) Cela a fait que Jean se dépêche. That has made that John hurry -up (ii) Yo hice que la máquina parara. I made that the machine stop The simplest hypothesis is that this is all that faire /hacer has to be subcategorized for. In some cases the sentential complement is non -finite, i.e., infinitival. These structures then have the properties they have because of completely independent considerations.
* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project