LEXICALIST GRAMMAR AND JAPANESE PASSIVES* Nobuko Hasegawa University of Washington The passive construction has been always a central issue in generaIn this paper, I will first review some of the major past tive grammar. analyses of passives, pointing out certain inadequacies. The analyses reviewed include Chomsky's movement analysis within the Extended Standard Theory (EST) (Chomsky (.1979a,1980a, 1980b)), Bresnan's (1980b) Lexical As for Japanese passives, a Theory, both of which concern English. traditional transformational analysis will be reviewed, which is represented by Kuno (1973, 1978), Harada (1973), Shibatani (1978a), etc. Then, I will propose a lexical analysis of passives within the framework advanced in Hasegawa (1981b). 1. Passives and the Extended Standard Theory The most important (and in fact the most attractive) claim made by the EST is that major syntactic operations, the passive being one of them, are controlled by three principles, namely the 8- criterion, Case Theory, and Binding Theory. However, I will show that this claim cannot be maintained, at least, as far as the passive is concerned. The relevant principles are presented here. (1) The 8- Criterion a. Every 8 -role must be filled by some lexical expressions and b. Each lexical expression must fill exactly one 8 -role. (Chomsky (1980b:16 -17)) (2) Case Theory a. NP -} Nominative if governed by Tense. b. NP 4 Objective if governed by [ -N] (V or P). c. NP + inherently case marked as determined by idiosyncratic property of [ -N]. d. Special rules like Of- Insertion in English. e. Exception: f. Case Filter: Participles, that are assign Case. *[NP [ +V] ( and [ -N] ?), do not phonetic matrix] -Case (Chomsky (1979a), Koster (1979)) (3) Binding Theory a. An anaphor is bound in its governing category, where an anaphor is a lexically specified anaphor (a reflexive and a reciprocal) and non -case marked traces. -25- 26 b. A pronominal is free in its governing category. c. An R- expression is free. (Chomsky (1979a)) Passive operations involve Move a, which is brought into play due to the stipulation imposed on 'participles.' According to (2e), participles cannot assign Case to an NP that they govern. The assumption here is that the same 8 -roles are assigned to both participles and ordinary (active) predicates. Hence, the verb kick and the participle kicked, for example, assign Agent on the subject and Theme on the object, but the object NP (Theme -role) does not receive Case (Objective) from kicked but only from Then a lexical item representing Theme must move to a case - marked kick. Since a noncaseposition; otherwise it is subject to Case Filter (2f). marked trace is an anaphor, it must be bound in S. In order for it to be bound in its governing category, the only position where it is allowed to This is how the EST describes the passive occur is the subject position. This procedure is summarized in (4). operation. (4) a. b. 8 -role John Agent Case Nom. (Active) (Passive) kicked the ball Theme Obj. ] was kicked the ball (participle) Theme 8 -role (Agent) Case Nom. 0 t_-_- Move a [ e The ball. was kicked Theme [.e ] Nom. First, the However, there are several points that can be criticized. assumption that a participle and its active counterpart assign the same 8 -roles to their arguments cannot be maintained. Notice that Theme as well as Agent must be present in 8 -role positions of an active verb kick as in On the other hand, Agent must be eliminated from the argument (4a). Unless it is eliminated, the 8- Critestructure of a participle, kicked. This means that the passive operation depends rion (la) is not satisfied. (i) a participle does not on two crucial stipulations about participles: assign Case to its object and (ii) a participle does not select the predicate external 8 -role (the 8 -role that is assigned to the subject of the active predicate.) Hence, the passive operation does not simply follow from the basic principles of the EST, but it follows from these two assumptions, which have to be stipulated in the lexical representations of a participle. In fact, these two stipulations characterize what the passive operation is, as will be seen in the fourth section.l Second, the claim that Move a must be invoked due to (2e) and (2f) cannot be true. The EST employs-Of- Insertion (2d), when a governor does not give Case to its arguments, though it governs them. Hence, nouns and adjectives employ Of- Insertion to assign Case to their objects as in I am fond of John and the destruction of the city. Then, the noncase- marked objectof a participle does not have to move to the subject position, if However, it can receive Case from of which is introduced by Of- Insertion. 27 such a sentence (It was kicked of the ball) is ungrammatical. Then, in order to rule this out, the EST must employ an ad hoc condition which prevents Of- Insertion from applying to participles. Hence, the passive operation (Move a) does not follow from the Case Theory, but the characteristic of the participle is responsible for it.2 Third, the above operation does not explain the case of passives with a sentential subject. [That John had kicked the ball] was believed (by his friend). (5) a. [That the earth is round] is not known (i b. tol 7 everyone). by D- Structure representations of these sentences are supposed to be those in (6) . (6) a. [NP e ] was believed [that John had kicked the ball]. b. [NP e ] is not known [that the earth is round]. Unless these embedded Ss are assumed to be dominated by an NP, there is no reason for them to move up to the subject position: Case is irrelevant to In fact, these Ss should not be analyzed as NPs, since the following S. sentences show that the embedded S does not have to move. (7) a. b. It was believed that John had kicked the ball. It is not known that the earth is round. There seems to be no way to describe the above phenomenon in a systematic way.3 2. Bresnan's Lexical Theory The passive operation of Bresnan's (1980b) Lexical Theory directly incorporates Perlmutter and Postal's (1977) claim that passivization (i) the subject of the active has the following two universal properties: appears as oblique in the corresponding passive and (ii) the object of the active appears as the subject of the corresponding passive. In Lexical Theory, the passive operation alters lexically encoded grammatical relations of the predicate argument structure (PAS) of lexical items. Her universal rule of Passive is (8). (8) Passive in UG (SUBJ) -> 0 (0JB) -} (SUBJ) / (OBJ) (Bresnan (1980b:8)) Besides this universal characteristic of the passive, language -particular operations have to be stated, which specify how the oblique function is expressed and how the passive morphology is assigned. In English, the oblique function is typically expressed by (BY OBJ) and the passive Her English passive rule is (9). predicate is of the form of a participle. LV (9) The Passive in English Functional Change: 0 (SUBJ) (OBJ) Morphological Change: -} V / (BY OBJ) (SUBJ) -r V[part] (Bresnan (1980b:9)) Rule (8) or (9) applies only to the predicate that has an object. Hence, if a sentential complement is considered to be (OBJ), sentences such as (5) can be easily accounted for. However, this analysis cannot explain the existence of sentences such as (7), where (OBJ), a sentential complement, is not realized as (SUBJ). Examples (7) can be derived from (5) by way of It- Extraposition, though within Bresnan's Lexical Theory, I believe, the phenomenon of It- Extraposition must be lexically accounted for (cf. Safir (1979)). A more problematic case is given in (10) and `I1), where it is shown that only 'extraposed' passives are allowed. )) a. John felt that it would rain. b. *That it would rain was felt (by John). c. (11) a. It was felt (by John) that it would rain. John reasoned that Mary has gone to Europe. b. *That Mary has gone to Europe was reasoned (by John). c. It was reasoned (by John) that Mary has gone to Europe. Nonextraposed sentences (lOb) and (lib) are ungrammatical (according to Williams). Hence, it is unlikely that a sentential complement of these verbs once becomes (SUBJ), and then it is extraposed to the end of the sentence. In the above examples, the passive seems to have applied without changing grammatical relations. Such cases cannot be described in Bresnan's Lexical Theory, because her passive rule must necessarily alter grammatical relations. 3. The Passive in Japanese4 In the traditional transformational framework, passives are derived from their active counterparts by Direct Passive Formation. This rule permutes the order of the subject NP and the object NP, and attaches a passive morpheme (r)are to the stem of the verb (i.e., inserts ( r)are between the verb stem and a tense element), and an agentive marker ni(yotte) to the original subject. There are problems in this analysis. First, Passive seems to apply to verbs which select sentential complements. Examples follow. (12) a. John -ga Tokyo -e Minna -ga to everyone -subj subj to sinzi-te-i-ru. go -past CMP believe -prog -pres. it -ta 'Everyone believes that John went to Tokyo.' 29 b. John -ga subj Tokyo -e to to (minna -ni) sinzi- rare -to -i -ru. go -past CMP everyone -by believe -pass. -prog -past it -ta 'It is believed (by everyone) that John went to Tokyo.' (13) a. Mokugekisya-ga eyewitness -subj John -ga subj Mary -o obj korosi -ta kill -past to CMP syoogen si -ta testify do -past 'The eyewitness testified that John had killed Mary.' b. John -ga Mary -o korosi -ta subj obj kill -past to CMP (mokugekisya -ni(yotte)) eyewitness -by syoogen s- are -ta. testify -passive -past 'It is testified (by the eyewitness) that John had killed Mary.' Passive examples (12b) and (13b) are supposed to be derived from their active counterparts, (12a) and (13a). Kuno (1976) comments on these passives as follows. Examples [such as (12b) and (13b) -NH] ... are pure passive sentences. I do not understand what status the to clauses have in these sentences because to clauses in general cannot be in the subject position. (Kuno (1976:46)) These examples are the same as their active counterparts, except that the passives do not have matrix 'subjects' and the actives lack the passive morpheme (r)are. There is no evidence that the sentential complement is an object in the active sentence and that it becomes a subject in the passive sentence, because it lacks case markers. These passives appear to be subjectless.5 Hence, the traditional Passive cannot explain these examples in a systematic way. Second, there are cases where Passive cannot apply. Not all the NP objects can be the subject of a passive predicate. In past transformational analyses of passives in Japanese, the following global condition has been proposed, to prevent Passive from applying to a derived object which is syntactically nondistinguishable from nonderived objects. (14) Harada (1973) and Kuno's (1978) Global Condition on Passive: Passive cannot subjectivize an NP that used to be a constituent of a sentence embedded in the sentence to which the rule applies. Presumably there are three cases where (14) operates: (i) the object which was raised by Subject -to- Object Raising; (ii) the derived object of the Ni- Causative structure; and (iii) the object of the embedded sentence in the causative structure. I will examine these three cases in turn. Case 1: Kuno (1976) argues that a certain class of verbs (thinking and feeling verbs) exhibits raising if the embedded predicate is an adjective or a nominal adjective (a 'nominal + copula' predicate). Thus, in his analysis (15b) is derived from (15a) by Raising. 30 (15) a. [S Mary -ga subj John -ga subj baka -da to] omot -ta. stupid CMP think -past 'John thought that Mary is stupid.' b. baka -da to omot -ta John -ga Mary -o obj 'John thought Mary to be stupid.' The raised object Mary-o in (15b) cannot be the subject of the passive structure as indicated in (16).6 John -ni(yotte) *Mary -ga subj (16) baka -da stupid by to omow- are -ta CMP think -passive -past 'Mary was thought to be stupid by John.' Since the transformational operation of Passive simply subjectivizes a syntactic object, (16) cannot be prevented, unless a condition such as In such an analysis, no explanation is provided (14) is postulated. for why the raised object does not act like an ordinary object. Case 2: A similar phenomenon is observed with respect to the object In Japanese, it has been argued that the causative of the Ni- Causative. constructions are divided in two types; the 0- Causative and Ni- Causative. Kuno (1973, 1978), Harada (1973), and Shibatani (1976), for example, pro For pose two different deep structures for these two types of causatives. the 0- Causative, a matrix object is postulated, while the Ni- Causative does not have it. (17) a. A deep structure of the 0- Causative John Mary [Mary Tokyo -e to b. go A deep structure of the Ni- Causative John c. (s)ase -ta. cause -past ik] [Mary Tokyo -e (s)ase -ta. ik] A surface structure of the 0- and Ni- Causatives John -ga Mary-Lid. Tokyo -e 'Johntiéte) ik- ase -ta. Mary go to Tokyo.' Surface structures of both types of causatives are identical except that the object (the causee) is marked by o in the 0- Causative and by ni in the Ni- Causative as shown in (17c). Due to the operation of Predicate Raising (PR) and S- Pruning, the derived surface structure (17c) is considered simplex. At this stage, Passive is applicable, which derives (18). (18) Mary -ga subj John -ni(yotte) by Tokyo -e to ik -ase- rare -ta go- cause -passive -past 'Mary was forced to go to Tokyo by John.' 31 What is interesting here is that (18) is not ambiguous between the NiIt is only considered to be the passive of the and 0- Causatives. 0- Causative. In other words, the object or.: the Ni- Causative, Mary -ni, which used to be the subject of the embedded clause, cannot be the subject of the passive. In order to describe this fact, (14) is employed in transformational analyses. Case 3: The last case where (14) is utilized is also relevant to As discussed directly above, the derivation of the causative structure. Hence, at the time Passive applies causatives involves PR and S- Pruning. to the causative predicate (s)ase, the structure is considered simplex. This means that if the embedded sentence contains an object, the derived structure cannot syntactically distinguish two objects, one being a matrix object (in the case of the 0- Causative) or a raised object through S- Pruning (in the case of the Ni- Causative) and the other being an object of the embedded sentence. (19) a. 0 John -ga Mary Mary-CI subj 'John( made) hon -o yom- ase -ta. read -cause -past Mary read a book.' John -ni(yotte) b. *Hon -ga book -subj by Mary -j ni yom -ase- rare -ta. read -cause -passive -past If the passive applies to (19a), making the object of the embedded verb, hon -o 'book -obj', the subject of the passive, an ungrammatical sentence (19b) results. To prevent this undesirable consequence, (14) is assumed to be operative here. Hon -o, which used to be the constituent of the embedded sentence, cannot be the subject of the passive. In the above, it is clear that the global condition (14) is employed for the sole purpose of describing the phenomena, without explaining why the derived object cannot be the subject of the passive. The significant generalization underlying the descriptive generalization (14) is this: As the assumed deep structures imply, the derived objects in question, the raised object, the object of the Ni- Causative, the object of the lower clause, are not semantically or thematically related to the verb to which the passive morpheme (r)are attaches and such objects cannot be the subject of the passive. In the traditional transformational framework, this lexical difference between two types of objects, one semantically related to the verb and the other not, cannot be incorporated into a syntactic passive operation. This difference can be incorporated in deep structure representations; however, the problem is that there is no means to preserve this lexical or deep structural information until Passive applies, except by postulating a global condition such as (14). 4. An Alternative Analysis In the above discussion, we have observed that neither EST nor 32 Bresnan's Lexical Theory can describe all the passive phenomena of English. The discussion on Japanese reveals that the passive in Japanese is sensitive to lexical information. The following is a summary of the above discussion. (20) a. the Extended Standard Theory Passive operation: Move a with the following stipulations. (i) a passive predicate is a participle. (ii) a participle does not assign Case to its object. (iii) a participle does not select the predicate external 8 -role. Problem: b. no explanation for the fact that a sentential complement can be moved to the subject position. Bresnan's Lexical Theory Passive operation: (i) the passive predicate is a participle. (ii) the active object becomes a passive subject. (iii) the active subject becomes null or oblique. Problem: c. no explanation for the fact that the passive does not always invoke (ii). the Passive in Japanese Passive operation: (i) the passive attaches ( r)are to the active predicate. (ii) the active object becomes a passive subject. (iii) the active subject becomes null or oblique. Problems: (i) no explanation for the passive on a verb with a sentential complement. (ii) no explanation for why the derived object cannot become the subject of the passive. Although the frameworks differ, (20) can be summarized in the following way. (21) The Passive Operation a. It involves participles in English and the (r)are attachment in Japanese. b. It eliminates the object function of the active predicate. c. It eliminates the predicate external argument (the subject) of the active predicate. Notice that in (21), whether the object becomes the subject of the passive is not specified. In fact, as discussed in previous sections, there are cases where the passive operation applies without making a sentential 33 object (or complement) a subject of the passive. I assume this change in grammatical function follows from something else. Based on (21), I propose a universal passive rule which operates in the lexicon as one of the lexical redundancy rules.? (22) Passive in UG a. OBJ + b. el -} :where e1 is the predicate external argument (the argument associated with the subject function in the active predicate). o Rule (22) is accompanied with morphological stipulations, which vary from language to language: in English, a participle is used, while in Japanese (r)are attaches to the original predicate. The following rules are proposed for English and Japanese. (23) a. The Passive in English Change in Functional Frame Morphological Change b. OBJ + 0 e - V } V[part] 0 The Passive in Japanese Change in Functional Frame OBJ -} 0 el } 0 Morphological Change V stem 4- [Vstem+ (r)are]. . v Hence, the lexical entry of kick (or ker in Japanese), for example, undergoes the following change if (23) applies. (24) a. kick: V; [Aglnt SUBJ b. kicked: Vpart; T eme] OBJ [i`getSUBJ Theme] Rule (23) erases the object function and the predicate external argument. As a result, the subject function (SUBJ) does not have a semantic role and a predicate internal argument (Theme) is not assigned to any grammatical function. Obviously, what needs to be done is to connect SUBJ and Theme. I employ the following convention. (25) In a lexical entry, if an argument of the predicate argument structure is free from a grammatical function and a grammatical function is free from an argument, they must be connected with each other. Due to (25), the object of the active predicate becomes the subject of the passive predicate. Thus, Perlmutter and Postal's generalization 34 follows from (23) and (25). At the same time, the operation of Move a can be done without syntactic movement.8 It is clear that the proposed passive rule and convention (25) can describe the passive phenomenon that involves an object. Now let us see how this analysis accounts for sentential objects or complements, which are problematic to the analyses reviewed in the above. If sentential complements of the verb such as believe and think are considered to be objects, it is clear that they would be realized as subjects in passives. The lexical entry of believe, for example, undergoes a similar change as (24). A more problematic case is feel or reason, where a sentential complement (or object) cannot be the subject of the passive, as shown in (10) and (11). What seems to be going on here is that the passive rule (23a) applies to feel or reason, but the convention (25) is not triggered. As a result, a participle is used, no thematic role is assigned to the subject, which is realized as it, and the comI assume that the sentential complement plement stays as a complement. of the verb of this class has dual grammatical function, OBJ -COMP, and that the passive rule (23a) erases OBJ and the external argument, but the COMP function still attaches to the argument. Thus, the convention (25) cannot be triggered. The difference between believe and feel is observed below. (For the expository purposes, the semantic roles of the PAS are expressed by 01, 62, .,., 0n instead of actual semantic roles such as Agent, Theme, Goal, etc.) (26) a. believe: [ 01 V; OBJ SUBJ b. (27) a. believed: feel: V part V; ; Passive (23a) Convention (25) 02] e2 [ [ 0l 02 Passive (23a) SUBJ OBJ-1 COMP b. felt: Vpart' ] SUBJ --, 92 [ ] SUBJ COMP In (27), since the deletion of OBJ does not trigger (25), the SUBJ function is left without an argument. A grammatical function whose predicate I will come argument is null is syntactically realized as it in English. back to this point shortly. According to (26), the sentential complenent of believe must be realized as the subject in the passive. However, as (7) shows, the subject of believed can be it and the complement can stay in the original position. Hence, I assume that verbs like believe, know, think, etc. This have dual lexical entries, one like (26) and the other like (27). explains why sentential complements of these verbs sometimes appear as subjects and sometimes as complements in passive constructions. The lexical entry such as (27) can be given to Japanese verbs with sentential complenents. Hence, sinzi 'believe', and syoogen su 'testify' 35 which appear in (13a) and (14a), respectively, have the following partial lexical representations. (28) a. sinzi: Vstem' e2 [ Omi-comp ST3BJ b. syoogen su: V ] il stem 1 02 StIBJ OBJICOMP [ ; ] Rule (23b) applies to these entries, deleting the external argument and the OBJ function. (29) a. sinzi -rare: V ; 82 [ stem SUBJ b. syoogen s-are: V stem ; COMP 82 ] [ SUBJ CMP Representations (29) are essentially the same as (27b) to the extent that both involve the SUBJ function without a semantic role. In Hasegawa (1981b) I argue that languages differ in the way to syntactically realize English, a typical 'configurational' language, an argumentless subject. must have a 'formal' subject in the syntactic representation of a senOn the other hand, Japanese, one of the free word order or 'non tence. configurational' languages, does not need a 'formal' subject. This difference is systematically observed in the following pairs. (30) a. Ame -da. rain -copula '(It) is raining.' b. (31) a. It rains. John -ga subj amerika -e to rasai. go -past seem it -ta '(It) seems that John went to the U.S.A.' b. It seems that John went to the U.S.A. Predicates such as rain and seem do not impose semantic roles on their In Japanese, sentences with such predicates do not have a subjects. subject, whereas in English, the subject is syntactically necessary to identify the sentence status of a string of words; hence, the existence of the expletive it in the subject position. This generalization also holds in the passive predicates such as (27b) and (29). Example (7b) results from (27b) and examples (12b) and Hence, this analysis is not only compatible with but (13b) from (29). also predicts the fact that the passive with a sentential complement, (12b) and (13b), does not have a subject. In the present framework, there are two ways to express the descriptive generalization that a derived object or the object that is not 36 thematically related to the passive verb cannot be the subject of the One obvious way is to incorporate this generalpassive in Japanese. The object must be semanization into a lexical passive rule (23b). tically or thematically related to the predicate. This is an approach If this approach is taken,rule (23b) must pursued in Hasegawa (1980). be replaced by (32).9 (32) The Passive in Japanese Change in Functional Frame OBJ 0 -} 81 4 0 Morphological Change Condition: 4 Vstem [Vstem + (r)are]V OBJ must be semantically related to the V stem stem. As has been discussed in the previous section, the existence of the global condition (14) is a direct consequence of treating a lexical In the present analysis, the passive, process as a syntactic one. being considered as a lexical process, can straightforwardly incorporate this information. The fact that thematic information is necessary to describe the phenomenon clearly indicates the passive in Japanese to be lexical. Given (32) and partial entries of omow 'think' and (s)ase 'cause' as in (33) and (34), respectively, the phenomenon that the raised object of omow or (s)ase cannot be the subject of the passive is easily explained.IU(Here, (30 stands for a nonargument position.) (33) omow: [ Vstem; eo 1 SUiJ (34) (s)ase: Vstem; a. OBJ [ ?1 SUBJ b. ] [ l SUBJ P 12E D 82 OBJ ¶0 OBJ e3 ] P1ED 2 PRED Lexical representations (33) and (34b) have objects which are not thematically related to the predicate. Hence, they are not subject to rule (32). However, (34a) can undergo (32), producing a passive predicate (35). (35) (s)ase-rare: Vstem' [ 82 SUBJ ] PLED RE The passivized causative (18) is given by this entry (35). Rule (32) explains the third case that is subject to (14); the Since object of the lower clause cannot be the subject of the passive. rule (32) applies to (s)ase, the relevant object is the object of (s)ase but not the object of the verb which serves the PRED function. Examples such as (19) would never be generated in the present analysis. 37 Thus, the problems with past analyses of Japanese passives are all eliminated. There is another way to explain the fact that a raised object or an object that is not thematically related to the predicate cannot undergo Passive. In the above, I have argued that in Japanese SUBJ with no thematic role is not realized in the syntactic representation. By imposing this principle on the passivized raising predicate, omow -are 'think -passive', for example, and (s)ase -rare 'cause -passive' with the SUBJ which is 60,the same result as the condition in (32) is obtained. Assume that the passive in Japanese is (23b) without the condition in (32). This rule applies to (33) and (34b), producing the following passive predicates. (36) a. omow-are: V [ stem' 60 SUBJ b. (s)ase-rare: 1 P ED $Q Vstem' SUBJ 2 P ED The nonargument subjects in (36) are not realized in the syntactic representation. Then, what these lexical representations produce is something like (37). (37) a. *Baka -da stupid to CMP omow -are -ta. think -passive -past - - -- was thought to be stupid.' b. *Hon -o yom -ase- rare -ta. book -obj read- cause -passive -past - - -- was forced to read a book.' In these examples the subject of baka -da 'is stupid' and yom 'read' would never be identified. They violate a general condition that all arguments in the predicate argument structure must be identified.11 The above two approaches can both explain the passive phenomena of Descriptively, they are equivalent. However, I prefer the latter approach, because it does not resort to any extra stipulation. The condition imposed on the Japanese passive (32) is superfluous, because it follows from the motivated principle that a SUBJ with 6a is not syntactically realized in Japanese. Japanese. From the above discussion, it can be concluded that the analysis proposed for the passive is superior to past analyses. It can account for a wider range of the data and it also accommodates universal characteristics of the passive. 38 FOOTNOTES I would like to thank Mike Brame and Joe Emonds for comments and discussions on an earlier version of this paper, and the participants at the conference for comments. I also thank Ann Farmer and Chisato Kitagawa for inviting me to the conference and providing me an opportunity to present the preliminary version of this paper. The research for this paper was supported in part by a Lockwood Foundation Fellowship. This paper is a slightly modified version of Chapter 5 of Hasegawa (1981b). 1In Chomsky (1980b), it is claimed that passives are characterized by these two stipulations. What is usually called "passive" seems to have two crucial pro perties: (42) (I) [NP,S] does not receive a 0 -role (II) [NP, VP] does not receive Case within VP, for some choice of NP in VP. (Chomsky (1980b:36)) 2Marantz (1980b) independently reaches the same argument as the one presented in this paragraph. 3One may assume that Move a optionally moves S into the subject (or COMP ?) position. Under this assumption, it is possible to obtain (5) and (7) from (6). However, this cannot explain the data in (10) and (11), where the movement of S is not allowed. This means that the passive involving a sentential complement (or object) is not controlled by any sub -theories in the EST, but it is relevant to the lexical characteristics of the verbs in question. The data in (5), (7), (10), and (11) can be possibly accounted for in the EST framework, if it is assumed that verbs such as believe and know have two different lexical entries; one with an NP object which exclusively dominates S and the other with an S complement, and that verbs such as feel and reason have only one type of lexical entry (with an S complement). Then, (5) is produced by believe' and know' (with an NP object) and (7) is obtained, if believe2 and e (with an S complement) are used. However, the existence of the PS rule NP + S must be independently motivated in such an analysis. 4Japanese has two types of passives: pure or direct passives and adversity or indirect passives. This paper deals with pure or direct passives. For the analysis of adversity or indirect passives in Japanese within the framework pursued here, see Hasegawa (1981b). 5I consider the embedded sentence in (12b) and(13b) to be a complement rather than a subject for the following reason. In Japanese, ordinary sentential subjects take a nominal element which is followed by the subject marker gaa as in (i). 39 (i) [Tikyuu -ga marui the earth -subj round (to yuu)] koto -ga akiraka -da. Nom. -subj obvious -is CMP 'That the earth is round is obvious.' There are no sentential subjects which are not marked by 2.2. If the embedded sentence is considered to be a subject in (12b) and (13b), we cannot explain why the sentential subject of the passive is exceptional to this generalization. To is exclusively used as a marker for a sentential complement. Of course, one can posit a rule of 22. deletion, which deletes ga after to, as suggested by Kuroda (personal communication). However, this rule seems to be a restatement of the problem. If we take the position that case - marked languages utilize case-markers to identify grammatical functions, rather than syntactic structures serve this end, the identification procedure is much simpler if to is used for a marker for sentential complements, along with for subjects and o for objects (though case arrays are not this simple). If to can also-le a subject marker, the identification procedure becomes complicated; to marks sentential complements when a verb is active and if the verb is passive to is a subject marker. 6Here, I follow Kuno's (1976) observation that (16) is ungrammatical as a pure or direct passive sentence, though it is grammatical as an indirect passive. See Kuno (1976) for the supportive arguments for this observation. Although Yamazaki (1979) challenges this claim, I do not find her arguments convincing. 7A thematic role which associates with the SUBJ function in the active (61 in (22)) may optionally be realized as a predicate internal argument with an oblique function. 8This is what Marantz (1980b) argues. The effect of Move a is obtained by 'intransitivization' (erasure of the object case) and de- 6 -ization' (erasure of the 6 -role of the subject). Rule (22) and convention (25) are essentially the same as what Marantz proposes. 9The condition in (32) may have to be more restricted, since there are some argument types which are thematically related to the predicate cannot appear as the subject of the passive. Such predicates are kat 'win', aw 'meet', etc. Objects of these predicates do not serve as Theme or Goal, which normally occurs as the subject of the passive. Hence, as an alternative to the condition of (32), (i) maybe proposed. (i) Condition: 10In OBJ = Theme or Goal of the V stem. representing omow 'think' and (s)ase 'cause', I use a PRED(ication) function. In the framework advanced in Hasegawa (1981b), the lexical item which serves as this function is subjectless and its subject is identified by an interpretive rule. See Hasegawa (1981b) for details. 40 11The difference between English and Japanese follows from the difference that English needs a syntactic or 'formal' subject even if no thematic role is imposed on the SUBJ function, while in Japanese SUBJ is In not syntactically realized if it does not have a thematic role. English the raising predicate believe, for example, produces a grammatical passive sentence such as (i). (i) John was believed to be stupid Example (i) involves the passivized believe (iib). (ii) a. believe: V; [ Q, 0J J b. believed: V part P 12E D 2] e0 ; [ SUBJ P D If The subject of the participle believed does not have a semantic role. its function is merely 'formally' filling a subject slot of the syntactic representation, the expletive it is sufficient. However, this nonargument subject'must also be interpreted as the subject of the following PRED function in the framework advanced in Hasegawa (1981b). Hence, the lexical item which occupies the subject slot of the sentence must be able to satisfy the subject argument of the predicate which serves as the PRED function of (iib).
* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project