対格言語と能格言語間の歴史的変化

全文

(1)Title. 対格言語と能格言語間の歴史的変化. Author(s). 菅野, 悟. Citation. 北海道教育大学紀要. 人文科学・社会科学編, 66(1): 63-77. Issue Date. 2015-08. URL. http://s-ir.sap.hokkyodai.ac.jp/dspace/handle/123456789/7845. Rights. Hokkaido University of Education.

(2) 北海道教育大学紀要（人文科学・社会科学編）第66巻第１号 Journal of Hokkaido University of Education（Humanities and Social Sciences）Vol. 66, No.1. 平成 27 年８月 August, 2015. 対格言語と能格言語間の歴史的変化＊菅野悟北海道教育大学旭川校英語学研究室. Diachronic Changes between Accusative and Ergative Languages KANNO Satoru Department of English Education, Asahikawa Campus, Hokkaido University of Education. ABSTRACT In this paper, I discuss diachronic changes between accusative and ergative languages. I base my discussion on the assumption that typological differences between accusative and ergative languages are reduced to distinct types of θ-feature inheritance. This assumption comes as a result of the combination of two Minimalist assumptions: Chomsky’s (2008) feature-inheritance mechanism and the recent feature-based θ-theory (see Bošković and Takahashi 1998; Hornstein 1999; inter alia). Based on this assumption, I show diachronic changes between accusative and ergative languages. It has been assumed that accusative languages change diachronically into ergative languages through the reanalysis of passive sentences while the reanalysis of anti-passive sentences is involved in a change from ergative languages into accusative languages. I show that these bi-directional historical changes are straightforwardly accounted for by the present assumption of θ-feature inheritance.. １．Introduction It has been said that languages in the world are divided into accusative and ergative languages, and ergative languages are estimated to occupy one-fourth of the world languages. In this type of languages, objects behave similarly to intransitive subjects and differently from transitive subjects. This situation is depicted as follows: ⑴ a. b.. AErg. OAbs. V. c.. ANom. SAbs. V. d.. SNom. OAcc. V V. (Dixon (1979)). 63.

(3) 菅野悟. Dyirbal is said to be a typical example of syntactically ergative language. See ⑵ and ⑶. ⑵. ŋuma. banaga-nyu. father-Abs. return-Nonfut. (Dyirbal) (Dixon 1994: 10). ‘Mother saw father.’ ⑶. ŋuma. yabu-ŋgu. father-Abs mother-Erg. bura-n see-Nonfut. ‘Mother saw father.’. (ibid.). In the intransitive construction in ⑵, the subject bears an absolutive case. This is the case of the object, not the subject, in the intransitive construction in ⑶. The subject in ⑶, on the other hand, appears with a different (ergative in this case) case. In this paper, I discuss diachronic changes between accusative and ergative languages. The main aim of this paper is to show that an accusative language changes into an ergative language through marked passive constructions while the opposite change (from ergative to accusative) occurs through marked anti-passive constructions. In order to give a syntactic account to these changes, I base my discussion on the assumptions proposed in Kanno and Kitada (2012), which show that syntactic differences between accusative and ergative languages are tantamount to structural differences in argument structure. In accusative languages, an Agent role is assigned to a higher argument located in [Spec, vP] while a Theme role is to a lower argument in the complement of V. On the other hand, under the assumptions in Kanno and Kitada (2012), in ergative languages, the positions between Agent and Theme arguments are inverted in accordance with the spirit of Marantz (1984). Adopting the assumptions, I argue that diachronic changes between accusative and ergative languages occur through the changes in assignment of these theta roles. The organization of this paper is as follows. In the next section, I lay out assumptions in this paper, especially focusing on the mechanism of theta role assignment proposed in Kanno and Kitada (2012). In Section 3, I discuss changes from accusative languages into ergative languages, resulting from changes in argument structure. In Section 4, the changes from ergative to accusative are discussed. Section 5 concludes the discussion of the paper.. ２．Theoretical background In this section, I introduce some assumptions on theta role assignment, especially focusing on Kanno and Kitada (2012), which propose the following differences between accusative and ergative languages: ⑷ In accusative languages, the [Th]-feature undergoes inheritance down to V, while in ergative languages, the [Ag]feature does so. ⑷ is based on two Minimalist assumptions: one is on feature inheritance and the other is on θ-roles as uninterpretable formal features. The feature-inheritance mechanism is first introduced by Chomsky (2008). He assumes that all uninterpretable features are an inherent property of a phase head. For example, the phase head C has uninterpretable features such as an Agree feature and a Tense feature at. 64.

(4) 対格言語と能格言語間の歴史的変化. the first stage of the derivation. Soon after C is introduced into the derivation, its Agree feature and Tense feature are passed down onto the non-phase head T. On the basis of the parallelism between the phase head C and the phase head v, Chomsky (2008) extends the inheritance mechanism to a vP domain and claims that an Agree feature is introduced on v and it is transferred onto V. The feature inheritance mechanism is illustrated in ⑸ and ⑹. ⑸. [CP [C’ C [Agree]/[Tense]. [TP [T’ T. ⑹. [vP [v’ v [Agree]. [VP [V’ V [Agree] ...]]]]. [Agree][Tense]. ...]]]]. These inheritance options in ⑸ and ⑹ make possible subject-verb agreement and verb-object agreement, respectively. The second assumption is the one made by Bošković and Takahashi (1998), Lasnik (1995), Hornstein (1999), Manzini and Rossou (2000), and Funakoshi (2009), among others. Their crucial argument can be paraphrased as follows: θ-roles are a type of uninterpretable formal features (hereafter, θ-features) to be checked off. To the extent that these assumptions hold at the same time, the following claim appears: that uninterpretable θ-features are introduced originally on the phase head v and inherited derivationally by V. Specifically, at the very beginning of the derivation, a vP phase has both the [Ag]-feature and the [Th]-feature on v, as in ⑺.1 ⑺. [vP [v’. v [Ag][Th]. [VP [V’V NP]]]]. Soon after this derivational stage, one θ-feature must be transmitted onto the lower head V owing to the application of feature inheritance. The question immediately arises as to which θ-feature undergoes inheritance. As an answer to this question, Kanno and Kitada (2012) propose Theta-Inheritance Parameter ⑷, repeated here as ⑻. ⑻. In accusative languages, the [Th]-feature undergoes inheritance down to V, while in ergative languages, the [Ag]feature does so.. This proposal maintains that a parametric difference between accusative and ergative languages can be reduced to distinct types of θ-feature inheritance: accusative languages have the structure in ⑼, whereas ergative languages have the structure in ⑽ after the application of feature inheritance. ⑼ [vP [v’. v[Ag]. [VP [V’ V[Th] NP]]]] =. accusative languages. 1 When I discuss the argument structure, I sometimes use the term NP. However, this is for convenience in the discussion, and I am agnostic to the exact projection of the nominal expressions. See Bittner and Hale (1996a, b) for related discussions.. 65.

(5) 菅野悟. ⑽. [vP [v’. v[Th]. [VP [V’ V[Ag] NP]]]] =. ergative languages. In accusative languages, the [Ag]-feature stays on v and the [Th]-feature is lowered down; by contrast, the inverse relationship is seen in ergative languages, where the [Th]-feature is on v and the [Ag]-feature on V.2 When I take the English sentence of ⑾ as a specific example, Mary checks off the [Th]-feature on V, and John checks off the [Ag]-feature on v, as illustrated in ⑾. ⑾. a.. John saw Mary.. (English). b.. [vPJohn[Ag] [v’ see1[Ag] [VP t1[Th] Mary[Th]]]]. This shows that the Agent argument stands higher than the Theme argument in the argument structure of the accusative language. In ergative languages, on the other hand, a different picture emerges due to the different feature-inheritance option. Take the Dyirbal sentence in ⑿ as an example. ⑿ a.. ŋuma. yabu-ŋgu. bura-n.. (Dyirbal). father-Abs mother-Erg see-Nonfut (Dixon 1994: 10). ‘Mother saw father.’ b.. [vP ŋuma[Th] [v’ [VP yabu-ŋgu[Ag] t1[Ag]] bura[Th]]] father-Abs. mother-Erg. see(ibid.). In ⑿, the [Ag]-feature is lowered down to V, which is checked off by yabu-ŋgu ‘mother,’ and the [Th]feature on v is checked off by ŋuma ‘father’ in [Spec, vP]. This example shows that the Theme argument stands higher than the Agent argument in the argument structure of the ergative language. In summary, I have laid out Kanno and Kitada’s proposal that typological differences between accusative and ergative languages can be captured by the parametric difference in inheritance of the θ-feature: the [Th]-feature undergoes inheritance in accusative languages, while the [Ag]-feature does so in ergative languages.. ３．Transition from ergative to accusative In this section, I discuss historical changes between accusative and ergative languages. It has generally been claimed that accusative languages become ergative ones through passive sentences, as in ⒀, and that ergative languages become accusative ones through anti-passive sentences, as in ⒁.3 2 There are some other possibilities that one might come up with under the current proposal. The first possibility is that both of these two features stay on v, and the other is that they both are passed down onto V. In this paper, I do not discuss these possibilities, which are pursued in Kitada (2012). 3 I am not claiming that this is the only source of accusativity/ergativity. Other sources are possessor phrases, topicalized phrases, and many others. However, given the limitation of space, this paper does not deal with the change from these sources. See Whitman and Yanagida (2012) for discussions on the development of ergativity from the non-. 66.

(6) 対格言語と能格言語間の歴史的変化. ⒀. Passives in accusative languages. →. Ergative languages. ⒁. Anti-passives in ergative languages. →. Accusative languages. I will argue that these changes between accusative and ergative languages can occur in either direction. This argument relies crucially on the following assumptions: the passive and the anti-passive constructions are regarded as marked constructions, and such marked constructions are reanalyzed as unmarked constructions. When the reanalysis has been completed, the“originally”marked constructions qualify as unmarked active transitive constructions. In this sense, such bi-directional changes apply again and again.4 To start with, this section focuses on the change from accusative to ergative languages. The change is shown in ⒂ to ⒄. [vP [v’ v [Th]. [VP V [Ag] NP]]]. (= ⑽). active transitives. ⒃. [vP [v’ v [Ag]. [VP V [Th] NP]]]. (= ⑼). anti-passives. ↓ ⒄. [vP [v’ v[Ag]. [VP V [Th] NP]]]. ergative type ↓. (= ⑼). active transitives. . ⒂. accusative type. The active transitive of ⒂ in ergative languages involves the inheritance of the [Ag]-feature from v to V. However, the anti-passive of ⒃ uses a different way of θ-feature inheritance. It is the [Th]-feature that is inherited from v by V; the [Ag]-feature stays on v. Then, this“marked”anti-passive construction undergoes reanalysis and becomes the“unmarked”active transitive construction. This process reflects the change from ergative languages into accusative languages. This claim leads me to predict. ⒅ for. ergative languages. ⒅. a.. The Theme argument in the active transitive has the same structural properties as the Agent argument in the anti-passive.. b.. The Agent argument in the active transitive has the same structural properties as the Theme argument in the anti-passive.. ⒅ claims that the Theme argument in the transitive and the Agent argument in the anti-passive occupy the higher [Spec, vP] position, while the other argument stays in the complement of the verb. In sections 3.1 to 3.3, I show that these are empirically supported. In section 3.4, I discuss Warrgamay and show that this language is an example that has moved toward an accusative type from an ergative type.. passive origin. 4 It could be the case that the completion of the reanalysis is not necessary for a language to undergo a further process toward accusativity/ergativity. Specifically, when an accusative language is in the process of transitioning toward an ergative language, the language gaining ergative properties can enter into the process of returning to an accusative type before the completion of the reanalysis of passives to ergative. Therefore, the discussions in the text are somewhat idealized.. 67.

(7) 菅野悟. 3.1．Missing arguments in conjuncts Firstly, the interpretation of missing arguments offers the similarity between the Theme argument in the active transitive and the Agent argument in the anti-passive. In other words, these arguments both show the same structural properties because they originate in a higher position than any other argument in the vP structure. As seen in ⒆ below, the interpretation of missing arguments is determined by the Theme argument ŋuma ‘father’ in the first conjunct in the active transitive. ⒆. [ŋumak. yabu-ŋguj. bura-n],. father-Abs. mother-Erg. see-Nonfut. banaga- nyu].. [tk/*j. return-Nonfut (Ura 2006: 115). ‘Mother saw father and returned.’. However, ⒇ shows that, when the anti-passive operation applies to the first conjunct, it is the matrix Agent argument ŋuma ‘father’ that determines the interpretation of the missing argument in the second conjunct. ⒇. [ŋumak. bural-ŋa-nyu. yabu-guj]. father.Abs. see-Antipass-Nonfut. mother-Dat and. [tk/*j. banaga-nyu] return-Nonfut (Dixon 1994: 165). ‘father saw mother and returned.’. In ⒇, the Agent argument is missing in the second conjunct, and its interpretation is determined by the matrix Agent argument ŋuma ‘father.’ Moreover, the Theme argument with a dative Case fails to determine the interpretation of the missing argument within the second conjunct. Therefore, I can analyze the parallelism between the Theme argument in the active transitive and the Agent argument in the anti-passive in terms of the ability to control the interpretation of the missing argument. At the same time, the Agent argument in the active transitive and the Theme argument in the anti-passive behave the same to the effect that they do not enter into any relationship with the semantic interpretation of the missing argument. 3.2．PRO in purpose clauses The second piece of evidence regards the licensing of PRO within purpose clauses in Dyirbal. Note first that it is only the Theme argument that serves as PRO in an active transitive within a purpose clause. As an example, I show below. . yabuk. ŋuma-ŋguj. mother-Absk father-Ergj. giga-n. [PROk/*j gubi-ŋgu. mawa-li]. tell to do-Nonfut. PROk/*j doctor-Erg. examine-Purposive. ‘father told mother to be examined by the doctor.’. (Dixon 1994: 169). In the purpose clause in , PRO is interpreted as Theme. On the basis of the fact that the Theme argument acts as PRO in purpose clauses, Next, let me take a look at where the anti-passive operation has applied to the purpose clauses.. 68.

(8) 対格言語と能格言語間の歴史的変化. . a.. ŋumak. banaga-nyu. father-Abs. return-Nonfut. [PROk. bural-ŋa-ygu. yabu-gu].. see-Antipass-Purposive. mother-Dat. ‘Father returned for the purpose of seeing mother.’ b.. yabul. ŋuma-ŋguk. giga-n. [PRO*k/l. mother-Abs. father-Erg. tell-Nonfut. bural-ŋa-ygu. jaja-gu].. see-Antipass-Purposive. child-Dat (Ura 2006: 115-116). ‘Father told mother for the purpose of seeing child.’. In these sentences, the element that works as PRO is the Agent argument. This shows the parallelism between the Theme argument in the active transitive and the Agent argument in the anti-passive in terms of the way that they act as PRO in the purpose clauses. 3.3．Relative formation The third piece of evidence comes from the relative formation, which connects two clauses and deletes the common argument in the relative clause. For example, contains two events: the woman’s sitting down and my watching of her. . palan. jukumpil. ngaja. purangu. nyinanyu. There-Nom. woman-Nom. I-Abs. see-Rel. sit-Nonfut (Levin 1983: 282). ‘the woman whom I am watching is sitting down.’. In these events, the common argument is jukumpil ‘woman,’ which is the Theme argument in the relative clause. On the other hand, when the common argument in the relative clause is the Agent, the application of the anti-passive operation is necessary. See . . payi. yara. pakalngangu. There-Nom. man-Nom spear-Antipass-Rel-Nom. pakul. yiriku. There-Dat kangaroo-Dat. ‘man who speared kangaroo is returning.’. panakanyu return-Nonfut (ibid.: 282). In , the matrix clause expresses the event of the man’s returning and the embedded clause expresses his spearing of a kangaroo. The presence of the anti-passive morpheme is necessary for the sentence to be fine. Contrasting with , I claim that deletion of the highest argument is allowed in the Dyirbal relative formation. In the active transitive, the Theme argument occurs higher in the argument structure than the Agent argument, and it can serve as a gap in the relative clause. On the other hand, in order for the Agent argument to be the gap, the anti-passive operation must be applied. Thus far, I have shown that the predictions in ⒅ are actually borne out. In view of this, I conclude that the anti-passive differs in argument structure from the active transitive. In the anti-passive, the Agent argument occupies [Spec, vP] as its base position, and in the transitive, the Theme argument stands there.. 69.

(9) 菅野悟. Furthermore, the following subsection will claim that the anti-passive becomes the active transitive through the reanalysis of its marked status to an unmarked one. I will show that this process actually occurred in Warrgamay. 3.4. On Warrgamay What I would like to show in this subsection is that one syntactic ergative language called Warrgamay reaches a stage close to an accusative language, through the reanalysis of the anti-passives as the active transitive.5 Before discussing some properties of Warrgamay, I review one important property in Dyirbal. In this language, the only argument that can serve as PRO is the Theme argument, as shows. rul boy-Abs. banaga-nyu. [gajin-du. nyunja-li]. return-Nonfut. girl-Erg. kiss-Purposive. (Dyirbal) (Dixon 1981: 104). ‘The boy returned to be kissed by the girl.’. In order for the Agent argument to act as PRO, the application of the anti-passive operation is necessary. Specifically, I get as a result of applying the anti-passive operation to the purpose clause. . rulgu. banaga-nyu. [nyunja-l(ŋ)a-ygu. gajin-gu ]. boy-Abs. return-Nonfut. kiss-Antipass-Purposive. girl-Dat. ‘The boy returned to kiss the girl.’. (ibid.). In , the anti-passive operation has applied to the purpose clause (we see this by the presence of the anti-passive morpheme l(ŋ)a). It is then the Agent argument that acts as PRO in the embedded clause. In contrast to Dirybal, Warrgamay shows a property crucially different from Dyirbal and other Australian languages. Specifically, I do not have to apply the anti-passive operation to the purpose clause even if I want the Agent argument to serve as PRO. This is shown in . . muyma bana-y. [nyuunja-lagu. gajiya-gu ]. boy-Abs return-Unmarked. kiss-Intransitive.Purposive. girl-Dat. ‘The boy returned to kiss the girl.’. (Warrgamay) (ibid.: 102). In , the Agent argument occurs as PRO. Importantly, the morpheme lagu no longer indicates the application of the anti-passive operation as it once used to. This is strong evidence that Warrgamay is very much closer to an accusative type than other Australian languages including Dyirbal. This is because the property of the Agent argument as PRO (without any morpheme) is the same as that found in accusative languages. Note that in the English. 5 Warrgamay belongs to the Dyirballic languages alongside Dyirbal. These two languages are very syntactically similar except in terms of the features discussed in this subsection.. 70.

(10) 対格言語と能格言語間の歴史的変化. sentence in , the agent argument serves as PRO. . Theyk hired Johnj [without PRO*j/k having to commit themselvesk/*himselfj to that salary].. (Ura 2006: 115). Let me discuss the more precise steps toward an accusative type in Warrgamay. In syntactic ergative languages such as Dyirbal, the Theme argument occurs higher than the Agent argument in the active transitive. This property is seen in Warrgamay as well. However, when I focus on purpose clauses, the Agent argument can stand higher than the Theme argument without the anti-passive morpheme in Warrgamay (see ). Therefore, Warrgamay has entered the stage where the Agent arguments can occur at [Spec, vP], so that they can serve as PRO. In other words, Warrgamay has already taken some steps forward to a syntactic accusative language, since the Agent argument stands in a higher position than any other argument. If this property is generalized to the active transitive and the process is completed, I can say that Warrgamay becomes an accusative language.6 It is important to note, at this point, that the change in Warrgamay from an ergative language to a (near-)accusative language occurs through the anti-passive operation. As I have claimed, the antipassive adopts a way of θ-feature inheritance that differs from the active transitive. That option of θ-feature inheritance from v to V used to be markedly specific to the anti-passive. However, this marked anti-passive has undergone reanalysis and becomes the unmarked active transitive, in accordance with the change of anti-passive morphemes into unanalyzable ones (see Dixon 1981 for detailed discussions on the morphology of this language). In summary, I have claimed that ergative languages become accusative languages through the antipassive. I have discussed the correctness of the predictions in ⒅. Moreover, one of the Australian languages, Warrgamay, is changing from an ergative type into an accusative type through reanalysis of the antipassive as the active transitive.. ４．Transition from accusative to ergative In this section, I discuss the other side of the changes between accusative and ergative languages: the change from an accusative type to an ergative type. I claim that this change occurs along the following process: [vP [v’ v [Ag]. [VP V [Th] NP]]]. (= ⑼). active transitives. . [vP [v’ v [Th]. [VP V [Ag] NP]]]. (= ⑽). passives. ↓ . [vP [v’ v [Th]. [VP V [Ag] NP]]]. Accusative type ↓. (= ⑽). active transitives. . . Ergative type. shows that the [Th]-feature is inherited from v by V in the active transitive in accusative languages, while shows that, in the passive, it is the [Ag]-feature that should be transmitted from v to V. When 6 However, this will unfortunately never be observed because Warrgamay is already a dead language.. 71.

(11) 菅野悟. the accusative language enters the stage where the former passive is reanalyzed as the active transitive and this reanalysis process is completed, I get an ergative language. This is because, in the active transitive, the [Th]-feature is checked off by the argument in [Spec, vP], and the [Ag]-feature is checked off by the complement argument of V. This claim on the θ-feature inheritance of the active transitive and the passive leads to the following prediction regarding accusative languages. . a.. The Agent argument in the active transitive has the same structural properties as the Theme argument in the passive.. b.. The Theme argument in the active transitive has the same structural properties as the Agent argument in the passive.. In section 4.1, I take up English as a representative of accusative languages and show that the active transitive and the passive differ in terms of the option of θ-feature inheritance in that the active transitive involves the inheritance of the [Th]-feature to V, while the passive involves the inheritance of the [Ag]-feature. In section 4.2, I discuss those languages that have changed from accusative to ergative through the passive, owing to the reanalysis of marked constructions. 4.1．Passive sentences in accusative languages: evidence from English To begin with, let me consider similarities in the base position between the Agent argument in the active transitive and the Theme argument in the passive. The current approach claims that they are base-generated in [Spec, vP]. Remember that accusative languages like English keep the [Th]-feature staying at v in the passive. This feature is checked off by the argument base-generated in [Spec, vP]. Since this position is located between T and v, the argument follows T but precedes v. . There are [many books] put on the table.. (Law 1999: 184). In , the [Th]-feature on v is checked off by many books, which follows are in T but precedes put in v. The same position is occupied by the Agent argument in the active transitive, as shown in . . [TP Johni [vP ti [v’ put many books on the table]]].. shows that the [Ag]-feature on v is checked off by John in [Spec, vP] (and then it moves to [Spec, TP]). In this regard, the Theme argument in the passive shows the same structural property as the Agent argument in the transitive. Next, VP-ellipsis provides evidence for the similarity between the Agent argument in the passive and the Theme argument in the active transitive (that is, the arguments in the complement of V). It has been widely acknowledged that, when VP-ellipsis applies in the active, all arguments within the verbal domain must be elided. Adjuncts, on the other hand, may be excluded from the application of the ellipsis. This is shown in . (I use the strikeout notation to show the part of the sentence that has been deleted.). 72.

(12) 対格言語と能格言語間の歴史的変化. . Did John [VP return the books]? a.. Yes, he did [VP return the books] on Tuesday.. b.. *Yes, he did [VP return the books] to the store.(Goodall 1997: 133). The contrast indicates that, different from the adjunct on Tuesday in (35a), the internal argument to the store in (35b) must be deleted along with VP-ellipsis. With the contrast between the internal argument and the adjunct in mind, consider the following sentences. . Will the books be [VP returned]? a.. Yes, they will be [VP returned] on Thursday.. b.. *Yes, they will be [VP returned] to the store.. c.. ?*Yes, they will be [VP returned] by John.(ibid.). Contrary to the adjunct in (36a), both the PP argument in (36b) and the by-phrase in (36c) must be included in the ellipsis. The grammaticality of (36c) shows that the by-phrase occurs within the verbal domain. In fact, if the by-phrase in the passive is elided, the sentence becomes grammatical. See . . a.. Sandy was attacked by a cougar.. b.. *?They said that Sandy would be attacked by a cougar, but she was [VP attacked] by a leopard.. c.. They said that Sandy would be attacked by a cougar, and she was [VP attacked by a cougar]. . (Culicover 2009: 225). In (37b), the Agent argument by a leopard stays outside of the application of VP-ellipsis. In this case, the sentence is ungrammatical. On the other hand, as (37c) shows, when VP-ellipsis applies to the Agent argument, the sentence is fine. This contrast in grammaticality shows that the Agent argument appears within the verbal domain. Specifically, the argument checks off the [Ag]-feature on V. This is possible because the [Ag]-feature is passed down from v to V in the passive. I have argued so far that the passive differs in θ-feature inheritance from the active transitive. Extending this further, I will claim below that the passive undergoes reanalysis and it changes into the active transitive. This is a process from an accusative type to an ergative one. 4.2．Passives to ergative This subsection shows that some accusative languages have changed into ergative ones through the passive. The proto-languages in are said to have been accusative types at one stage. a.. Proto-Australian (Hale 1970; Anderson 1977). b.. Proto-Polynesian (Chung 1978; Hohepa 1969; Hale 1970). c.. Proto-Oceanic (Pawley 1973; Wouk 1996). d.. Proto-Austronesian (Estival and Myhill 1988). 73.

(13) 菅野悟. However, some of their descendants or modern languages show syntactic ergativity in some constructions. Moreover, the researchers in assert that the change occurs through the passive. For example, let me take a look at a sentence from East Futunan. This sentence expresses the syntactic ergativity. . na. ’amo. e. Past carry. tagata Alo. Erg man. a. Alo. le. Abs Def. fonu. (East Futunan). turtle. o. ave. e. lātou. ki. le. Puke. Comp. take. Erg. 3Pl. Loc. Def. Puke (Moyse-Faurie 2003: 11). ‘The men from Alo carried the turtle to bring it to Mount Puke.’. In , the embedded clause is the purpose clause, expressing the reason for the matrix event of the men’s carrying of the turtle. The key points to the current discussion are: ⒤ the Theme argument in the matrix determines the interpretation of the missing argument in the embedded purpose clause, and ⅱ the Theme argument serves as PRO. Neither of them is observed in accusative languages such as English. Rather, these two properties are found in syntactic ergative languages like Dyirbal. I show , which shows that the theme PRO is bound by the matrix Theme argument. . yabuk. ŋuma-ŋguj giga-n. [PROk/*j gubi-ŋgu. mawa-li]. mother-Absk. father-Ergj tell to do-Nonfut. PROk/*j doctor-Erg. examine-Purposive (Dixon 1994: 169). ‘father told mother to be examined by the doctor.’. Therefore, in contrast to accusative languages such as English, East Futunan behaves as a syntactic ergative language in this construction.7 The question is whether or not the change of East Futunan from an accusative to an ergative language occurred through the passive. Hohepa (1969) and Chung (1978) argue that it did. For example, Chung (1977, 1978) argues that the Proto-Polynesian language used to be an accusative type, although its modern descendants are divided into ergative and accusative types. In an accusative type, the active transitive occurs in the following structure (where the morpheme i is a case particle): . Tns Verb. Subj. i DO. (Proto-Polynesian). (Chung 1977: 6). Although there are some exceptions, most Polynesian languages are verb-initial, and Tense (abbreviated as Tns) precedes the verb. When the passive operation applies to the structure in , the structure is changed into . . Tns Verb-Cia. e Agent (=underlying Subj). Subj (=underlying DO). (ibid.). 7 Note that East Futunan belongs to Polynesian languages, which belong to Austronesian, and the Proto-Polynesian language used to be accusative at one stage.. 74.

(14) 対格言語と能格言語間の歴史的変化. The passive morpheme Cia is attached to the verb and the morpheme e is given to the Agent argument. On the other hand, in an ergative type, the sentence structure of transitive sentences is . . Tns Verb. e Subj. DO(ibid.: 7). At this point, I find parallelism between the passive structure of an accusative type in and the active transitive structure of an ergative type in . In both structures, the Agent argument is marked with the morpheme e, and the Theme argument is unmarked. The difference is attributed to the presence or absence of the passive morpheme Cia. Chung further argues that Proto-Polynesian used to be accusative with the active transitive structure in and the passive structure in . Some descendant branches of Proto-Polynesian subsequently have developed the transitive structure in through the reanalysis of . However, not all descendants have undergone this change. Therefore, the modern descendants of Proto-Polynesian are divided into the accusative type and the ergative type. The claim of Chung summarized here is consistent with the present claim. I have argued that the changes in type occur through the reanalysis of marked sentences as unmarked ones. I have shown that Warrgamay has taken some steps forward to an accusative type in accordance with the disappearance of the anti-passive morpheme. Similarly, in the case of the change from accusative to ergative languages, some descendants of Proto-Polynesian have undergone this change in accordance with the disappearance of the passive morpheme Cia. Therefore, I conclude that changes in both directions actually occur in language change. I would like to offer two more languages that behave the same as East Futunan in terms of these issues. Tagalog, often analyzed as a morphological ergative, shows syntactic ergative behavior in some particular constructions.8 For example, consider . . a. tindahan. b-in-ili. ng. lalake ang. in. store. IN-buy. AG. man PAT. diyaryo. at. newspaper and. b-in-asa. niya. IN-read. 3sg-AG. (ito). (Tagalog). this. (Foley and Van Valin 1977: 302). ‘In the store the man bought the newspaper and read it.’. In , the Theme argument diyaryo ‘the newspaper’ controls the interpretation of the missing argument in the second conjunct when the Theme argument ito is deleted. Tagalog used to be an accusative type and the change occurs through the passive (see (37b)). Indonesian shows syntactic ergative behavior in some constructions as well. For example, see . . Ahmad. me-. nemu-i. Ahmad. MEN meet CF. Ali,. lalu. di- pukul -nya.(Indonesian). Ali. then. DI beat. 3sg. 8 I assume that the properties of the syntactic ergative are not mutually exclusive to those of the morphological ergative within one language. It is possible for one language to show both syntactic and morphological ergative behaviors.. 75.

(15) 菅野悟. ‘Ahmad ran into Ali, and beat him up.’. (Verhaar 1988: 355). In , the two events are coordinated. The missing argument in the second conjunct is the Theme argument that is controlled by the matrix Theme argument.9 Finally, let me mention split ergativity in Dyirbal. This language is not a pure syntactic ergative language. Its pronouns inflect in an accusative paradigm. I claim that this is a residue of the earlier accusative pattern. As is well known, when the Agent is a pronoun, the application of the passive operation causes pragmatically unnatural sentences. As a result, the accusative pattern remains even in the modern language when the Agent arguments are pronouns. To summarize the discussion thus far, I have argued in this section that passives differ in θ-feature inheritance from the active transitive and that the change from an accusative type to an ergative one occurs when the passive having a particular type of θ-feature inheritance is reanalyzed as unmarked. As specific examples, I have offered East Futunan, Tagalog, and Indonesian. I have argued that, in modern descendants, syntactic ergative behavior is observed in particular constructions.. ５．Conclusion In this paper, I have provided a new minimalist approach to diachronic changes between accusative and ergative languages. I have adopted assumptions on θ-feature inheritance, and claimed that the uninterpretable θ-features originate on the phase head v and are passed down onto the lower V head. On the basis of these assumptions, I have proposed that the [Th]-feature goes down onto V in an accusative type and the [Ag]-feature undergoes inheritance to V in an ergative type. I have discussed the bidirectional change between the two. For example, Dyirbal is at the stage of the ergative type and English is at the stage of the accusative type. Thus, accusativity/ergativity can be captured in terms of the cyclic change between the accusative and ergative patterns. ＊. This paper is a part of joint work with Shin-Ichi Kitada and an earlier version of this paper was. presented at the 144th Meeting of the Linguistic Society of Japan held at Tokyo University of Foreign Studies in June, 2012 (Kanno and Kitada (2012)). I would like to express my deepest gratitude to Cedric Boeckx, and John Whitman for their invaluable comments and suggestions. This work is supported in part by a Grant-in-Aid for Young Scientists (B) (No. 26770169). All remaining errors and inadequacies are my own.. References Anderson, S. 1977. On mechanisms by which languages become ergative. In Mechanisms of syntactic change, ed. C. Li, 317-. 9 The data for Tagalog and Indonesian are rather controversial. I am not claiming that the data conclusively support the current analysis. Rather, the data are suggestive and could be analyzed under the current proposal. Needless to say, a precise analysis is left for further research.. 76.

(16) 対格言語と能格言語間の歴史的変化. 363. Austin: University of Texas Press. Bittner, M. and K. Hale 1996a. Ergativity: Toward a theory of a heterogeneous class. Linguistic Inquiry 27: 531-604. Bittner, M. and K. Hale 1996b. The structural determination of case and agreement. Linguistic Inquiry 27: 1-68. Bošković, Ž. and D. Takahashi 1998. Scrambling and last resort. Linguistic Inquiry 29: 347-366. Chomsky, N. 2008. On phases. In Fundamental issues in linguistic theory: Essays in honor of Jean-Roger Vergnaud, ed. R. Freidin, C. Otero, and M. Zubizarreta, 133-166. Cambridge, MA: MIT Press. Chung, S. 1977. On the gradual nature of syntactic change. In Mechanisms of syntactic change, ed. C. Li, 3-55. Austin: University of Texas Press. Chung, S. 1978. Case marking and grammatical relations in Polynesian. Austin: University of Texas Press. Culicover, P. 2009. Natural language syntax. Oxford: Oxford University Press. Dixon, R. 1981. Grammatical reanalysis: An example of linguistic change from Warrgamay (North Queensland). Australian Journal of Linguistics 1: 91-112. Dixon, R. 1994. Ergativity. Cambridge: Cambridge University Press. Estival, D. and J. Myhill 1988. Formal and functional aspects of the development from passive to ergative systems. In Passive and voice, ed. M. Shibatani, 441-491. Amsterdam: John Benjamins. Foley, W. and R. Van Valin 1985. Information packaging in the clause. In Language typology and syntactic description, volume 1: Clause structure, ed. T. Shopen, 282-364. Cambridge: Cambridge University Press. Funakoshi, K. 2009. Theta-marking via agree. English Linguistics 26: 1-32. Goodall, G. 1997. Θ-alignment and the by-phrase. CLS 33: 129-139. Hale, K. 1970. The passive and ergative in language change: The Australian case. In Pacific linguistic studies in honour of Arthur Capell, ed. S. Wurm and D. Laycok, 757-781. Canberra: The Australian National University. Hohepa, P. 1969. The accusative-to-ergative drift in Polynesian languages. Journal of the Polynesian Society 78: 295-329. Hornstein, N. 1999. Movement and control. Linguistic Inquiry 30: 69-96. Kanno, S. and S. Kitada. 2012. Synchronic differences and diachronic changes between accusative and ergative languages. paper presented at the 144th conference of the Linguistic Society of Japan. Kitada, S. 2012. Feature inheritance and four types of argument structure. paper presented at the 13th conference of the English Linguistic Society of Japan. Lasnik, H. 1995. A note on pseudogapping. MIT Working Papers in Linguistics 27: 143-163. Law, P. 1999. On the passive existential construction. Studia Linguistica 53: 183-208. Levin, B. 1983. On the nature of ergativity. Doctoral dissertation, MIT. Manzini, R. and A. Roussou 2000. A minimalist theory of A-movement and control. Lingua 110: 409-447. Marantz, A. 1984. On the nature of grammatical relations. Cambridge, MA: MIT Press. Moyse-Faurie, C. 2003. The ergative features of Papuan and Austronesian languages. Ergatividade na Amazônia II, Working Paper, 1-20, Paris: Centre d’Etudes des Langues Indigènes d’Amérique. Pawley, A. 1973. Some problems in proto-Oceanic grammar. Oceanic Linguistics 12: 103-188. Ura, H. 2000. Checking theory and grammatical functions in universal grammar. Oxford: Oxford University Press. Ura, H. 2006. A parametric syntax of aspectually conditioned split-ergativity. In Ergativity: Emerging issues, ed. A. Johns, D. Massam, and J. Ndayiragije, 111-141. Dordrecht: Springer. Verhaar, J. 1988. Syntactic ergativity in contemporary Indonesian. In Studies in Austronesian linguistics, ed. R. McGinn, 347-384. Ohio: Ohio University. Whitman, J. and Y. Yuko 2012. The formal syntax of alignment change. In Diachronic syntax: Parameter theory and dynamics of change, ed. J. Avelar, S. Cyrino, C. Galves and R. Lopez. Oxford: Oxford University Press. Wouk, F. 1996. Voice in Indonesian discourse and its implications for theories of the development of ergativity. Studies in Language 20: 361-410.. （旭川校准教授）. 77.

(17)