7# ȋdH) %:_(ˇˇ<x ; [̣*7 ̣s0̣̣)̣̣̣̣̣̣Ergative Subjects* Colin Phillips MIT The most striking feature of .i.ergative ;systems is the fact that subjects of transitive and intransitive verbs behave differently for purposes of case and agreement. Why should this be so? One possible answer is that the transitivity of the verb directly affects how its subject behaves syntactically. That is, transitive and intransitive subjects have differing D-structure positions (Marantz 1984, B. Levin 1983), or conditions on case assignment force intransitive subjects to bear a particular case (Bobaljik 1993, Campana 1992, Chomsky 1993, Laka 1993, J. Levin & Massam 1985). Alternatively, the differential behaviour of subjects is not directly determined by the transitivity of the verb: an independent syntactic requirement, such as the .i.Extended Projection Principle ;(EPP, Chomsky 1982), causes subjects to behave differently in ergative systems. This paper argues for the second of these two possibilities. If the transitivity of the verb is directly responsible for the fact that transitive and intransitive subjects behave differently, then we expect the two kinds of subjects to always behave differently for case and agreement. On the other hand, if the EPP is responsible for the contrast between transitive and intransitive subjects, then we allow for intransitive subjects to behave just like transitive subjects provided that the EPP can be taken care of by another element, such as a non-argument. In other words, we allow for the possibility of variability in the behaviour of intransitive subjects. In section 1 I show that the Papuan language .i.Yimas ;(Foley 1991) provides an example of an ergative system with precisely this kind of variability: when the EPP is satisfied by a non-argument, all subjects are marked by ergative agreement. Section 2 shows how the variable marking of intransitive subjects, and an ergative pattern in which absolutive is assigned higher than ergative, follows from a theory of .i.economy of derivation;. Section 3 gives evidence from the distribution of .i.anti-agreement ;effects in subject extraction in Yimas, which support the idea that arguments may A-bar move for Case as a last resort. In section 4 I discuss the consequences of intransitive ergative subjects in Yimas for the analysis of subject-oriented phenomena in .i.ergative ;languages. Phenomena which treat subjects as a natural class are widely attested in ergative languages; most of the examples in the literature involve binding-related phenomena. It has been claimed that subjects are picked out as a natural class only by virtue of sharing the same VP-internal position. The subject property from Yimas that Im highlighting is a Case-related phenomenon, which implicates a position outside VP. This suggests a different account of which positions are targeted by subject-oriented phenomena. 1. Ergativity and the Extended Projection Principle The aim of this section is to show that the appearance of an ergative agreement system in Yimas is due to the effects of the .i.Extended Projection Principle;. Intransitive subjects are normally marked absolutive, satisfying the demands of the EPP. However, when the EPP can be independently satisfied, intransitive subjects are not absolutive but ergative identical to transitive subjects. Case distinctions are marked by inflectional affixes on the verb in Yimas. Independent nominals are caseless, and they are both liberally ordered and freely omitted. As a result, almost all of the sentences that follow consist of just an inflected verb. Case differences are encoded by different sets of agreement inflections. Agreement marking for 3rd person arguments follows a classic ergative-absolutive pattern. Objects of transitive verbs and subjects of intransitive verbs are marked identically, by absolutive markers (1a-b). There is a separate ergative inflectional paradigm for subjects of transitive verbs (1c). (1) a. pu- n- tay 3pl-Abs 3sg-Erg see He saw them. b. pu- wa -t 3pl-Abs go Perf They went. c. na- mpu- tay 3sg-Abs 3pl-Erg see They saw him. For our purposes here the most interesting property of Yimas is the fact that a given argument can be marked by different agreement affixes, depending on what other arguments and functional elements are in the clause. For example, a 2nd person subject is marked either nominative or absolutive. The factors governing the choice are given in (2). They appear complicated at first, but they reduce to a simple generalization. (2) 2nd person subjects are marked: Nominative - when there is a 3rd person object, marked absolutive. Absolutive - when the 2nd person subject is the sole argument. Absolutive - when there is a 1st person object, marked Accusative. (3) a. na- n- tay 3sg-Abs 2sg-Nom see You saw him. b. paNkra- wa -t 2pc-Abs go Perf You few (paucal) went. c. ma- Na- tay 2sg-Abs 1sg-Acc see You saw me. This alternation is explained by the general requirement that any finite verb form in Yimas be marked by absolutive inflection. This requirement overrides other principles of agreement marking. When another argument is marked absolutive, as in (3a), 2nd person subjects are marked by nominative agreement; but in the absence of another absolutive marked argument, 2nd person subjects satisfy the requirement for an absolutive. (4a-b) shows an alternation due to the same requirement, this time between between ergative and absolutive. (4a) repeats (1a), and shows that a 3rd person subject is marked ergative when there is a 3rd person object marked absolutive. In (4b), however, the object is 2nd person, and therefore marked accusative. In this case the 3rd person subject becomes absolutive, in order to make the verb well-formed. (4) a. pu- n- tay 3pl-Abs 3sg-Erg see He saw them. b. na- nan- tay 3sg-Abs 2sg-Acc see He saw you. I take the requirement for an absolutive agreement marker to be the reflex in Yimas of a version of the .i.Extended Projection Principle; (EPP), which requires that the head Agr10 be governed at S-structure. I assume that Agr1P is the XP immediately below CP, and that Agr1 is associated with absolutive case. Therefore, movement of a pro argument to the specifier of Agr1P both satisfies the EPP and triggers the realization of absolutive agreement. Crucially for the argument being developed, the presence of an absolutive agreement marker is only one of the ways of satisfying the EPP in .i.Yimas;. The EPP is also satisfied when the verb is prefixed by one of a small class of prefixes, which I assume to be of category C0. I assume that these become attached to the verb root and its inflectional material by head movement of V0 to C0 via Agr20 and Agr10. (5) Complementizer prefixes: m- relativizing complementizer ta- negation ant- modal - potential ka- modal - likely (6a-b) shows that these prefixes have the same effect on subject marking as the presence of a 3rd person object: the 2nd person intransitive subject is marked nominative, and the 3rd person transitive subject is marked ergative. (6) a. ta- nan- wa -r -um Neg 2pl-Nom go Perf PL You all didnt go. b. ka- mpu- Na- tput-n LIKE 3pl-Erg 1sg-Acc hit Pres They are going to hit me. Although the Complementizer prefixes do away with the need for an absolutive marker, they are not in competition with absolutive agreement. (7) shows that a Complementizer prefix and an absolutive agreement marker may cooccur. This also shows that what I have been calling absolutive case is not just a word-initial allomorph of the normal agreement markers. (7) ta- pu- n- tpul-c -um Neg 3Abs 3sg-Erg hit Perf PL He didnt hit them. With these preliminaries in mind, we can now ask the main question of this section: what happens to the .i.ergative ;agreement system in Yimas when the effects of the EPP are controlled for? In other words, does the ergative system arise because of the EPP? We have already seen what happens to 3rd person transitive subjects and objects when the EPP is independently satisfied. (6b) and (7) show that transitive subjects are ergative, and objects are absolutive. All that is missing are the intransitive subjects. To test how intransitive 3rd person subjects are marked when the EPP is independently satisfied, we need to look at examples of intransitive verbs with complementizer prefixes. Unfortunately, this test is not easy to apply, due to the following confound: most of the complementizer prefixes are part of a complex C0, containing an agreement suffix, such as the suffix -um in (7), which is in competition with Absolutive and Ergative prefixes. This agreement suffix agrees in number with the leftmost agreement prefix, which in turn leads to the deletion of that prefix, if it is Absolutive or Ergative. Nevertheless, we can show for each complementizer prefix, that it causes intransitive subjects to be treated as ergative rather than as absolutive. This confirms the claim that the .i.ergative ;agreement system is an artifact of the EPP. a. ka- ka- is the one complementizer prefix which does not cause the deletion of ergative or absolutive prefixes: this is because is does not introduce an agreeing suffix. Hence it provides the clearest test of how the EPP affects intransitive subjects. (8) shows an intransitive subject marked ergative following ka-. (8) balus-an ka- Nkl- ya -ka -arm -n airplane-Obl LIKE 3pl-Erg come Seq board Pres Those few will board the plane now. b. ta-/ant- With these complementizer prefixes it is impossible to see directly whether an intransitive subject is marked ergative or absolutive, due to competition with the complementizer agreement suffixes which these prefixes introduce. However, the form of the complementizer agreement suffixes used with instransitive subjects indicates that they are being treated as ergatives. For most numbers, the .i.complementizer agreement ;suffixes do not encode case differences, i.e. the form of agreement for absolutive dual is identical to the form for ergative dual, as in (9). (9) a. ta- mpu- tpul -c -rm Neg 3pl-Erg hit Perf DL They didnt hit those two. b. ta- pu- nan- tpul-c -rm Neg 3Erg 2pl-Acc see Perf DL Those two didnt hit you. Case differences are encoded for singulars, however: singular transitive objects (absolutive) are marked -ak, whereas singular transitive subjects (ergative) are marked by zero agreement (10a-b). In this respect, the intransitive subject in (10c) patterns just like the transitive subject in (10b). (10) a. ta- - mpu- tay -c -ak transitive object NEG 3sg-Abs 3pl-Erg see perf SING They didnt see him. b. ta- - kra- tpul - transitive subject NEG 3sg-Erg 1pl-Acc hit SING He didnt hit us. c. anan- - mal - intransitive subject POSS 3 die SING He almost died. This shows that, although not overtly marked, intransitive subjects behave as ergatives when ta- or ant- satisfies the EPP. c. m- The relative complementizer m- also introduces an agreeing suffix, which leads to deletion of absolutive or ergative markers in the same way as ta- and ant- do. So again we cannot directly see the form of intransitive subject agreement. But there is independent evidence that all 3rd person subjects are treated alike when m- takes care of the EPP. In this case, the evidence comes from wh-question formation. A clefting strategy is used for extraction of transitive or intransitive subjects (11a-b), but is not required for object extraction (11c). I assume that the clefting strategy, which supplies the C-prefix m-, is used as a last resort when there is no potential absolutive marker. Again the parallel between transitive and intransitive subject extraction indicates that 3rd person intransitive subjects are being treated as ergatives. (11) a. nawm m- - kul- cpul -um? transitive subject who-pl Comp 3pl-Erg 2pl-Acc hit PL question Who hit you all? b. nawn m- - na- ya -n -? intransitive subject who-sg Comp 3 DEF come Pres SING question Who is coming? c. nawn impa- - tpul? transitive object question who-sg 3sg-Abs 3sg-Erg hit Who did he hit? Reiterating the main point of this section: 3rd person agreement in Yimas follows an .i.ergative;-absolutive system, in which transitive and intransitive subjects are normally marked differently. We asked whether the contrast between transitive and intransitive subjects was a direct consequence of the transitivity of the verb, or whether it is the result of an independent phenomenon. We showed that case alternations in Yimas are explained by the demands of the EPP, and furthermore, that once the EPP is controlled for, transitive and intransitive subjects behave identically. Therefore, ergative case patterns are not directly determined by verb .i.transitivity;. Bobaljik (1993) claims that: given two structural ... Cases, languages must determine which will be realised on the sole argument of an intransitive clause. ...this is the result of a very simple parameter, the .i.Obligatory Case Parameter; On the contrary, weve seen that there is no obligatory Case in Yimas. This opens up the possibility that grammars in general do not need to include conditions which force a special treatment of intransitive subjects for Case purposes. However, we have only partly explained the behaviour of subjects in .i.Yimas ;so far. In transitive clauses, why do objects satisfy the EPP, rather than subjects, in contrast with English? And why are intransitive subjects marked ergative when the EPP is independently satisfied. Put in slightly more leading terms: why do subjects appear to prefer ergative to absolutive? The next section addresses these issues. 2. Deriving Agreement Alternations I assume a version of the Case theory developed by Shlonsky (1987) and Baker (1991) known as Generalized Visibility. The key claim of this approach is that Case licensing is a precondition for interpretation, where interpretation means phonetic interpretation at PF, and semantic interpretation at LF. (12) .i.Generalized Visibility If X is a potential Case-bearing element, X can be interpreted at level a only if X is Case-marked at level a. This implies that an expletive element, like there, which is overtly realized, but is presumably not interpreted, requires Case licensing only at PF. pro, on the other hand, which is interpreted, but not overtly realized, requires Case licensing only at LF. I also assume that agreement heads are potential Case bearing elements, and following reasoning from Baker (1991), that overt case-bearing heads force nominal arguments to be phonologically null. If we assume that Spec-head agreement can Case-license one element at a time, then a conflict arises in the situation where both the Specifier and the head require Case-licensing. Generalized Visibility offers a solution to this conflict, provided that the argument in the specifier position is pro. In such a situation, the agreement head but not pro requires Case-licensing at SS/PF, and pro, but not the agreement head, requires Case-licensing at LF. Since all of the actual arguments of the clause are pro, any overt nominal expressions that we see must be coindexed adjuncts. This explains the free ordering and omission of overt nominal expressions in a rich agreement language like Yimas. Generalized Visibility predicts which elements require Case-licensing, and at which levels, but does not predict where arguments will move for Case. I assume that syntactic movement conforms to a relativized notion of .i.economy of derivation;, in which the candidate set of operations, from which the shortest is chosen, consists of all operations which immediately satisfy the same requirement. The element which moves may satisfy a requirement of its own, or a requirement on the site which it moves to. So, if an NP requires Case itself, it moves to the closest position which can satisfy its Case requirement. On the other hand, if an agreement head needs a specifier to agree with, the closest available NP to that head is moved to its specifier. This view of economy differs from more familiar versions, in that it assumes no component of greediness (cf. Chomsky 1993). In (13a-b) I combine this version of economy with the assumption that lower elements in tree get the first opportunity to satisfy their requirements, and illustrate the derivations predicted for transitive clauses in a language with overt NP arguments (13a, eg. Icelandic), and a language with pro arguments and overt agreement heads (13b, Yimas). Depending on whether movement is driven by XPs or by heads, nested or crossing paths of Case movement result (cf. Murasugi 1992, this volume).   (13b) shows that in a transitive clause in .i.Yimas;, the EPP is automatically satisfied by the object moving to [Spec,Agr2]. This is not the case in an intransitive clause. Economy dictates that the subject move to [Spec,Agr2], and be marked ergative, just as the transitive subject does in (13b); this is only possible, however, when the EPP is independently taken care of (cf. 8, 10c, 11b above). When the EPP is not independently satisfied, the subject is forced to move to [Spec,Agr1], and it is marked by absolutive agreement. 3. Anti-Anti-Agreement The previous sections set out one argument for the transitive clause structure in (13b), which was based on the assumption that the EPP is a requirement on the head Agr1. In this section I present another argument for the structure in (13b), based on observations of where .i.agreement ;is and is not permitted in questions. In addition, this provides a reason to assume that XPs move for Case by S-structure in .i.Yimas;, contrary to recent proposals of Campana (1992) and Murasugi (1992); finally, Case-motivated movement turns out not to be uniformly A-movement. One of the most striking features of the wh-questions in (11) is that extracted arguments are not marked by agreement prefixes, as we have come to expect. This phenomenon is known as .i.anti-agreement;, and is found in subject extraction contexts in many languages with rich subject agreement (cf. Ouhalla 1993). In Yimas, anti-agreement is found with both subject and object extraction, as can be seen in (11a) and (11c), repeated below. (11) a. nawm m- - kul- cpul -um? subject extraction who-pl Comp 3pl-Erg 2pl-Acc hit PL Who hit you all? c. nawn impa- - tpul? object extraction who-sg 3sg-Abs 3sg-Erg hit Who did he hit? There is, however, one environment where extraction does not trigger .i.anti-agreement;. (14) is a question in which the extracted argument is marked by a normal agreement prefix. The only difference between (11a) and (14) is that the non-extracted argument, the object, is 2nd person in (11a), and 3rd person in (14). (14) nawn pu- n- tpul who-sg 3pl-Abs 3sg-Erg hit Who hit them? We might term the unexpected agreement in (14) .i.anti-anti-agreement;. I'll assume here that this effect is essentially the same as the effect of negation on subject extraction found in some anti-agreement languages, discussed by Ouhalla (1993). Ouhalla observes that in languages in which negation intervenes between subject position and [Spec,CP], it blocks the anti-agreement normally found with subject extraction in the language: .i.Welsh;, .i.Breton ;and .i.Berber ;show this interaction. On the assumption that negation occupies an A-bar position between IP and CP, anti-anti-agreement is triggered when subject extraction violates Relativized Minimality. The pair of examples in (15) come from Breton: (15a) shows anti-agreement with subject extraction; (15b) shows the reappearance of subject agreement in a negated relative. (15) a. Ar vugale a lenne (*lennent) al levrio the children COMP read read-3pl the books The children who read the books. b. Ar vugale ne (*lenne) lennent ket al levrio the children NEG read read-3pl NEG the books The children who did not read the books. (Hendrick 1988) Ouhallas explanation of anti-agreement focusses on the .i.A-bar Disjointness Requirement (ABDR) ;of Aoun & Li (1990, 1993), which requires roughly that a pronoun must not be bound by the most local A-bar binder. On the assumption that rich agreement licenses pro, and that pro may be the legitimate tail of an A-bar chain, pro cannot be the tail of an A-bar chain that satisfies Minimality: for if pro is the tail of such a chain, the ABDR will be violated. Impoverishing agreement is a way of preventing the licensing of pro in such contexts, and of thereby ensuring that the ABDR is satisfied. Rich subject agreement becomes possible in negated sentences like (15b), according to Ouhalla, because negation is the closest A-bar binder of pro. I suspect that the effects are incorrectly attributed to the ABDR. First, Aoun & Li (1990, 1993) argue that the ABDR must hold only at LF, since in .i.Chinese ;an illicit sequence *quantifieri...pronouni is improved by the insertion of a wh-operator which does not shield the pronoun until LF: quantifieri ... pronouni ... wh-phrasej. Meanwhile, related facts force Ouhalla to assume that the ABDR holds at S-structure: wh-in-situ appears not to trigger anti-agreement effects. This makes it difficult to account for anti-agreement effects and Chinese pronoun binding effects using the same principle. Second, Ouhalla claims that .i.anti-agreement ;involves impoverishing subject agreement in order to avoid licensing pro. This seems to be descriptively incorrect. .i.Yimas;, .i.Palauan ;(Georgopoulos 1991) and the languages cited by Ouhalla delete agreement with extracted subjects where at all possible. Default 3rd person singular agreement appears only where tense and agreement are spelled out as a portmanteau, i.e. only in cases where it is impossible to delete agreement without also deleting tense. If the function of anti-agreement is merely to avoid the licensing of pro, as Ouhalla claims, then we should not expect anti-agreement to impose so strong a requirement as complete deletion on agreement morphemes. As an alternative, I suggest that anti-agreement effects reflect different ways in which A-bar traces can be licensed. Where possible, familiar antecedent government relations hold; but antecedent government fails when an A-bar specifier intervenes. Where .i.antecedent government ;fails, a head chain may mediate the relation between [Spec,CP] and the extraction site. Each specifier is governed by its head, and the two heads are related by Bakers (1988) .i.Government Transparency Corollary;, or an analogue. The two ways of licensing A-bar traces are shown in (16a-b): I assume that licensing via a head chain is exploited only where normal antecedent government fails (16b: pro in Agr1 occupies an A-bar position), due to the extra chain-links involved, which are dispreferred for reasons of representational economy.  (16) a. [CP whi C+Agr2 [Agr2P ti tAgr2 ...  b. [CP whi C+Agr1+Agr2 [Agr1P proj tAgr1+Agr2 [Agr2P ti tAgr2 ...  But why does the person specification of an argument that is not extracted appear to trigger this effect? In both (11a) and (14), the site of wh-extraction is presumably [Spec,Agr2], but only in (14) does any material intervene between Agr2P and CP there is an object pro in [Spec,Agr1] in (14). For (11a) I assume that the 2nd person object is Case-licensed without needing to exit VP (see Phillips, to appear). If the object in [Spec,Agr1] occupies an A-bar position, then we predict the same anti-anti-agreement effect found in negated questions in .i.Breton;, .i.Welsh ;or .i.Berber;. Two questions arise at this point: first, why should the object in [Spec,Agr1] be in an A-bar position, given that the same position is generally taken to be an A position in familiar accusative languages? Second, why do we not find .i.anti-anti-agreement ;in the object question (11c), which also seems to involve extraction across a filled [Spec,Agr1]? For a possible reason why objects can only reach [Spec,Agr1] by A-bar movement, we can look to the criticisms of the derivation in (13b): Bobaljik (1993) and Chomsky (1993) argue that the nested paths derivation of a transitive clause in (13b) is impossible, since no pattern of A-movements will yield it, given the conditions on movement which they assume. We might in fact assume that Bobaljik and Chomsky are correct in their claim that objects cannot A-move to [Spec,Agr1] across a subject, but incorrect in their assumption that impossible A-movement entails impossible movement for Case. A-bar movement for Case is also possible, but only as a last resort, as argued in Miyagawa (1993). Also, assuming something akin to Chomskys (1993) conditions on A-movement offers an answer to the second question: why does extraction across a subject allow anti-agreement (11c)), whereas extraction across an object does not (14)? Objects can only reach [Spec,Agr1] by A-bar movement, which is why they induce Relativized Minimality effects for extraction, including .i.anti-anti-agreement;. Subjects can reach [Spec,Agr1] by A-movement, which is why they do not interfere with A-bar movement across them, and hence why they do not interfere with .i.anti-agreement;. In effect, what I am claiming here is that [Spec,Agr1P] in Yimas is mixed, in just the same way as Diesing has claimed for .i.Yiddish ;(Diesing 1990). A further consequence of the intervention effect diagnosed in (14) is that the Case-movements shown in (13b) must take place by S-structure, and cannot be delayed until LF. This conclusion is based on an observation of Ouhalla, who notices that anti-agreement effects are not found in languages which lack overt wh-movement in subject questions. Therefore, he argues, anti-agreement effects must reflect S-structure configurations. If anti-agreement effects reflected LF configurations, we would not expect variation in when wh-movement takes place to predict variation in the presence of anti-agreement effects. The conclusion that objects move to [Spec,Agr1] at S-structure in ergative systems conflicts with claims of Campana (1992) and Murasugi (1992), who both assume that at least .i.Case-movement ;of objects must be delayed until LF. 4. Subject Properties In the remainder of the paper I discuss possible cross-linguistic implications of the following findings about Yimas from sections 1 and 2: i. Transitive and intransitive subjects may occupy the same case position (Spec,Agr2) in an apparently ergative system. ii. In transitive clauses, objects are higher than subjects at S-structure. iii. [Spec,Agr1] is a mixed position. In an important 1976 paper, Anderson points out that despite the differences in case and agreement inflection between ergative and accusative languages, there are striking parallels across the two language types in binding phenomena: in transitive clauses subjects can bind object anaphors, but objects cannot bind subjects. Andersons response to the contrasting natural classes of arguments picked out by case and binding was to assume that the binding phenomena reflect true syntactic configurations, and that the case phenomena are merely morphological in nature. Faced with various kinds of evidence that the distribution of Case-marking does reflect syntactic properties, and armed with a wider range of syntactic positions, recent authors such as Campana (1992) and Murasugi (1992) have claimed that Andersons contrasts arise because different syntactic phenomena are sensitive to different classes of syntactic positions. Case sensitive phenomena involve Case positions like [Spec,Agr1] and [Spec,Agr2] in (13b) above. Subject oriented phenomena, on the other hand, such as binding, involve primarily thematic positions inside VP. Neither Campana nor Murasugi stipulate directly that different phenomena look at different syntactic positions. Instead, they achieve the contrast by assuming that different classes of phenomena are sensitive to different syntactic levels. Binding conditions crucially apply at S-structure, at which point objects are still inside VP, and hence asymmetrically c-commanded by subjects. This derives Andersons observations about .i.binding ;in .i.ergative ;languages. Case-sensitive phenomena refer to LF structures, in which objects now asymmetrically c-command subjects. We may term these approaches to Andersons Problem separation approaches. The findings about .i.Yimas ;listed in (i-iii) above challenge Campana and Murasugis assumptions in two respects. First, we found evidence in section 2 that objects move to [Spec,Agr1] by S-structure. Second, evidence from sections 1 and 2 shows that transitive and intransitive subjects have more in common than just originating in [Spec,VP]: section 1 showed that both transitive and intransitive subjects may occupy [Spec,Agr2]; a consequence of section 2 is that even when objects move across subjects for Case, the highest A-position in a clause will always be filled by a subject. We may then assume that subject-oriented phenomena target the highest A-position in a clause. Taken together, these findings remove the need for a separation approach to Andersons Problem. The binding facts noticed by Anderson follow straightforwardly from the standard assumption from accusative languages, that binding conditions apply to all and only A-positions. 5. Conclusions This paper has discussed the implications of a subject property rather different from most examples of subject properties in the ergativity literature. First, in Yimas it is possible to control for the effects of the Extended Projection Principle on agreement marking patterns, and it can be shown that it is the EPP that is responsible for the ergative agreement system. When the EPP is independently satisfied, all subjects are marked ergative. This has the consequence that grammars do not need to include conditions which apply specifically to intransitive verbs, or requirements that a given case be obligatorily assigned this conclusion diverges from a good deal of recent literature on ergativity. Second, alternations in agreement marking for subjects due to the EPP motivate an approach to Case movement in ergative and accusative systems based on a greed-free notion of economy of movement. Third, alternations in agreement marking for extracted subjects, due to the person of the object, lend support to the assumption that in ergative systems objects are case-marked (a) higher than subjects, and (b) in an A-bar position. References Anderson, Stephen. 1976. On the Notion of Subject in Ergative Languages. In C.N. Li (ed) Subject and Topic. New York: Academic Press. Aoun, Joseph & Yen-hui Audrey Li. 1990. Minimal Disjointness. Linguistics 28, 189-204. Aoun, Joseph & Yen-hui Audrey Li. 1993. Syntax of Scope. Cambridge, MA: MIT Press. Baker, Mark. 1988. Incorporation: A Theory of Grammatical Function Changing. Chicago: University of Chicago Press. Baker, Mark. 1991. On some Subject/Object Non-Asymmetries in Mohawk. Natural Language and Linguistic Theory 9, 537-76. Bittner, Maria. 1994. Case, Scope and Binding. Dordrecht: Kluwer. Bobaljik, Jonathan. 1993. Ergativity and Ergative Unergatives. In C. Phillips (ed) Papers on Case & Agreement II, MITWPL 19, 45-88. Campana, Mark. 1992. A Movement Theory of Ergativity. Doctoral dissertation, McGill University. Chomsky, Noam. 1982. Some Concepts and Consequences of the Theory of Government and Binding. Cambridge, MA: MIT Press. Chomsky, Noam. 1993. A Minimalist Program for Linguistic Theory. In K. Hale & S.J. Keyser (eds) The View from Building 20. Cambridge, MA: MIT Press. Diesing, Molly. 1990. Verb Movement and the Subject Position in Yiddish. Natural Language and Linguistic Theory 8, 41-79. Foley, William. 1991. The Yimas Language of New Guinea. Stanford: Stanford University Press. Georgopoulos, Carol. 1991. Syntactic Variables: A-bar Chains and Resumptive Pronouns in Palauan. Dordrecht: Kluwer. Hendrick, Randall. 1988. Anaphora in Celtic and Universal Grammar. Dordrecht, Kluwer. Laka, Itziar. 1993. Unergatives that Assign Ergative, Unaccusatives that Assign Accusative. In J. Bobaljik & C. Phillips (eds) Papers on Case & Agreement I, MITWPL 18, 149-172. Levin, Beth. 1983. On the Nature of Ergativity. Doctoral dissertation, MIT: MITWPL. Levin, Juliette & Diane Massam. 1985. Surface Ergativity: Case/Theta Relations Reexamined. In S. Berman (ed) Proceedings of NELS XV. Amherst, MA: GLSA. Marantz, Alec. 1984. On the Nature of Grammatical Relations. Cambridge, MA: MIT Press. Miyagawa, Shigeru. 1993. Case Checking and Minimal Link Condition. In C. Phillips (ed) Papers on Case & Agreement II, MITWPL 19, 213-254. Murasugi, Kumiko. 1992. Crossing and Nested Paths: NP Movement in Accusative and Ergative Languages. Doctoral dissertation, MIT: MITWPL. Murasugi, Kumiko. 1994. This volume. Lexical Case and NP-Raising. Ouhalla, Jamal. 1993. Subject Extraction, Negation and the Anti-Agreement Effect. Natural Language and Linguistic Theory 11, 477-518. Phillips, Colin. 1993. Conditions on Agreement in Yimas. In J. Bobaljik & C. Phillips (eds) Papers on Case & Agreement I, MITWPL 18, 173-213. Phillips, Colin. to appear. Verbal Case and the Nature of Polysynthetic Inflection. Proceedings of CONSOLE II. Shlonsky, Ur. 1987. Null and Displaced Subjects. Doctoral dissertation, MIT: MITWPL. * I would like to thank Mark Baker, Jonathan Bobaljik, Andrew Carnie, Ken Hale, Heidi Harley, Alec Marantz, Kumiko Murasugi, David Pesetsky and Andrea Zukowski for valuable discussion of the material in this paper. All of the Yimas examples and translations are taken from Foley (1991), to which I am greatly indebted. The glosses are my own interpretation of the data, and differ from Foleys in a number of places. All errors are entirely my own. Bobaljik, Chomsky and Laka assume that in an intransitive clause a setting of the Obligatory Case Parameter dictates the case of the subject. Campana assumes a similar requirement, which is not parameterized. Yimas distinguishes 4 different numbers: singular, dual, plural, and paucal, which generally refers to groups of 3-7. See Phillips (1993, to appear) for discussion of the person-based ergative split in Yimas. Roughly, 1st and 2nd person arguments may be Case-licensed through incorporating into the verb. pu- is glossed as a number neutral form here, although it is normally the 3rd person absolutive plural marker. This is because pu- can be used for all numbers in negated verbs. In these cases, the number suffix disambiguates the number of the object. Apart from combinations with ka-, which never involve deletion of agreement prefixes, the only exceptions involve combinations of the negative marker ta- with 3rd person agreement markers: the agreement prefixes are not deleted, rather they are replaced by the number neutral form pu-. See Phillips (1993, to appear) for discussion of why complementizer agreement competes with absolutive and ergative agreement. The oblique marker -an is the only case-marker found on independent nominals in Yimas. Nominals marked with -an are never associated with agreement inflections on the verb. It might be objected that the suffixes are encoding a thematic contrast rather than a case contrast in (10). This hypothesis is reinforced by the observation that singular 1st person subjects (i.e. nominatives) are also marked by a zero agreement suffix. However, if the suffixes reflect thematic distinctions, then we might expect agreement with a 2nd person object to be identical to that with a 3rd person object. (i) shows that singular 2nd person objects are also marked by a zero agreement suffix. Therefore, the suffix -ak is restricted to absolutives. i. ipa ta- mpan- tpul - 1pl Neg 1ag/2sg-Acc hit SG We didnt hit you. [The prefix (ka)mpan is a portmanteau used for combinations of a 1st person agent (any number) with a 2nd person singular patient. The free pronoun in (i) shows that the agent is plural, and therefore that zero agreement is with the object. Agreement with the plural subject would be -um.] Again, we need to exclude the possibility that Yimas simply chooses a different question form for subject and object questions, in which case the facts in (11) would not be telling us anything about Case distinctions. However, subject questions do not always require clefting: transitive subject questions in which the object is a 3rd person providing an absolutive to satisfy the EPP are not clefts. Therefore, the clefting strategy is not simply a property of subject questions. i. nawn pu- n- tpul who-sg 3pl-Abs 3sg-Erg hit Who hit them? [The lack of anti-agreement in this wh-question is an independent effect, discussed in section 3 below.] This prevents overt nominals from appearing in A-positions; it does not prevent them from being realized altogether. Both adjunct and A-bar positions are available for overt nominals. In fact, the specifier could fail to be pro, provided that it is some other phonologically null element, eg. wh-trace. This version of economy owes a lot to ideas in Murasugi (1992). They assume that no A-movement may cross two intervening A-specifiers. Miyagawa (1993) shows that although there is a scope interaction in Japanese between a head noun and the NP which it exceptionally Case marks genitive, the interaction disappears when a third nominal intervenes and is in an A-position. Miyagawa attributes the loss of scope interaction to the fact that the genitive NP is forced to A-bar move for case in the case with an intervening A-position. Examples are Basque (Control: Anderson 1976); West Greenlandic Inuit (Bittner 1994: Control, Reflexive & Pronominal Binding, Switch Reference); Abkhaz (Murasugi 1992: Reflexive Binding). This is not strictly correct for Murasugis theory: she assumes that subjects do raise to [Spec,Agr2] at S-structure, and only objects remain in situ. However this difference is not important to the points that follow.   In Grammatical Relations: Theoretical Approaches to Empirical Issues. C. Burgess, K. Dziwirek & D. Gerdts (eds.). Stanford: CSLI. (in press) uvpp}}4d WORD4 4"u #"v #"% #"% #"; #"; #" #" #"R #"R #"i #"i # ,Times .+gVP +V' +V (t (~t`W3"~`1# ( jAgr1P +Agr1' (:dAgr1 (9Agr2P +Agr2' (gAgr2 (Q| accusative ("> nominativeaG[0"V#aN-b0"(X#  ( (13) a.  +uagent +themeZad WORDa  a,Times .+jVP +V' +V  (agent  +theme"0## #"H## #"`## #"w## #"## #"r## #`,~#`l`#ac.w0")m#aH\0"[#  ( }Agr1P + Agr1' +Agr2P +Agr2' ( b. +Z.Agr1 +80Agr2  (#Z absolutive  +/t +t;-''d WORD' '"=##^#####"# ,Times .+@ antecedent government('s'd WORD' '"># >#"###"# ,Times .+s antecedent government" #"#IONw-3i-3Y-3d WORD-3 -3"<#" #"#" <#" #" #" #"M#"M#"M#" #" #" #  ### ,Times .+S 1" X +y2" ( 3"  ()hhead-mediated government"),!U"$1$%2&R&++\//00s0117122@2353- C+ *!+  C  ZGJSTuv=>|Z]cd   * + ( + 4 5 G {  6>(/1347CJghuwef. J   @ J    P.5QSnu%HIop9<HLbelo _ b k l !*!4#8#9#n#q$[$\%%% %!%Q%T%V%Y&&&&&&  J J  @    P&&&&&&'8'@(((())))***"*$*U*X**+++++++,+,,,C,F,I,L,S,,,,,-?-B-F-J-]-^-c-e-----...//000001 1 1'1:1B1h1i11112262S2Z2[2  J   @ J  R2222555555667;7>7D7E8899599::[:\::::;;J>M? ??z?????AACCDmDnDoDpDDDDFFGGGGHCHFHKHLIIIIJJJ H @ J   @  QJJJKKfKwKKKKL LLLMMMMO OOOOOO!O"O&O)O0O1OOQRRR8R@RARRRRS$S'SoSrSSTTU`UcUkUlUUUUUUUUUV V VVVW*WFWIWXWYWbWoWWWWWWWWWWXX#  @ @    @VX#YoYr[[[[[[[[\Y\\\\]<]=]G]J]N]O]]]b]f]g]m]q]w]x]~]]]]]]]]]]]]]]]]]]]^^_____```` ` ``aaaabbd2d3eeeef\f_fmfn J H8 @  H  Hh  @ Nfnffffjj!j/j0jWjmm.m/qqrrrrr r#r,r-sss s vvvw{8{C{D{{|||R|a||}<}b}}~ ~#~$~&~Q~p~~z"EeG&AAP7-F^y @  @ J   Tfg:;pqtop #$%'~GImuBC&)vwMN*+,./14u  H   @ J J @  M @#'()* E F G { |Q|}4FG]-AUqg$<Ml\  h@   h@ 8 0 h h h h h   <8Gaxy^_"# !%M%N%U%V&&&' '4'5'A'B()y)z)))**%*C*D+p+q++,,D,k,,,,,   h@ hh!@@  h!@@  h!@@ h  @h h   h@ h 8,-_-`-f-g001 1C1X111272_2s2t568888899:::;? B|DlDmDoDqDrɻ|vo! ! } h h  h  h!@@   h@  h!@@   h@   h@ hh!@@h &DrDsFFFFFHJJJKK*KyKKKLLM$MDMUMVPPPQ Q.QhQQQTW*Y];]<]>]v]w]yøø蔔荍zoi  h  hh   h@  @  h!@@   h@   h@ hh!@@h   P!p(]y]]`azd5gipjUjVjWjmjnjjkukkkm0n#paruvvvvvw[yzL{6{7{8{C{D{|(|~|}n}~:~-]h^wEƽ趯ۨ۟hh h h h 8 8 h    hh 8  h;Hef9:opno$J`anoABuvLM*+-.01yνηԱ hh 1h hh:18body" ' uge#[%,]/;IPz@.&2JX#fnUVWXYZ[\],Dr]y^_`abc chser? S w  HH(FG(HH(d'@=/ @H-:LaserWriter 8 TimesCourierSymbolIPATimesvvvEE)Ergative SubjectsMITWPL RTG Grant