The Anaphor Agreement Effect and Georgian Anaphors* Nino Amiridze Abstract Standard Binding Theory does not give a straightforward account for the subject anaphor gap. Rizzi’s (1990) Anaphor Agreement Effect and its modified version (Woolford (1999)) were designed to account for that gap. The paper deals with Georgian anaphors representing counter-examples for both versions of the Anaphor Agreement Effect. It argues that the explanation for the presence of subject anaphors in Georgian can be given neither within Reinhart and Reuland’s (1993) “Reflexivity” framework as it was previously suggested in Anagnostopoulou and Everaert (1999) and Everaert (2001) for Greek and Georgian. In fact there is more than only the internal structure of anaphors and the related anaphoric properties responsible for the existence of subject anaphors. 1. Anaphor Agreement Effect explaining a gap in the paradigm of anaphors Rizzi (1990) offers evidence from Italian and Icelandic to argue that anaphors that are in agreement with the verbal complex are ungrammatical. The Icelandic data in (1a) and the Italian example in (2a) illustrate that when an anaphor is marked by nominative and, thus, triggers agreement like any nominative NP does in these languages, the sentences are ungrammatical. But if the anaphors are marked by a case other than nominative (i.e., * I am grateful to Martin Everaert for his help on every stage of writing this paper. I have benefited from comments from Rusudan Asatiani, Olivier Bonami, Patricia Cabredo-Hofherr, Peter Cole, Gabriela Hermon, Torsten Leuschner, Lea Nash, Eric Reuland, Kakhi Sakhltkhutsishvili, Kevin Tuite. I would also like to thank Xabier Artiagoitia and Itziar Laka for their help with Basque data and also everyone who participated in the discussion on Basque subject anaphors during December 2001 on the Basque linguistics e-list. This research was made possible through a grant of the Utrecht institute of Linguistics OTS in the context of the Language in Use project. 1 accusative in (1b) and genitive in (2b)) and, thus, do not trigger agreement the sentences are grammatical: (1) a. *Siggai telur aδ SIGi Sigga elski María. [Icelandic, Maling (1984)] thinks that REFL-NOM loves-SUBJ Maria-ACC “Sigga thinks that he loves Maria.” b. Siggai telur aδ María elski SIGi. Sigga thinks that Maria-NOM loves-SUBJ REFL-ACC “Sigga thinks that Maria loves him.” (2) a. *A loroi interessano solo se stessii. [Italian, Rizzi (1990)] to them.DAT interest.3PL only REFL.3PL.NOM “They’re only interested in themselves.” b. A loro importa solo di se stessi. to them.DAT matters only of REFL.3PL.GEN “They only matter to themselves.” Rizzi (1990) argues that what precludes a nominative anaphor is the presence of agreement and formulates a generalization which “holds quite systematically in natural languages” the following way: (3) The Anaphor Agreement principle Anaphors do not occur in syntactic positions construed with agreement The anaphor in (1a) is a subject and the sentence is ungrammatical while the same anaphor in object position (cf. (1b)) is fully grammatical. This must not give us a wrong idea that there are certain positions (i.e., object position) associated with anaphors where they can appear as opposed to other positions where one can never find anaphors. Rizzi (1990) gives instances of anaphors appearing as subjects (cf. (4)) and whenever they do they never trigger agreement with the finite verb: (4) Hún sagdi aδ she said sig vantadi peninga. [Icelandic, Maling (1984)] that REFL-ACC lack-SUBJ money “She said that she lacked money.” If an anaphor does trigger agreement, as is always the case for the subject position in languages like English, Italian and Icelandic, and is sometimes the case for an object 2 position (cf. (2a)), the sentences are ungrammatical. Therefore, not the subject/nonsubject status of an anaphor is crucial but the presence/absence of agreement (Rizzi 1990: 33). In order to explain the incompatibility of anaphors and agreement, Rizzi (1990) (following Picallo (1985)) claims that the agreement affix, assumed to be pronominal (Chomsky (1981)), and the anaphor will clash whenever forming a chain. Both members of the chain are never able to satisfy binding conditions simultaneously. The agreement being pronominal is subject to the Principle B and requires to be locally free while the anaphor has to satisfy principle A and requires to be locally bound. Under these considerations agreeing anaphors are correctly predicted to be ungrammatical. Since anaphors are incompatible with agreement, (subject) anaphors in the absence of agreement should be grammatical (unless blocked by other aspects of the binding theory) (Chinese, Korean, Malayalam, Khmer, Thai, Vietnamese, etc.), as predicted by the principle. The languages lacking agreement completely are a support for the hypothesis since they allow subject anaphors: (5) Zhangsan shuo ziji hui lai. [Chinese, Huang (1982)] Zhangsan said himself will come “Zhangsan said he will come.” 2. Modified Anaphor Agreement Effect 2.1 Languages with object agreement The Anaphor agreement principle implies not only that subject anaphors are ungrammatical but that object anaphors will also be excluded in languages with object agreement (Woolford (1999)). Data from object agreement languages support this prediction. These languages show that the normal object agreement cannot be used when the object is an anaphor. Some of them behave like, for instance, Swahili where anaphors can only trigger anaphoric agreement (cf. (6a)) but not the normal object agreement (cf. (6b)) while others, like Inuit, employ intransitive constructions reflecting only subject agreement for expressing reflexivity (cf. (7)): (6) a. Ahmed a-na-ji-penda mwenyewe. [Swahili, Vitale (1981)] Ahmed 3SUBJ-PRES-REFL-love himself (emphatic reflexive) “Ahmed loves himself.” 3 b. *Ahmed a-na-m-penda Ahmed mwenyewe. 3SUBJ-PRES-3OBJ-love himself (emphatic reflexive) “Ahmed loves himself.” (7) a. *Hansiupi immii asap-puq. [Inuit, Bok-Bennema (1991)] Hansi.ERG himself.ABS wash-IND.3SG.3SG. “Hansi washed himself.” b. Asap-puq. wash- IND.3 SG “He washed himself.” Woolford (1999) argues that if a language still allows agreement triggered by an anaphor, like the languages with object agreement do, then the agreement will exclusively be special - not normal object agreement characteristic to these languages but a special anaphoric form of agreement (cf. (6a)). Woolford (1999) thus modifies Rizzi’s principle as follows: (8) The Anaphor Agreement Effect (modified) Anaphors do not occur in syntactic positions construed with agreement unless the agreement is anaphoric. 2.2 Some counter-examples As Woolford (1999) notes, there are counter-examples to the modified version of the principle as formulated in (8). There are languages with object agreement that still have object anaphors, such as Tamil, Modern Greek, Albanian, Jacaltec, Salayarese and Georgian. Woolford (1999) tries to show that, for different reasons, the languages in question are not real counter-examples. She argues that, for instance, in Georgian the features of the agreement morpheme do not match those of the anaphor and the agreement is default form of agreement (thus, in the technical sense, no agreement). For instance, a reflexive phrase in Georgian always triggers 3rd person agreement even when it is bound by 1st or 2nd person antecedent (cf. (9)). Thus, only the possessor within the reflexive NP is coindexed with the antecedent while the agreement between the object NP and the verb is triggered by the whole phrase: 4 (9) (me) čem-s I.NOM my-DAT tav-s vakeb. head-DAT I.praise.him1 [Georgian, Harris (1981)] “I praise myself.” Woolford (1999) groups Albanian and Georgian together as using a default agreement strategy to avoid the violation of the Anaphor Agreement effect. In Albanian, like in Georgian, the anaphors trigger 3rd person agreement whatever the person of the antecedent: (10) a. Vetja më dhimset. [Albanian, Hubbard (1985)] REFL.NOM CL.1SG.DAT feel-sorry-for.3SG.PRES.NONACTIVE “I feel sorry for myself.” b. Dritës i Drita.DAT dhimset vetja. CL.3SG.DAT pity.3SG.PAST.NONACTIVE [Albanian, Massey (1990)] REFL.NOM “Drita pities herself.” Woolford (1999) argues that the subject-verb agreement relation in (10) is a manifestation of “default” agreement, and that such agreement does not count for the Anaphor Agreement principle because the agreement morpheme is an inert slot filler, not indicating actual agreement. According to Woolford (1999), Greek uses a different strategy to avoid the Anaphor Agreement effect. In Greek a reflexive NP consist of a definite determiner (i.e., ton (cf. (11)), a possessor (mu) and a head (eafton) and triggers some kind of nonanaphoric agreement (the accusative, masculine, singular clitic ton), contrary to the predictions of the both versions of the Anaphor Agreement Effect (cf. (3), (8)): (11) Egho ton xero ton eafton mu. I know DET.ACC.MASC.SG REFL POSS.GEN.1SG CL.ACC.MASC.SG “I know myself.” [Greek, Iatridou (1988)] But following Iatridou’s (1988) analysis, the element that agrees with the antecedent is, 1 Unless otherwise required, in the glosses only masculine singular forms (he/him/his) will be given; i.e., a verb form like vk′lav “I kill him/her/it/them”, will be glossed simply as “I kill him” for reasons of space. 5 in fact, the embedded possessive. Woolford (1999), thus, argues that the Greek case is no counter-example for the Anaphor Agreement Effect. In (11) not the anaphor (mu) is doubled by a clitic but the whole NP (ton eafton mu). Although the way the verbal complex agrees with subjects/objects is different in Georgian and Greek (person marking vs. clitic doubling) they both employ the same mechanism with regard to the reflexive NP—the whole phrase gets doubled in Greek (cf. (11)) / marked (in Georgian (cf.(9)) but not the possessive. The agreement mechanism remains the same irrespective of whether we adopt the same treatment of Georgian possessors as embedded anaphors and thus put Georgian into the same group as Greek according to the coindexing below: (9′) [(me)]i [čem-si tav-sj]j vakeb. (11′) [Egho]i tonj xero [ton eaftonj mui]j. We could also argue that Greek employs, in fact, the same “default agreement” strategy as Georgian. In both languages the whole reflexive phrase (but not its possessive) is in an agreement relation with the verbal complex (clitic in the case of Greek and a person marker in the case of Georgian). We could call the agreement a default agreement in the sense that these agreement markers are constant and do not vary with regard to the person feature of the antecedent. For instance, in Greek the clitic “triggered” by the reflexive with the 1st person antecedent in (11) will be the same Accusative, Masculine, Singular as the clitic triggered by the reflexive with 3rd person, feminine antecedent in (12): (12) I Maria ton DEF.NOM.FEM.SG Maria thavmazi ton CL(ACC.MASC.SG) admires DET.ACC.MASC.SG eafton tis. REFL [Greek, Iatridou (1988)] POSS.GEN.FEM.SG “Maria admires herself.” The same holds for Georgian, the reflexive phrase with 1st person antecedent (cf. (9)) will trigger the same agreement as the reflexive with 2nd (cf. (13a)) or 3rd person 6 antecedent (cf. (13b))2: (13) a. šen šen-s tav-s akeb. you.NOM your-DAT head-DAT you.praise.him “You praise yourself.” b. is tavis tav-s akebs. he.NOM POSS.REFL head-DAT he.praises.him “He praises himself.” 3. Does Georgian have default agreement? I will show that there is no reason to assume that Georgian has a default agreement strategy. If the language had the strategy, default agreement would hold only in the cases where the phrase šen-s tav-s is a reflexive like it is in (13). Although (14) shows that the same phrase also triggers 3rd person agreement when used non-anaphorically: (14) (me) sark′e- ši šen-s tav-s vxedav, t′an-s I.NOM mirror-in your-DAT head-DAT I.see.it k′i body-DAT while/but - vera. cannot “I see your head in the mirror, but I cannot see the body.” Therefore, the agreement is regular, triggered by the phi-feature specification of the head of the phrase tav-: 3rd person, singular. It is clear that not the anaphoric status of the phrase is responsible for triggering the 3rd person agreement in (9), (13) but the referential properties of the whole phrase, namely the person feature (3rd) of the head of the phrase (tav-). The fact that the agreement both in Greek and Georgian is always 3rd person singular irrespective of the phi-features of the antecedent is due to the morphosyntactic 2 Greek and Georgian reflexive constructions are also treated alike by Everaert (2001) calling them “inalienable possession anaphors” (cf. Anagnostopoulou and Everaert (1999)). Everaert (2001) argues that the possessive itself could not be called an anaphor (for the Binding Theory), only the combination of the possessive and the head noun have anaphoric properties. Following Anagnostopoulou and Everaert (1999) who showed that “α is an anaphor” is not equivalent to “α agrees with its antecedent” Everaert (2001) argues that the element that agrees with an antecedent as a result of an anaphoric relation should be distinguished from the anaphoric element itself. 7 properties of the specific type of anaphoric expressions these languages use not to their referential (anaphoric) status (cf. Everaert (2001)). Therefore, Georgian still remains a counter-example for the Anaphor Agreement Effect. The language has object agreement and also object anaphors which trigger neither anaphoric agreement, nor default agreement but normal object agreement, contrary to the Anaphor Agreement Effect (cf. (3)) or its modified version (cf. (8)). Observe that Georgian not only allows agreeing object anaphors but also agreeing anaphors in subject position, contrary to what Woolford (1999) claims. In discussing Georgian data Woolford (1999) mentions that the language does not allow subject anaphors and relates their ungrammaticality to the fact that Georgian allows “default” agreement in the construction where the anaphor is an object but not in the construction where it is a subject (Woolford (1999):272). However, certain verbs in Georgian do not show the usual subject-object asymmetry for binding (cf. (15a) vs. (15b)) and the reflexive phrase in subject position is perfectly grammatical with them (cf. (15a)): (15) a. šen-ma tav-ma gac′ama (šen). [Amiridze and Everaert (2000)] your-ERG head-ERG he.tormented.you you.NOM “It was yourself who tormented you / made you suffer.” b. šen you.ERG ac′ame šen-i tav-i. you.tormented.him your-NOM head-NOM “You tormented yourself.”; “You made yourself suffer.” 4. Alternative explanations 4.1 Binding Theory and the Anaphor Agreement effect It has been established that standard Binding Theory does not give a straightforward account for the nominative/subject anaphor gap. Rizzi’s hypothesis and its modified version (Woolford (1999)) were designed to account for that gap. However, the Georgian reflexive expression is a problem for both the original Anaphor Agreement Effect (Rizzi (1990)) and its modified version (Woolford (1999)) since the expression does trigger a regular agreement (not a default one as argued by Woolford (1999)) and contrary to both versions can appear as an object (cf. (15b)) and even as a subject (cf. (15a)). 4.2 Explanations within the “Reflexivity” framework In Anagnostopoulou and Everaert (1999) an explanation is given for the fact that Greek 8 allows a nominative/subject anaphor with certain verbs. Treatment of the Georgian data through the reflexivity framework (cf. Anagnostopoulou and Everaert (1999)) also seems to be possible. Let us, therefore, look into their analysis in some detail. Their explanation is formulated within the Reinhart and Reuland (1993) “Reflexivity” framework. In the Reflexivity framework, binding is not directly about the relative distribution of anaphors vs. pronominals but about reflexive predicates:3 (16) a. A reflexive-marked syntactic predicate is reflexive. b. A reflexive semantic predicate is reflexive-marked. Unlike standard Binding Theory, in Reflexivity the configurational effects of anaphora are not attributed to the Binding Conditions – which do not refer to the notion ccommand - but to the Condition on Chain Formation which is given in (17)4: (17) a. Condition on A-chain A maximal A-chain (α1,..,=αn) contains exactly one link—α1—which is +R. b. An NP is +R iff it carries full specification for phi-features (number, gender, person) and structural Case. In Reinhart and Reuland’s view, every lexical element is subject to A-chain formation under the conditions set out in (17). The chain condition interacts with the reflexivity conditions, and in the example 3 The notions reflexive marking and reflexive predicate are defined in (i): (i) a. A predicate (of P) is reflexive-marked iff either P is lexically reflexive or one of P's arguments is a SELF-anaphor;. b. A predicate is reflexive iff two of its arguments are coindexed. 4 The definition is based on the notion of Generalized chain condition as defined in (i): (i) C=(α1,...,αn) is a chain iff C is the maximal sequence such that (i) there is an index i such that for all j, 1≤j≤n, αj carries that index, and (ii) for all j, 1≤j<n αj governs αj+1 9 below I will explain how. Nominative Anaphors are excluded by the Chain Condition (17a) under the assumption that anaphors are typically -R. To illustrate this, consider the examples in (18): (18) a. The mani pleases himselfi. b. *Himselfi pleases the mani. (19) a. [(the man)i (+R), himselfi (-R)] b. [himselfi (-R), the mani (+R)] In (18) the predicates are both reflexive and reflexive-marked satisfying (16a,b). In other words, as far as the reflexivity conditions in (16) are concerned both (18a) and (18b) are grammatical. The ungrammaticality of (18b) is due to a violation of the Condition on A-chains: in (19b) the head of the chain is -R since the English anaphor himself is not fully specified for phi-features (cf. (17b)). So let us turn the argumentation around. If in a language (18b) would be grammatical, the chain condition, apparently, has not been violated. But that is only possible if the anaphor is marked (+R). Anagnostopoulou and Everaert (1999) claim that this is the case in Greek. O eaftos tu is headed by a noun (eaftos) which acts as a SELF-element while, at the same time, it is fully specified for phi-features ([masc],[3pers],inflected for [numb/case]), thus being +R. In fact Anagnostopoulou and Everaert (1999) argue that such an anaphor has the structure of an inalienable possession NP. (20) a. [O eaftosj tui]j ton The self his CL.ACC provlimatizi [ton Petro]i. puzzle-3SG the Petros.ACC “Himself puzzles Petros.” b. [(O eaftos tui)j (+R)] [(ton Petro)i (+R)] In (20) there can be no violation of the chain condition because the two coindexed elements tu and ton Petro do not form an A-chain (cf. (20b). 5. How Georgian tavis- tav- fits the typology of reflexive NPs 5.1 The relevance of the form of the anaphor In (21) the predicate is both reflexive and reflexive-marked satisfying Binding 10 conditions A and B. Neither Chain condition is violated there since given the structure of the anaphor there is no chain formation in Georgian - the two coindexed elements tavis- and president- do not form an A-chain: [tavis-mai tav-maj]j ixsna (21) POSS.REFL-ERG [p′rezident′-i]i. head-ERG he.saved.him president-NOM “The president was out of the hard situation only because of himself (his past doings, personal charm, etc.)” Therefore, analogically to Greek (cf. Anagnostopoulou and Everaert (1999)) we could argue that subject anaphors are allowed in Georgian thanks to their structure and properties as [+SELF,+R] elements which are able to escape Chain Formation due to their internal structure. However, it is not clear how it is possible that the same type of anaphor, for instance, in Georgian with the same ability to escape the chain formation is grammatical in subject position only with some verbs (cf. (15a), (21)) and not with others (cf. (22)): (22) *tavis-ma tav-ma daxat′a POSS.REFL-ERG REFL-ERG k′ac-i. he.drew.him man-NOM “Himself drew the man.” The same is true for the languages with the same type of anaphor, for instance, Greek and Basque where only some verbs allow reflexives as subjects (for Greek cf. (20a) vs. (23) and for Basque cf. (24a, b) vs. (24c)): (23) *Ton eafto tu the tu REFL POSS.ACC CL.DAT aresi o Petros.5 [Greek] like-3SG DEF Petros.NOM “Himself likes Petros.” 5 aresi is a subject experiencer verb (cf. (Anagnostopoulou (1999)) taking a dative experiencer as a subject and a nominative theme as an object. Thus, the anaphor in (23) is the subject argument of the verb. 11 (24) a. Niri batez ere neure I.DAT above buru-a-k ematen dit beldurra.6 all my.own head-DET-ERG [Basque] giveAUX “Above all it is myself that causes me fear.” b. Egunotan, neure buru-a-k kezkatzen nau. day.DET.in my.own head-DET-ERG worry.HAB me.has.it “These days, my(own)self worries me.” c. *Neure buru-a-k ikusi nau ni. my.own head-DET-ERG seen me.has.it I “Myself has seen me.” If only the structure of an anaphor matters (enabling to escape the chain formation) then the anaphors in these languages (or in any language with the same type of anaphor) have to be grammatical in any position (subject or object position) in any context irrespective of the verb semantic class. While the data shows that this is not the case in either of the languages. As Georgian (cf. (32)), Greek (cf. (20a)) and Basque data (cf. (24b)) suggest anaphoric expressions are allowed in subject position with object experiencer verbs while neither of the languages have reflexives as subjects with, for instance, subject experiencer verbs (cf. (25), (23), (26)): (25) *tavis tav-s POSS.REFL head-DAT uqvars ivane. 7 [Georgian] he.loves.him Ivane.NOM “Himself loves Ivane” (26) *Bere buru-a-k her Miren maite du. head-DET-ERG Miren.ABS love [Basque, Itziar Laka, p.c.] has “Herself loves Miren.” Therefore, I conclude that it is not (only) the structure of the anaphor which is responsible for the grammaticality of subject anaphors. 6 (24a) is taken from Hualde and de Urbina (in press) and (24b), (24c) are from I.Laka’s Basque Grammar Page. 7 Georgian subject experiencer verbs take a dative experiencer as a subject and a nominative theme as an object. The reflexive in (25) is the subject argument of the corresponding verb. 12 fear 5.2 Reciprocals as subjects in Georgian If the form of an anaphor (“inalienable possession” type (cf. Anagnostopoulou and Everaert (1999)) matters we should expect the reciprocals to be barred from subject positions since they have a different structure (i.e., they are not of the “inalienable possession” type, cf. (27)) than reflexives in Georgian: (27) a. ertmanet-i [Modern Georgian] RECIPROCAL-NOM “each other” b. ert-man-ert-i [Middle Georgian] one-ERG-one-NOM “each other” For Basque (cf. (29) taken from Hualde and de Urbina (in press)) and Greek (cf. (30a)) this prediction is borne out. However, Georgian, allows reciprocals as subjects (cf. (28a))8: (28) a. ertmanet-i RECIPROCAL-NOM gvaocebs [Georgian] čven. it.surprises.us we.DAT “We are surprised by (something related to) each other.” b. čven vaocebt ertmanet-s. we.NOM we.surprise.them RECIPROCAL-DAT “We surprise each other.” (29) a. *Jon-i eta Miren-i bat-a Jon-DATand Miren-DAT beste-a-k ematen die beldurra. one.DET other-DET-ERG give AUX [Basque] fear “One another cause Jon and Miren fear.” b. *Jon eta Miren bat-a beste-a-k kezkatzen ditu. Jon.ABS and Miren.ABS one-DET other-DET-ERG worry AUX “One another worry Jon and Miren.” (30) a. *O enas ton allo anisixun to Yani kje ti Maria. 9 The one the other worry [Greek] the Yanis and the Maria “Each other worry Yanis and Maria.” 8 Tuite (1998) also mentions the phenomenon. 9 I am grateful to Alexis Dimitriadis for the Greek sentences in (30). 13 b. O Yanis kje i Maria anisixuno The Yanis and the Maria worry enas ton allo. the one the other “Yanis and Maria worry each other.” All this might suggest that there is not a simple primitive concept anaphor and the distinction between reflexives and reciprocals has to be made (for the similar view cf. also Everaert (2000)) just like it is made for different types of reflexives (Reinhart and Reuland (1991, 1993), Safir (1996), Anagnostopoulou and Everaert (1999)). If we accept this, we should distinguish between reflexives and reciprocals and agree that in the distribution of the subject reflexives their form is a relevant factor: (31) If a language has a subject anaphor, the anaphor will be “non-pronominal”. i.e. its form (structure, properties) is relevant (Anagnostopoulou and Everaert (1999)). This is in accordance with the data since all the languages reported in the literature having subject reflexives have non-pronominal anaphors (Basque (cf. (24), (26)), Modern Greek (cf. (11), (12), (20a), (23)), Dargwa (Kibrik (1997)), Nepali (Bickel and Yadava (2000)), Albanian (cf. (10) and also Hubbard (1985), K. Williams (1988)), Toba Batak (Schachter (1984), Cole and Hermon (in preparation)). Although it is not clear and requires a cross-linguistic examination whether the reverse is also the case - whether languages having non-pronominal anaphors always show the phenomenon of subject anaphors. If the answer is “yes”, then the reason of having subject anaphors could lie in their structure. If not then either the structure of the anaphor is not relevant at all or there is also something else apart from the structure which needs further investigation. 6. The relevance of the verb classes allowing anaphors in subject position My hypothesis, still to be made more precise, is that it is not the structure of the anaphor but the semantic/structural properties of the classes of verbs that make it possible for reflexives to appear as subjects. No matter what the structure of the anaphor is, whether it is an inalienable type reflexive or non-inalienable type reciprocal, it can surface as a subject only with verbs of certain characteristics. 14 At least Georgian data shows that only causatives, both morphological (cf. Asatiani (1982), Boeder (1989)) and lexical (cf. Amiridze and Everaert (2000))10 are able to take reflexives as subjects. Apart from causatives any object experiencer predicate being treated as causative constructions (Pesetsky (1995)) can also take both reflexives (cf. (32)) and reciprocals (cf. (28a)) as subjects: (32) [tavis-mai tav-maj]j gaaoca [k′ac-i]i. POSS.REFL-ERG he.surprised.him head-ERG [Georgian] man-NOM “The man got surprised only because of something related to himself.” It is noteworthy that none of the subject experiencer verbs are able to have reflexive (cf. (25) or reciprocal subject (cf. (33)): (33) *ertmanet-s sz′ult ivane RECIPROCAL-DAT they.hate.them da meri. [Georgian] Ivane.NOMand Meri.NOM “Each other hate Ivane and Meri.” The same is true for Greek and Basque. In both languages reflexive phrases appear as subjects only with object experiencer verbs while the subject experiencer verbs are ungrammatical with reflexive subjects (cf. (34a) vs. (34b) (taken from Anagnostopoulou (1999)) for Greek and (24b) vs. (26) for Basque): (34) a. Tin Maria tin DEF Maria.ACCCL-ACC provlimatizi/enoxli/anisihi puzzles/bothers/worries o eaftos tis.[Greek] DEF REFL-NOM her “Maria is puzzled/bothered/worried with/at/by herself.” b. *Tin Maria DEF den tin thavmazi/aghapai o eaftos Maria.ACC not CL-ACC admires/likes tis. DEF REFL-NOM her “Herself doesn’t admire/like Mary.” 7. Summary 10 Here “lexical” causative refers to those transitive verbs which can be interpreted as “cause-to-Intransitive” no matter whether they have morphological causative marking or not. 15 According to Rizzi’s (1990) Anaphor Agreement Effect, anaphors do not occur in syntactic positions construed with agreement. And if they do the agreement is exclusively either anaphoric or “default” (cf. Woolford (1999)). The present paper argues that Georgian represents a counter-example for both versions of the principle. It shows that the agreement triggered by anaphors in Georgian is neither anaphoric nor default but normal person agreement. The paper tries to show that apart from agreeing object anaphors Georgian also allows agreeing anaphors in subject position with certain verbs / verb readings. It argues that it is not sufficient to relate the distribution and the form of the anaphors in subject position. Rather it seems necessary to take into the consideration the verb classes allowing the phenomenon. References AMIRIDZE, N. and M. EVERAERT. 2000. On subject anaphora in Georgian. Paper presented at the 10th Caucasian Colloquium, Munich. ANAGNOSTOPOULOU, E. 1999. On experiencers. In A. Alexiadou, G. C. Horrocks and M. Stavrou (Eds), Studies in Greek Syntax, pp. 67–93. Kluwer, Dordrecht. ANAGNOSTOPOULOU, E. and M. EVERAERT. 1999. Towards a more complete typology of anaphoric expressions. Linguistic Inquiry 30: 97–118. ASATIANI, R. 1982. Typological Analysis of Simple Sentence. Mecniereba, Tbilisi (in Georgian). BICKEL, B. and Y.P.YADAVA. 2000. A fresh look at grammatical relations in IndoAryan. Lingua 110: 343–373. BOEDER, W. 1989. Verbal person marking, noun phrases and word order in Georgian. In L. Marácz and P. Muysken (Eds), Configurationality. Foris, Dordrecht. BOK-BENNEMA, R. 1991. Case and Agreement in Inuit. Foris, Dordrecht. CHOMSKY, N. 1981. Lectures on Government and Binding. Foris, Dordrecht. COLE, P. and G. HERMON. (in preparation) Reflexives and c-command in Toba Batak. Ms. University of Delaware. EVERAERT, M. 2000. Types of anaphoric expressions: Reflexives and reciprocals. In Z. Frajzyngier and T.S. Curl (Eds) Reflexives: Forms and Functions. Benjamins, Amsterdam and Philadelphia. EVERAERT, M. 2001. Paradigmatic restrictions on anaphors. In K. Meegerdoomian, 16 L.A. Bar-el. Somervill, WCCFL 20 Proceedings, pp. 178–191.Cascadilla Press. HARRIS, A. C. 1981. Georgian Syntax. Cambridge University Press, Cambridge. HUALDE, J.I. and J. O. DE URBINA, eds. (in press) A grammar of Basque. Mouton de Gruyter, Berlin. HUANG, J. 1982. Logical Relations in Chinese and the Theory of Grammar. Doctoral dissertation . Massachusetts Institute of Technology, Cambridge, MA. HUBBARD, P.L. 1985. The Syntax of the Albanian Verb Complex.Garland, New York. IATRIDOU, S. 1988. Clitics, anaphors and a problem of coindexation. Linguistic Inquiry 19: 698–703. KIBRIK A.E. 1997. Beyond subject and object: Toward a comprehensive relational typology. Linguistic Typology 1: 279–346. MALING, J. 1984. Non-clause-bounded reflexives in Modern Icelandic. Linguistics and Philosophy 7: 211–241. MASSEY. V. 1991. Experiencers, themes, and c-command in Albanian. Ms. University of North Carolina at Chapel Hill. PESETSKY, D. 1995. Zero Syntax. MIT Press, Cambridge, MA. PICALLO, C. 1985. Opaque Domains. PhD dissertation. CUNY, New York. REINHART, T. and E.J. REULAND. 1991. Anaphors and logophors: An argument structure perspective. In J. Koster and E.J. Reuland (Eds), Long Distance Anaphora. Cambridge University Press, Cambridge. REINHART, T. and E. REULAND. 1993. Reflexivity. Linguistic Inquiry 24: 657–720. RIZZI, L. 1990. On the anaphor-agreement effect. Revista di Linguistica 2: 27–42. SAFIR, K. 1996. Semantic atoms of anaphora. Natural Language and Linguistic Theory 15: 545–589. SCHACHTER, P. 1984. Semantic-role-based syntax in Toba Batak. In P. Schachter (Ed), Studies in the Structure of Toba Batak. UCLA Occasional Papers in Linguistics 5. Los Angeles. TUITE, K. 1998. Kartvelian Morphosyntax: Number Agreement and Morphosyntactic Orientation in the South Caucasian Languages. LINCOM Europa, München. VITALE, A.J. 1981. Swahili Syntax. Foris, Dordrecht. WILLIAMS, K. 1988. Exceptional behaviour of anaphors in Albanian. Linguistic Inquiry 19, 161–168. WOOLFORD, E. 1999. More on the anaphor agreement effect. Linguistic Inquiry 30, 257–287. 17