OBJECT AGREEMENT, GRAMMATICAL RELATIONS, AND INFORMATION STRUCTURE IRINA NIKOLAEVA ABSTRACT
Transcription
OBJECT AGREEMENT, GRAMMATICAL RELATIONS, AND INFORMATION STRUCTURE IRINA NIKOLAEVA ABSTRACT
OBJECT AGREEMENT, GRAMMATICAL RELATIONS, AND INFORMATION STRUCTURE1 IRINA NIKOLAEVA University of California, San Diego ABSTRACT Northern Ostyak (Uralic) has optional object agreement. This paper analyzes the grammatical behavior of objects that trigger agreement and objects that do not, and demonstrates that while the former participate in certain syntactic processes, the latter are syntactically inert. The asymmetry cannot be explained with reference to semantics or argument status, as both objects bear an identical argument relationship to the predicate. Following the functional approach to language, under which the clause has three independent representational levels (syntax, semantics, and information structure), I suggest that the two objects differ in their information structure status. The object that does not trigger agreement bears the focus function, and systematically corresponds to the focus linear position. It is further argued that virtually all grammatical relations in Ostyak demonstrate reduced syntactic activity when they are in focus. This leads to a search for an information structure-driven motivation for certain behavioral properties. 1. Introduction This paper focuses on the syntactic characteristics of direct objects in a language with optional object agreement. It makes two major points. First, most current research on object agreement refers to the domain of the semantics-syntax interface: objects that trigger agreement and objects that do not are associated with distinct structural positions, and this asymmetry is essentially semantically driven. Below I analyze optional object agreement in Northern Ostyak, a Uralic language spoken in western Siberia. Contrary to current assumptions, in this language the two objects do not differ semantically or in terms of their argument relation to the predicate. This 2 raises the question of the possible motivation for object agreement. The paper suggests that the sole motivation is the information structure of the sentence, without any reference to semantics. Second, in Northern Ostyak the two objects exhibit different behavioral profiles, and the same discrepancy is observed in other grammatical relations. This indicates that certain syntactic properties do not necessarily follow from the grammatical relations born by the corresponding NPs, but arguably from their communicative characteristics. Thus, the paper presents evidence for direct syntax-information structure mapping, a type of relationship often ignored in syntactic research. My analysis crucially involves an approach allowing the clause to have three autonomous levels of representation, which are related via certain linking principles: semantic structure, information structure, and syntactic structure. Roughly speaking, the units of the semantic level are semantic roles and verbal semantics. Information structure is represented by the information structure statuses of the corresponding sentence elements, such as topic and focus. This paper concentrates on the syntactic level, represented by grammatical relations. Following a long tradition in linguistics, I identify grammatical relations as clusters of grammatical properties of two types: coding properties and behavioral properties.2 Coding properties pertain to the form and the position of the corresponding nominal itself, while behavioral properties concern its ability to determine the encoding of other elements, either within the clause or clause-externally. The nominals that can determine the encoding of other elements are syntactically active. Using the terminology of Relational Grammar they are called “terms”. Normally, only terms trigger anaphoric processes such as coreferential deletion, verbal agreement, and reflexivization. The following hierarchy of terms is assumed: subject, direct object, indirect object, and oblique object. Terms are opposed to syntactically inactive non-terms (or adjuncts), which lack behavioral properties. In section 2 I introduce the phenomenon of object agreement in Ostyak, its morphology, and previous proposals. Since the previous account, based on semantics, has not proved satisfactory, I will identify some particular problems. Central to the paper is section 3, where the grammatical relations within the clause are studied on the basis of syntactic tests, and direct objects are shown to display two different clusters of behavioral properties depending on whether or not they trigger agreement. Section 4 focuses on word order and shows that object agreement correlates with the linear position of objects. This leads to the question of the argument status of the two objects, which is addressed in section 5. It is shown that the two objects do not differ in their argument relationship to the predicate. Section 6 looks towards possible future analysis, and concludes the article by speculating that the two objects must differ in their information structure 3 status. It is further hypothesized that Ostyak has a structural position associated with the focus function, and that focus elements are characterized by a lower syntactic activity than the corresponding non-focus elements. 2. Object agreement in Northern Ostyak Among the Uralic languages, object agreement is present in Hungarian, Mordvinian, Ob-Ugrian (Ostyak and Vogul) and Samoyedic (Nenets, Enets, Nganasan, and Selkup). In all these languages, intransitive and transitive verbs display subject agreement affixes. Such forms are traditionally called the “subjective conjugation.” In addition, transitive verbs have another paradigm, called the “objective conjugation” (Collinder 1960; Sinor 1988).3 Verbs in the objective conjugation agree both with the subject and the direct object. Transitive verbs occur in either the subjective or the objective conjugation, so object agreement is optional, while subject agreement is obligatory. Morphologically, Uralic languages exhibit three types of objective conjugation:4 (i) In Hungarian and Samoyedic (see É.Kiss 1987 for Hungarian; Hajdú 1968 for Samoyedic), a verb in the objective conjugation does not take a distinct OM and SM, but one unsegmentable affix which both agrees with the subject and encodes the definite status of the object (the person and the number of the object is not marked). (ii) Mordvinian has the most complex paradigm: the boundaries between the OM and SM are clear-cut enough, and the OM marks both the person and number of the object (Keresztes 1989). (iii) In Ob-Ugrian (Honti 1984, Kálmán 1965), the objective conjugation forms contain the OM and SM, the OM indicating the number of the object (but not the person). The OM and SM exhibit different degrees of agglutination, depending on dialect. The syntax of object agreement also varies. This paper studies type (iii) as represented by the Northern Ostyak dialects.5 Northern Vogul seems to be similar to Northern Ostyak, while the other Ob-Ugrian dialects may differ. 2.1. Morphology As in other Uralic languages, in Northern Ostyak intransitive verbs agree with the subject (the subjective conjugation). The SM marks the subject for three persons and three numbers: SG, DU, and PL (example (1a)). Transitive verbs can agree with the subject alone (the subjective 4 conjugation, example (1b)), or with the subject and the object (the objective conjugation, examples (1c)-(1e)). In the latter case they display the SM in combination with the OM, which marks the number of the direct object (SG, DU, and PL) and precedes the SM. (1) a. ma jel∂n om∂s-l-∂m I at.home sit-T-1SG I am sitting at home. b. ma t m k laη wel-s-∂m I this reindeer kill-T-1SG I killed this reindeer. c. ma t m k laη wel-s-∅-em I this reindeer kill-T-SG-1SG I killed this reindeer. d. ma t m k laη wel-s∂-l-am I these reindeer kill-T-PL-1SG I killed these reindeer. e. ma t m k laη wel-s∂-ηil-am I these reindeer kill-T-DU-1SG I killed these (two) reindeer. The morphological templates of the subjective and objective conjugations are as follows: (2) a. subjective conjugation: stem-T-SM (intransitive and transitive verbs) b. objective conjugation: stem-T-OM-SM (transitive verbs) The objective and subjective conjugations differ formally in two respects: first, the objective verbs as opposed to the subjective mark the object number (the OM); second, the subject markers in the two conjugations are phonologically different. For example, in (1) the subjective SM for 1SG is -∂m, while the objective SM for 1SG is -em with the SG object and -am with the DU and the PL object. Similar distinctions can be made for the whole paradigm. In (3a) I present the agreement affixes of the subjective conjugation and in (3b) the agreement affixes of the objective conjugation. 5 (3) a. Subjective conjugation: SM SG DU PL 1 -∂m -(∂)m∂n -uw 2 -∂n -(∂)t∂n -(∂)ti 3 ∅ -(∂)η∂n -∂t b. Objective conjugation: OM + SM object SG DU PL 1 -∅-em -ηil-am -l-am 2 -∅-en -ηil-an -l-an 3 -∅-(l)li -ηil-li -l-∂lli 1 -∅-em∂n -ηil-m∂n -l-∂m∂n 2 -∅-l∂n -ηil-l∂n -l-∂ll∂n 3 -∅-l∂n -ηil-l∂n -l-∂ll∂n 1 -∅-ew -ηil-uw -l-uw 2 -∅-l∂n -ηil-l∂n -l-∂ll∂n 3 -∅-el -ηil-al -l-al subject SG DU PL Note that the OM for the SG object is always phonologically null, so the corresponding SM affix (for example, -em for 1SG subject) can be regarded as a portmanteau morpheme referring both to the subject and the direct object. Since in most of my examples the object is in SG, I will not mark zeros and will conventionally gloss the SM in the subjective paradigm as SUBJ, and the combination of the OM and SM in the objective paradigm as OBJ, without indicating the OM specifically. Subjective and objective conjugations are also opposed in the Imperative, not presented here. Independent of conjugation type, subjects are always encoded by the unmarked nominative, pronominal objects by the morphologically marked accusative, nominal objects by the unmarked accusative homonymous with the nominative. 6 2.2. Previous proposals In previous descriptions of Ostyak the presence of object agreement (the objective conjugation) has been stated to be conditioned by the definiteness of the direct object (Rédei 1965; Honti 1984, and many others). The term “definite,” in its turn, is taken to denote the formal properties of object NPs. According to Honti (1984, 99-100), the definite objects that trigger agreement in Ostyak are: a possessive NP, a personal pronoun, an embedded complement clause, and a null object. This list (sometimes augmented by nouns modified by a demonstrative pronoun) is traditionally repeated in almost all descriptions of Ostyak. The main reason for this account seems to be the influence of the Hungarian grammatical tradition. As mentioned above, in Hungarian the verbs in the objective conjugation do not actually show agreement with the object, but simply mark it for definiteness. They occur under strict conditions for the formal organization of the object NP. The list of definite direct objects that trigger the objective conjugation in Hungarian includes referential NPs of the following types: proper nouns, the nouns determined by a demonstrative pronoun or a definite article, possessive NPs, 3rd person personal pronouns, reflexive, demonstrative and some other types of pronouns, as well as embedded complement clauses, and null objects. The Hungarian objective conjugation is used if and only if the object belongs to one of the aforementioned formal categories while with all other types of objects only the subjective conjugation occurs. For Ostyak, the formal types of the object NPs are usually described just as is customary in grammars of Hungarian. But, as Honti himself notices, although the aforementioned types of definite objects can indeed trigger agreement, this is not necessarily the case. All of them (except for the null object) are compatible with the subjective conjugation as well. As demonstrated in (1b) and (1c), a direct object modified by the demonstrative pronoun t m 'this' may or may not trigger agreement. The same is true for pronominal objects or possessive NPs: in (4a) the verb is not marked for object agreement (it is in the subjective conjugation); in (4b) the objects are formally the same as in (4a), but they trigger agreement (the objective conjugation). (4) a. ma n η-en / n η xot-en I wan-s-∂m you-ACC you house-2SG see-T-1SG.SUBJ I saw you/your house. b. ma n η-en / n η xot-en I wan-s-em you-ACC you house-2SG see-T-1SG.OBJ I saw you/your house. 7 On the other hand, object types other than those listed above can also trigger agreement, such as the nominals modified by a weak quantifier in example (5). (5) a. l w amuj k laη ux∂l-s-∂lli he pa n wel-s-∂lli some reindeer follow-T-3SG.OBJ and not kill-T-3SG.OBJ He followed a/some reindeer but did not kill it/them. b. l w kat k laη ux∂l-s-∂lli he two reindeer follow-T-3SG.OBJ He followed two reindeer. So in Ostyak, objects identical in their formal expression may trigger agreement, or they may not. A purely semantic account based on the notion of definiteness, although well supported cross-linguistically, is also unsatisfactory: in both cases objects may be definite and indefinite, and sentences that minimally differ in the presence of object agreement have the same (truthconditional) meaning. Neither does specificity affect object agreement (at least in a direct way): although non-specific objects never trigger agreement, specific objects may either trigger agreement, or they may not (see the examples above). All this leads Honti to the conclusion that strict rules of object agreement are altogether lacking in Ostyak (Honti 1984, 1987). The same is stated in Gulya (1967), Bese, Dezs and Gulya (1970), Wickman (1970), Ganschow (1972), and Szalacsek (1984). Some additional remarks have also been made. Ganschow (1972) notices that in most cases (80% according to his statistics) the object that does not trigger agreement tends to be strictly preverbal. Gulya (1967), and later Szalacsek (1984), state that the definite status of an object in Ostyak does not depend on formal conditions, but on whether it has been previously mentioned. Szalacsek has also suggested that object agreement normally occurs with perfective verbs. De Groot (1983) and Marcantonio (1989) have argued that diachronically object agreement has to do with the topicalization of the object. These suggestions capture a certain descriptive intuition, but no detailed analysis has been provided, and how the aforementioned factors interact with each other remains unclear. This paper examines only the syntactic aspect of object agreement, completely ignored in previous studies. The crucial problems for me are the grammatical relations borne by the two types of direct objects, the object that does not trigger agreement (called O1 hereafter), and the object 8 that triggers agreement (called O2), together with the syntactic structures represented by corresponding constructions. In particular, the following questions will be addressed: (i) Which syntactic properties characterize O1 and O2? Note that the behavioral properties of elements of the Ostyak clause have neither been spelled out nor accounted for in previous work. (ii) Do O1 and O2 indeed have different linear positions, as noticed in Ganschow (1972), and how is this reflected in the syntactic configuration? (iii) Do O1 and O2 differ in their argument relationship to the predicate? Questions (i), (ii), and (iii) are studied in sections 3, 4, and 5 respectively. 3. Grammatical relations In previous descriptions of Ostyak the identification of grammatical relations relied only on the morphological coding, the ability to trigger agreement, and partly on word order. In this section I will discuss their behavioral properties.6 I begin with the identification of the subject because, as further analysis will show, some subject properties are shared by O2. Then, crucially to the argument of this paper, I will demonstrate that O1 and O2 are characterized by different grammatical behavior. 3.1. Subject properties Subjects are identified by a number of syntactic effects not shared by other grammatical relations. Control of coreference in converbial clauses. The predicate of the embedded clause is a non-finite verb (participle or converb).7 For some types of embedded clauses the only possible controller of the subject of the embedded clause is the subject of the matrix clause, i.e. the subjects of the two clauses must be coreferential. These are adverbial clauses of manner or concomitant action with the converb ending in –man. The embedded clause is shown in square brackets, with the symbol ∅ indicating a zero subject coreferential with the subject of the matrix clause: (6) p r∂ imiI m ηi-luwj mon -∂tk law-∂l [∅ i/*j/*k ari-man] old woman we-DAT tale-PL tell-T-3SG.SUBJ The old woman tells us tales while she is singing. sing-CONV 9 Example (7) indicates that subjects of the two clauses can be non-coreferential if the subject of the embedded clause bears the relation of inalienable possession to the subject of the matrix clause (null subject in this case). (7) [s m-∂l m n-man] ∅ jel∂ wantilij-∂s heart-3SG go-CONV ahead look-T.3SG.SUBJ Shei looked ahead being afraid (literally: heri heart being left). As shown by (6), lower grammatical relations cannot control the converb with –man. 8 Control of coreference in relative clauses. The predicate of the relative clause is a participle with the suffix –ti or –(∂/u)m, which precedes the head noun. Ostyak uses the so-called gap relativization strategy (Comrie 1989, 151-153), i.e. the grammatical role of the relativized noun is not overtly indicated within the relative clause. The relative clause itself, therefore, may be ambiguous and its meaning may only be recoverable through the context. Thus, (8) a e-m p t∂rt-∂m x tl father-1SG talk-PART day means both ‘the day my father was talking about’ and ‘the day when my father was talking’. As far as my material shows, every syntactic position is accessible to relativization. Crucially, the coreference of the relative clause subject with the subject of the matrix clause is obligatorily marked by the possessive affix attached to the relativized head noun. (9) ratp∂rxo-xišp∂rxo [[∅ x em-um] t xe-l ew∂lt] kim et-∂s Ratp∂rxo-xišp∂rxo hide-PART place-3SG from out come-T.3SG.SUBJ Ratp∂rxo-xišp∂rxoi left the place where hei/*j had hidden. In (9) the 3SG possessive marker -l on the head noun t xel indicates the coreference of the subjects in the matrix and the embedded clauses. In the absence of the possessive marker attached to the relativized noun the two subjects cannot be interpreted as coreferential, compare (9) with (10): 10 ratp∂rxo-xišp∂rxo [[∅ x em-um] t xi ew∂lt] kim et-∂s (10) Ratp∂rxo-xišp∂rxo hide-PART place from out come-T.3SG.SUBJ Ratp∂rxo-xišp∂rxo left the place where somebody had hidden. The possessive affix on the head noun indicates only the subjects’ coreference, the coreference of elements other than subject not being overtly marked.9 Coreference across clauses. The control of zero anaphora across clauses, namely between two conjoined clauses, identifies subjects rather than other grammatical relations. As shown in (11), the null copy can be controlled by the subject of the first clause, but not by the direct object, independent of whether it is O2 (as in (11a)) or O1 (as in (11b), and not by the indirect object (as in (11a)). (11) a. Petrai Juwan-aj ewe-lk Peter pa ∅ i/*j/*k m n-∂s m -s-li John-DAT daughter-3SG give-T-3SG.OBJ and go-T-3SG.SUBJ Peter gave his daughter to John and left. b. xsar i sow∂r j musm∂lt-∂s fox hare pa ∅ i/*j m n-∂s wound-T-3SG.SUBJ and leave-3SG.SUBJ The fox wounded the hare and left. Coreference with grammatical relations other than the subject is established through the repetition of the full NP in the conjoined clause or its substitution for the anaphoric pronouns it∂l, tumel 'that one, the latter'. Compare (11b) and (12): (12) xsar sow∂r musm∂lt-∂s fox hare pa tumel wound-T-3SG.SUBJ and that m n-∂s leave-3SG.SUBJ The fox wounded the hare and he (the hare) left. If the subject of the first clause corresponds to the non-Agent argument in the conjoined clause (Patient, Recipient, or some other roles), the mechanism of passivization is used to maintain subject coreference, as in (13). (13) xsar sow∂r musm∂lt-∂s fox hare lta ∅ mojpar-na x l-na m -s-a wound-T-3SG.SUBJ then bear-LOC fish-LOC give-T-3SG.PAS The fox i wounded the hare j, then the bear gave it i/*j some fish. 11 The passive construction in Ostyak is not possible with a pronominal Agent. The next example shows that if the Agent of the conjoined clause is a personal pronoun, passivization does not apply, but the first clause subject controls the null copy in the object position. (14) Juwan Petra resk∂-s-li John Peter itna ma ∅ wost∂-s-em hit-T-3SG.OBJ therefore I throw.out-T-1SG.OBJ Johni hit Peterj, therefore I threw himi/*j out. As can be seen from (14), the object null copy is controlled by the subject of the first clause (Juwan), but not by its object (Petra). As in the case of the null copy in the subject position, to establish coreference with the object its repetition or substitution for the anaphoric element is required: (15) Juwan Petra resk∂-s-li John itna ma tumel wost∂-s-em Peter hit-T-3SG.OBJ therefore I that throw.out-T-1SG.OBJ Johni hit Peterj, therefore I threw himj/*i out. The examples cited here demonstrate that subjects have more controlling properties than other grammatical relations: they control coreference in converbial embedded clauses and the null copy across clause; they also trigger possessive marking on the relativized noun. 3.2. Split in object properties The behavioral difference between O1 and O2 can be shown by the following tests. Control of coreference in participial clauses. As argued in section 3.1, in an embedded clause with the converbial predicate (-man converb) the two subjects must normally be coreferential. In the other type of embedded clauses, those with a participle ending in –ti or (∂/u)m, the two subjects do not have to be coreferential. The semantic type of such an embedded clause is marked either by the locative case marker –na/-(∂)n on postposition following it, for example: the participle, or by a 12 (16) a. [x l wel-ti jox nik j x∂t-m-el-na / j xt-∂m-na] a e-l xop-na fish kill-PART people down come-PART-3PL-LOC/come-PART-LOC father-3SG boat-LOC om∂s-l sit-T.3SG.SUBJ When fishermen came down to the river, his father was sitting in a boat. b. [kaš-em m n-t-al / m n-ti j pijn] li-ti pain-1SG go-PART-3SG /go-PART after pit-l-∂m eat-PART start-1SG.SUBJ I will start eating after my pain stops. As (16) indicates, the participle either takes a possessive affix agreeing with the subject of the embedded clause (j x∂t-m-el, m n-t-al), or it does not (j xt-∂m, m n-ti). The presence of such an affix is optional and conditioned communicatively, namely by the information structure status of the embedded subject. Basically, the possessive affix marks its topicalization. A discussion of this matter is irrelevant for the present paper (see Nikolaeva, Kovgan and Koshkareva 1993 for details), the important point being that the possessive affix on the participle is optional, as in (16). However, when the matrix and the embedded clause subjects are coreferential, the possessive affix must attach itself to the participle. Thus, in (17), as opposed to (16), the possessive affix is strictly obligatory (17) [a e-m t tj x sew∂r-m-al/*sewr-∂m sis] l w m sa nom∂lm-∂s father-1SG wood cut-PART-3SG/*cut-PART when he something remember-T.3SG.SUBJ When my father was cutting wood, he remembered something. This property also characterizes O2, the object that triggers agreement in the matrix clause. The examples in (18) demonstrate that the subject of the embedded clause can be controlled by the O2 and this determines the obligatory possessive affix on the participle. (18) a. [∅i ta -l sawit-t-al / *sawit-ti s xat] ittam jox-lal imi-xiliI pa herd-3SG graze-PART-3SG/*graze-PART when this people-3PL Imi-Xili again u -l∂-lli find-3SG.OBJ These people found Imi-Xili again when he was grazing his herd. 13 b. [x li un ul-m-al / *ul-∂m p ta] x l / ∅i n x n tal-s-em fish large be-PART-3SG/be-PART because fish out not carry-T-1SG.OBJ I didn’t take out the fish because the fish/it was large. So the possessive affix that is attached to the participle normally indicates the topicalization of the embedded clause subject. In the case that the embedded clause subject is coreferential with the subject or O2 of the matrix clause, such a possessive affix is strictly obligatory. Crucially, O1, i.e. the object that does not trigger agreement in the matrix clause, cannot be coreferential with the subject of the embedded clause at all. Thus (19) is ungrammatical with any form of the embedded predicate, whether it has a possessive affix or not. The difference between (19) and (18b) is the presence or absence of object agreement in the matrix clause. (19) * [∅ / x l un ul-m-al/ul-∂m p ta] x l / fish large be-PART-3SG/be-PART because fish n x n tal-s-∂m out not carry-T-1SG.SUBJ I didn't take out the fish because it was large. To conclude, when the subject and O2 control the subject of the adverbial participial clause, they always trigger possessive affixes on the participle, as opposed to other grammatical relations. O1 cannot control the subject of the embedded clause. Possessive reflexivization. In Ostyak, reflexive possessive affixes are phonologically identical to non-reflexive possessive affixes. The control of possessive reflexivization is characteristic of subjects, as is illustrated in (20). (20) a i p x-∂l resk∂-s-li father son-3SG hit-T-3SG.OBJ The fatheri hit hisi son. The O2 object is a potential reflexivization controller as well, so from this point of view it does not differ from the subject. In (21) the O2 object (s rt) controls the reflexive possessive affix in k tpe-l. 14 (21) ittam s rt k tpe-l ew∂lt m w-na l sk∂-s-li this pike middle-3SG from ground-LOC throw-T-3SG.OBJ He threw this pikei to the ground (holding it) in the middle (in itsi middle) (Pápay 2). Unlike the subject and O2, O1 cannot control reflexivization, so in (23) the object p x∂l cannot be interpreted as being referred to by the possessive affix in xot-∂l-na. (22) a i xot-∂l-na p x-∂l want-∂s father house-3SG-LOC son-3SG see-T-3SG.SUBJ The fatheri saw hisi sonj in hisi/*j house. Note that reflexivization is not ruled out in (22) because of the absence of a precedence relationship between O1 and the possessive NP xot-∂l-na. The subject and O2 can control a reflexive element even if they follow it in the linear representation of the clause, as shown in (23a) and (23b) respectively: (23) a. a e-l p x-∂l resk∂-s-li father-3SG son-3SG hit-T-3SG.OBJ Hisi fatherj hit hisj soni. b. [p x-∂l-na wos-na tuw-∂m unti] l w kolxoz-na rupit-∂s son-3SG-LOC city-LOC take-PART before she farm-LOC work-T.3SG.SUBJ Before heri son took her to the city, shei she was working on the farm. Thus, the ungrammaticality of (22) is not conditioned by the linear order, and indicates that O1, unlike the subject and O2, does not have controlling properties. Quantifier float. The universal quantifiers itliji/xoliji, sa/isa 'all'10 can be floated to the immediately postnominal position (as far as my material shows, no other constituents can intervene between the nominal and the quantifier). Both the subject and the O2 condition the quantifier float, i.e. the universal quantifiers either precede or follow them. See example (24a) for the subject and (24b) for O2. 15 (24) a. ( sa) awrem ( sa) or-na all child m n-∂s all forest-LOC go-T-3SG.SUBJ All the children went to the forest. b. l w ( sa) an∂t ( sa) il paj∂ t-s∂-lli he all cups all down drop-T-3SG.OBJ He dropped all the cups. On the other hand, O1 does not have this property. As shown by (25), with O1 the universal quantifier occurs only pre-nominally: (25) l w sa an∂t (* sa) il pajt-∂s he all cups all down drop-T-3SG.SUBJ He dropped all the cups. So the subject and O2 are opposed to O1 by this property as well. Possessor topicalization. In Ostyak, the dependent NP always precedes the head noun, so in the possessive phrase the possessor precedes the possessed. With the pronominal possessor the construction is head-marked, the possessive relationship being encoded by the possessive suffix on the possessed NP: n η jernas-en 'your dress-2SG', l w jernas-∂l 'his/her dress-3SG', etc. The pronominal possessor is easily dropped, therefore the possessive suffix is often the only overt marker of the possessor: jernas-∂l 'his/her dress'. With the nominal possessor, the possessive marker on the possessed noun is absent, the possessive relationship is marked neutrally from the morphological point of view and is encoded only by word order: Maša jernas 'Masha's dress'. The possessive affix is impossible in this case: *Maša jernas-∂l 'Masha's dress-3SG’. The possessor can be topicalized and fronted to the clause-initial position. In this case it can be separated from the possessed NP by other constituents, which is absolutely impossible in the case of the regular possessive construction. Since discontinuous constituents are otherwise not present in Ostyak, we might expect then that the possessive relationship should be marked locally. Indeed, even if the fronted possessor is a noun, the possessed obligatorily has a possessive affix attached, which anaphorically refers to it. Thus, in (26), in the construction imi lik-∂l ‘woman’s anger’, the 3SG possessive affix is obligatory because the possessor imi is topicalized and separated from the possessed by the adjunct ijolti ‘always’. 16 (26) imi ijolti lik-∂l/*lik et-∂l naw∂rniη pela woman always anger-3SG/*anger come-T.3SG.SUBJ frog to The woman, she is always angry with the frog (her anger always comes to the frog). Possessor topicalization occurs if the possessed noun bears the grammatical relation of the subject, as in example (26), or of O2, as shown in (27). (27) Juwan motta xot-∂l k al∂-s-em John before house-3SG see-T-1SG.OBJ I saw John's house before. Alternatively, O1 cannot license possessor topicalization, compare (27) with the ungrammatical example (28), where object agreement is not present. (28) *Juwan motta xot-∂l k al∂-s-∂m John before house-3SG see-T-1SG.SUBJ I saw John's house before. 3.3. Summary of behavioral properties The properties of the subject, O1, and O2, discussed in sections 3.1 and 3.2, are summarized below: (29) Property S O2 O1 Verbal agreement + + - Control of coreference in converbial clauses + - - Control of coreference in relative clauses + - - Control of the null copy across clause + - - Control of coreference in participial clauses + + - 17 Quantifier float + + - Control of possessive reflexivization + + - Possessor topicalization + + - The following generalizations arise from the data in (29). First, as expected, the most easily identified and syntactically active grammatical relation is the subject, which is superior to other grammatical relations in its controlling properties. Second, and crucially for the purposes of this paper, O1 and O2 have radically different syntactic characteristics. O2 shares certain subject properties. In addition to determining verbal agreement, it triggers quantifier float and possessor topicalization, and can control both possessive marking in the adverbial participial clause and possessive reflexivization. O1 lacks all these properties. Since it is syntactically inert, it should be qualified as a non-term. 3.4. Argument against O1 being a non-term The question addressed in this section is whether O1 can indeed be identified with the nonterms. In Ostyak the following elements can be described as non-terms: the passive agent; the instrumental, locative or manner phrase encoded by locative case –na/-(∂)n; the locative phrase encoded by the locative or lative/translative case –a/-ji; and postpositional phrases with different semantics. These elements are syntactically inactive in Ostyak: normally they do not trigger possessor topicalization and quantifier float and do not have any controlling properties (some exceptions will be discussed in section 6). However, there is a significant difference between these elements and O1 in at least one respect. An important, perhaps definitional feature of non-terms (adjuncts) is that they are not subcategorized for by a lexical head, and therefore can in principle be removed without being anaphorically controlled and without affecting well-formedness. In other words, non-terms crosslinguistically license the non-specific null word with an existential or universal interpretation (see Polinsky 1993 on the licensing of referential and non-specific null word by different grammatical relations). For example, as in a large family of languages, in Ostyak non-terms that correspond to passive agents license the non-specific null word, compare (30a) and (30b): 18 (30) a. a e-l-na kesi jel l sk∂-s-a father-3SG knife away throw-T-3SG.PAS The knife is thrown away by his father. b. kesi jel l sk∂-s-a knife away throw-T-3SG.PAS The knife is thrown away (by anybody). Other types of non-terms also license non-referential null. Crucially, unlike regular non-terms, O1 must be overtly present in the clause. Sentences with the transitive verb in the subjective conjugation without an overt object are strictly ungrammatical. See (32b), which demonstrates that O1 can license neither a referential nor nonreferential null word, in contrast with O2, which licenses the referential anaphoric null in (31a): (31) a. ma ∅ wan-s-em I see-T-1SG.OBJ I saw it/him/*something. b. * ma ∅ wan-s-∂m I see-T-1SG.SUBJ I saw something/it/him. This immediately qualifies O1 as different from regular non-terms. Thus, although O1 demonstrates the loss of certain term properties, there seems to be a basic difference between it and non-terms: unlike non-terms, O1 must be overtly present in the clause. This implies that both O2 and O1 are terms, although they are characterized by different syntactic behavior. Some alternative analyses will be discussed in section 5. In the next section I will identify a regular distinction between the linear positions of O1 and O2. 4. Constituency Object agreement correlates in a particular way not only with the behavioral properties of the object, but also with word order. To demonstrate this, I present the statistics on the position of O1 and O2 with respect to another sentence element, denoted here as X. By X, I mean any element other than those which can be identified as the subject and direct object, excluding also functional and parenthetical words. Roughly speaking, X stands for indirect objects, oblique objects, and 19 adjuncts. Northern Ostyak is a fairly rigid SOV language that exhibits the typological features of the Greenberg XXIII type (Greenberg, 1966): adjectives and relative clauses preceding the head noun, postpositions, etc. The SOV order is the most common statistically, but some elements, including what I call here X, can violate verb finality in certain pragmatically strong contexts. Subject initiality can be violated by ‘scene-setting’ spacio-temporal elements or topics. In normal cases element X follows or precedes the direct object, so the most frequent orders are SOXV or SXOV. I will concentrate on the clauses with one element X, which constitute the overwhelming majority in the texts. The following table is based on the word orders attested for transitive clauses in Pápay (1906-1908).11 (32) 403 sentences with O1 611 sentences with O2 sentences % sentences % SOV(X) 329 81 199 32.4 SXOV 39 10 14 2 SOXV 35 9 155 25.5 OS(X)V 0 0 10 1.6 S(X)VO 0 0 7 1 Table (32) clearly shows different word order patterns in clauses with O1 and O2. For O1 the most common pattern is SOV, while types SOXV and SXOV are fairly rare, and all the other types are altogether impossible. O1 tends to be strictly preverbal (this is compatible with the suggestion in Ganschow 1972); in my material the non-preverbal O1 is observed only in 9% of cases in pattern SO1XV, and this will be discussed specially in section 5. The statistics also confirm that O1 cannot be dropped from the clause: in the analyzed corpus of texts, the null object was attested 223 times (in clauses S(X)V(X)), and in all of them the verb had the objective conjugation form (36.5% of the overall number of objective conjugation clauses). On the other hand, for O2 all word orders are available (though some of them are less frequent). In the clauses with object agreement the object can be absent, separated from the verb by another constituent X, and sometimes postponed or fronted before S. So O2 does not have to be overtly expressed, and, lacking a fixed linear position, is more independent of the verb than O1. Modern syntactic research usually connects object agreement with structural position: the object that triggers agreement is analyzed as VP-external, and the object that does not trigger agreement as VP-internal. The independence of the object that triggers agreement from the verb is 20 explained by syntactic mechanisms associated with semantic effects (Mahajan 1989, 1991; Diesing 1992; Georgopoulos 1991, 1992; Diesing and Jelinek 1993; Woolford 1995; among others). Certain semantic features of the object (such as specificity, definiteness, humanness, animacy) condition its lack of abstract Case (De Hoop 1992), its particular phrasal scope properties (Diesing 1992; Runner 1995), or, under the minimalist assumptions, a strong Case feature in the agreement phrase (Chomsky 1995). In some languages this forces the object to be overtly excluded from the VP and be moved to (or generated in) the VP-external position. In languages without object agreement this merely results in the so-called “scrambling” or object shift (Webelhuth 1992; de Hoop 1992; Holmberg and Platzack 1995; Vikner 1995, 97-100, and references therein), while in languages with overt agreement morphology the object triggers agreement from that position. Although semantically driven analysis is not satisfactory for Ostyak (see section 2.2), the data in (32) also leads to the conclusion that O1 forms a single constituent with the verb, while O2 does not. In conformity with the cross-linguistic data, O1 can be described as VP-internal and O2 as VP-external. The VP-external status of O2 follows from the presence of the lexical material intervening between O2 and the verb, while the VP-internal status of O1 follows from the O1-V adjacency restriction. Such an analysis is strengthened by the following two considerations. First, under the configurational approach, the VP-internal position must be structurally subordinate to the positions of the other NPs in the clause (which arguably represents an otherwise flat structure in Ostyak) so that the VP-internal element does not bear command relations to them. This may explain why O1 is inert with respect to at least some of the syntactic facts discussed in section 3, i.e. those properties that are structure-sensitive (such as quantifier float and binding). Second, no phenomenon in Ostyak is known that treats O2 and the verb as a constituent, but the constituency of O1 and the verb is supported by the following tests. (i) The sequence O1 + V can be questioned, while O2 + V cannot. Compare (33b) and the ungrammatical (33c), both being a potential answer to the question in (33a): (33) a. Petra m j wer-∂s? Peter what do-T.3SG.SUBJ What did Peter do? b. l w t m keši wo k-∂s he this knife throw-T.3SG.SUBJ He threw this knife. c. * l w t m keši wo k-∂s-li 21 he this knife throw-T.3SG.OBJ (ii) Verbs in the subjective form cannot coordinate, leaving the object stranded (as in 34a), while verbs in the objective form can (as in 34b). (34) a. * ni w -l-∂t pa lot-l-∂t woman take-T-3PL.OBJ and sell-T-3PL.OBJ They take a woman and sell her. b. ni w -l-el pa lot-l-el woman take-T-3PL.OBJ and sell-T-3PL.OBJ They take a woman and sell her. (iii) The sequence O2 + V cannot be coordinated while O1 + V can. In (35a) two verbal constituents are coordinated and the manner adverb takes scope over the two conjuncts; alternatively, (35b) demonstrates that the two strings O2 + V do not coordinate at all. (35) a. jam∂ well woj wel-∂l pa x l wel-∂l animal kill-T.3SG.SUBJ and fish kill-T.3SG He kills animals and fish well. b. *jam∂ woj wel-l∂-lli well pa x l wel-l∂-lli animal kill-T.3SG.OBJ and fish kill-T.OBJ (iv) As mentioned above, O1 must be overtly present in the clause, therefore the deletion of O1 in the stripping structures in not possible, as indicated by (36c). O1 can only be deleted together with the verb, as in (36b). (36) a. Petra woj wel-∂l pa Juwan woj wel-∂l Peter animal kill-T.3SG.SUBJ and John animal kill-T.3SG.SUBJ Peter kills animals, and John kills animals b. Petra woj wel-∂l pa Juwan i Peter animal kill-T.3SG.SUBJ and John too Peter kills animals, and John too c. * Petra woj wel-∂l pa Juwan i wel-∂l Peter animal kill-T.3SG.SUBJ and John too kill-T.3SG.SUBJ 22 Alternatively, O2 alone can be deleted in the same structure: (37) Petra woj wan-s∂-lli pa Juwan i wan-s∂-lli Peter animal see-T.3SG.OBJ and John too see-T.3SG.OBJ Peter saw the animal and John saw it too. To sum up, the VP-internal status of O1 and the VP-external status of O2 allows us to account for a number of facts discussed above, namely the adjacency restriction, and under the formal approach, some structure-sensitive properties of O1 and O2. It is also confirmed by a number of constituency tests. So in what follows I will refer to O1 as VP-internal and to O2 as VP-external (the actual status of what I call here a “VP” will be addressed later in section 6). The next question then is, which syntactic position is associated with O2? Current syntactic theories provide different positions for the VP-external object that triggers agreement, either Spec of VP (Georgopoulos) or under a more recent approach Spec of AgrO (Woolford; Mahajan; Chomsky 1995). O2 in Ostyak is not limited to a single position, so the approach which relies on movement to a particular position would have to make additional claims explaining the difference between the orders O2S(X)V, SO2XV, and S(X)VO2 (see table (30)). The free position of the VP-external object in some languages is captured through the mechanism of (topic)-adjunction to any maximal projection outside the VP, rather than placement to a particular category (Bresnan and Mchombo 1987; Choi 1996). This mechanism seems to describe the facts of Ostyak word order better than movement to a specific position, because it can account for the relatively free place of O2. 5. Argument status of O1 and O2 In the previous section I have argued that O1 is VP-internal and O2 is VP-external. However, the relationship between the VP-external element and the verb, on the one hand, and the VP-internal element and the verb, on the other, leaves a number of alternatives. In section 5.1 I focus on the syntactic and argument status of O1, and in section 5.2 on that of O2. 23 5.1. Arguments against the incorporation analysis An immediate suggestion about the status of O1 might be that it undergoes a process of incorporation into the verb. Incorporation is here defined as the grammatical process of the formation of a compound element (the verb in this case) from two or possibly more lexical roots. The incorporated nominal normally loses most of the syntactic functions that the free element has in the corresponding construction, while the head element (the verb) retains its properties. Under the lexical approach to incorporation, it has been argued that this operation might change the argument structure but does not necessarily do so (Rosen 1989; Spencer 1995). For O1, the following facts, demonstrated above, would be consistent with incorporation analysis. First, O1 was shown to have limited term properties. Although not every syntactically inert element presents an instance of noun incorporation, every incorporated element is characterized by the absence or at least the reduction of syntactic activity. Next, O1 demonstrates a great degree of fusion with the verb, in the sense that it must follow the adjacency requirement and cannot be suppressed in clause structure (in that it differs from non-terms). Finally, the description of O1 as an incorporated noun is indirectly confirmed by the fact that the subjective verb, co-occurring with O1, has fewer surface transitive features than the objective verb, cooccurring with O2, because O1 is not marked on the verbal form. The verb that co-occurs with O1 does not differ formally from an intransitive verb. This, in principle, might suggest an analysis whereby O1 undergoes incorporation altering the argument structure, which results in the detransitivization of the verb. Below I present evidence that in Ostyak certain nouns are indeed subject to incorporation and that they differ significantly from O1 in their syntactic properties. As follows from table (33), in 9% of clauses O1 is separated from the verb by another element and this fact has not been accounted for so far. Importantly, the element that can intervene between O1 and the verb must belong to the particular closed class and form a so-called “complex predicate” with the verbal head. Complex predicates are usually treated as a category that is transitional between lexical compounds and syntactic phrases. They have not been studied for Ostyak in detail (see, however, the analysis of a similar phenomenon in Hungarian, with some discussion of the Ostyak data in Ackerman 1987), so I introduce here their basic classification. With respect to their semantics and formal expression two types of complex predicates in Ostyak are observed: conventionally speaking, adverbial and nominal. The first component of the adverbial complex predicate is expressed by verbal prefixes that bear a separate phonological stress and usually contribute additional locational or directional semantics to the verb, like n x 'up' 24 in example (38a), as well as by certain non-directional adverbials elaborating on the meaning of the verb, such as t laηti, telna 'wholly', jan∂s 'separately', artan 'a lot', and others (example (38b)). (38) a. t m xoljoη xoj aηk-∂l n x am∂t-s-∂t this thirty man mother-3SG up seat-T-3PL.SUBJ These thirty men lifted up their mother (Pápay 3) b. sox t laηti kaw∂r-s-∂m sturgeon wholly cook-T-1SG.SUBJ I cooked the whole sturgeon. In Ostyak, like in Hungarian, adverbial components of the complex predicate can be separated from the verb under certain syntactic conditions, but in general the order is more stable than in Hungarian and the adverbial component tends to be strictly preverbal. The nominal type of complex predicates involves semantically light nominals with a directional or object-like meaning (sometimes called “reduced complements”) marked with the locative case –na/-(∂)n or, less frequently, the lative case –a/-ji. This type is very restricted lexically; sometimes it exhibits non-compositional semantics and involves idiomatic (as in (39a)) or cognate objects (as in (39c)). (39) a. ma n η-en nem-na p n-l-∂m I you-ACC name-LOC put-T-1SG.SUBJ I will give you a name (Pápay 3) b. ma i itta n η-en l lpi-na sos∂m-l-∂m I so well you-ACC lead-LOC pour-T-1SG.SUBJ So I will pour lead on you (or: pour you with lead). c. jiηk-em nom∂s-na num∂si-l-∂m water-1SG mind-LOC think-T-1SG.SUBJ I am thinking about my water. Unlike the adverbial components of a complex predicate, reduced complements never leave the immediately preverbal position, and as shown in (39) intervene between O1 and the verb, although otherwise O1 must be strictly preverbal. Like O1, reduced complements do not have any controlling properties. Based on their distribution and semantics they may be analyzed as 25 instances of “compound” noun incorporation (in the sense of Rosen 1989), i.e. incorporation that changes the argument structure of the verb (no assumption is made here regarding the argument structure of adverbial complex predicates). Importantly, certain syntactic properties are only limited to reduced complements, which means they should be classified differently to other grammatical relations, including O1. First, in Ostyak, direct objects, indirect objects, as well as certain oblique objects and adjuncts can be promoted to subject through passivization (Kulonen 1989). Example (40a) demonstrates the passivization of the construction with the locative non-term. Example (40b), however, is ungrammatical, which shows that passivization of the reduced complement is not allowed. (40) a. ar woj t m j x-na om-∂s many bird this tree sit-T.3SG.SUBJ Many birds sat on this tree b. ma n η-en nem-na p n-l-∂m t m j x ar woj-na om-s-a this tree many bird-LOC sit-T-3SG.PAS This tree was sat on by many birds. *nem n η-en p n-l-a I you-ACC name-LOC put-T-1SG.SUBJ name you-ACC put-T-3SG.PAS I will give you a name A name is given to you. Second, the reduced complement cannot be questioned separately from the verb, while O1 and non-terms can. Compare (41a) which questions O1, and (41b) which questions a non-term (a passive agent), with the ungrammatical example (41c) construed as a potential question to (39c): (41) a. m la wan-s-∂n? xot wan-s-∂m what see-T-2SG.SUBJ house see-T-1SG.SUBJ What did you see? I saw a house. b. x j-na wan-s-a? Juwan-na wan-s-a who-LOC see-T-3SG.PAS John-LOC see-T-3SG.PAS Who saw (this)? John saw (this). c. *ji k-en m laj-na num∂si-l-∂n? water-2SG what-LOC think-T-2SG Third, the reduced complement does not act as a head, i.e. it cannot be modified by an adjective or a relative clause and cannot be the head of a possessive phrase, as opposed to O1 and non-terms. Example (42a) demonstrates the ability of O1 to be modified by a relative clause and 26 be a head of a possessive phrase, and example (42b) shows that the oblique can be modified by a relative clause, while modifying a reduced compliment results in ungrammaticality, as in (42c). (42) a. kaš∂η woj torum p rt-um wer-l each wer-l animal God order-PART buiseness-3SG do-T.3SG.SUBJ Each animal does its business ordered by God. b. nin awrem-ηil-∂n-na jont-ti low-na wos-ew sukat-s-a you child-DU-2DU-LOC play-PART horse-LOC city-1PL break-T-3SG.PAS Our (toy) city is destroyed by the horse with which your two children are playing. c. *ma n η-en torum p rt-um nem-na p n-l-∂m I you-ACC God order-PART name-LOC put-T-1SG.SUBJ I will give you a name ordered by God. The fact that O1 can be modified does not present independent evidence for or against the incorporation analysis for O1 because incorporated elements in some languages allow modifier and possessor stranding (Baker 1988, 92-96) if the incorporated element retains its argument function (Rosen 1989). But it confirms the regularities concerning the incorporated status of reduced complements. Finally, the verb adjacency restriction can be violated for O1 under certain conditions. Parentheticals law∂l, ik∂n a, tti, nti 'so to say'; m la, toηxa, m lajna, (i)lampa, mos∂η 'perhaps', x ti 'indeed' and functional elements i emphatic particle; ki 'if'; negations n, al, nta; emphatic question particle x n(ti) can easily float in the sentence, although they do not interrupt such constituents as AP, NP, or PP. Given this, they are expected not to intervene between the incorporated nominal and the verb. However, O1 can be separated from the verb by a parenthetical or functional word, or both. (43) a. x-∂n sem-∂n toηxa jow∂r-l-∂η∂n head-3DU eye-3DU perhaps wrap-T-3DU Perhaps they wrap their heads and eyes. b. tor∂m man-em m sa i God I-DAT p rt-∂s something EMPH order-T-3SG.SUBJ God has ordered me (to do) something (Pápay 1) c. wer ilampa i wer-s-∂m 27 business perhaps EMPH do-T-1SG.SUBJ Perhaps I have finished the business. The data in (43) disclaim the generalization that no lexical material intervenes between O1 and the predicate. It is not sufficient by itself to rule out the incorporation analysis of O1, given the possibility of abstract incorporation (Baker 1988, 202) whereby the incorporated element forms a discontinuous constituent with the head, but considerably weakens it. Importantly, inserting parentheticals or functional words between the reduced complement and the verb is impossible, reduced complements are always characterized by immediate adjacency to the verb. Since they differ from O1 in this respect, this also presents further evidence for their special status. In sum, reduced complements are characterized by greater fusion with the verb and more apparent syntactic inertness than O1, on the one hand, and non-terms, on the other hand. Unlike other nominals they are not accessible to passivization and relativization and cannot be questioned. Notice also that in complex predicates the “objecthood” of a reduced complement is suppressed which is overtly expressed by the locative marking as opposed to the accusative marking of O1. All these syntactic properties make reduced complements better candidates for incorporation analysis than O1. 5.2. Arguments against anaphoric agreement The next step is to argue about the argument status of O2. O2 has been shown to have a relatively free position in the clause. Since O2 is either “doubled” by OM, or is absent from the clause and anaphorically referred to by OM, one might expect that the OM itself bears an argument function, while O2 is in fact some kind of adjunct. Such a situation has been reported, for example, for the Bantu language Chiche a, analyzed within the LFG framework by Bresnan and Mchombo (1987), which has revealed numerous similarities to Ostyak. As in Ostyak, in Chiche a object agreement is optional. Inside the VP the object noun phrase and the OM are distributed complimentarily. In the presence of the OM the object must be generated VP-externally in an adjoined topic position, while the VP-internal object does not trigger agreement. The crucial distinction is further made between two object types with respect to their argument relationship to the predicate. The basic suggestion of Bresnan and Mchombo is that the OM is an incorporated pronoun that has a referential function and therefore satisfies the argument structure of the verb. When the overt object is present in addition to OM, it cannot bear the argument object function due to the LFG functional uniqueness principle, 28 which prohibits the same function being assigned twice. The occurrence of such an object inside the VP is blocked by the OM because only arguments can appear in the local domain of the verb (Bresnan and Mchombo, 1987, 745), so the object that triggers agreement must be generated VP-externally as a floating topic. The anaphoric relationship between the OM and the floating topic is called "anaphoric agreement." The minimal difference between the anaphoric and grammatical agreement affixes then lies in their argument value: the grammatical agreement marker unlike the anaphoric agreement marker does not bear an argument relationship to the predicate. It can be revealed by a number of syntactic tests. Remarkably, the predictions Bresnan and Mchombo make with respect to the anaphoric object agreement turn out to be wrong for Ostyak, suggesting that Ostyak object agreement is of a grammatical character. As Bresnan and Mchombo argue, verbal case government is completely inconsistent with the OM being an incorporated argument anaphorically linked to topic. Since the topical “object” NP is related to the verb not by government but by anaphora with OM, the verb cannot assign it a case under government (Bresnan and Mchombo 1987, 765). If the object case and the anaphoric OM coexist in a language, they are predicted to be in complimentary distribution (as attested, for example, in Arabic). But in Ostyak, at least personal pronouns in the object function bear the morphologically overt accusative regardless of object agreement (see examples (4) above) and there is no reason to think that the case is not governed by the verb. The situation becomes more complicated with nominal objects, because in modern Northern Ostyak they do not show an overt accusative marker. Nevertheless, they can be reasonably assumed to have a zero-marked accusative.12 Following Bresnan and Mchombo's logic, the case governed object must bear an argument relationship to the verb even in the presence of OM. Ostyak facts can be naturally explained only if both O2 and O1 are analyzed as having the same argument function, i.e. as real objects.13 The second property predicted for languages with anaphoric object agreement is the contrastive discourse role of the independent object pronouns. The anaphoric agreement marker refers only to an established discourse topic, therefore the contrastive (replaced) topic must be introduced by other means, namely, by independent pronouns. The independent pronouns are morphologically different from agreement markers and have only "a contrastive use that makes them incompatible with anaphora to the topic" (Bresnan and Mchombo 1987, 752). Thus in Chiche a independent pronouns in the object position are only interpreted contrastively (as replaced topics), while a discourse-old topic in an object function is referred to by the anaphoric OM. 29 Unlike this situation, independent personal pronouns in Ostyak do not necessarily entail a contrastive interpretation and might serve for non-contrastive anaphora. In example (44) the 3rd person object pronoun l wel does not seem to bear the contrastive function since, according to the text where the example is found, the controller of l wel can be established without ambiguity. (44) iti woj-lal um a-l∂-lli. wan um a-l∂-lli x w um a-l∂-lli, sat so animal-3PL caress-T-3SG.OBJ short caress-T-3SG.OBJ long caress-T-3SG.OBJ all l w-el wek wanna t j-ti he-ACC always close pit-s-el have-PART start-T-3PL.OBJ So he was caressing his animals. He was caressing them for a while, they all. started being close to him forever Furthermore, the 3rd person pronoun is required when used reflexively, compare (45a) and (45b): (46) a. Petra l w-el resk∂-s-li Peter he-ACC hit-T-3SG.OBJ Peter hit him/himself. b. Petra resk∂-s-li Peter hit-T-3SG.OBJ Peter hit him/*himself. The topic-anaphoric role of independent pronouns is inconsistent with anaphoric agreement under the theory suggested by Bresnan and Mchombo, and indicates the grammatical character of object agreement in Ostyak. Even more important is that anaphoric agreement as distinct from grammatical agreement is predicted to be non-local, i.e. it can carry over between elements of distinct simple clauses. The reasoning supporting this claim is that the arguments must be expressed within the phrase structures headed by the predicate, be marked on the predicate itself, or otherwise remain unexpressed and be anaphorically controlled by non-local elements (Bresnan and Mchombo 1987, 752). Verbs agree grammatically only with their governable arguments, so grammatical agreement must be structurally local to the verb. In contrast, the antecedent of the anaphoric (non-reflexive) relationship is not related to the argument position by government, 30 therefore there is no requirement for it to be structurally local. Thus, in Chiche a, the relation between the OM and the floating topic NP can be non-local: see the following example borrowed from Bresnan and Mchombo 1987, 754 (number 7 shows agreement in class, the bracketing is mine): (46) [asilikálí á gânyu chigawéngá ichi a-na-úz-á mercenaries mtsogoleri wâthu terrorist(7) this SM-PAST-tell-INDIC leader [kutí s-a-ngath-é our [ku-chí-gwír-a ]]] that not-SM-can-SUBJN INF-OM(7)-catch-INDIC This terrorist, the mercenaries told our leader that they cannot catch him. In (46) the floating topic chigawéngá ‘terrorist’ located in the matrix clause is anaphorically related to the OM in the second down embedded clause. As typical of all Uralic languages with object agreement, in Ostyak the agreement with the object of the subordinate clause is encoded on certain auxiliary-like matrix verbs (for Hungarian see Ackerman 1987, 341-350; Farkas and Sadock 1989, 326-327). This is demonstrated in (47). (47) [ma werit-l-em [n η-en jont-ti jernas]] I can-T-1SG.OBJ you-DAT sew-CONV dress I will be able to sew a dress for you. However, this does not constitute a strong argument in favor of non-local agreement, as it is not entirely obvious whether constructions like (47) are monoclausal or multiclausal. In the former case, the auxiliary and the regular verb represent a single syntactic unit, as in the analysis of monoclausal multipredicate constructions by Davies and Rosen (1988). But even in the latter case, under some assumptions they might be viewed as the surface realization as a single clause of what consists underlyingly of two separate clauses (the “clause union” approach going back to Aissen and Perlmutter (1983)). Independent of the approach taken, examples such as (47) do not apparently present a regular instance of non-local object agreement in the sense suggested by Bresnan and Mchombo. Other than in constructions with auxiliaries, O2 cannot cross over the clause boundary, as is illustrated in (48). (48) *[maw-lal a ke-l (maw-lal) want-∂s [ p x-∂l ∅ li-s-li]] candy-3PL mother-3SG candy-3PL see-T.3SG.SUBJ boy-3SG eat-T-3SG.OBJ 31 The candies, the mother saw that the boy ate them. Next, according to Bresnan and Mchombo, the NP object argument is absent when the agreement marker bears a fully specified set of noun features, which cannot be duplicated by the overt NP. This explains the complimentary distribution of the agreement marker and the object NP. The Ostyak OM, although it marks the number of the object, does not specify its person (as independent pronouns do). Because the OM does not supply all the necessary referential features of the object, they must be specified elsewhere. This conditions the presence of the overt object NP, which is not incompatible with the OM in Ostyak. Therefore Ostyak is different from Chiche a, where the OM bears all the noun features relevant in the language, so that the doubling of the OM by overt argument with the same set of features is prohibited. Finally, languages with incorporation of object pronouns are predicted to have anaphoric deletion in the object position as a natural consequence of the fact that the OM bears the object function and the lexical object does not have to be overtly present (Bresnan and Mchombo 1987, 765). However, this property by itself does not constitute a piece of evidence for the presence of anaphoric object agreement in the language. Cross-linguistically it is quite common that elliptic mechanisms (empty pronouns) are independent of agreement (cf. Rizzi 1986). Null objects are found in Chinese, Japanese, Korean, Portuguese, Quechua, informal German (Huang 1984, 1991), Italian (Rizzi 1986), Turkish (Kornfilt 1987, 637), and are possible in Russian under certain discourse conditions. None of these languages have object agreement. Thus, although languages with pronominal incorporation are expected to have a null object, the presence of anaphoric object deletion cannot be an argument in favor of the anaphoric character of object agreement.14 Therefore Ostyak does not display the clustering of properties predicted by Bresnan and Mchombo for languages with anaphoric object agreement: it exhibits case government in the object position regardless of agreement, agreement is local, independent pronouns are used topic-anaphorically, and the OM does not bear the fully specified set of noun features. Object ellipsis does not seem to be a decisive argument in supporting or opposing anaphoric agreement. This implies that the OM in Ostyak lacks pronominal reference and argument function, and can be characterized as a simple grammatical agreement morpheme. Consequently, O2 must be analyzed as a verbal argument. 32 5.3. Conclusion The tests used in this section reveal the absence of difference in argument relationship to the verb between O1 and O2. Although O1 fails to retain some term properties, it cannot be identified as a non-term because, unlike clear non-terms, it is not optional. On the other hand, in a number of syntactic effects O1 differs from incorporated elements or at least demonstrates a lower degree of fusion with the verb than the undoubtedly incorporated nouns in Ostyak. This leads to the conclusion that O1 is a verbal argument that does not form a compound predicate with the verb, but rather a higher level constituent, arguably a VP. I have also demonstrated that in constructions with O2, the OM cannot be viewed as an incorporated pronoun, so O2 itself must bear an argument function. 6. Towards an analysis We are left with the conclusion that object properties in Ostyak can split between the VPinternal and VP-external positions without affecting the truth-conditional meaning of the sentence and the argument structure of the predicate. Two questions arise at this point: what motivates the presence of two alternative transitive constructions within one language, and, since both O1 and O2 were shown to be terms, do they bear an identical grammatical relation? One possibility is to explain the asymmetry of the objects by their linear position alone. But even then one would have to make further assertions about the conditions determining the positional difference. The approach adopted in this paper is based on the assumption that functional factors are not additional to structure-internal explanations, but might bear the primary importance in motivating certain syntactic generalizations. As shown above, the motivation for object agreement in Ostyak cannot be captured semantically in an obvious way, so it might prove useful to turn to other levels of clause representations, in particular the level of information structure. A likely hypothesis is that the positional and behavioral split between O1 and O2 is primarily motivated by their status in information structure. No attempt is made here to define information structure status in detail, nor to provide an information structure analysis of the constructions in question (this is done at length elsewhere (Nikolaeva, in preparation)). I will only tentatively formulate the information structure motivation and concentrate on the grammatical consequences of such analysis, which makes the two following predictions. First, since O1 and O2 systematically correspond to different linear positions, and since information structure typically affects the linearization rules of the “free word order” languages, it may be that these positions are 33 themselves associated with certain information structure functions. This specifically applies to the immediately preverbal position of O1 (O2 is relatively free, as argued above). Second, information structure and syntax are logically independent levels, but their units map onto each other in a particular way. This predicts that the information structure statuses associated with O1 and O2 can not only be borne by direct objects. If they affect certain behavioral properties, the behavioral split might (though need not) be observed for other grammatical relations as well. These predictions are discussed in sections 6.1 and 6.2. 6.1. Focus position Much recent research has presented evidence that SOV languages are characterized by a preverbal focus position, which is the last linear position available for non-verbal elements given verb finality (Kim 1988; É. Kiss 1995; Choi 1996, among others). Without going into detail, focus is here defined after Lambrecht (1984, 213) as “the semantic component of a pragmatically structured proposition whereby the assertion differs from the presupposition.” This definition operates on the level of the pragmatic structuring of the information, and does not refer to syntax. However, under the approach taken in this paper information structure functions can be associated with particular syntactic units and positions. A full analysis of focus in Ostyak goes beyond the scope of this paper. Here I will only illustrate the point that in Ostyak it is also the element immediately adjacent to the verb (or, more accurately, to the predicate) that bears the focus status. Cross-linguistically, focus manifests a strong functional and formal correlation with wh-question words and their targets. Remarkably, the wh-phrase systematically corresponds to the immediately preverbal position in Ostyak, independent of the grammatical relation it bears (the direct object in example (49a), the non-term in (49b), and the indirect object in (49c)). Separating it from the verb by some element yields ungrammaticality (this option is shown in brackets in (49)). (49) a. (*m la) awrem-l-am m la wer-l-∂t? what child-PL-1SG what do-T-3PL.SUBJ? What are my children doing? b. ma (*xota ) n η-en xota wel-l-em? I how you-ACC how kill-T-1SG.OBJ How shall I kill you? c. (*x j elti) t m an x j elti m -s-en? 34 who to this cup who to give-T-2SG.OBJ To whom did you give this cup? In the answers to wh-questions the target phrase must be in the immediately preverbal position, see (50) as an answer to question (49c): (50) (*Juwan-a) t m an Juwan-a m -s-em John-LAT this cup John-LAT give-T-1SG.OBJ I gave this cup to John. Other elements universally known to be associated with focus (for example, a nominal under the scope of the focus particles only or even) also immediately precede the predicate. The focus position accommodates only one constituent. A crucial piece of evidence for this comes from multiple wh-questions, as is shown by the next example. In (51) one of the whquestion words (x l a ‘where’) is situated in the preverbal position, while another one (x n i ‘when’) is clause-initial and cannot be adjoined to the focus position. (51) x n i n η m η-iluw (*x n i) x l a want-l∂-lan? when you we-ACC when where see-T-PL-2SG.OBJ When did you see us where? (Pápay 7) Focus elements other than wh-words demonstrate the same regularity. Thus the focus position, immediately adjacent to the predicate, is unique. Further evidence for the association of this position with focus comes from object agreement. Remarkably, in questions to the object the verb does not show object agreement (example (49a)), and in questions to any other element the object is always encoded as O2 (examples (49b)-(49c)). The opposite distribution is ungrammatical. These facts can be accounted for if object agreement is taken to mark the non-focus status of the object. Since O1 must be adjacent to the predicate (except that parentheticals and functional words intervene between them), it can be described as bearing the focus status in the information structure. When the wh-word bears the object role, it assumes the focus position and does not trigger agreement. On the contrary, there is no such requirement for O2. When the wh-word bears a grammatical relation other than that of direct object, it corresponds to the focus position and excludes the object from it. The object cannot be strictly preverbal and in this case it is encoded 35 as O2. Answers to wh-questions demonstrate the same distribution of object agreement: in the answer to questions to the object, the object is encoded as O1. This is shown in (52), construed as an answer to (49a). (52) awrem-l-am xot wer-l-∂t / *wer-l-el child-PL-1SG house do-T-3PL.SUBJ / do-T-3PL.OBJ My children are building a house. See also example (33) above. In the answer to a question to any other element the object is encoded as O2 (as in (50)). The analysis of O1 as bearing the focus status in the information structure and of O2 as not being in focus accounts for the requirement that O1 be overly present in the clause. Indeed, the focus position must be filled. No focused element can be omitted from the clause because this would make the assertion impossible. If focus is the element that minimally distinguishes assertion from presupposition (see the definition of focus given above), then in the absence of the focus element assertion becomes equivalent to presupposition, and this makes the whole proposition redundant.15 Therefore, when the object is focused (is O1) it is not accessible to deletion. I have argued that the focus function is grammaticized in a structural position in Ostyak. Non-focus elements in the clause are situated relatively freely, as long as the requirements on subject-initiality and verb-finality are followed. Hence, the focus element forms a constituent with the verb to the exclusion of other arguments. The status of this constituent might be the subject of a special discussion, regarding whether it can be described as a VP or as a focus phrase headed by the functional focus node, as has been proposed for “discourse-configurational” languages (É. Kiss 1995). I leave this question open. In any case, the preverbal position is filled without any reference to semantics, so elements with a different semantic content map onto it. This explains the differences with Chiche a discussed in section 5.2. A necessary part of Bresnan and Mchombo’s analysis of Chiche a is the phrase structure rule which generates the object within a VP, in the post-verbal position. A VP-external NP cannot bear the object function because the phrase structure rules require a VP-internal object, so this NP is assigned a non-governable topic function. The situation in Ostyak is radically different because there is no initial requirement for the object to be generated VP-internally: the object, as well as other grammatical relations, is mapped onto the linear order depending on its information structure function. Therefore, there is no reason to postulate a difference in argument function for the two types of objects based on word order.16 36 6.2. Other grammatical relations To explore the second prediction of my analysis, I will look at certain behavioral properties of lower grammatical relations. I conventionally unify them here by the general term "oblique” which includes indirect objects, oblique objects, and non-terms. For the sake of space I present only the examples for locative non-terms, but the same tests apply to most other obliques (although potentially they might be characterized by distinct properties not discussed here). Possessive reflexivization. Normally obliques do not license reflexives, as illustrated in (53), where the possessive affix on jik-∂l cannot be interpreted as coreferential with the oblique (k laηna). (53) *ap e-l aη∂t-ηil-al ew∂lt niη k laη-na katl∂-m-al brother-3SG horn-DU-3SG by female reindeer-LOC hold-EV-3SG His brother held the female reindeer (on the female reindeer) by its horns. The sentence, however, becomes grammatical if the oblique does not occupy the immediately preverbal position, compare (53) with (54): (54) ap e-l niη k laη-na aη∂t-ηil-al ew∂lt katl∂-m-al brother-3SG female reindeer-LOC horn-DU-3SG by hold-EV-3SG His brother held the female reindeer (on the female reindeer) by its horns (Pápay 5). The contrast between (53) and (54) indicates that the ability to control reflexivization is sensitive to the position of an oblique rather than to its oblique status. However, as was shown in section 3.2, precedence relations as such do not affect the property of controlling reflexivization, therefore the contrast between (53) and (54) cannot be explained in terms of linear precedence. The illformedness of (53) apparently has to do with the fact that the oblique niη k laηna occupies the focus position. Possessor topicalization. With obliques that are strictly preverbal the possibility of possessor topicalization is questionable. Example (55a), where the possessed noun xotl takes the oblique role, is only marginally acceptable, and the regular possessive construction without the possessive suffix on the head noun (as in (55b)) is strongly preferred. 37 (55) a. ?/* Juwan xot-l John x a m n-∂s house-3SG to go-T-3SG.SUBJ He went to John's house. b. Juwan xot x a m n-∂s John house to go-T-3SG.SUBJ He went to John's house. The next example shows that obliques which are not immediately preverbal, and particularly those fronted before the subject, can easily trigger possessor topicalization. (56) as i u wus-∂l ew∂lt … jiηk l η-ti pit-∂s bear mouth hole-3SG through water come-PART start-T.3SG.SUBJ Water started coming through the mouth of the bear (Pápay 2). Quantifier float. Example (57) illustrates the case where the oblique adjacent to the verb cannot trigger quantifier float: (57) ma itliji wos-∂t-na (*itliji) u-s-∂m I all city-PL-LOC all be-T-1SG.SUBJ I have been to all cities. The data on non-preverbal obliques is missing, but (57) demonstrates that with regard to this property preverbal obliques are inactive and therefore pattern together with O1 and differ from subjects and O2. I have shown that obliques that are not immediately preverbal trigger possessor topicalization and control possessive reflexivization, as does O2, whereas obliques in the focus position do not have these properties and do not license quantifier float, in the same way as O1. Thus, both direct objects and obliques demonstrate a systematic behavioral split at least with respect to two properties (the data on the control of the embedded clauses for obliques are missing). Non-focused elements are superior to focused elements: terms associated with focus lose some of their properties; conversely, non-terms that are not in focus acquire certain term properties. This confirms the prediction that behavioral difference is affected by the information structure statuses borne by the respective nominals. 38 This brings us to the second question formulated at the beginning of section 6, namely, whether O1 and O2 bear an identical grammatical relation. In Ostyak most grammatical relations demonstrate a hybrid syntactic behavior to a certain extent (although subjects and some types of obliques have not been studied in this respect). Given that most grammatical relations, although superficially distinguished only by position, are regularly associated with two sets of grammatical properties, the description of each of them as two distinct grammatical units would not be economical. This would lead to the doubling of virtually all the grammatical relations. As the behavioral split is systematic, the obvious alternative is to recognize the possibility that certain properties are not peculiar to a particular grammatical relation, since it can acquire or lose them while remaining identifiable. This possibility is partly based on the fact that a change in the behavioral properties of grammatical relations is not accompanied in Ostyak by a change in their semantics and coding (with the exception of linear order). Furthermore, syntactic properties other than those discussed here do not relativize with respect to the position of the corresponding nominal. For example, neither objects nor obliques have any of the subject properties discussed in section 3.1; some non-terms (instrument and manner phrases) are not accessible to passivization (Kulonen 1989); subjects, direct objects and indirect objects can control the purpose clause with the participle ending in –ti(ji)/-ta, unlike lower grammatical relations (this property has not been discussed here). Thus, my analysis suggests that the alternation in grammatical properties of nominals within the clause does not necessarily entail a change in grammatical relations. Depending on the framework, this may be taken to indicate that grammatical relations are relative to a particular stratum of grammatical description (in the derivational frameworks such as Relational Grammar or some versions of GB); or that they cannot be classified as clear-cut categories (Keenan 1976). The functional approach based on parallel levels of clause representations explicitly recognizes that certain grammatical processes might be affected by the information from one of these three levels, leaving the other two unchanged. The syntactic properties that were discussed in this section seem not to be associated with grammatical relations per se, but are in fact sensitive to the information structure. A full explanation of this would go beyond the scope of this paper (see a thorough analysis of information structure-based motivation for certain syntactic properties of nominals in double-object constructions in Polinsky (forthcoming)). My purpose was to demonstrate that the behavioral asymmetries between the two types of objects in Ostyak cannot be accounted for in terms of syntactic or semantic effects, and seem to be determined solely by their information structure status. 39 NOTES 1. I would like to thank all my Ostyak informants, who are too numerous to mention here. I am also grateful to Farrell Ackerman, David Perlmutter, and Maria Polinsky for their helpful comments on earlier versions of this paper. Abbreviations: ACC 2. - accusative OM - object marker CONV - converb PART - participle DAT - dative PAS - passive DU - Dual PL - Plural EMPH - emphatic S - subject EV - Evidential SG - Singular INDIC - Indicative SM - subject marker INF - infinitive SUBJ - subjective conjugation LOC - locative SUBJN - subjunctive O - object T - Tense OBJ - “objective” conjugation V - verb These labels go back to Keenan 1976. In his paper on a universal definition of subject he discusses the third type of properties, the semantic. However, it is not clear to what extent semantic properties are definitional for grammatical relations, and, in any case, in Northern Ostyak they remain unchanged in the constructions in question. 3. The objective and subjective conjugations are sometimes called the “definite” and “indefinite” conjugations respectively. 4. There has long been a vigorous debate in historical Uralistics concerning whether the objective conjugation in the different Uralic languages has a common origin. According to the standard view, the “premises” for the objective conjugation (object agreement) were present as early as Proto-Uralic, but thereafter the modern systems developed independently of each other (Collinder 1960; Serebrennikov 1961; Hajdú 1966). Helimski (1982) convincingly proves that Hungarian and Samoyedic objective paradigms go back to the same archetype, which he reconstructs as an areal property within Proto-Uralic. 5. Most of the examples come from my field notes, recorded in 1989-1993 among the northern Ostyaks. Nearly all of them are in the dialect of Obdorsk, the rest come from the dialects of Shuryshkary and Kazym. Some examples originate from Pápay's folklore collection (Pápay 1906-1908); in this case the numerals given in the examples refer to the corresponding text. I use the transcription traditionally employed to render Ostyak data. 40 6. The list of relevant tests presented here is not intended to be exhaustive, due to lack of material. 7. Converb is defined here as a non-finite verbal form that does not attach personal affixes (in fact, any affixes) and normally expresses adverbial subordination. Converbs differ from participles, which can have a personal and case marking in Ostyak. 8. The status of the dative corresponding to the semantic role of experiencer found in some “subjectless” constructions is not entirely clear in this respect and needs further investigation. Presumably the dative can also control –man converbs. At least one example is found in which the subject of the embedded clause is in an inalienable possessive relationship with the “subjectoid” dative argument of the matrix clause: n η-en [ n η sem-en want-man] i arat xor∂η xoj wel-ti-na you-DAT you eye-2SG see-CONV so many xor∂η man kill-PART-LOC al nt u-l sorry NEG be- T.3SG.SUBJ With your eyes seeing, you won’t feel sorry for so many xor∂η people being killed. (Pápay 12). 9. The subject of the relative clause might not be overtly expressed by a lexical item. In this case the head noun must attach the possessive affix referring to the subject of the relative clause, even if the two subjects are not coreferential. As shown in Ackerman and Nikolaeva (1997), the possessive affix is then the only way to encode the relative clause subject and has an agentive rather than a possessive interpretation. 10. The quantifiers isa/ sa, on the one hand, and itliji/xoliji, on the other hand, are dialectal variations (shown here and afterwards with a slash). The discussion of the semantic difference between isa/ sa and itliji/xoliji goes beyond the scope of the paper. 11. The number of examples with subject-drop is not shown specifically. 12. The Uralic accusative affix has been lost in most Ostyak dialects for phonological reasons. Note, however, that the overt accusative for nominals is still present in some Vogul and Ostyak dialects, as well as in related Samoyedic languages where it marks direct objects independent of object agreement. 13. The absence of an overt morphological distinction between O1 and O2 would also present difficulties for those theories that crucially rely on the difference in (structural) case assignment for VP-internal and 41 VP-external objects (for example, de Hoop 1992). At the very least, this fact would make the theory more abstract: one would need some additional arguments for opposing the structural and morphological case. 14. Within the GB framework Huang (1984, 1991) shows that the null object can neither be identified as PRO, nor as a NP-trace, nor as pro (the latter possibility is ruled out precisely because it does not meet the requirement of licensing and identification by agreement). Huang argues that the null object can be analyzed akin to VP-ellipsis as a variable bound by a null operator (possibly a null topic). 15. Of special interest are sentences where the focus status is borne by the predicate itself to the exclusion of other elements. This case is not addressed here. 16. In addition, the difference between Ostyak and Chiche a discussed in section 5.2 seems to be motivated by a subtle distinction in the information structure status of the object that triggers agreement. In Chiche a this object is described as the topic, which is defined by Bresnan and Mchombo (1987, 746) as the element “under discussion.” It can be shown (Nikolaeva, in preparation) that O2 in Ostyak is not a topic in this sense but rather can be characterized as a “secondary topic” (the term is from Givón 1984). Roughly speaking, the secondary topic is an element in information structure that is characterized by certain topic properties but is less “important” than the primary topic (or simply topic in the aforementioned sense). The secondary topic always stands in a certain relationship to the primary topic and the sentence makes an assertion about this relationship. This difference seems to be crucial. Since Li and Thompson 1976 it is well known that the topic (the primary topic) can be external to the clause: it might not bear the argument function and can be generated as an adjunct. It is not at all obvious whether the same observation applies to the secondary topic. Further research is needed in this direction. REFERENCES Ackerman, Farrell. 1987. Miscreant morphemes: Phrasal predicates in Ugric. Ph.D. diss. University of California, Berkeley. Ackerman, Farell; and Nikolaeva, Irina. 1997. Similar forms and different functions: The person/number paradigm in Western Armenian and Northern Ostyak. Paper presented at the International LFG conference. University of California at San Diego, La Jolla, June 1997. Aissen, Judith; and Perlmutter, David. 1983. Clause reduction in Spanish. In: Perlmutter, David (ed.), Studies in Relational Grammar 1, 360-403. Chicago: University of Chicago Press. Baker, Mark. 1988. Incorporation: A theory of grammatical function changing. Chicago: University of Chicago Press. 42 Bese, László; Dezs , László; and Gulya, János. 1970. On syntactic typology of the Uralic and Altaic languages. In: Dezs , László and Hajdú, Péter (eds), Theoretical problems of typology and the northern Eurasian languages, 113-128. Amsterdam: B. R. Gruner. Bresnan, Joan; and Mchombo, Sam. 1987. Topic, Pronoun, and Agreement in Chiche a. Language 63: 741-783. Choi, Hye-Won. 1996. Optimizing structure in context: Scrambling and information structure. Ph.D. diss., Stanford University. Chomsky, Noam. 1995. The Minimalist Program. Cambridge, Massachusetts: MIT Press. Collinder, Bjorn. 1960. Comparative grammar of the Uralic languages. Stockholm: Almqvist and Wiksell. Comrie, Bernard. 1989. Language universals and linguistic typology. Chicago: The University of Chicago Press. De Groot, Caspar. 1983. Verb agreement and ergativity in the Ugrian languages: A reconstruction. Lingua 61: 209-230. De Hoop, Helen. 1992. Case configuration and noun phrase interpretation. Ph.D. diss., University of Groningen. Davies, William; and Rosen, Carol. 1988. Unions as multi-predicate clauses. Language 64: 52-88. Diesing, Molly. 1992. Indefinites. Linguistic Inquiry Monograph 20. Cambridge, Massachusetts: MIT Press. Diesing, Molly; and Jelinek, Eloise. 1993. The syntax and semantics of object shift. Working papers in Scandinavian syntax 51. Farkas, Donka F.; and Sadock, Jerrold M. 1989. Preverb climbing in Hungarian. Language 65: 318-338. Ganschow, Gerhard. 1972. Az osztják tárgyas igeragozás használatának kérdéséhez. Nyelvtudomány Közlemények 74: 183-185. Georgopoulos, Carol. 1991. On A- and A'-agreement. Lingua 85: 135-169. ____________. 1992. Another look at object agreement. Proceedings of NELS 22, 165-177. GLSA, Amherst: University of Massachusetts Press. Givón, Talmy. 1984. Direct object and dative shifting: Semantic and pragmatic case. In: Plank, Frans (ed.) Objects: Towards a theory of grammatical relations, 151-182. London, Orlando: Academic Press. Greenberg, Joseph H. 1966. Some universals of grammar with particular reference to the order of meaning elements. In: Greenberg, Joseph H. (ed.) Universals of Language, 73-113. Cambridge, Massachusetts: MIT Press. 43 Gulya, János. 1967. A tárgyas ragozás az Osztjákban (a vahi nyelvjárás alapján). Nyelvtudomány Közlemények 69: 389-394. Hajdú, Péter. 1966. Bevezstés az uráli nyelvtudományba. Budapest: Tankönyvkiadó. Hajdú, Péter. 1968. Chrestomathia Samojedica. Budapest: Tankönyvkiadó. Helimski, Eugen A. 1982. Vengersko-samodijskije lexi eskije paralleli. Moskow: Nauka. Holmberg, Anders; and Platzack, Christer. 1995. The role of inflection in Scandinavian Syntax. New York, Oxford: Oxford University Press. Honti, László. 1984. Chrestomathia ostiacica. Budapest: Tankönyvkiadó. __________. 1987. K probleme vozniknovenija objektnogo spriaženija ugorskix jazykov. Nyelvtudományi Közlemények 85: 341-346. Huang, James C.-T. 1984. On the distribution and reference of empty pronouns. Linguistic Inquiry 15: 531-574. ___________ . 1991. Remarks on the status of the null object. In: Freidin, Robert (ed.) Principles and parameters in comparative grammar, 56-76. Cambridge, Massachusetts: MIT Press. Kálmán, Béla. 1965. Vogul chrestomathy. Indiana University Uralic and Altaic Series. 46. Bloomington: Indiana University Press. . Keenan, Edward. 1976. Towards a universal definition of "subject." In: Li (ed.), 303-333. Keresztes, László. 1989. Chrestomathia Mordvinica. Budapest: Tankönyvkiadó. Kim, Alan. 1988. Preverbal focus position in type XIII languages. In: Hammond, Michael; Moravcsik, Edith; and Wirth, Jessica (eds) Studies in syntactic typology, 148-171. Amsterdam, Philadelphia: John Benjamins. É. Kiss, Katalin. 1987. Configurationality in Hungarian. Dordrecht: Foris. __________. 1995. (ed.) Discourse configurational languages. New York, Oxford: Oxford University Press. Kornfilt, Jaklin.1987. Turkish and the Turkic languages. In: Comrie, Bernard (ed.) The world’s major languages, 619-644. London, Sydney: Croom Helm. Kulonen, Ulla-Maija. 1989. The Passive in Ob-Ugrian. Helsinki: Suomalais-Ugrilainen seura. Lambrecht, Knut. 1994. Information structure and sentence form: Topic, focus, and the mental representations of discourse referents. Cambridge: Cambridge University Press. Li, Charles N. (ed.) 1976. Subject and topic. New York, San Francisco, London: Academic Press. Li, Charles; and Thompson, Sandra. 1976. Subject and topic: A new typology of languages. In: Li (ed.), 457-490. 44 Mahajan, Arnoop. 1989. Agreement and agreement phrases. In: Laka, Itziar; and Mahajan, Arnoop (eds) Functional heads and clause structure. MIT Working Papers in Linguistics 10, 217-252. Cambridge, Massachusetts: MIT Press. ________. 1991. Clitic doubling, object agreement and specificity. Proceedings of NELS 21: 263277. Marcantonio, Angela. 1989. On the case of the object in Finnish: a typological, diachronic and comparative analysis. Finnisch-Ugrischen Forschungen 48: 130-170. Nikolaeva, Irina; Kovgan, Elena; and Koshkareva, Natalia. 1993. Communicative roles in Ostyak syntax. Finnisch-Ugrischen Forschungen 51: 125-167. Nikolaeva, Irina. In preparation. “Secondary topic as a relation in information structure.” Pápay, József. 1906-1908. Északi-osztják nyelvtanulmányok. Nyelvtudományi Közlemények 3638: 345-398, 52-79, 164-195, 258-275, 111-150, 313-329. Polinsky, Maria. 1993. Ponaexalo tut vas: Russian distributive clauses with bare genitive. In: Avrutin, Sergei; Franks, Steven; and Progovac, Ljiljana (eds) Annual Workshop on Formal Approaches to Slavic Linguistics. The MIT Meeting, 374-399. Ann Arbor: Michigan Slavic Publications. ___________, forthcoming. A non-syntactic account of some asymmetries in the double object constructions. Rédei, Károly. 1965. Northern Ostyak chrestomathy. The Hague, Bloomington: Indiana University Press. Rizzi, Luigi. 1986. Null objects in Italian and the theory of pro. Linguistiv Inquiry 17: 501-557. Rosen, Sara. 1989. Two types of noun incorporation: a lexical analysis. Language 65: 294-317. Runner, Jeffrey T. 1995. Noun phrase licensing and interpretation. Ph.D. diss. University of Massachusetts at Amherst. Serebrennikov, Boris A. 1961. Osnovnyje linii razvitija padežnoi i glagolnoi sistemy v uralskix jazykax. Moscow: Nauka. Sinor, Denis. 1988. (ed.) The Uralic languages: description, history, and foreign influences. Leiden, New York: Brill. Spencer, Andrew. 1995. Incorporation in Chukchi. Language 71: 439-489. Szalacsek, Margit. 1984. The problem of the definite and indefinite conjugational forms in the Northern-Ostyak Language. Nyelvtudományi Közlemények 21: 426-430. Vikner, Sten. 1995. Verb movement and expletive subjects in Germanic languages. New York, Oxford: Oxford University Press. 45 Webelhuth, Gert. 1992. Principles and parameters of syntactic saturation. New York, Oxford: Oxford University Press. Wickman, Bo. 1970. Über die Verwendung der objektiven Konjugation. In: Schlachter, Wolfgang (ed.) Symposion über Syntax der uralischen Sprachen, 216-235. Göttingen: Vandenhoeck, Ruprecht. Woolford, Ellen. 1995. Object Agreement in Palauan: Specificity, Humanness, Economy and Optimality. In: Beckman, Jill N.; Walsh, Laura; and Urbanczyk, Suzanne (eds) Papers in Optimality Theory. University of Massachusetts Occasional Papers in Linguistics 18, 655700. Cambridge, Massachusetts: University of Massachusetts Press. 46 NOTES 1 I am grateful to Farrell Ackerman, David Perlmutter and Maria Polinsky for helpful discussions of earlier versions of the paper. Abbreviations used are as follows: ACC OM - object marker CONV- converb PART - participle DAT - Dative PAS - passive DU - Dual PL - Plural EMPH - emphatic S - subject EV - Evidential SG - Singular INDIC - Indicative SM - subject marker INF - infinitive SUBJ - subjective conjugation LOC - Locative SUBJN - subjunctive O - object T - Tense OBJ - “objective” conjugation V - verb 2 - Accusative These labels go back to Keenan 1976. In his paper on a universal definition of subject he discusses the third type of properties, semantic properties. It is not clear, however, to which extent the semantic properties are definitional for grammatical relations, and in any case, in Northern Ostyak, they remain unchanged in the constructions in question. 3 Objective and subjective conjugations are sometimes called “definite” and “indefinite” conjugations respectively. 4 There has long been a vigorous debate in historical uralistics on the issue of whether the objective conjugation in the different Uralic languages has a common origin. According to the most commonly accepted view, the “premises” for objective conjugation (object agreement) were present as early as in Proto-Uralic, but the modern systems developed basically independently of each other (Hajdú 1966, Collinder 1960, Serebrennikov 1961). Helimski (1982) convincingly proves that Hungarian and Samoyedic objective paradigms go back to the same archetype, which he reconstructs as an areal property within Proto-Uralic. 5 Most of the examples come from my field notes recorded in 1989-1993 among northern Ostyaks. I am greateful to all my informants who are too numerous to mention here. Unless indicated otherwise, the examples represent the dialect of Obdorsk. The Shuryshkary and Kazym dialect examples are marked as Sh. and Kaz. respectively. Some examples are taken from Pápay's folklore collection (Pápay 1906-1908); in this case the number in the reference indicates the number of the corresponding text. I use the transcription traditionally practiced for rendering Ostyak data. 47 6 The list of relevant tests presented here is not supposed to be exhaustive. I do not have enough material to make definite judgments on some types of extractions, multiple wh-questions, etc. 7 Converb is defined as a non-finite verbal form that does not attach personal affixes (in fact, any affixes) and normally has adverbial meaning. Converbs therefore differ from participles which can have personal and case marking in Ostyak. 8 The status of the dative corresponding to the semantic role of experiencer found in some “subjectless” constructions is not entirely clear in this respect and needs further investigation. Presumably dative can also control converbs with –man and –ti(ji) affixes, see the following example: luw-el want-ti al-li ji-s-?t he-DAT look-CONV sorry-TRANS become-T-3PL.SUBJ He started feeling sorry for them looking [at them] (literally: they became pity for him to look). (Pápay 15) At least one example is also found in which the subject of the embedded clause is in inalienable possessive relationship with the “subjectoid” dative argument of the matrix clause: n ?-en [ n ? sem-en want-man] i arat xor?? xoj wel-ti-na you-DAT you eye-2SG see-CONV al nt u-l so many xor?? man kill-PART-LOC sorry NEG be- T.3SG.SUBJ With your eyes seeing, you won’t feel sorry for so many xor?? people being killed. (Pápay 12). 9 The subject of the relative clause might not be overtly expressed with a separate lexical item, in this case the head noun must attach the possessive affix referring to the relative clause subject, even if the two subjects are not coreferential. As we show in Ackerman and Nikolaeva (1987), the possessive affix then is the only way to encode the relative clause subject and has agentive rather than possessive interpretation. 10 Quantifyiers isa/ sa, on the one hand, and itliji/xoliji, on the other hand, are dialectal variations (shown here and hereafter with slash). It is beyond the scope of the paper to discuss the semantic difference between isa/ sa and itliji/xoliji. 11 The number of examples which exhibit subject-drop is not shown specifically. 12 The Uralic accusative affix has been lost in most Ostyak dialects for phonologial reasons. Notice, however, that the overt accusative for nominals is still present in some Vogul and Ostyak dialects, as well as in related Samoyedic languages where marks direct objects independently of object agreement. 48 13 The absence of overt morphological distinction between O1 and O2 can also present difficulties for the theory that crucially relies on difference in (structural) case assignment for VP-internal and VP-external object (like in de Hoop 1992) or, at least, makes it more abstract: one would need some additional arguments for opposing structural and morphological case. 14 Within the GB framework Huang (1984, 1991) shows that the null object in this case is neither PRO, nor NP- trace, nor pro, the latter possibility is ruled out precisely because it does not meet the requirement of licensing and identification by agreement. Huang argues that it can be analyzed akin with VP-ellipsis as a variable bound by a null operator (possibly a null topic). 15 Of special interest are sentence where the focus status is borne by the predicate itself with the exclusion of other elements. This case is not addressed here. 16 Ostyak demonstrates an interesting parallel to Chichewa in interaction of focus with object agreement. Like in Chichewa, in questions to the object the verb does not show object agreement (example (55a)) because agreement generally marks the non-focus status of the object (O2) and in questions the focus status is borne by the object wh-word. On the contrary, in the questions to oblique (examples (55b)-(55c)), the object is always encoded as O2. The reason for this is that the oblique occupies the focus position, so that the object must be located outside it and trigger agreement. In spite of striking similarities between the two languages, the difference in object agreement between Ostyak and Chichewa discussed in section 5.2., seems to be motivated by subtle difference in information structure status of object that trigger agreement. In Chichewa is object is described as topic defined by Bresnan and Mchombo (1987, 746) as the element “under discussion”. In Ostyak, as I will show (Nikolaeva, in preparation), O2 is not a topic in this sense but rather can be characterized as “secondary topic” (the term is found in Givón 1984). Roughly, secondary topic is an element in information structure that stands in a certain relationship to the primary topic (or simply topic in the aforementioned sense) but is not important enough to become a primary topic. The secondary topic, therefore, is impossible in the absence of the primary topic in the sentence. This difference seems to be crucial. Since Li and Thompson 1976 it is well known that the topic (the primary topic) can be external to the clause: it might not bear the argument function and be generated as an adjunct. It is not at all obvious whether the same can apply to the secondary topic. Further research is needed in this direction.