Anaphora in Ossetic correlatives and the typology of clause combining*

Oleg Belyaev Institute of Linguistics of the Russian Academy of Sciences & Sholokhov Moscow State University for the Humanities

Ossetic, an Iranian language of the North Caucasus, makes extensive use of correlatives in its system of subordination. I argue that most subordinate clause types in Ossetic belong to this construction, with their function being derived from the combination of the meaning of the subordinating element and the form of the correlate. I further argue that the link between the relative phrase and the correlate may involve indirect relations such as bridging and split antecedence. They must therefore be viewed as obligatory pronominal coreference, where correlates are semantically no different from ordinary anaphoric pronouns or definite descriptions.

Keywords: clause combining; subordination; relative clauses; correlatives; anaphora; syntax; Iranian languages; Ossetic

1. Introduction

Ossetic is a Northeastern Iranian language spoken by about half a million people in the Republic of North Ossetia, part of Russia, in the North Caucasus, and in the disputed region of South Ossetia in Transcaucasia. Ossetic has two main dialects: Iron and

* I am thankful to Maria Kholodilova, Natalia Serdobolskaya, Andrej Sideltsev, Dag Haug, Ash Asudeh, Tatiana Nikitina, Misha Knyazev, the participants of the neo-LENCA Work- shop held in Stockholm in 2012 and three anonymous referees for the helpful discussion and comments on earlier versions of this paper. I would also like to thank my consultants: Madina Darchieva, Irina Mirikova, Marina Gurieva and Fatima Aguzarova, for their transla- tions and grammaticality judgements. I am also grateful to Aslan Guriev, Tamerlan Kambolov, Zalina Kusaeva and Kharum Takazov for their assistance in the work on Ossetic. Finally, I thank Pirkko Suihkonen and Lindsay Whaley for their editorial work. All errors are mine. This ­research has been carried out with the support of the Russian Foundation for the Humanities, projects No. 12-34-01345 and 13-04-00342, and the Presidium of the Russian Academy of ­Sciences (programme “Corpus linguistics”).  Oleg Belyaev

Digor, both of which have literary forms. This paper deals with the Iron dialect, which is spoken by the majority of Ossetians. Sourced examples supplied by the abbreviation “ONC” are taken from the Ossetic National Corpus 〈http://corpus.ossetic-studies.org/ en〉, with the source given as in the Corpus; unsourced examples are from my field- work in Vladikavkaz and Alagir (North Ossetia). Ossetic grammar is characterized by fusional morphology, rather typical for Indo-European languages, and by agglutinative nominal morphology, with 9 cases and 2 numbers. The unmarked is SOV, although all other possibilities, especially SVO, are also quite frequent and reflect various information structure con- figurations. Ossetic has a highly grammaticalized preverbal position, where focused elements and negative pronouns are obligatorily placed (Lyutikova & Tatevosov 2009). Alignment is fully accusative, with no traces of ergativity in spite of prolonged contact with Caucasian languages. One of the most unusual features of Ossetic grammar is its system of subordina- tion. In this paper, I will argue that most of Ossetic finite subordinating constructions are to be classified as subtypes of the so-called correlative construction, where the subordinate clause is co-indexed by a demonstrative pronoun or in the main clause, without the two forming a constituent. The functions of particular subordinate clauses are derived from the forms of the subordinator (relative phrase) and the corre- late. I will further argue that the relation between the relative phrase and the correlate in Ossetic relative correlatives may involve bridging and split antecedence. This rela- tion must, therefore, be described as a case of obligatory pronominal anaphora, with the relative phrase serving as the antecedent. The correlate, in this treatment, is an ordinary anaphoric pronoun, serving as the referring expression.

2. Correlatives and other subordinate clause types in Ossetic

Most subordinate clauses in Ossetic are constructed according to the model demon- strated in the following example, which illustrates a relative clause: (1)1 [didinȝ̌-ətɜ sə čəžg-ɜn ba-lɜvar kod-t-aj], flower-pl what girl-dat pv-present do-tr-pst.2sg

. The transcription of Ossetic closely followsDzaxova (2009), with the following differ- ences: I use diacritical marks instead of digraphs for affricates (i.e. /č/ instead of /tš/, etc.) and /ə/ instead of /ɘ/. I transcribe geminate consonants using the /ː/ sign unless there is an explicit­ morpheme boundary. The names of Ossetic sources are transliterated (not transcribed) using a simpler system. The glosses follow the Leipzig rules( http://www.eva.mpg.de/lingua/ resources/glossing-rules.php). I consistently differentiate between grammatical abbrevia- tions, which use small caps (e.g. gen, pst.1sg), and lexical abbreviations (e.g. 1Sg ‘me’, Dem Dist ‘that’). Anaphora in Ossetic correlatives and the typology of clause combining 

fed-t-on wəj fəd-ə2 see.pfv-tr-pst.1sg DemDist[gen] father-gen ‘I saw the father of the girl who you gave flowers to’. Example (1) is a typical instance of a correlative construction (Lipták 2009): the sub- 3 ordinate clause is to the left of the main clause, the subordinator (sə čəžgɜn), or NPrel in the terminology of Andrews (2007), is obligatorily preverbal, and the subordinate clause is coindexed by a freely positioned demonstrative pronoun (wəj). In this exam- ple, the pronoun cannot be omitted and its position in the main clause is free (I will call such pronominal elements correlates in the rest of this paper; they correspond to NPmat in Andrews 2007). These features obtain in most of the Ossetic subordinate clauses, although all have exceptions: a few subordinators are not obligatorily prever- bal, the correlate need not be present in certain special cases, and the subordinate clause can be embedded in the main clause or be in postposition. All of this will be described in more detail in the following sections. Apart from this class of constructions, Ossetic has some less frequent types of subordination. First, it also possesses postnominal externally-headed relative clauses, although these are used very infrequently, mostly for appositive (non-restrictive) rela- tive clauses: (2) χorž ɜfšin šatana, [nart-ɜn žond či aməd-t-a]… good lady Satana Narts-dat knowledge who.nom show-tr-pst.3sg’ ‘The good lady Satana,who showed knowledge to the Narts…’ (ONC: Max dug4 1, 2008) Postnominal relative clauses, like correlatives illustrated in (1), also use an interrogative-­ derived relative pronoun (či ‘ w h o’, sə ‘what’ or kɜsə ‘which’) in the preverbal position, but they cannot contain an internal head, and the subordinate clause forms a constitu- ent with the head. In the written language, a different kind of postnominal relative clause is found, introduced by the pronoun kɜsə ‘which’ in clause-initial position. It is unanimously identified by native speakers as an artificial Russian-influenced construction (Gagkaev 1956: 221).

. Ossetic possesses differential marking. As a rough generalization, nominal direct objects are usually marked by genitive if human and by nominative if inanimate, although the actual distribution is somewhat more complicated (Kulaev 1961). Direct objects expressed by personal pronouns and enclitics are always genitive-marked. . I will be using the cover term “subordinator” both for noun phrases like in (1) and non- phrasal subordinating elements (kɜj, kʷə, sɜmɜj, etc.). When discussing NPs, I will also employ the term “relative phrase” in the same function. . “Max dug” (‘our era’) is an Ossetian literary magazine.  Oleg Belyaev

Ossetic also has so-called pseudocoordination (Yuasa & Sadock 2002), where the semantically subordinate clause is introduced by the coordinating conjunction ɜmɜ ‘and’: (3) birɜ-tɜ ɜnqɜl wəd-əštə, [ɜmɜ χušːar irəšton many-pl (think) be-pst.intr.3pl and south Ossetia nɜ ba-χaw-zɜn značit-ə k’uχ-t-ə] neg pv-fall-fut[3sg] Znachit-gen hand-pl-in ‘Many thought that South Ossetia would not fall into Znachit’s hands.’ (ONC: Max dug 5, 2006) Such constructions have mixed status between coordination and subordination (­Belyaev forthcoming), similar to the English “left-subordinating AND” described in Culicover & Jackendoff (1997). A special case of pseudocoordination which I will call “coordinating inversion” will be briefly discussed in this paper. Some complement clauses are asyndetic: (4) žɜrond uš =dɜr ɜnqɜl u, [či- dɜr old woman add (think) be.prs.3sg who-indef =ɜj ba-fχɜrd-t-a] 3SgEncl.gen pv-offend-tr-pst.3sg ‘And the old woman thinks that someone has offended her.’ (ONC: Max dug 5, 2002) Direct speech is often marked by the participle-converb of the verbž ɜʁən ‘to say’, or by the citation particle =dam: (5) [ratː, žɜʁ-gɜ], žaχt-a give.pfv[imp.2sg] say-pcvb say-pst.tr.3sg ‘He said, “Give”.’ (ONC: Iron adæmon sfældystad. Dykkag tom, 2007) Finally, non-finite subordination is also available in Ossetic, although it is a minor strategy compared to finite subordinate clauses. Non-finite clauses are used for clausal complements in control constructions (6) and for converbal subordination (7), which is very infrequent in the spoken language (for more info on converbs cf. Belyaev & Vydrin 2010; on infinitives and nominalizations cf. Lyutikova & Tatevosov to appear). (6) mɜn5 dɜ= sard-ə ni-sə fɜnd-ə iv-ən 1Sg.gen 2SgPoss life-in neg-what want-prs.3sg change-inf ‘I don’t want to change anything in your life.’ (ONC: Max dug 2, 1996)

. The verb fɜndən ‘to want’ in Ossetic requires genitive marking of the Experiencer and nominative marking of the Stimulus; the verb itself agrees with the Stimulus, and thus is in 3rd person singular in this case. Anaphora in Ossetic correlatives and the typology of clause combining 

(7) dɜš-ə χištɜr jɜ= wɜχšč-ət-əl ten-gen senior 3SgPoss shoulder-pl-super š-χɜs-gɜ-jɜ ičo-mɜ ba-kašt pv-raise-pcvb-abl Icho-all pv-look[pst.intr.3sg] ‘The leader of the ten-men group looked at Icho,raising his shoulders.’ (ONC: K’æbysty Zauyr, Æcægælon arvy byn, 2010) In what follows I will only discuss the major types of canonical finite subordination in Ossetic. Specifically, I will exclude non-finite clauses, pseudocoordination (except for coordinating inversion, as it is important for delimiting the boundaries of correla- tives), direct speech marked by citative particles, asyndetic subordination and post- nominal relative clauses, as all of these are either comparatively infrequent, limited to specific conditions such as same-subjecthood, or have unclear status with respect to the coordination-subordination distinction.

2.1 Subordinators and correlates As stated above, most subordinators in Ossetic are always preverbal (PV), or, more precisely, are located in the preverbal area, as they can only be separated from the verb by negation, negative pronouns, and certain (Abaev 1959: 163; Lyutikova & Tatevosov 2009). This is identical to the behaviour of interrogatives, which, as focused elements, obligatorily stand in the same preverbal position: (8) ‹*či›6 žawər-ə ‹či› fed-t-a? who.nom Zaur-gen who.nom see.pfv-tr-pst.3sg ‘Who saw Zaur?’ In fact, most of the canonical subordinators in Ossetic, preverbal or not, except for iwgɜr ‘if’ and, possibly, səma ‘as if’, are either homonymous with interrogatives or eventually diachronically derived from them. Five subordinators in Ossetic are not obligatorily preverbal (nPV), and can stand in any position preceding the verb: (9) mɜn fɜnd-ə, [‹sɜmɜj› žawər ‹sɜmɜj› 1Sg.gen want-prs.3sg Purp Zaur rajšom ‘sɜmɜj› ɜrba-sɜw-a ‹*sɜmɜj›] tomorrow pv-go-sbjv.3sg ‘I want Zaur to come7 tomorrow.’

. Expressions marked by angular brackets are mutually exclusive, i.e. it is implied that, if they appear in one of the specified positions, they do not appear in the others. . The meaning ‘to come’ in Ossetic is expressed by using the ɜwən ‘to go’ together with the appropriate preverb.  Oleg Belyaev

Due to this variability in positioning nPV subordinators, Erschler (2012) refers to them using the term “floating”. In general, the subordinate clause can include only one subordinator.8 In particular, both a PV and an nPV subordinator cannot be present in the same subordinate clause, even if their meaning is the same. The only exception is sɜmɜj ‘in order that’ and kʷəd ‘how’ that can be both simultaneously used in a purpose or a complement clause: (10) detːa-jɜn žaχt-on, [sɜmɜj =mɜ Dedta-dat say-pst.tr.1sg Purp 1SgEncl.gen am kʷəd waz-a], aftɜ DemProx.in how pv-go-sbjv.3sg so ‘I told Dedta to leave me here.’ (ONC: Max dug 6, 2004) Another use of kʷəd ‘how’ is as a particle in independent sentences, also preverbally, thus (10) could be analyzed as only having sɜmɜj as a real subordinator. However, this interpretation is impossible, as the correlate in (10) is aftɜ ‘so’, which is typical for kʷəd, but not for sɜmɜj (see Table 1). Different subordinators use different correlates, but there is no fixed connection between the two. Rather, the choice of the correlate reflects the syntactic position the subordinate clause occupies in the matrix clause. For example, kʷə marks both tem- poral and complement clauses; in the former case, the correlate is wɜd ‘then’, while in the latter, it is wəj ‘that’ (distal demonstrative). Thus, it is only from the combination of the subordinator and the correlate that the meaning of the whole construction can be deduced. Complement clauses and relative clauses are resumed by the correlate wəj ‘that’ or an NP containing a demonstrative determiner in different case forms, cf. (1), (11b), (13); temporal clauses are resumed by wɜd ‘then’ or by time-referring NPs (which includes the pronoun wəj ‘that’) in appropriate case forms, cf. (11a), (15), (17), manner clauses are resumed by aftɜ ‘thus’, etc. (11) a. [dɜ= nəχaš =dən kʷə a-jqʷəšt-on], 2SgPoss speech =2SgEncl.dat when pv-hear-pst.tr.1sg wɜd ba-sin kod-t-on then pv-happiness do-tr-pst.1sg ‘When I heard you speak, I became happy.’ b. [iron-aw =ɜj kʷə žɜʁ-id], Ossetian-equ 3SgEncl.gen if say-opt.3sg wəj =jɜ fɜnd-ə… DemDist[nom] 3SgEncl.gen want-prs.3sg ‘He wants him to say it in Ossetic…’ (ONC: Max dug 2, 2001)

. With the exception of multiple correlatives, which I treat as a separate construction (see Section 2.6). Anaphora in Ossetic correlatives and the typology of clause combining 

Moreover, the list of possible correlates is not fixed for each subordinator: in many cases a full NP with the appropriate noun can be used in place of a more commonly employed pronoun or adverb, as in (12), where the pronoun wəj either in the comita- tive form (‘with that’) or as a dependent of a head noun (‘the wish for that’) would be sufficient for reference to a proposition. (12) a-sə ɜʁdaw [tər sɜmɜj ra-jgʷər-a], DemProx-attr tradition baby Purp pv-be_born-sbjv.3sg wə-sə bɜlːicː-imɜ bašt u DemDist-attr wish-comit linked be.prs.3sg ‘This tradition is connectedto the wish that a baby be born.’ (ONC: Max dug 2, 2007) A lexical head can be present inside either the subordinator or the correlate. It may also be found in both positions, and the nouns used can even be different: (13) [uroč-ə qɜr-ɜj sə lɜpːu zərd-t-a], lesson-in shout-abl what boy speak-tr-pst.3sg wə-sə fədwaȝ̌-ə nəjːarȝ̌-ət-imɜ DemDist-attr hooligan-gen parent-pl-comit a-nəχɜš-tɜ kɜn-zən-ɜn pv-speech-pl do-fut-1sg ‘I will speak to the parents of the misbehaving boy who spoke loudly at the lesson.’ (lit. ‘Which boy spoke loudly at the lesson, I will speak to the parents of that hooligan.’)

The correlate must contain a distal demonstrative or a “property” demonstrative aχ( ɜm ‘such’), and it is usually impossible for the subordinator to be coindexed by an enclitic, a possessive pronoun, a zero , or an NP without a demonstrative. The exception is when the correlate takes part in a construction where using a possessive proclitic is strongly preferred, such as with tɜkːɜ ‘most’:

(14) ɜmɜ [žɜχː-əl sə χɜšt-ətɜ wəd-iš], and earth-super what war-pl be-pst.intr.3sg šɜ= tɜkːɜ karž-dɜr təʁd bədər-ə š-šəʁd-i 3PlPoss most violent-compar wide field-in pv-burn-pst.intr.3sg ‘And the most violent of the wars that were fought on Earth began in an open field.’ (ONC: Plity Gris,Uacmystæ , 2004)

The most widely used Ossetic subordinators are summed up in Table 1. The brackets around a correlate signify that, under certain circumstances (to be specified below), the correlate may be omitted. The column “pos” specifies the position of the subordi- nator: preverbal (PV) or non-preverbal (nPV). The column “wh” specifies whether the  Oleg Belyaev

subordinator can also function as an interrogative in independent sentences. Wher- ever wəsə ‘that’ is specified,aχ ɜm ‘such’ can be used.

Table 1. Major subordinators and their correlates9

subordinator correlate meaning pos wh

sə ‘what’ wəj ‘that’ relative PV + či ‘who’ wəsə ‘that’ + NP sə ‘which’ + NP kʷə ‘when’ wɜd ‘then’ time, condition PV – wəsə + time-NP time wəj + postpos. wəj complement kʷəd ‘how’ aftɜ ‘thus’ manner, time, purpose PV + (wəj) complement kɜm ‘where’ wəm ‘there’ location, condition PV + wəsə + place-NP location wəj + postpos kɜj (wəj) complement PV - wəj + postpos. cause and other adverbial meanings all interrogatives (wəj) embedded question, PV + multiple correlative wəsə + NP sal ‘how many’ (+NP wal ‘that many’ (+NP) quantity PV + saš ‘how much’ (+NP) wəjaš ‘that much’ (+NP) PV + savɜr ‘what kind of’+NP aχəm ‘such’+NP quality PV + sɜmɜj ‘in order that’ (wəj təχːɜj) ‘for that’ purpose nPV −10 (wəmɜn) ‘to that’ (wəj) complement kɜd wɜd condition nPV − iwgɜr wɜd condition nPV − salənmɜ ‘while’ wɜdmɜ ‘meanwhile’ time limit nPV − walənmɜ ‘meanwhile’ səma ‘as if’ (aftɜ) manner, complement nPV −

. Two of the nPV subordinators are actually homonymous with interrogatives: sɜmɜj is the ablative form of sə ‘ w h a t ’, a n d kɜd means ‘when’. However, both their position and their seman- tics in subordinating constructions are different from their usage in independent sentences. Anaphora in Ossetic correlatives and the typology of clause combining 

An important observation is that the semantics of PV subordinators does not always strictly correspond to the semantics of the relevant interrogatives. The most telling example is with temporal clauses: the interrogative for ‘when’ is kɜd, but tempo- ral subordinate clauses are marked by either kʷə ‘when’ or kʷəd ‘how’, never with kɜd. For its part, kʷə is never used in interrogatives (also cf. Erschler 2012). (15) a. [kʷə/*kɜd =dɜ10 fed-t-on ], wɜd… when 2SgEncl.gen see.pfv-tr-pst.1sg then ‘When I saw you…’ b. kɜd/*kʷə =mɜ fed-t-aj? when 1SgEncl.gen see.pfv-tr-pst.2sg ‘When did you see me?’

Contrast this with embedded questions, where only kɜd can be used: (16) žon-ən, [də kɜd/*kʷə know-prs.1sg 2Sg.nom when ɜrba-səd-tɜ], wəj pv-go-pst.intr.2sg DemDist[nom/gen]11 ‘I know when y ou c am e .’ Thus, the subordinators used in simple correlatives must in principle be distinguished from interrogatives, constituting a separate class of lexical items. Embedded questions and multiple correlatives behave somewhat differently, as will be discussed below.

2.2. The position of the subordinate clause In addition to being preposed, each subordinate clause can also be center-embedded inside the main clause, in which case it must immediately precede the correlate: (17) tajmuraz-əl, [sɜf kʷə fɜ-dɜn], Taymuraz-super injured when pv-be.prs.1sg wəj fɜštɜ wɜrtɜ qɜw-gɜron fe-mbɜld-dɜn DemDist[gen] after over_here village-end pv-meet-pst.intr.1sg ‘I met Taymuraz over here, at the outskirts of the village, after I was injured.’ (ONC: Cægæraty Maksim. Æxsyzgon ulæft, 2002)

. Since the enclitic is second-position (Wackernagel-like) and does not count as a full constituent, the subordinator in this example is in preverbal position. . The distal demonstrative does not obligatorily distinguish between nominative and geni- tive cases, which is why I gloss it as nom/gen in direct object position.  Oleg Belyaev

In this configuration, the subordinate clause and the correlate form a constituent, which is demonstrated by the fact that two such pairs can be coordinated:

12 (18) alan kɜš-ə [NP [sə činəg =ən Alan read-prs.3sg [NP what book 3SgEncl.dat ba-lɜvar kod-t-a aslan], wəj], pv-present do-tr-pst.3sg Aslan DemDist[nom/gen] j ɜmɜ [NP [sə stat ja nə-ffəšt-a žawər], wəj] and [NP what article pv-write-pst.tr.3sg Zaur DemDist[nom/gen] ‘Alan is reading the book that Aslan presented him with and the article that Zaur wrote.’

Normally, the subordinate clause cannot follow the main clause: the correlate must be present, even if it is the last word in the sentence (19); the exceptions will be discussed in the next section.

(19) ɜž χɜr-ən, [sɜmɜj sɜr-on], 1Sg.nom eat-prs.1sg Purp live-sbjv.1sg wəj təχːɜj, inːɜ-tɜ =ta sɜr-əns, DemDist[gen] for other-pl Contr live-prs.3pl [sɜmɜj χɜr-oj], wəj təχːɜj Purp eat-sbjv.3pl DemDist[gen] for ‘I eat in order to live, while others live in order to e at .’ (ONC: Max dug 6, 2004)

2.3 “Omission” of the correlate In the majority of subordinate clause types, the correlate can never be omitted. It can only be absent if both of the following are true:

1. The subordinate clause is either a complement clause (correlate wəj) or a purpose clause (correlate wəj təχːɜj or wəmɜn) (Abaev 1959; Serdobolskaya to appear).13 2. The subordinate clause is located after the main clause.

. The relative phrase sə činəg in this example is in preverbal position: the verb is a complex predicate whose nominal part is marked by a preverb, which makes it behave as a single element indivisible by anything except for enclitics. On the enclitic see Footnote 11. . In Serdobolskaya (forthcoming), it is also demonstrated that the presence of the correlate with certain complement clauses is related to factivity. This is never a strict rule, however: the correlate can always be restored. Anaphora in Ossetic correlatives and the typology of clause combining 

This is illustrated in (20)–(22): (20) a. ɜž žon-ən, [žawər šajɜgoj kɜj 1Sg.nom know-prs.1sg Zaur liar Comp u], (wəj) be.prs.3sg DemDist[nom/gen] b. [žawər šajɜgoj kɜj u ], *(wəj) ɜž žonən ‘I know that Zaur is a liar.’ (21) a. žawər ɜrba-səd-i, [sɜmɜj də Zaur pv-go-pst.intr.3sg Purp 2Sg.nom jemɜ zur-aj], (wəj təχːɜj) 3Sg.comit pv-speak-sbjv.2sg DemDist[gen] for b. [sɜmɜj də jemɜ zur-aj], *(wəj təχːɜj) žawər ɜrba-səd-i ‘Zaur came in order for you to speak to him.’ (22) a. [də kʷə rba-səd-tɜ], *(wɜd) 2Sg.nom when pv-go-pst.intr.2sg then ba-sin kod-t-on pv-joy do-tr-pst.1sg b. basin kodton, [də kʷə rbasədtɜ], *(wɜd) ‘I was happy when y ou c am e .’ There are, in principle, two possible interpretations of this phenomenon:14 either examples with such “omission” (20a)–(21a) contain a zero pronoun as a correlate, or they do not belong to the correlative type at all, i.e. the subordinate clauses with “zero” correlates occupy the appropriate structural positions without the mediation of a pronoun that we observe in other cases. The latter explanation appears to be more plausible, because it would be ad hoc to postulate zero pronouns that are restricted to correlative constructions, only appear in a specific linear position, and can only be subjects, objects or purpose adjuncts. Furthermore, complement and purpose clauses without correlates are open for certain long-distance dependencies that cannot cross the boundary of a correlative clause, in particular, for leftward topicalization:

(23) koloni-jə fɜštɜ bəsəqɜni qɜw-sovet ɜmɜ šk’ola prison-gen after B. village-soviet and school

. An anonymous reviewer suggests that an analysis in terms of ellipsis is also possible. However, I am aware of no known ellipsis types that display similar behaviour, and postu- lating a special kind of ellipsis for this construction would be ad hoc. Furthermore, the con- trast between (20)–(21) and (22), as well as between (a) and (b), is inexplicable. Also hard to explain is the correlation of the presence of wəj with factivity (Serdobolskaya to appear).  Oleg Belyaev

sə nɜ kod-t-oj, [sɜmɜj Øi aχʷər what neg do-tr-pst.3pl Purp study kod-t-a-id], fɜlɜ jɜ= k’ɜχ-tɜ nə-bːəsɜw kod-t-a do-tr-cntrf-3sg but 3SgPoss leg-pl pv-fight do-tr-pst.3sg ‘After (he had spent time in) prison, whatever the village soviet and the school did in order for Bitsiqan to study, he resisted.’ (lit. ‘After prison Bit-

siqani what didn’t the village soviet and school do [so that Øi learns], but he resisted.’) (ONC: Xæmycaty Albeg, Bonivajænty, 1982) (24) a. ɜž žon-ən, [žawər tuʁan-ə kɜj 1Sg.nom know-prs.1sg Zaur Tugan-gen Comp šːard-t-a], wəj find.pfv-tr-pst.3sg DemDist[nom/gen] b. *tuʁan-ə ɜž žonən, [žawər kɜj šːardta], wəj ‘I know that Zaur has found Tugan.’ Another argument against the idea of a zero correlate is that a subordinate clause lack- ing a correlate can be coordinated with a “subordinate clause+correlate” group: (25) mušːɜ šɜ= wɜlquš lɜwːəd-iš ɜmɜ Mussa 3PlPoss near stand-pst.intr.3sg and gɜχːɜtː-əl fəšt-a, [sal bričkɜ-jə paper-super write-pst.tr.3sg how_many britzka15-gen a-rvəšt-a] ɜmɜ [=šɜ či pv-send-pst.tr.3sg and 3PlEncl.gen who.nom lašt-a], wədon nɜm-tːɜ drive-pst.tr.3sg DemDist.pl[gen] name-pl ‘Mussa stood near them and wrote down, how many britzkas he sent and the names of those, who drove them.’ (ONC: Mamsyraty Dæbe, Cardy k’æpxæntyl, 1988) It would be unusual if an NP with a zero head were coordinated with a full NP, as null elements cannot generally undergo coordination with overt ones. It is more probable that in (25) we deal with the coordination of a CP and an NP (such coordination of unlikes is expected in this position, cf. Sag et al. 1985). These arguments make a “zero” or “optional” correlate analysis of examples like (20a)–(21a) very implausible. If such an analysis is ruled out, there seems to be no other logical possibility than to consider complement and purpose clauses lacking cor- relates to have different syntactic structure from those clauses that do have correlates.

. A type of horse-drawn carriage. Anaphora in Ossetic correlatives and the typology of clause combining 

In case a correlate is present, it occupies the complement position of the main clause, while the subordinate clause is an adjunct either to the main clause or to the correlate. In case there is no correlate, the complement position is occupied by the subordi- nate clause itself, without the mediation of a pronominal correlate, just as what we find in languages like English or Russian (according to traditional analyses). It should be mentioned that the same conclusion has been reached based on different data in ­Gareyshina et al. (2011), which provides independent support for this analysis.

2.4 Coordinating inversion Every correlative construction has an alternative form which I will call “coordinat- ing inversion”: the correlate is in a focus position (either pre- or postverbal),16 the subordinate clause stands after the main clause and the two clauses are linked by the coordinating conjunction ɜmɜ ‘and’: (26) … dišːag wəj u, ɜmɜ [strom 48 wonder DemDist[nom] be.prs.3sg and Strom až-ə dɜrʁ-ə senator kɜj u!] year-gen length-in senator Comp be.prs.3sg ‘What is a wonder is that Strom is senator for 48 years!’ (ONC: Max dug 2, 2003) (27) wəj aχɜm lɜg u, ɜmɜ [ɜpːəndɜr DemDist[nom] such man be.prs.3sg and at_all jɜ= koj kɜm-ɜn nɜ fe-qʷəšt-oj] 3SgPoss rumour who-dat neg pv-hear-pst.tr.3pl ‘He is a man about whom no one has heard.’ (Sjögren 1844: 487)17

Some native speakers actually allow omitting the conjunction ɜmɜ, which makes the construction look similar to simple postposition of the subordinating clause. Such examples should not be analyzed in this way, however, for two reasons. First, not all native speakers allow omitting the conjunction. Second, when the conjunction is

. An anonymous reviewer observes that some native speakers allow any position of the correlate in examples like (26). However, this construction is heavily dependent on pragmatic conditions that are not yet fully understood, and eliciting grammaticality judgments of iso- lated sentences is thus unreliable. For instance, artificially constructed examples analogous to (27) are often rejected by native speakers. As for corpus data, I have been unable to find a single example which does not conform to this generalization; therefore, if it is not a rule, then at least it is a very strong tendency. . Andreas Sjögren explained the existence of such constructions by “the clumsiness of a people still crude and uneducated” (ibid.).  Oleg Belyaev

­omitted, the requirement for the correlate to be focused still remains. Such a require- ment would not be expected were this simply an alternative variant of positioning the subordinate clause. Presumably, the function of coordinating inversion in Ossetic is to focus the sub- ordinate clause in such a way as to make it possible to situate it after the main clause – a configuration that is disallowed for ordinary correlatives. This construction is thus functionally, though probably not syntactically, akin to a cleft. Coordinating inversion is equally possible with PV and nPV subordinators: (28) lɜpːu-jɜn jɜ= χi =dɜr wəj boy-dat 3SgPoss Refl[gen] add DemDist fɜnd-ə, ɜmɜ [=jən sɜmɜj iwːəl =dɜr want-prs.3sg and 3SgEncl.dat Purp all add žon-oj jɜ= waršt-ə təχːɜj] know-sbjv.3pl 3SgPoss love-gen about ‘What the boy himself wants is for everyone to know about his love.’ (ONC: Max dug 2, 2001)

The conjunctionɜ mɜ ‘and’ in this construction retains many coordinating properties. For example, unlike PV and nPV subordinators, it must be positioned strictly between the two clauses. Detailed discussion of the syntactic properties of coordinating inver- sion is beyond the scope of this paper. In general, it appears to be similar to other “pseudo-coordinating” constructions in Ossetic, which display mixed coordinating- subordinating properties (Belyaev forthcoming). Like these constructions, it involves the conjunction ɜmɜ ‘and’ and is always encountered in postposition. This kind of drift from subordination to coordination may be related to the fact that focused subordi- nate clauses are used in assertion (Diessel & Hetterle 2011). The fact that coordinating inversion is applicable to different subordinators in Table 1, regardless of the clause type they introduce (e.g. (26) and (28) are complement clauses, while (27) is a relative clause), is one of the strongest reasons to consider all of these clause types to belong to one and the same syntactic construction. However, the exact circumstances under which coordinating inversion is possible are yet to be studied.

2.5 “Asyndetic” correlatives It has been mentioned above that Ossetic has true asyndetic constructions, with the subordinate clause being positioned after the main clause and no element marking subordination (4). However, there is another type of construction which may be called asyndetic in the sense that it has no subordinator, but does have a correlate: (29) nɜ =mɜ wərn-ə, [šərχ-ətɜ čeka-jə neg 1SgEncl.gen believe-prs.3sg red-pl Cheka-in Anaphora in Ossetic correlatives and the typology of clause combining 

zɜʁɜl χʷəmɜtɜȝ̌-ə urš aficer-ə ra-waz-oj], wəj in_vain simple-gen white officer-gen pv-leave-sbjv.3pl DemDist ‘I do not believe that the Reds in the Cheka would let go of a white officer for no reason.’ (ONC: Qajttaty Sergej, Fædisontæ, 1984) As seen from (29), these constructions follow exactly the same rules of positioning as ordinary correlatives: the subordinate clause is embedded to the left of the correlate. Thus, they have to be classified as correlatives, even if not completely canonical; it may be said that they employ a null subordinator. It is even possible that the relation between “normal” asyndetic subordination like (4) and “correlative” asyndetic subordination like (29) is the same as between complement clauses with and without correlates (20). This possibility requires further elaboration that is beyond the scope of this paper.

2.6 Multiple correlatives A widespread feature of correlative constructions is the possibility of multiple relativ- ization (Lipták 2009: 5). This is also possible in Ossetic: (30) čəžg-ɜj lɜpːu-jɜ, [či kɜj song-əl18 kʷəd girl-abl boy-abl who.nom who.gen arm-super how fɜ-χɜšt], təmbəl-sɜg-ɜj ɜr-χɜsəd-əštə pv-hold[pst.intr.3sg] round-ring-abl pv-hold-pst.intr.3sg šɜ= kɜrɜzi-jɜ wɜχšč-ət-əl 3PlPoss Recip-gen shoulder-pl-super ‘Girls and boys started to hold each other’s shoulders in the same way as they held each other’s hands.’ (lit. ‘Boys and girls, who held whose hand how, held each other’s shoul- ders.’) (ONC: Qazity Meliton, Æuuændyn adæjmagyl, 2005) This is an extreme example of multiple correlatives, with three positions being simul- taneously relativized. In line with the overall uniformity of Ossetic finite subordination, multiple rela- tivization is possible not only for relative clauses proper, but for other clause types, cf.: (31) [lɜpːu-t-ɜj či kʷəd səd-i], aftɜ boy-pl-abl who.nom how go-pst.intr.3sg so =iw wɜrtɜmɜn-ə raž ba-lɜwːəd… iter Wartaman-gen before pv-stand[pst.intr.3sg] ‘As the boys went, they stood before Wartaman.’ (lit. ‘How who of the boys went, so (he) stood before Wartaman.’) (ONC: Xuygaty Sergej, Narty farnæg, 2005)

. When part of an NP or a PP is relativized, the rest undergoes pied piping, because neither NPs nor PPs can ever be split in Ossetic.  Oleg Belyaev

There are three important differences between multiple and simple correlatives that require treating them as two separate constructions.

2.6.1 Restrictions on the use of subordinators The first difference is that multiple correlatives can only be formed using interroga- tives; nPV subordinators, as well as those PV subordinators that, while not being inter- rogatives, function in simple correlatives (kʷə and kɜj), cannot be used in multiple relativization. Instead of kʷə, kɜd – the word used for ‘when’ in independent interroga- tive sentences – has to be used: (32) [či kɜd/*kʷə ɜrba-səd-i], wɜd who.nom when pv-go-pst.intr.3sg then ba-šiχor kod-t-a pv-lunch do-tr-pst.3sg ‘Everyone ate lunch when they c am e .’ (lit. ‘Who came when, then (they) ate lunch.’)

Similar examples where kɜd is used are also found in written texts: (33) [kɜd kɜj ba-fɜnd-ə], wɜd =ɜm when who.gen pv-want-prs.3sg then 3SgEncl.all ba-χɜšː-ə jɜ= nwažinag ɜmɜ jɜ= χɜrinag pv-bring-prs.3sg 3SgPoss drink and 3SgPoss food ‘Whenever whoever wants, then (they) bring him their food and drink.’ (ONC: Xuygaty Sergej, Narty farnæg, 2005) The following demonstrates the ungrammaticality of nPV subordinators: (34) *[salənmɜ či rba-səd-i], wɜd while who.nom pv-go-pst.intr.3sg then ba-šiχor kod-t-a pv-lunch do-tr-pst.3sg ‘Everyone ate lunch while they came.’ (lit. ‘While who came, then (they) ate lunch.’) Therefore, given that the classes of subordinators and interrogatives are formally distinct in Ossetic (15)–(16), it is interrogatives, not subordinators, that are used in multiple correlatives, in spite of the fact that on surface the two are in most cases indis- tinguishable from each other.

2.6.2 Restrictions on the correlate The second important difference between simple and multiple correlatives lies in the acceptable forms of the correlate. Recall that the restrictions that simple correlatives Anaphora in Ossetic correlatives and the typology of clause combining  impose in this respect are very rigid: the correlate is obligatory and must contain either a distal demonstrative or aχɜm ‘such’, save for a few exceptional cases. Multiple correla- tives display no such constraints: in addition to the forms that can be used with simple correlatives, the correlate may also be a pronominal enclitic (35) or even a zero (36).

(35) wɜdɜ [čii səj ra-mbul-a], thus who.nom what pv-win-sbjv.3sg

wəji =jɜj jɜ= mad-ɜn ra- pːɜl-ɜd! DemDist[nom] 3SgEncl.gen 3SgPoss mother-dat pv-boast-imp.3sg

‘Thus let whoeveri has won somethingj boast of itj before their mother!’ (ONC: Bulk’aty Mixal, Terkæj Turkmæ, 2011)

(36) [čii kɜjj wardon-ə a-bad-ə], { Øi / who.nom who.gen cart-in pv-sit-prs.3sg

wəjj } žarȝ̌-ətɜ fɜ-kɜn-ə DemDist[gen] song-pl pv-do-prs.3sg

‘In whosej cart onei sits, that person’sj songs (they)i sing.’ (Ajlarty Izmail, Madælon ævžaȝ̌y tyxxæj)19 A similar observation is found in Bhatt (1997) for multiple correlatives in Hindi, which can have zero correlates even when corresponding simple correlatives cannot. Another difference concerns obligatoriness of the correlate. Like simple correla- tives, multiple correlatives cannot be used when no (even covert) correlate at all is present: (37) *[či sə rba-χɜšː-a], dwar ba-jgom kɜn who.nom what pv-bring-sbjv.3sg door pv-open do[imp.2sg] (‘Whoever brings whatever, open the door.’)20 However, multiple correlatives only require the last of the preverbal interrogatives to have some correlate in the main clause; the rest can behave as ordinary indefinites or free-choice items (Vendler 1967): (38) bɜrɜg nɜ-w, kɜd =šɜ known neg-be.prs.3sg when 3PlEncl.abl fəšt-a, wə-sə rɜštɜg, wə-mɜ write-pst.tr.3sg DemDist-attr time DemDist-all gɜšgɜ amənd sɜw-ə, ɜrmɜšt according_to show.ptcp go-prs.3sg only

. 〈http://allingvo.ru/LANGUAGE/madalon_avzadxy_tyhhaj.htm〉 . From here on, intended translations of examples will be given in parentheses.  Oleg Belyaev

[məχʷər-ə =šɜ či kɜdi print-in 3PlEncl.abl who.nom when

ra-səd], wə-sə rɜštɜgi pv-go[pst.intr.3sg] DemDist-attr time ‘The time when he wrote them is unknown, according to this it is only the time when each [lit. which] of them went in press that is indicated.’ (ONC: Basity Zælinæ. Dyuuæ zærdæjy, 2010) Thus, in a sense, we can say that in multiple correlatives only the immediately prever- bal interrogative element serves for establishing the relation between the two clauses. The rest of the interrogatives, although essential for the construction itself, do not directly participate in linking the two clauses, and behave like indefinite pronouns. How exactly this behaviour is to be explained is beyond the scope of this paper.

2.6.3 Semantics The third major difference between simple and multiple correlatives is that the latter can only have universal semantics. Specific reference of the participants is prohibited: (39) [kɜm-ɜn sə čəžg21 jɜ= žɜrdɜ-mɜ who-dat what girl 3SgPoss heart-all fɜ-səd-iš], wəj-ə ra-kʷərd-t-a pv-go-pst.intr.3sg DemDist-gen pv-ask-tr-pst.3sg ‘Everyone married whatever girl they liked.’ ‘Someone married a girl he liked.’ There is no logical necessity for multiple correlatives to involve such a semantic restric- tion, e.g. it is not observed in Hindi (Srivastav 1991). However, semantic restrictions are not unusual for correlatives in general. For example, contemporary standard (writ- ten) Russian only admits correlatives without a lexical head with universal semantics, both single and multiple (Nikunlassi 2008: 264–265).

2.7 Embedded questions Embedded questions in Ossetic share some similarities with multiple correlatives. The most obvious one is that embedded interrogatives can also be multiple (as can questions): (40) [innɜ pišmo-t-ɜj či kɜd fəšt other letter-pl-abl who.nom when write.ptcp

. In this example, sə čəžg ‘what girl’ is in preverbal position, because žɜrdɜmɜ sɜwən ‘to like’ (lit. ‘to go to one’s heart’) is a complex predicate. Anaphora in Ossetic correlatives and the typology of clause combining 

ɜr-səd], wəj bɜrɜg nɜ-w pv-go[pst.intr.3sg] DemDist[nom] known neg-be.prs.3sg ‘It is unknown when which of the other letters was written.’ (ONC: Max dug 10, 2000) The second similarity is that in embedded questions, like in multiple correlatives, interrogatives are used, and not subordinators, i.e. kɜd, and not kʷə, is used for tempo- ral reference (16). This is also not surprising, as the semantics of embedded questions are derived from the meanings of interrogative clauses; what is “relativized” are not the arguments expressed by interrogative pronouns themselves, but the content of the question. At the same time, embedded questions are similar to complement clauses utilizing lexical subordinators (20) in that the main clause must coreference the ques- tion by the distal demonstrative wəj (in the appropriate case form governed by the verb) if the subordinate clause is not sentence-final: (41) a. [kɜm sɜr-ə], wəm-ɜj =jɜ ba-faršt-on where live-prs.3sg DemDist-abl 3SgEncl.gen pv-ask-pst.tr.1sg b. *[kɜm sɜr-ə], ba-faršt-on =ɜj (=zə) where live-prs.3sg pv-ask-pst.tr.1sg 3SgEncl.gen 3SgEncl.abl ‘I asked him/her where s/he lives.’

2.8 Summary: Types of finite subordination in Ossetic Major canonical finite subordinate clauses (in the sense specified above, in the intro- duction to Section 2) in Ossetic can be divided into four types. The schemes below illustrate each of the types for clarity. The abbreviations SubCl and MainCl stand for “subordinate clause” and “main clause”; PV and nPV stand for “preverbal subordi- nator” and “non-preverbal subordinator”; V stands for “verb”; DemCorr stands for “demonstrative correlate”.

–– simple correlatives, used for the majority of subordinate clause types;22 these have two structural manifestations:

a. [(nPV) … (PV) V …]SubCl [… DemCorr …]MainCl b. [… [(nPV) … (PV) V …]SubCl DemCorr …]MainCl –– sentence-final “true” (non-correlative) complement and purpose clauses:

[ … V [(nPV) … (PV) V …]SubCl ]MainCl

. The idea that most of finite subordination in Ossetic, including complementation, is -es sentially relativization is originally due to Andreas Sjögren, the author of the first grammar of Ossetic (Sjögren 1844: 516). Unfortunately, this insight has largely been ignored in subsequent work on Ossetic grammar.  Oleg Belyaev

–– embedded questions (external distribution the same as simple correlatives or non-correlative complement clauses, but possibly multiple preverbal interroga- tives are used instead of PV or nPV); –– multiple correlatives (external distribution the same as simple correlatives, but multiple preverbal interrogatives used instead of PV or nPV, and no restrictions on the correlate).

Because syntactic and semantic differences between simple and multiple correlatives are by no means unusual (Bhatt 2003), the Ossetic situation seems to be the norm rather than the exception. However, the fact that different subordinators are used in single and multiple correlatives is unexpected; at least, I am not aware of any compa- rable distinction that has been described in the literature on correlatives or on relative clauses in general. All subordinate clauses that use the subordinators and correlates in Table 1 behave uniformly with respect to the position of the subordinate clause and the restrictions on the correlate23 and can participate in coordinating inversion regardless of their seman- tic type (relative, adverbial, or complement). For these reasons, all of them should be analyzed as belonging to the same syntactic construction. This may appear simi- lar to the claim in Arsenijević (2009) and Kayne (2010) that, in fact, all complement clauses are relative clauses, or the analysis of certain complement clauses in Adyghe as relativization in Gerasimov and Lander (2008) and Caponigro and Polinsky (2011). However, it is not clear whether Ossetic correlatives can be called relative clauses in the standard sense of the term. From the discussion above, it is certain that they are quite syntactically different from the latter; it will also be demonstrated below that the meanings of Ossetic correlatives cannot be described using the semantics typically used for canonical relative clauses. Therefore, even though Ossetic correlatives are functionally close to relative clauses and share with them a few structural similarities, whether the two can be considered to belong to the same kind of subordination from a cross-linguistic perspective is an open question.

3. Non-total coreference in Ossetic correlatives

Since Partee (1975), the semantics of relative clauses is described as set intersec- tion. Specifically, the meaning of the relative NPgirl who was standing is said to be [λx.girl’(x) & stand’(x)]; combining it with a definite determiner gives us, using the

. With the exception of complement and purpose clauses, which allow an alternative, non- correlative configuration; but this configuration is also available under regular and restricted conditions. Anaphora in Ossetic correlatives and the typology of clause combining 

­simplest account of definiteness, [ιx.girl’(x) & stand’(x)], which, using the iota-operator­ (Whitehead & Russell 1925–1927), refers to the unique individual at the intersection of the set of girls and the set of standing persons. Since correlatives are, at least on surface, syntactically different from canonical relatives, one can expect semantic differences as well. In the literature, the semantic analysis of correlatives is usually closely related to the syntactic analysis. Two main approaches that are most prominent can be called, following Lipták (2009), the “high adjunction & binding” analysis (Srivastav 1991; Dayal 1995; Davison 2009) and the “low adjunction & movement” analysis (Bhatt 2003). The former approach treats the relative clause as being base-generated in the left periphery and anaphorically linked to the correlate, while the latter situates it in the DP-adjoined position from where it can optionally move to the IP or CP-adjunct position. These two approaches obviously entail different ways of treating the way the semantics of correlatives is derived. However, what is fairly uncontroversial is the result of this semantic derivation, which generally leads to the same meaning that canonical relative clauses possess. For example, Srivastav (1991), who is a proponent of the peripheral adjunction and anaphoric linking approach, suggests that the correla- tive clause which girl is standing has the following meaning: (42) λP.P(ιx.girl’(x) & stand’(x)) The correlate is interpreted as a gap, which leads to the main clause being an expres- sion of type ‘e, t’, say [λx.tall’(x)], which is applied as an argument to (42); the result of this functional application is exactly the same as would have happened if we had used an ordinary definite relative clause, i.e. [tall’(ιx.girl’(x) & stand’(x))], which means ‘a unique entity at the intersection of the set of girls and the set of standing people is tall’. The semantics proposed in Dayal (1995) and Grosu and Landman (1996) involves an additional maximalization operation but does not otherwise differ from (42). This view of the semantics of correlatives does not distinguish between the rela- tive phrase and the correlate as separate entities or discourse referents and therefore does not permit any relation between them other than total identity. In what follows I will argue that, contrary to this prediction, this relation in Ossetic can involve the kind of indirect correspondences that is typical for pronominal anaphora, and must thus be analyzed as obligatory coreference rather than variable binding.

3.1 Split antecedents Unlike Hindi and other languages where stacking of correlatives in the left periphery is prohibited, there can be several subordinate clauses corresponding to the same cor- relate in Ossetic:

(43) fɜlɜ [iv-ən kɜji nɜ qɜw-ə] ɜmɜ but change-inf who.gen neg need-prs.3sg and  Oleg Belyaev

[wərəšːag jɜ= rajgʷərd-ɜj čii u], Russian 3SgPoss birth-abl who.nom be.prs.3sg

wədon-mɜi zašːoχ-ə fərt itːɜg bɜlːicːag-ɜj DemDist.pl-all Dzassokh-gen son quite desirable sə šːard-t-a? what find.pfv-tr-pst.3sg ‘But what did Dzassokh’s son find quite desirable for those who don’t need to change and who are Russians from birth?’ (ONC: Max dug 5, 1998) In (43), the correlate and each of the subordinators refer to the same participant, which is what one would expect given the standard semantics of relative clauses. However, very often we find a different kind of relationship where a single plural correlate -cor responds to several relative phrases that are not coreferent to each other:

(44) [sard-ɜj sə konflikt-tɜi išt-a], [sə life-abl what conflict-pl take-pst.tr.3sg what

χarakter-tɜj ɜvdəšt-a], wədoni+j character-pl show-pst.tr.3sg DemDist.pl wəd-əštə kadɜȝ̌-ə ɜrmɜg be-pst.intr.3pl legend-gen material

‘What conflictsi he took from life, what charactersj he demonstrated, theyi+j were legendary material.’ (ONC: Gædiaty Sek’a, Uacmystæ, 1991) (45) dɜ= fəd-mi-jɜn ɜvdišɜn wə-zɜn, [sə 2SgPoss bad-thing-dat witness be-fut[3sg] what arv-ə bən sɜr-əš], [sə žɜχː-əl sky-gen under live-prs.2sg what ground-super sɜw-əš], wədon go-prs.2sg DemDist.pl ‘The sky that you live under, the ground that you walk on will be witnesses to your wrongdoing.’ (lit. ‘To your wrongdoing will be witnesses, under which sky you live and on which ground you walk, them.’) (ONC: Qajttaty Sergej, Fædisontæ, 1984)

24 ?? This kind of meaning is not typical for canonical relative clauses, cf. I saw the meni+j whoi walked the dog (and) whoj walked the cat), but is normal for pronominal anaphora

. Canonical relative clauses have also been observed to involve split antecedents (Perlmutter­ & Ross 1970). This phenomenon occurs when one relative pronoun coreferences two external heads found in coordinate clauses (Mary met a mani and John met a womani whoi+j know each other well). This is completely opposite to what is found in Ossetic, where it is the correlate that has split antecedents in the form of relative pronouns/subordinators. Anaphora in Ossetic correlatives and the typology of clause combining 

(cf. Johni hit Davidj, and I called themi+j), where the term “split antecedents” is used to characterize such phenomena (Giorgi 1984). Split antecedents are also allowed in multiple correlatives:

(46) [čii kɜ-imɜj fɜ-χəl iš], wədoni+j who who-comit pv-fight Exst DemDist.pl ba-fidəd-t-oj pv-make_peace-tr-pst.3pl ‘Those who fought against each other have made peace.’

(lit. ‘Whoi fought with whomj, theyi+j made peace.’)

3.2 Bridging Even more convincing than split antecedence is the fact that the relation between the subordinator and the correlate may involve no coreference in the strict sense at all, but only an associative connection:

(47) [pišmo sə boni ra-jšt-a], wə-sə letter what day pv-take-pst.tr.3sg DemDist-attr

ižɜrj∈i ješt’a ɜmɜ verɜ-mɜ nə-fːəšt-a evening Ye. and V.-all pv-write-pst.tr.3sg ladinɜ-jə χabar bɜlvərd-dɜr ba-žon-ən-ə təχːɜj L.-gen story precise-compar pv-know-inf-gen for

‘On what dayi he received the letter, that eveningj∈i he wrote to Yesta and Vera in order to find out Ladina’s story better.’ (ONC: Ældattaty Viktor, Ladinæ, 1998)

(48) [afɜz-ɜj afɜz-mɜ =šɜ sə kɜnd-tɜi year-abl year-all 3PlEncl.gen what commemoration-pl

qɜw-ə], wə-sə χɜrz-tɜj~i mɜ= need-prs.3sg DemDist-attr expense-pl 1SgPoss χi-mɜ iš-ən Refl-all take-prs.1sg

‘What commemorationsi they need from year to year, I take the expensesj~i on myself.’ (ONC: Gaglojty Vladimir, Ġe, marʒæ, isči!.., 2009)

(49) danel, [fɜštag-mɜ jɜ= χo-t-ɜm sə k’am i Danel last-all 3SgPoss sister-pl-all what photo j a-rvəšt-a], wə-sə kost um-əj~i, urš pv-send-pst.tr.3sg DemDist-attr suit-in white χɜdon ɜmɜ tar galstuč-ə lɜwːəd shirt and dark tie-in stand[pst.intr.3sg]

‘Danel, which photoi he had sent his sisters last time, stood in that suitj~i, white shirt and dark tie.’ (ONC: Ældattaty Viktor, Ladinæ, 1998)  Oleg Belyaev

(50) [adɜjmag χʷəždɜr sə vžag-əli zur-a], person better what language-super speak-sbjv.3sg

wə-sə adɜməχatː-ɜjj~i u DemDist-attr nationality-abl be.prs.3sg be.prs.3sg

‘Which languagei a person speaks best, from that nationj~i they are.’ (ONC: Ajlarty Izmail, Gæȝ̌ynaty Rimæ, Kcojty Rimæ, Iron diššægtæ æmæ æmbišændtæ, 2006) This phenomenon is well-known in the literature on definiteness and anaphora under the term “bridging” (Clark 1975; Heim 1982). With definite descriptions, it occurs whenever the antecedent implies the existence of the entity denoted by the referring NP in some way, cf. When I go to a bar, the barkeeper always throws me out, but #When I go to a playground, the barkeeper always throws me out (examples from Bos et al. 1995). Examples (47)–(50) are all cases of bridging. In (47), “evening — day” is a text- book example of a part-whole relation. In (48), what is translated as “commemora- tions” refers to a meal that is held in order to honour a dead person. This implies expenses involved in organizing such an event, hence the connection between the rela- tive NP and the correlate. In (49), the relationship is less direct and consists in the fact that the person in question, Danel, had sent his sisters a photo in which he was wearing a suit, and this is the same suit as this person is described as wearing at present. Finally, in (50), a language is perceived to be necessarily associated with a nation that speaks it. Clearly, bridging cannot be accommodated into the standard semantics of rela- tive clauses in any way: the latter requires identity of the main clause participant and the subordinate clause participant, while bridging is by definition a relation which precludes such identity. Bridging is possible for multiple correlatives just as well as for simple ones:

(51) [čii səj akːag w–a], aχɜm χisawk~j who.nom what deserving be-sbjv.3sg such master

=əni vɜjː-ə 3SgEncl.dat be.hab-prs.3sg

‘Whateverj onei deserves, theyi have such a masterk~j. (ONC: Ajlarty Izmail, Gæȝ̌ynaty Rimæ, Kcojty Rimæ, Iron diššægtæ æmæ æmbišændtæ, 2006)’

4. The anaphoric analysis

As I have stated above, neither split antecedence nor bridging can be reconciled with the traditional semantics of relative clauses and require a less direct linking of some kind between the subordinator and the correlate. The semantic literature distinguishes between coreference and variable binding (Bach & Partee 1980; Büring 2004: 81–96, Anaphora in Ossetic correlatives and the typology of clause combining 

104–117; Reinhart 1999). The semantic difference between the two can be illustrated by the following examples:

(52) a. Peteri showed Johnj a photo of themi+j.

b. *Peteri showed Johnj a photo of themselvesi+j. Pronominal anaphora, as in (52a), can involve split antecedents, while reflexivization, as in (52b), cannot. The difference between the corresponding semantic representa- tions can, as a simplification, be represented in the following way:

(53) a. Johni wants to wash himselfi. ∃x. John(x) & want(x, wash(x, x))

b. Johni wants David to wash himi. ∃x. ∃y. ∃z. John(x) & David(y) & want(x,wash(y, z)) & ant(z, x) In (53b), the relation [ant(z, x)] denotes the identity of John (x) and him (z), but noth- ing in principle prevents the relation between z and x to be something else than total coreference. For example, we may say that the relation [ant(x,y)] implies either [x = y], [y ∈ x] or [bridge(x, y)]. Split antecedents are described via multiple antecedence rela- tions, i.e. e.g. [ant(x,y) & ant(x,z)] means that x has two antecedents, y and z, and the expression can be resolved to [y ∈ x & z ∈ x]. In contrast, in (53a) x is substituted into both predicates, thereby making anything other than total coreference logically impossible. Due to the possibility of bridging and split antecedence, the relation between sub- ordinators and correlates in Ossetic correlatives is clearly analogous to (52, 53a) rather than (52, 53b), i.e. involves coreference rather than variable binding. A schematic and simplified representation of (48) is the following: (54) ∃x. ∃y. commemorations(x) & needed(x) & expenses(y) & take(me, y) & ant(y, x) Paraphrased in natural language terms, (54) means something like “Commemora- 25 tionsi are needed. I take the expensesj (on myself)”. Correlatives are thus seman- tically close to parataxis with obligatory anaphora, where the subordinator is an indefinite and the correlate a pronoun or a definite description. The only difference is that in correlatives, the subordinate clause is in presupposition. How exactly the semantic derivation that leads to the result like (54) is structured is a question for further research. A dynamic model like DRT (Kamp & Reyle 1993) is probably required in order to provide a full formal account of the semantics of Ossetic

. Note that the translation, which is an acceptable sentence in English, involves exactly the same kind of bridging as the Ossetic original.  Oleg Belyaev

correlatives, especially in light of examples like (51), which involve universal quanti- fication and implication and are thus instances of “donkey anaphora” (Geach 1962). A very clear parallel to Ossetic correlatives in terms of semantics are relative clauses with resumptive pronouns, cf. the following example from Persian: (55) mærd-i [ke u ra diruz molaqat man-restr Comp he acc yesterday meeting kærd-id] aqa-ye Bayat bud. do.pst-2pl Mr-ezf B. be.pst[3sg] ‘The man that [you met (*him) yesterday] was Mr. Bayat.’ (Taghvaipour 2004: 277) Just like correlates, resumptives have been argued to be ordinary pronominals (­McCloskey 2002). In Asudeh’s (2012) analysis, so-called syntactically active resump- tives (SARs) contribute an anaphoric relationship while retaining the standard seman- tics that all pronominals have. There are even attested examples from spoken English that involve bridging:

(56) You assigned me to a paperi which I don’t know anything about the subjectj~i. (Prince 1990) The caveat, though, is that resumption in English is better understood as a process- ing strategy related to universal parsing constraints rather than as a grammaticalized means of relativization, i.e. so-called “intrusion” (Asudeh 2012: 279–325). In those languages where genuine resumption is found, bridging has not been attested (Ash Asudeh, p.c.). In general, only pronouns may be used as resumptives, with the excep- tion of Lebanese Arabic, where epithets can be used (Aoun et al. 2001). Thus, while both resumptives and correlates are anaphoric, the relation between the head of the relative clause and the resumptive is much more restricted than the relation between the relative phrase and the correlate in Ossetic correlatives. There are two other important differences between resumptives and correlates. The first difference is that in correlatives, binding proceeds from the subordinate clause into the main clause, while in resumption it is the other way around. In this sense, cor- relates can be called “backward resumptives”. Second, resumption in Asudeh’s (2012) analysis is just an additional device of establishing coreference; the standard semantics of relative clauses is retained. As I have argued above, this semantics does not work for Ossetic, and, therefore, coreference is the sole device that establishes the linking between the relative phrase and the correlate. Another parallel are control constructions, specifically so-called split control, where the zero subject (PRO) in the subordinate clause can have two arguments of the main clause as its split antecedents (Stiebels 2007: 5):

(57) Johni invited Maryj [PROi+j to go to the movies together]. Anaphora in Ossetic correlatives and the typology of clause combining 

However, control constructions involve zero pronouns, making phenomena such as bridging untestable. Thus, of special interest are the few languages where so-called “copy control” has been attested. For example, it is available in Telugu adverbial con- structions, which at first sight look rather similar to Ossetic correlatives:

(58) [Kumaar-kii koopam wacc-i], aa Kumar-dat anger.nom come-part.cnp that

pichooDui akkadi-nunci weLLipoyinaa-Du idiot.nom there-from left-3.m.s

‘Kumari got angry, and the idioti left.’(Haddad 2009: 84) Here we also find a free-form anaphoric NP in the main clause, possibly even contain- ing an epithet which is obligatorily coreferent with the subject of the subordinate clause. However, the controlled NP in the subordinate clause is unmarked in Telugu, and the only constraint on its form is that it must not be pronominal (ibid.: 85). Furthermore, unlike Ossetic, in Telugu full coreference of the two NPs is required (ibid.: 86–87). Therefore, based on the available data, we can say that Ossetic correlatives repre- sent a special construction involving obligatory anaphorically mediated coreference that is distinct from relativization or control. Ossetic correlates share a number of properties with resumptive pronouns, but there are also important differences which do not allow completely equating the two. The fact that Ossetic correlatives are used for such a wide range of clause types follows naturally from this analysis. Indeed, if Ossetic correlatives are only a case of obligatory pronominal coreference, it can be expected that whatever can be referred to via anaphora can also be expressed by the correlative construction, given the appropri- ate subordinator. Anaphoric reference to events and propositions is a well-known and relatively well-understood phenomenon (Asher 1993). It is also natural that “asyndetic correlatives” (29) are available for complement clauses. While anaphorically referring to an argument or adjunct requires this argument or adjunct to be explicitly spelled out in the sentence, the event argument and similar entities are present and available for reference in every finite verbal predication, making an explicit subordinator redundant. It is notable that bridging and split antecedents are also observed in correlatives that immediately precede the correlate:

(59) kadɜȝ̌ə ɜrmɜg wədəštə, [sardɜj sə konflikttɜi išta], [sə χaraktertɜj ɜvdəšta], wədoni+j (modification of 44) ‘‘What conflictsi he took from life, what charactersj he demonstrated, theyi+j were legendary material.’

(60) mɜ= χimɜ išən, [afɜzɜj afɜzmɜ =šɜ sə kɜndtɜi qɜwə], wəsə χɜrztɜj~i (modification of 48)

‘ What commemorationsi they need from year to year, I take the expensesj~i on myself.’  Oleg Belyaev

Thus, perhaps somewhat surprisingly, we need the same anaphoric semantics even for correlatives left-adjoined to their correlates. These are syntactically local and are, there- fore, not usually treated anaphorically even in analyses like that of Davison (2009).

5. Anaphoric correlatives and similar constructions beyond Ossetic

While I am claiming that Ossetic correlatives represent a special semantic type of sub- ordination that is distinct from all other clause types, I am not suggesting that they are the unique representative of this type. While detailed data on the syntax of correlatives is unfortunately lacking for the majority of languages that possess them, a number of languages of the world display such transparent similarities between correlatives and ordinary pronominal anaphora that an analysis along the lines outlined in Section 4 is the most natural initial hypothesis.

5.1 Iranian languages While no other New Iranian languages except for Ossetic use correlatives as the main strategy of clause combining or relativization, a few allow a correlative-like strategy as an alternative to canonical postnominal relative clauses: (61) [pārd ya dəraxt dər-əm tsə tu], ya̱w-ī ko̱t-əv ‘[last year (pārd) that tree that was here], they dug it up’, i.e. ‘They dug up the tree that was here last year.’ (Wakhi, Bashir 2009: 850) (62) [a ceni wext-ê cı çinêvı] nıka a do fina biro that woman time-ez her was.not now she will again come ‘The woman that didn’t have time will now come again.’ (lit. ‘That woman, she didn’t have time; she will now come again.’) (Dimili/ Zazaki, Todd 2008: 112) These relative clauses, if they can be so called, are described as such in grammars, but since there is no overt marking of subordination, it is not clear whether (61) and (62) simply represent paratactic constructions that are functionally similar to rela- tivization. If this is so, then the linking can only be anaphoric; such constructions would thus represent a first step in the grammaticalization of correlatives of the kind we observe in Ossetic.

5.2 Mande languages Correlatives where the relative phrase is marked in the same way as deixis are attested in a number of Mande languages of West . For instance, Makeeva (2013) pro- vides the following example of correlatives in Kla-Dan: Anaphora in Ossetic correlatives and the typology of clause combining 

(63) [Lè kéȅ lṵ̏ wá ya̋ ɓe̋ȅ kpȁ ] ȁlṵ̏̏ woman rel pl 3pl.jnt food good cook.jnt 3pl.nsbj lɔ̏ yȅ gɔ̰̀ kʌ̋ love 3sg.exi man do.ntr ‘Men love women who cook well.’ (Kla-Dan, Makeeva 2013: 79) The marker glossed asrel is homonymous with a deictic marker (ibid.: 78). Unlike Wakhi and Dimili/Zazaki, in Kla-Dan the subordinate status of correlative clauses is clear: they can be embedded into the main clause as adjuncts to VP (ibid.). Similar conclusions on the anaphoric status of correlatives are reached for other Mande languages in Nikitina (2012: 320) and Khachaturyan (2013: 5); in particular, Mano and Wan use ordinary demonstrative pronouns for relativization: (64) gà màŋ yā [kó yī lē é plá ] à lɛ̀ŋ go rice with 1+2du slept woman def at 3sg to ‘Take some rice to the woman at whose place we slept.’ (lit. ‘Go with rice – we slept at the woman’s place – to her.’) (Wan, Nikitina 2012: 320) (65) [ē nɔ́ɔ̄ ɓɛ̄ lɛ́ è nàà 3sg.refl child.pl dem foc 3sg.dipfv love:ipfv ō ká ā] ē nū ō sāmā ká 3pl.nsbj with term 3sg.pret come 3pl.nsbj gift with ‘He brought gifts to his children that he loves.’ (Mano, Khachaturyan 2013: 5) While the possibility of phenomena such as those described above in Ossetic remains an open question for Mande, Examples (63)–(65) clearly demonstrate that correla- tives in these languages involve the same material that is used in ordinary pronominal anaphora, and it may be hypothesized that the semantic mechanisms that are used are also the same.

5.3 Interrogatives as indefinites A final parallel to Ossetic correlatives concerns the indefinite use of interrogatives in subordinate clauses in some Indo-European languages, e.g. Russian:26 (66) [Esli b kto s sekundomerom tut if sbjv who.nom with stop_watch.m:sg.inst here stojal], to my by stand.ipfv:pst[sg.m] then we.nom sbjv

. The transliteration of Russian follows the orthography.  Oleg Belyaev

mirovoj rekord zafiksirovali. world:sg.m.acc record.m[sg.nom] observe.pfv:pst:pl ‘If someone had stood here with a watch, we would have observed a world record.’ (lit. ‘If who…’) (Russian, Zevaxina & Oskol’skaja 2013) In this example, kto ‘who’ is used in indefinite function — a meaning that is only possible in subordinate clauses in Russian (Tret’jakova 2009). Mitrenina (2012: 69) hypothesizes that correlatives in modern Russian could have appeared from such interrogatives used in indefinite function. A similar situation is observed in Hittite. Interrogatives do not normally function as indefinites, but they have this function in conditional clauses: (67) nu =kan mān AWAT NARARI kuwapi conn loc.ptcl if summons assistance when šarā išparza-zi # up come-prs.3sg ‘If at some point (lit. when) a summons for assistance comes up…’ ’ (Hittite, NH/lNS (CTH 106.A.1) Bo 86/299 rev. iii 44–46; Andrej Sideltsev, p.c.) In addition, Hittite, just as Ossetic, has a highly grammaticalized preverbal ­position. The normal order of elements is “interrogative – negative (pronoun or particle) – indefinite pronoun – verb” Goedegebuure( 2003, 2009; Sideltsev 2002; Sideltsev forthcoming). Hittite also employs correlatives as the main strategy of relativization; interrogatives are used as relative pronouns and occupy the same linear position as interrogatives in independent sentences. An intriguing fact is that there are some examples where the relative pronoun in such constructions follows the negation, i.e. behaves like an indefinite pronoun: (68) nu KUR.KUR ‹MEŠ›LÚKÚR ŪL =ya conn country enemy neg and kui-ēš kūruriyaẖẖe-šk-ir # which-nom.pl.c be_hostile-ipfv-pst.3pl ‘And even the enemy countries which did not (yet) start hostilities…’ (Hittite, NH/NS (CTH 61.II.5.B) KBo 3.4+ obv. I 8–9, Andrej Sideltsev, p.c.) This parallel with Russian and Hittite may explain what is otherwise a somewhat puzzling fact about Ossetic correlatives. Namely, all the other examples of putative anaphoric correlatives that we have observed above involve relative markers that go back to . In such cases, the connection with anaphora is transpar- ent, and the diachronic pathway from parataxis and anaphoric binding to relativiza- tion is widely accepted (Givón 2009: 97–120). Ossetic subordinators, however, are derived from interrogatives — words that one can hardly expect to act as antecedents Anaphora in Ossetic correlatives and the typology of clause combining  of pronominals.­ 27 Interrogatives may thus have become grammaticalized as a sort of obligatorily coreferenced indefinites by going through a stage like the one exemplified in (64)–(66) and the one reconstructed for Russian in Mitrenina (2012). This seems to be especially convincing if we compare (66) and (67) with the almost identical use of the non-final interrogative či ‘who’ as an indefinite in (38). Ossetic does not allow interrogatives in indefinite function outside of multiple correlatives (69), even with negation (70), but it is plausible to assume that it used to make more extensive use of such items at an earlier stage of its history.28 (69) #či rba-səd-i who.nom pv-go-pst.intr.3sg (‘Someone c am e .’ ) (OK with interrogative intonation: ‘Who came?’) (70) #sə nɜ fed-t-a what neg see.pfv-tr-pst.3sg (‘He didn’t see anything.’ ) (OK with interrogative intonation: ‘What didn’t he see?’)

6. Conclusions

In this paper, I have provided a general description of the syntax and semantics of Ossetic correlatives and have demonstrated the following typologically interesting and, in combination, cross-linguistically unusual properties of Ossetic subordination:

–– most of the semantic types of subordination are expressed by the simple correla- tive construction, with very few minor syntactic variations; –– the only non-correlative finite subordinating constructions in Ossetic are post- posed complement and purpose clauses (which also have correlative variants),

. The scenario for the transition from interrogatives to relative pronouns in Heine & Kuteva (2006) is inapplicable to Ossetic for two reasons: first, it is incompatible with the anaphoric analysis; second, it involves indirect questions as an intermediate stage, which is improbable for Ossetic due to indirect questions being syntactically different from simple correlatives. . Such an assumption is especially plausible if we consider the form of negative pronouns, all of which are derived from interrogatives using the negative prefix ni- (Pir. *nai-, Abaev 1973: 179). In Slavic, where negative pronouns are formed in exactly the same way, they are reconstructed as being derived from the combination of clausal negation with an interrogative used as an indefinite (Willis 2013: 378).  Oleg Belyaev

direct speech and two minor strategies: pseudocoordination and externally- headed canonical relativization; –– multiple correlatives differ from simple correlatives in using interrogative pro- nouns instead of specialized subordinators; –– the relation between the subordinator/relative phrase and the correlate in both single and multiple correlatives may involve split antecedence and bridging, nei- ther of which can be accounted for using the standard semantics of relativization; –– semantically, Ossetic correlatives are best described as a construction distinct from relativization: in correlatives, obligatory coreference is established between the relative phrase, acting as a sort of indefinite, and the correlate, which is an ordinary pronominal or definite description; correlates are thus a kind of “back- ward resumptives”; –– based on available data, one may conjecture that correlatives involving this kind of anaphorically mediated coreference are not unique to Ossetic but occur in a number of languages of the world.

Abbreviations

abl = ablative, acc = accusative, add = additive particle (‘also’, ‘even’), all = alla- tive, attr = attributive, c = common gender (in Hittite), comit = comitative, cnp = conjunctive participle (in Telugu), Comp = complementizer, compar = comparative, conn = clause-initial connective (in Hittite), Contr = contrastive particle, cntrf = counterfactual, dat = dative, def = definiteness, dem/Dem = demonstrative, dipfv = dependent imperfective (in Mande), Dist = distal deixis, du = dual, Encl = enclitic pronoun, equ = equative case, exi = existential series of predicative pronouns (in Mande), ez = ezafe, foc = focus, gen = genitive, fut = future, hab = habitual, imp = imperative, in = inessive-illative, indef = indefinite pronoun, inf = infinitive, inst = instrumental, intr = intransitivity, iter = iterative, jnt = joint form of pronominal- predicative markers, change of verbal in joint constructions (in Mande), loc. ptcl = locative particle (in Hittite), m = masculine, neg = negation, nom = nomina- tive, nPV = non-preverbal subordinator, nsbj = non-subject pronoun (in Mande), ntr = change of tone contour in neutral aspect construction (in Mande), O = object, opt = optative, part = participle (in Telugu), pcvb = participle-converb, pfv = perfective, pl/Pl = plural, Poss = possessive pronoun, pret = preterite, Prox = proximal deixis, pst = past tense, ptcp = participle, Purp = purpose, pv = (perfectivizing) preverb, PV = preverbal subordinator, refl/Refl = reflexive, rel = relative marker, s/S = subject, sbjv = subjunctive, sg/Sg = singular, super = superessive-superlative case, term = terminal marker (in Mande), tr = transitivity, V = verb. Anaphora in Ossetic correlatives and the typology of clause combining 

References

Abaev, Vasilij I. 1959. Grammatičeskij očerk osetinskogo jazyka (A grammatical sketch of Ossetic). Ordžonikidze: Severo-Osetinskoe knižnoe izdatel’stvo. Abaev, Vasilij I. 1973. Istoriko-ètimologičeskij slovar’ osetinskogo jazyka (A historical and etymo- logical dictionary of Ossetic), Vol. 2, L–R. Leningrad: Nauka. Andrews, Avery. 2007. Relative clauses. In Language Typology and Syntactic Description, Vol. 2: Complex Constructions, Timothy Shopen (ed.), 206–236. Cambridge: CUP. Aoun, Joseph, Choueiri, Lina & Hornstein, Norbert. 2001. Derivational economy. Linguistic Inquiry 32(3): 371–403. DOI: 10.1162/002438901750372504 Arsenijević, Boban. 2009. Clausal complementation as relativization. Lingua 119(1): 39–50. DOI: 10.1016/j.lingua.2008.08.003 Asher, Nicholas. 1993. Reference to Abstract Objects in Discourse [Studies in Linguistics and Philosophy 50]. Dordrecht: Kluwer. DOI: 10.1007/978-94-011-1715-9 Asudeh, Ash. 2012. The Logic of Pronominal Resumption [Oxford Studies in Theoretical Linguis- tics 35]. Oxford: OUP. DOI: 10.1093/acprof:oso/9780199206421.001.0001 Bach, Emmon & Partee, Barbara H. 1980. Anaphora and semantic structure. In Papers from the Parasession on Pronouns and Anaphora, Jody Kreiman & Almerindo E. Ojeda (eds), 1–28. Chicago IL: Chicago Linguistic Society. Bashir, Elena. 2009. Wakhi. In The Iranian languages [Routledge Series], Ger- not Windfuhr (ed.), 825–858. London: Routledge. Belyaev, Oleg. Forthcoming. Systematic mismatches: Coordination and subordination at three levels of grammar. To appear in Journal of Linguistics. Belyaev, Oleg & Vydrin, Arseniy. 2011. Participle-converbs in Iron Ossetic: Syntactic and seman- tic properties. In Topics in Iranian Linguistics [Beiträge zur Iranistik 34], Agnes Korn, Geof- frey Haig, Simin Karimi & Pollet Samvelian (Eds), 117–134. Wiesbaden: Reichert. Bhatt, Rajesh. 1997. Matching effects and the syntax-morphology interface. MIT Working Papers in Linguistics 31: 53–68. Bhatt, Rajesh. 2003. Locality in correlatives. Natural Language & Linguistic Theory 210: 485–541. DOI: 10.1023/A:1024192606485 Bos, Johan, Buitelaar, Paul & Mineur, Anne-Marie. 1995. Bridging as coercive accommodation. Paper presented at CLNLP Workshop, Edinburgh, April 3–5, 1995. 〈http://arxiv.org/abs/ cmp-lg/9508001〉 (16 January 2014). Büring, Daniel. 2004. Binding Theory[Cambridge Textbooks in Linguistics]. Cambridge: CUP. Caponigro, Ivano & Polinsky, Maria. 2011. Relative embeddings: A Circassian puzzle for the syntax/semantics interface. Natural Language and Linguistic Theory29(1): 71–122. DOI: 10.1007/s11049-011-9121-9 Clark, Herbert H. 1975. Bridging. In Proceedings of the 1975 Workshop on Theoretical Issues in Natural Language Processing, Roger Schank & Bonnie L. Nash-Webber (eds), 169–174. Stroudsburg PA: Association for Computational Linguistics. DOI: 10.3115/980190.980237 Culicover, Peter & Jackendoff, Ray. 1997. Semantic subordination despite syntactic coordina- tion. Linguistic Inquiry 28(2): 195–217. Davison, Alice. 2009. Adjunction, features and locality in Sanskrit and Hindi/Urdu correlatives. In Correlatives Cross-linguistically [Language Faculty and Beyond 1], Anikó Lipták (ed.), 223–262. Amsterdam: John Benjamins.  Oleg Belyaev

Dayal, Veneeta. 1995. Quantification in correlatives. In Quantification in Natural Languages, Vol. 2, Emmon W. Bach, Eloise Jelinek, Angelika Kratzer & Barbara H. Partee (eds), 179–205. Dordrecht: Kluwer. Diessel, Holger & Hetterle, Katja. 2011. Causal clauses: A cross-linguistic investigation of their structure, meaning, and use. In Linguistic Universals and Language Variation [Trends in Linguistics: Studies and Monographs 231], Peter Siemund (ed.), 21–52. Berlin: Mouton de Gruyter. Dzaxova, Veronika T. 2009. Fonetičeskie xarakteristiki fonologičeskoj sistemy sovremennogo ose- tinskogo (ironskogo) literaturnogo jazyka (The phonetic characteristics of the phonological system of contemporary Iron Ossetic). Vladikavkaz: Izdatel’stvo SOGPI. Erschler, David. 2012. From preverbal focus to preverbal “left periphery”: The Ossetic clause architecture in areal and diachronic perspective. Lingua 122(6): 673–699. DOI: 10.1016/j. lingua.2012.01.009 Gagkaev, Konstantin U. 1956. Sintaksis osetinskogo jazyka (The syntax of Ossetic). Ordžonikidze: Severo-Osetinskoe knižnoe izdatel’stvo. Gareyshina, Anastasia, Grashchenkov, Pavel, Lyutikova, Ekaterina & Tatevosov, Sergei. 2011. Sentential proforms and complementation in Ossetian. In 33. Jahrestagung der Deutschen Gesellschaft für Sprachwissenschaft. Text: Strukturen und Verarbeitung. 〈http://www.zas. gwz-berlin.de/fileadmin/veranstaltung_zas/workshops/archiv/proforms/gareyshina_osse- tian.pdf〉 (3 January 2014). Geach, Peter. 1962. Reference and Generality [Contemporary Philosophy]. Ithaca NY: Cornell University Press. Giorgi, Alessandra. 1984. Toward a theory of long distance anaphors: A GB approach. The Lin- guistic Review 3: 307–361. DOI: 10.1515/tlir.1984.3.4.307 Givón, T. 2009. The Genesis of Syntactic Complexity: Diachrony, Ontogeny, Neuro-cognition, Evo- lution. Amsterdam: John Benjamins. Gerasimov, Dmitry V. & Lander, Yury A. 2008. Reljativizacija pod maskoj nominalizacii i fak- tivnyj argument v adygejskom jazyke (Relativization under the guise of nominalization and the factive argument in Adyghe). In Issledovanija po glagol’noj derivacii, Vladimir A. Plung- ian & Sergei G. Tatevosov (eds), 290–313. Moscow: Jazyki slavjanskoj kul’tury. Goedegebuure, Petra. 2003. Reference, Deixis and Focus in Hittite: The Demonstratives ka- “this”, apa- “that” and asi- “yon”. Ph.D. dissertation, University of Amsterdam. Goedegebuure, Petra. 2009. Focus structure and Q-word questions in Hittite. Linguistics 47(4): 945–969. DOI: 10.1515/LING.2009.033 Grosu, Alexander & Landman, Fred. 1998. Strange relatives of the third kind. Natural Language Semantics 6(2): 125–170. DOI: 10.1023/A:1008268401837 Haddad, Youssef A. 2009. Copy control in Telugu. Journal of Linguistics 45(1): 69–109. DOI: 10.1017/S0022226708005525 Heim, Irene R. 1982. The Semantics of Definite and Indefinite Noun Phrases. Ph.D. dissertation, University of Massachusetts, Amherst. Heine, Bernd & Kuteva, Tania. 2006. The Changing .Oxford: OUP. DOI: 10.1093/acprof:oso/9780199297337.001.0001 Kamp, Hans & Reyle, Uwe. 1993. From Discourse to Logic [Studies in Linguistics and Philosophy 42]. Dordrecht: Kluwer. Kayne, Richard. 2010. Why isn’t this a complementizer? In Comparisons and Contrasts [Oxford Studies in Comparative Syntax], 190–227. Oxford: OUP. Anaphora in Ossetic correlatives and the typology of clause combining 

Khachaturyan, Maria. 2013. Between correlative clauses, pronoun retention strategy, paratac- tic construction and cleft: Relative clauses in Mano. Presentation at 10thBiennial Confer- ence of the Association for Linguistic Typology (University of Leipzig, August 15–18, 2013). 〈https://www.academia.edu/5287182/Between_correlatives_internally_headed_clauses_ and_cleft_relative_clauses_in_Mano〉 (5 February 2014) Kulaev, N.X. 1961. K voprosu o probleme padežej v osetinskom (Towards the problem of case in Ossetic). In Voprosy sostavlenija opisatel’nyx grammatik, S.G. Barxudarov, N.A. Baskakov, A.A. Reformatskij (eds), 245–252. Moscow: Izdatel’stvo AN SSSR. Lipták, Anikó. 2009. The landscape of correlatives. InCorrelatives Cross-linguistically [Language Faculty and Beyond 1], Anikó Lipták (ed.), 1–48. Amsterdam: John Benjamins. Lyutikova, Ekaterina & Tatevosov, Sergei. 2009. The clause internal left edge: Exploring the pre- verbal position in Ossetian. Presentation at Third International Conference on Iranian Lin- guistics. University of Paris III: Sorbonne Nouvelle, September 11–13. 〈http://otipl.philol. msu.ru/staff/people/tatevosov/clause_internal_left_edge_4.pdf〉 (4 February 2014). Lyutikova, Ekaterina & Tatevosov, Sergei. To appear. Nominalization and the problem of indi- rect access: Evidence from Ossetian. The Linguistic Review. 〈http://otipl.philol.msu.ru/ staff/people/tatevosov/lyutikova_tatevosov_revised.pdf〉 (4 February 2014). Makeeva, Nadežda V. 2013. Kommunikativnye strategii i korreljativnaja konstrukcija v jazyke kla-dan i drugix jǔznyx mande (Discourse strategies and correlatives in Kla-Dan and other Mande languages). Voprosy jazykoznanija 1: 77–94. McCawley, James D. 2004. Remarks on adsentential, adnominal, and extraposed relative clauses in Hindi. In Clause Structure in South Asian Languages [Studies in Natural Language and Linguistic Theory], Veneeta Dayal & Anoop Mahajan (eds), 291–312. Dordrecht:­ Kluwer. DOI: 10.1007/978-1-4020-2719-2_10 McCloskey, James. 2002. Resumption, successive cyclicity, and the locality of operations. In Derivation and Explanation in the Minimalist Program, Samuel David Epstein & T. Daniel Seeley (eds), 184–226. Malden MA: Wiley-Blackwell. DOI: 10.1002/9780470755662.ch9 Mitrenina, Olga V. 2012. The syntax of pseudo-correlative constructions with the pronoun ­Kotoryj (‘Which’) in Middle Russian. Slov̌ene 1: 61–73. Nikitina, Tatiana. 2012. Clause-internal correlatives in Southeastern Mande: A case for the propagation of typological rara. Lingua 122: 319–334. DOI: 10.1016/j.lingua.2011.12.001 Nikunlassi, Ahti Olavi. 2008. Primestoimenno-otnositel’nye konstrukcii v sovremennom russkom jazyke (Pronoun-adjoined relative clauses in contemporary Russian). Helsinki: Helsinki University Press. Partee, Barbara H. 1975. Montague Grammar and transformational grammar. Linguistic Inquiry 6: 203–300. Perlmutter, David M. & Ross, John R. 1970. Relative clauses with split antecedents. Linguistic Inquiry 1: 350. Prince, Ellen. 1990. Syntax and discourse: A look at resumptive pronouns. In Proceedings of the Sixteenth Annual Meeting of the Berkeley Linguistics Society, Kira Hall, Jean-Pierre Koenig, Michael Meacham, Sondra Reinman & Laurel Sutton (eds), 482–497. Berkeley CA: Berke- ley Linguistics Society. Reinhart, Tanya. 1999. Binding theory. In The MIT Encyclopedia of the Cognitive Sciences, ­Robert A. Wilson & Frank C. Keil (eds), 86–88. Cambridge MA: The MIT Press. Reuland, Eric. 2001. Primitives of binding. Linguistic Inquiry 32(3): 439–492. DOI: 10.1162 /002438901750372522  Oleg Belyaev

Sag, Ivan A., Gazdar, Gerald, Wasow, Thomas & Weisler, Steven.1985. Coordination and how to distinguish categories. Natural Language and Linguistic Theory3: 117–171. DOI: 10.1007/ BF00133839 Serdobolskaya, Natalia. To appear. Semantics of complementation in Ossetic. In Semantic Func- tions of Complementizers in European Languages [Empirical Approaches to Language Typology], Petar Kehayov & Kasper Boye (eds). Berlin: Mouton de Gruyter. Sideltsev, Andrej. 2002. Inverted word order in Middle Hittite. In Anatolian Languages [Associa- tion for the History of Language Studies in the Science & History of Language 6], Vitaly V. Shevoroshkin & Paul J. Sidwell (eds), 137–188. Canberra: Association for the History of Language. Sideltsev, Andrej V. Forthcoming. Clause internal and clause leftmost verbs in Hittite. To appear in Altorientalische Forschungen. Sjögren, Andreas J. (Andrej Mixajlovič Šëgren) 1844. Osetinskaja grammatika s kratkim slova- rem osetinsko-rossijskim i rossijsko-osetinskim (Ossetic grammar, with a short Ossetic-­ Russian and Russian-Ossetic dictionary), Vol. 1: Osetinskaja grammatika (Ossetic gram- mar). St.Petersburg: Tipografija Imperatorskoj Akademii Nauk. Srivastav, Veneeta. 1991. The syntax and semantics of correlatives.Natural Language and Lin- guistic Theory 9: 637–686. DOI: 10.1007/BF00134752 Stiebels, Barbara. 2007. Towards a typology of complement control. In Studies in Complement Control [ZAS Papers in Linguistics 47], Barbara Stiebels (ed.), 1–80. Berlin: ZAS. Taghvaipour, Mehran A. 2004. An HPSG Analysis of Persian Relative Clauses. In Proceedings of the 11th International Conference on Head-Driven Phrase Structure Grammar, Stefan ­Müller (ed.). Stanford CA: CSLI. 〈http://www.stanford.edu/group/cslipublications/csli- publications/HPSG/2004/taghvaipour.pdf〉 (5 February 2014). Todd, Terry Lynn 2008. A grammar of Dimili, also known as Zaza. Electronic publication: http:// www.zazaki.de/english/T.L.Todd-AGrammarofDimli.pdf. Accessed 19 January 2014. Tret’jakova, Ol’ga D. 2009. Funkcii neopredelennyx mestoimenij bez pokazatelja neopredel- ennosti (na materiale slavjanskix jazykov) (The functions of indefinite pronouns with- out indefinite marking in Slavic languages).Vestnik Moskovskogo universiteta. Serija 9. Filologija 1: 134–147. Vendler, Zeno. 1967. Linguistics in Philosophy. Ithaca CA: Cornell University Press. Whitehead, Alfred North & Russell, Bertrand. 1925–1927. Principia Mathematica, 3 Vols, 2nd edn. Cambridge: CUP. Willis, David. Negation in the history of the Slavonic languages. In The History of Negation in the Languages of Europe and the Mediterranean, Vol. 1: Case Studies, David Willis, Christopher Lucas & Anne Breitbarth (eds), 341–398. Oxford: OUP. Yuasa, Etsuyo & Sadock, Jerry M. 2002. Pseudo-subordination: A mismatch between syntax and semantics. Journal of Linguistics 38(1): 87–111. DOI: 10.1017/S0022226701001256 Zevakhina, Natalia A. & Oskolskaya, Sof’ja A. 2013. Neopredelënnye mestoimenija bez markera neopredelënnosti v russkom jazyke (Indefinite pronouns without indefiniteness marking in Russian). Presentation at Tret’ja konferencija «Russkij jazyk: konstrukcionnye i leksiko- semantičeskie podxody» (12 September – 14 September 2013), St. Petersburg, Institute of Linguistic Studies of the Russian Academy of Sciences.