1 Introduction

This paper investigates contact-induced change in the grammar of relative clauses (RCs) in Khanty, an endangered Ob-Ugric (Finno-Ugric, Uralic) language spoken along the river Ob and its tributaries in Western Siberia. Khanty is best characterized as a dialect continuum with three main varieties: Northern, Eastern, and the by now extinct Southern Khanty. There are significant phonological, morphological and lexical differences between these dialects making mutual intelligibility difficult (often impossible) between Northern and Eastern Khanty (Schmidt 2006 [1973]: 28–36). ^[1] Here we focus on Surgut Khanty, one of the two major Eastern dialects. Surgut Khanty has approximately 2,800 speakers (Csepregi and Onina 2011) and is mutually unintelligible with Vakh-Vasyugan Khanty, the other major Eastern dialect (Schön 2017: 12).

Khanty is a highly agglutinative language with SOV word order. ^[2] Traditionally, Khanty employs just one finite verb per clause, and makes widespread use of non-finite subordination; cf. Nikolaeva (1999: 45–46) and Schmidt (2006 [1973]: 69) on Northern Khanty and Filchenko (2007: 435) on Eastern Khanty. The first Khanty texts were collected from the Northern dialects in 1844; the first texts of the Surgut dialect were collected in 1901 and were published as Paasonen and Vértes (2001). In these texts relative clauses with an external head are categorically pre-nominal and participial in line with the strong preference for one finite verb per clause. These RCs employ the gap strategy: they do not contain an internal head or a relative pronoun (and have no complementizer either). ^[3] This pattern is likely to be the original way of expressing RCs in Uralic (Nikolaeva to appear).

These RCs are in many ways similar to the pre-nominal participial noun-modifiers in the well-known Indo-European languages, e. g. English the slowly falling/fallen leaves.

Correlative clauses (sometimes also called co-relatives) are a type of RC with an internal head. They occur on the left periphery of the main clause and are linked to the main clause via a noun (phrase), the correlate, which must contain or correspond to a demonstrative. ^[5] The correlative clause and the nominal correlate pick out the same referent and occupy the same argument slot (Lipták 2009: 2). Correlative clauses appear already in the first recorded Khanty texts. They employ a finite verb and a relative pronoun which is form-identical to the corresponding interrogative pronoun. We shall refer to such a relative pronoun as an ‘interrogative-based relative pronoun’.

It is generally assumed that Proto-Uralic had very little finite embedding, and no complementizers or other left-peripheral sentence connectors (such as relative pronouns) at all (Hajdú 1966: 82; Bereczki 1996: 94, among others). If this is so, then correlatives represent an innovation in Khanty. They likely developed under the strong influence of Russian as a more prestigious contact language (see also Potanina 2008: 78). The fact that correlatives represent a relatively rare construction in late-nineteenth century Khanty fits well with this picture.

Since they are already present in the first texts, we cannot tell exactly when correlatives first appeared in Khanty. What matters for us, however, is that in the earliest Surgut Khanty texts there are only two types of RCs. Externally headed RCs are participial and pre-nominal while correlatives are finite and have a relative pronoun. Crucially, there are no externally headed RCs which are finite or which feature a relative pronoun. This state of affairs was probably stable until recently. In his description of Vakh Khanty, another Eastern dialect, Gulya (1966) mentions the existence of relative pronouns, but all of his examples are correlatives. ^[6]

In the recent past, however, under the growing influence of Russian, new types of RCs have become possible. Based on work with a native speaker consultant, Csepregi (2012) reports three new types of externally headed RCs in Surgut Khanty. The first is the post-nominal participial RC with no relative pronoun or other sentence connector, as in (3):

Compared to the externally headed RC inherited from Proto-Uralic (see [1]), (3) reverses the word order of the modifier and the head but introduces no additional changes. The second new type, as in (4), is the post-nominal participial RC introduced by what Csepregi calls the ‘proto-relative pronoun’ ťu (form-identical with the distal demonstrative ‘that’):

Finally, the third type, shown in (5), is the finite post-nominal RC featuring the interrogative-based relative pronouns which also occur in correlatives:

The focus of Csepregi’s study was the pre-nominal participal RC; therefore, several questions regarding the new RC types in (3) through (5) necessarily remained open. We list some of these here:

1. Pre-nominal RCs: Do they also admit the new relative pronouns or ťu?

2. Post-nominal non-finite RCs: Do they show structural changes with respect to pre-nominal RCs, or does the change only affect the order of the noun and the RC? Do they admit relative pronouns?

3. ťu in post-nominal RCs: How similar is it to the relative pronouns used in correlatives and externally headed finite RCs? Can it appear in finite RCs?

4. Relative pronouns: Do all interrogative pronouns have a use as a relative pronoun? Can the relative pronouns of finite RCs also appear in non-finite post-nominal RCs? Can anything precede the relative pronoun within the RC? Does the type of the external head (lexical noun or pronoun) influence the choice of relative pronoun?

The first aim of this paper is to answer these empirical questions. ^[7]

Our second aim is to determine to what extent language contact with Russian is responsible for the rise of the new RCs and, relatedly, what types of contact-induced change are attested in the domain of Khanty RCs. Khanty and Russian have a long history of language contact: according to Potanina and Filchenko (2016: 27), Russian has been the dominant language in the Eastern Khanty region for over a period of “at least 150 years, and markedly so within the recent 50–60 years”. ^[8] Most Khanty speakers living today went to boarding school, which sped up their assimilation to Russian language and culture. As a result, today wide-spread diglossia (ca. 100% unidirectional bilingualism) characterizes Khanty-speaking communities. The younger generations are either balanced bilinguals or their dominant language is Russian. Children learn Khanty only if their parents have traditional jobs such as fishing (Csepregi and Onina 2011). In this situation, Russian (SVO) is exerting a strong influence on both the lexicon and the syntax of the language. Here we seek to identify the depth of this influence in the realm of RCs.

In order to gain insight into the new RCs exemplified in (3)–(5), we worked closely with two speakers. They are both Khanty-Russian bilinguals but their language acquisition before school was monolingual (Khanty). Our primary consultant was a fluent speaker of the Yugan variety. ^[9] She was born in 1966 and has worked as a teacher, a journalist and a collector of Khanty texts in the field. She provided grammaticality judgments on an initial written questionnaire in the autumn of 2017 in Nefteyugansk (Siberia). Additionally, we worked with her in Budapest over a two-month period between January and March 2018. During this period she explained and clarified the judgments in the questionnaire in detail and provided grammaticality judgments on further sentences. Additionally, she performed a picture prompt based spontaneous sentence production task as well as several directed sentence production tasks. The latter involved arranging Khanty words printed to flashcards into the most neutral word order. Data provided by her are marked as (Yg.). Our other informant was a fluent speaker of the Tromagan variety. She was born in 1949 and uses Khanty in (part of) the family. She provided grammaticality judgments on a subset of the issues investigated here during our fieldwork in Kogalym (Siberia) in June 2017. Data collected from her are marked as (Tra.). ^[10] During the grammaticality judgment tasks both consultants were provided with Khanty sentences; they were not asked to translate Russian sentences into Khanty. Both speakers were tested on the target constructions on multiple different occasions. The interviews were conducted in Russian by Katalin Gugán. ^[11]

The paper is structured as follows. The discussion begins in Section 2 with post-nominal participial RCs without a relative pronoun or other sentence connector. In Section 3, we turn to post-nominal participial RCs introduced by ťu. Finite RCs whose external head is a lexical noun will be the topic of Section 4 while in Section 5 we zoom in on finite RCs with a pronominal head and on correlatives. Section 6 concludes our discussion.

2 Post-nominal participles without a connecting element

As already mentioned above, the original externally headed RCs in Khanty are participial, pre-nominal, and employ the gap strategy (i. e. they have no relative pronoun). Example (6) is illustrative:

More recently, however, participles can also occur post-nominally. This requires an intonational break both before and after the participle:

The contact language Russian allows both the participle-N and the N-participle order, in relatively free variation:

With (7) as a new possibility, Khanty has thus taken over the flexibility of participle placement with respect to the head noun from Russian.

Although the Khanty and Russian patterns in (7) and (8b) exhibit the same head-modifier order, they also differ in three important respects. Firstly, while N-participle is a neutral word order in Russian, in Khanty this order is clearly marked and less preferred than participle-N. Secondly, Russian participles in both pre-nominal and post-nominal position exhibit gender, number and case concord with the head noun. This is shown in (9) for post-nominal participles:

In Khanty, on the other hand, post-nominal participles remain uninflected (just like their pre-nominal counterparts, cf. [6] above):

The morphological dependency between the noun and the post-nominal participle in Russian is thus not replicated in Khanty (even though, as we will see below, adjectives and numerals in post-nominal position do bear agreement). At the same time, this situation yields a parallel between the two languages on a more abstract level: in both cases, pre-nominal and post-nominal participles differ from each other only in their placement with respect to the head noun, without any other observable differences.

The lack of number and case on Khanty post-nominal participles stands in an interesting contrast with data from Hungarian, a close relative of Khanty. Similarly to Khanty, participial RCs in Hungarian are pre-nominal by default and show no number or case concord with the head noun: ^[12]

Participles can also appear post-nominally, between two intonational breaks, as in Khanty. In this case, however, the number and case marking of the head noun must appear on the participle: ^[13]

It is generally agreed that number and case marking on Hungarian post-nominal participles is obligatory because in these examples the participle, in fact, stands in a pre-nominal position with respect to an elided head noun. The number and case inflection belong morphosyntactically to the noun, but after N-ellipsis they must attach to the participle for phonological support (Dékány 2011; Lipták and Saab 2016, among others). The structure of the relevant part of (12) is thus (13):

We suggest that the contrast between (10) and (12) shows that in Khanty there is no elided noun after the participle; instead, genuine post-nominal placement of participles (head-modifier order) is becoming possible.

Khanty post-nominal participles also have different distributional properties from post-nominal numerals and adjectives. Numerals and adjectives, like all Khanty noun modifiers, are pre-nominal by default (and show no concord with the head):

They can appear post-nominally, enclosed by intonation breaks, but crucially, in this case they must have the same number and case marking as the head noun:

Lack of number and case concord in post-nominal position produces ungrammaticality:

The post-nominal numerals and adjectives in (15) and (16) thus find close counterparts in Hungarian post-nominal N-modifiers (participles, adjectives and numerals), and they likely also involve an elliptical structure similar to (13). Khanty post-nominal participles, on the other hand, involve no elliptical noun; following the Russian pattern, Khanty is at an early stage of developing a new head-modifier order for participial RCs. ^[14]

The third difference between post-nominal participles in Russian and Khanty concerns participle-internal word order. Khanty participles in pre-nominal position are strictly head-final. If the agent is expressed, it is marked by locative case and is preferred to be the first constituent within the participle, as in (6). These properties also characterize post-nominal participles; see (7) and (10). In Russian, on the other hand, the head-final order is dispreferred for post-nominal participles; the agent (bearing instrumental case) follows the participial verb, as in (9) (Irina Burukina, p.c.). That is, the order of the head noun and the RC can now follow the Russian model, but the word order within the participle does not change.

There are no other structural changes to Khanty post-nominal participles either. When in a contact situation the recipient language has predominantly non-finite subordination while the model language employs wide-spread finite subordination and left-peripheral sentence connectors, the result may be that the non-finite clauses of the recipient language start admitting sentence connectors (complementizers, relative pronouns, etc.) while at the same time keeping the non-finite verbal form. An example of this is seen in Dolgan (Siberian Turkic) purpose clauses (17). The purposive relation in Dolgan is expressed by the future participle bearing possessive accusative case (cross-referencing the subject of the non-finite clause). As a result of language contact, however, purposive participles now admit the complementizer štobï ‘in order to’ borrowed from Russian (Stapert 2013: Ch. 8.3). Importantly, in Russian štobï ‘in order to’ occurs in finite embedded clauses and infinitives but not in participles.

As pointed out in Section 1, Khanty, based on the Russian model, employs relative pronouns (form-identical to interrogative pronouns) in correlatives and more recently also in post-nominal finite RCs. These relative pronouns, however, are ungrammatical in post-nominal participles (18) as well as in pre-nominal participles (19): ^[15]

Sentences like (18) have grammatical alternatives that involve a pre-nominal participle (without a relative pronoun, cf. [20a]) or a post-nominal finite RC (with the relative pronoun retained), as in (20b):

Khanty participles are thus not undergoing the type of structural change that Dolgan purposive participles did: they do not admit a sentence connector that is typical of finite clauses.

To summarize, Khanty participles can be placed in post-nominal position, but this remains a marked word order. At the same time, the relevant examples are not just “Russian sentences spoken with Khanty words”: neither the morphological dependency between N and the post-nominal participle, nor the participle-internal word order is copied from Russian. There are no other structural changes to the participle either. The only parameter that is affected by the change is the order of N and the participle.

3 Post-nominal participles with ťu

In Section 2 we saw that interrogative-based relative pronouns cannot appear in post-nominal participial RCs. Csepregi (2012), however, reports one example in which such a participle is introduced by ťu, a pronoun form-identical to the distal demonstrative ‘that’. The relevant example, given in (4), is repeated in (21):

Csepregi calls ťu a ‘proto-relative pronoun’, but it remains unclear exactly what this means. In this section we aim to determine how to best characterize ťu in post-nominal participial RCs. After providing a background to Khanty demonstratives in general and to the use of ťu in particular, we will discuss five logically possible structures for (21) and distil the underlying structure.

Let us begin with a brief description of how ťu fits into the system of Khanty demonstratives. Khanty makes a formal distinction between demonstratives that modify a noun (adnominal demonstratives) and demonstratives that stand in for a whole noun phrase (pro-nominal demonstratives); the latter are morphologically more complex than the former. Within both the adnominal and the pronominal series, demonstratives show an opposition between proximal vs. distal as well as between the referent being present or visible vs. not being present or visible in the context. This is summarized in Tables 1 and 2.

As shown in Table 1, ťu is an adnominal distal demonstrative which is used when the referent of the noun phrase is not present or visible in the context. While pronominal demonstratives inflect for the appropriate number and case, adnominal demonstratives remain uninflected. This means that in contrast to its pronominal counterpart ťut, as in (22b), ťu is invariant in form, as in (22a):

In addition to its function as a distal demonstrative, ťu is also used as an emphatic discourse particle, meaning ‘alas, behold, lo, then’ (Csepregi 2001 [1998]: 23). This is illustrated in (23):

With this background in place, let us now turn to the analysis of (21). Two structures can be excluded immediately. Firstly, ťu cannot form a constituent with the nominal that follows it, as personal pronouns cannot be modified by demonstratives. Thus the structure in (24), with ťu being an adnominal modifier of łüw, can be safely put aside:

Secondly, it also cannot be the case that in (21) ťu is the pronominal head of the participial RC and also an appositive modifier of ‘old man’ (‘old man, that, whose daughter I am thinking of’). We have seen that ťu is strictly an adnominal demonstrative. As it has no pronominal use, it cannot be the pronominal head of a participial RC. Thus (25), too, is excluded as a plausible analysis:

Having excluded (24) and (25) as possible parses of (21), three possibilities remain that require closer scrutiny. The first is that in (21) ťu is an adnominal demonstrative modifier of ‘old man’, exceptionally occurring in post-nominal position. In this case ťu is string-adjacent to the participle but is structually not part of it, as in (26):

The second possibility is that in (21) ťu is a discourse particle, as in (23), rather than a demonstrative. ^[16] Finally, it may be the case that ťu in (21) is structurally internal to the participle, functioning as a grammaticalized connective element (relative particle or pronoun), as in (27):

Before we investigate these possibilities in detail, it is worth asking whether it is plausible at all that in addition to its interrogative-based relative pronouns (used in correlatives and externally headed finite RCs), Khanty would also grammaticalize a demonstrative into a relative pronoun. The question is all the more relevant because Russian does not offer a model for this: all Russian relative pronouns are interrogative-based.

Forest Enets data show that, unexpected (or even unlikely) as it may be, this scenario can indeed materialize. Forest Enets is a moribund Northern Samoyedic language of Western Siberia. As a result of massive Russian-Enets bilingualism, relative pronouns have appeared in the language. But while the relative pronouns of correlatives are form-identical to interrogative pronouns (in line with the Russian pattern, cf. [28a]), Khanina and Shluinsky (2008) report that the relative pronoun of post-nominal finite RCs is form-identical to the Forest Enets demonstrative čiki ‘this’, as in (28b–c). ^[17] A language thus may use relative pronouns with different origins at the same time.

With this in mind, let us now return to Khanty and the analysis of ťu in (21). Several considerations suggest that ťu is not a grammaticalized relative pronoun in this context, and so (27) is not the underlying structure. Firstly, relative pronouns are expected to be inflected for number (in agreement with the number of the external head), and they are expected to occur with the case or postposition that is appropriate for the gap-site in the RC. One might argue that since ťu is not inflectable for number or case as a demonstrative (cf. [22]), it is not reasonable to expect that it would be inflectable for these categories as a relative pronoun either. Even if this is granted, however, a relative pronoun should be able to occur as a complement of a postposition. As shown in (29), this is not the case for ťu: ^[18]

In the grammatical version of (29) the personal pronoun łin ‘they(du)’ appears between ťu and the postposition. Łin serves as the complement of ‘between’, and ťu is interpreted as a discourse particle. (On this use of ťu cf. also [23].)

Secondly, if ťu had a relative pronoun use, then we could reasonably expect it to also occur in post-nominal finite RCs (as these RCs do admit interrogative-based relative pronouns, see Section 4). This expectation is not borne out, however: a finite RC introduced by ťu is ungrammatical:

(31) can be improved into an acceptable sentence by either changing the finite verb to a participle or by replacing ťu with the interrogative-based relative pronoun mǝtapi ‘which’.

Relatedly, if ťu was a grammaticalized relative pronoun in post-nominal participles, then we would expect that these participles can also admit interrogative-based relative pronouns. However, as pointed out in connection with (18), this is not possible. We are not aware of any language in which externally headed finite and non-finite RCs feature different types of relative pronouns (interrogative- vs. demonstrative-based), ^[19] therefore we take the ungrammaticality of (31) as evidence that ťu has no relative pronoun use. ^[20]

Thirdly, when a post-nominal participle with ťu is paraphrased with a pre-nominal participle, ťu is retained pre-nominally, as in (32b):

If ťu was a relative pronoun in post-nominal participles, then we would expect it to disappear from pre-nominal paraphrases as pre-nominal participles cross-linguistically very strongly resist relative pronouns (see fn. 15). The fact that ťu is retained in (32b) (and is interpreted as an adnominal demonstrative modifying the head) shows that it is not a relative pronoun in (32a).

We conclude from the discussion above that ťu appearing between the head noun and a post-nominal participle is not a relative pronoun (not even a proto-relative pronoun). This leaves us with two analytical possibilities: ťu in this position is either a discourse particle or an adnominal modifier of the head noun exceptionally standing in post-nominal position.

In certain examples the discourse particle analysis is surely on the right track. In (33), for instance, the head noun has a (pre-nominal) proximal demonstrative modifier (ťi). Therefore it could not have a post-nominal demonstrative modifier as well, especially not one that is distal and so yields a semantic clash with ťi.

In (34) the head noun is both preceded and followed by ťu. We are not dealing with two demonstrative tokens here, either: discussion of (34) with our informant reveals that the first ťu is interpreted as a demonstrative, while the second is a discourse particle.

We have not found any cases in which post-nominal ťu is interpreted as an adnominal demonstrative of the head, and there is no evidence that any other adnominal demonstrative could exceptionally be post-nominal either. In (35) ťi appears between the head noun and the post-nominal participle, but according to our consultant, it is interpreted as an emphatic particle ‘just now, behold’ rather than as a demonstrative modifier of ‘food’. (See also Csepregi 2001 [1998]: 23 on the use of ťi as an emphatic particle.)

In (36) the demonstratives tom ‘that’and tem ‘this’ (both used when the referent is present or visible in the context) find themselves between the head and the post-nominal participle. While these orders are grammatical, the demonstratives crucially modify the agent of the participle rather than the head noun. It is thus not possible for any adnominal demonstrative to appear post-nominally.

We conclude that a ťu that appears to introduce a post-nominal participle is neither a relative pronoun nor an adnominal demonstrative in exceptional post-nominal position. This use involves the discourse particle ťu. (21) and other examples like it are actually post-nominal participial RCs without a connecting element; that is, they instantiate the type discussed in Section 2. ^[21]

4 Post-nominal finite RCs with a lexical head

As mentioned in Csepregi (2012), post-nominal finite RCs have also started to appear in Surgut Khanty. In this section we look at post-nominal finite RCs with a lexical noun in the head position. (For brevity’s sake, we shall call them ‘lexically headed finite RCs’.) Other types of finite RCs will be the topic of Section 5.

5 Other finite RCs and relative pronouns

In Section 4.2 we saw that there is speaker variation in the use of interrogative-based relative pronouns in lexically headed finite RCs. While the older consultant rejects relative pronouns in this environment (42), the relatively younger one requires them (41). But even in the grammar of the latter speaker, qŏjaγi ‘who’ and müwǝłi ‘what’ have no relative pronoun use in these RCs; the head noun in the relevant examples is referred back to with mǝtapi ‘which’. This is repeated here for the reader’s convenience:

Lexically headed finite RCs, however, appear to be more conservative in their use of relative pronouns than other types of finite RCs.

6 Conclusion

This paper investigated the grammar of three new types of RCs in Surgut Khanty: the participial post-nominal RC with no connecting element, the participial post-nominal RC preceded by ťu and the post-nominal finite RC. We have uncovered the following answers to the empirical questions posed in Section 1:

1. Pre-nominal RCs have not changed; they do not admit interrogative-based relative pronouns (19). They can be preceded by ťu, but ťu is then an adnominal demonstrative modifier of the head rather than part of the participle (32).

2. Post-nominal non-finite RCs do not exhibit any structural changes with respect to pre-nominal RCs (10) and do not admit relative pronouns (18). The change only affects the order of the noun and the RC.

3. The ťu preceding post-nominal non-finite RCs shares no characteristics with interrogative-based relative pronouns (29–34). Rather, it is a discourse particle (and so it cannot act as a relative pronoun in finite RCs [31]).

4. Relative pronouns are syncretic with interrogative pronouns. They only appear in finite RCs (18), and in lexically headed RCs they can only very marginally be preceded by other RC-internal material (54). All interrogative pronouns have a relative pronoun use in correlatives, free relatives and light-headed relatives, but not all of them can be used in lexically headed RCs (44–45).

The new RCs emerged in Surgut Khanty as a result of language contact with Russian. While the changes affecting RCs run deep, there are also clear limitations on grammatical convergence with the model language. These are most clearly observable in post-nominal participial RCs. While participles can now be placed in post-nominal position following the Russian pattern, neither the morphological dependency between the noun and the participle, nor the internal structure of the participle are taken over from Russian. This situation is best characterized as adaptation of an existing structure towards the Russian model.

In the case of externally headed finite RCs, on the other hand, we can observe wholesale borrowing of the Russian structure. This includes: (i) order with respect to the head noun, (ii) finite verb inflection, (iii) syncretism between interrogative and relative pronouns, (iv) high preference for the relative pronouns to be the first element in the RC, and (v) the exclusion of certain relative pronouns from lexically headed finite RCs but not from light-headed finite RCs, free relatives and correlatives. The only limitation here is that there is no borrowing of phonetic material: Khanty replicated the interrogative-relative syncretism of Russian with its own lexical resources. As relative pronouns first appeared in correlatives and were extended to externally headed RCs only later, we are faced here with an instance of replica grammaticalization rather than polysemy copying.

The fact that finite externally headed RCs were taken over from Russian “as is” while post-nominal participles still differ from their Russian counterparts in many respects, shows that changing already existing structures (in this case participles) is more difficult or slow in Khanty than adopting a completely new structure.