Possessive pronouns in Russian-German language contact: variation or change?

: The article addresses the question of what happens to internal linguistic variability in a language contact situation by looking at how ﬁ rst-generation Russian immigrants in Germany use two variants of possessive pronouns (re ﬂ exive and non-re ﬂ exive) for referencing the subject. To what extent does the frequency of the variants, as well as linguistic constraints on their use, remain the same as in the monolingual variety, or change to converge with the structures of the majority language? A forced choice sentence completion task was conducted with 96 adult monolingual Russian and 96 adult bilingual Russian-German speakers to measure the frequency of both variants and the e ﬀ ect of animacy and referentiality of the possessum. The results reveal an increased frequency of the non-re ﬂ exive variant in bilingual Russian, indicating convergence with German. At the same time, bilingual speakers observe the same animacy and referentiality constraints on the possessive choice as monolinguals. The e ﬀ ect of linguistic constraints, however, is smaller than the in ﬂ uence of individual speaker-related variables. Overall, our study indicates that convergence in language contact a ﬀ ects the frequency but not the conditioning of the variants and only in some bilingual individuals.


Introduction
Studies on language contact generally acknowledge that language variability, i.e., alternative expressions conveying the same meaning, plays a role in shaping outcomes of language contact (Heine and Kuteva 2005;Thomason and Kaufman 1988).It has been observed that contact-induced grammatical changes often involve "changes in which an inherently variable system in the receiving language undergoes changes that make that part of the language converge with the contact language, usually by increasing the occurrence of the category that happens to be most like the equivalent one in the contact language" (Backus 2005: 333).A well-known example of convergence is an increase in overt subject pronouns in pro-drop languages in contact with a non-pro-drop language (Poplack and Levey 2010: 393).Although distributional convergence is widely attested, it is unclear whether it only affects variant frequency or also usage conditions (Backus 2005: 322).This has been addressed by two approaches: experimental and comparative variationist sociolinguistic.
The experimental approach investigates L1 attrition using experimental methods comparing late (postpuberty) bilinguals with monolinguals residing in a home country (Schmid and Köpke 2019).Language variation has not been an explicit focus here but has emerged as an important issue; attrition is often observed in grammatical areas where the monolingual variety has variants constrained by syntax and pragmatics (Chamorro and Sorace 2019).Tsimpli et al. (2004), for example, showed that late Greek-English bilinguals use preverbal subjects more frequently than L1 Greek speakers, observing definiteness constraints, whereas monolinguals prefer postverbal subjects irrespective of definiteness.Gürel (2004) found changes in the binding conditions for the Turkish pronoun o 's/he' in late Turkish-English bilinguals, indicating English influence, while the conditioning of kendisi 'self' and the null pronoun remained stable.Vulchanova et al. (2022) examined Spanish deictic demonstratives in late Spanish-Norwegian bilinguals, finding a higher frequency of ese at the expense of este and aquel among bilinguals.Factors constraining the usage of these forms, however, were identical in monolinguals and bilinguals.The results of experimental studies are, therefore, mixed: the usage conditions of morphosyntactic variants in the L1 of late bilinguals in a contact situation can remain stable in some cases but change in others.
Studies bridging contact and variationist sociolinguistics focus on differences in linguistic constraints on variation between two communities through a comparative apparent time analysis of spontaneous spoken data (e.g., Léglise and Chamoreau 2013;Meyerhoff 2009;Meyerhoff and Nagy 2008;Poplack 2008).They investigate whether grammatical variant frequency and linguistic and sociolinguistic usage constraints change during language contact.Findings show that, contrary to the convergence hypothesis, grammatical variation patterns either remain stable, even if the structures of the languages in contact favour convergence (stability hypothesis), or diverge from the monolingual variety but not in the direction of the majority language.Instead, changes in a contact variety may be caused either by the acceleration of internal developments already present in the monolingual variety (acceleration hypothesis), or from the failure of the contact variety to participate in an ongoing change of the monolingual variety (divergence hypothesis) (Dumont and Wilson 2016;Lealess and Smith 2011;Poplack 2008;Poplack and Levey 2010).
Lealess and Smith (2011) investigated subject relative pronouns in mainstream Canadian and Quebec English and found the same overall rate of who in Canadian English and the late bilinguals' contact variety.Early bilinguals tended to use who more, possibly indicating convergence with French.However, substantial speaker variation suggests an influence of speaker-related variables.The importance of sociolinguistic factors, social class in particular, is also emphasized in Poplack and Levey (2010: 402-404).
With regard to variant conditioning, Lealess and Smith (2011) found the contact and mainstream varieties to be influenced by two linguistic factors (the presence of intervening material and the type of antecedent) to the same extent, whereas the third factor (sentence type) affected both varieties differently.They rejected the influence of French as a contact language, as the influence of sentence type in Quebec English matched patterns found in other English varieties, suggesting that maybe mainstream Canadian English changed, Quebec English lagging behind.Other examples of the contact variety failing to participate in ongoing mainstream variety change are need used as a deontic modal and singular concord with existentials in English (Poplack 2008).
Overall, both approaches indicate at least some changes in the L1 of first-generation speakers in grammatical areas where variation exists in the monolingual variety.However, it still needs to be clarified why and how far these changes go beyond usage frequencies in affecting variant conditioning.
In this study, we ask whether the relationship between two grammatical alternatives has changed in the firstgeneration immigrant variety of Russian in Germany compared to Russian spoken in Russia.We focus on two alternative pronominal forms expressing the 1st person reflexive possessive function: reflexive possessive (henceforth RP) and non-reflexive possessive (NRP).The choice between them is known to be difficult for second language and heritage learners of Russian (Comer 2009;Perevozchikova forthcoming-1).No study has so far investigated possessive variants in first generation Russian immigrants, although Gushchina (2013: 121) cites one example from spontaneous speech of a late Russian-German bilingual (ja by moim detjam tože khotela by èto peredat' 'I would also like to give it over to my.NREFL children'), where a NRP is used instead of a RP preferred in this case in the monolingual variety.
The two possessive forms in Russian have been competing for centuries, typically with the RP as the default option for referencing the subject (Pekelis 2021; Večerka 1993).Currently, possessive choice is constrained by syntactic, pragmatic, and semantic factors (Perevozchikova forthcoming-2), among which the referentially of the possessum and its animacy were suggested as the most important ones (Timberlake 1980).
The aim of this study is to compare the frequency of the possessive variants and the influence of two factors, referentiality and animacy, on the possessive choice in monolingual and bilingual varieties of Russian by using experimentally elicited data.Although not representing the vernacular, an experimental approach enabled control over the linguistic context and ensured comparability of the bilingual and monolingual material.The data were analysed using linear mixed models to account for inter-speaker variability and to assess the relative weight of linguistic and individual speaker-related factors.
We first briefly introduce variation in the Russian possessive system with a focus on animacy and referentiality as background for the hypotheses outlined in Section 3. The methodology of the study is presented in Section 4, followed by the discussion of results.
2 Russian 1st person possessives as a grammatical variable Russian, along with other Slavic languages, belongs to the reflexive type of possessive adnominal modifiers, i.e., it has an RP to distinguish a coreferential from a non-coreferential pronominal possessor in all persons and numbers (Manzelli 1990).Unlike in the 3rd person (1), this distinction is neutralized in 1st person possessives (2), which are considered free variants in reflexive possessive reference functions (Barnetová 1979: 355;Padučeva 2010: 191;Rozental' et al. 2015: 280;Timberlake 2004: 252).The NRP moj can therefore be used both for its core function of indicating possession non-reflexively and for reflexively referring to the grammatical subject, similar to the German possessive mein, which is not specified for reflexivity and is used for reflexive and non-reflexive possessive reference. (1) Oleg 1 uechal v otpusk.
The choice between the two possessives in Russian is conditioned by an interplay between syntactic, semantic, and pragmatic factors (Perevozchikova forthcoming-2).The referential status of the possessive phrase has been argued to be the major semantic factor: referentiality is claimed to inhibit the RP; non-referentiality to favour it (Šmelev 1988;Timberlake 1980).An expression is considered referential if "it is used to refer to a particular individual whose identity is established independently of a given proposition" and non-referential (attributive) if "it describes properties or characteristics identifying an individual appropriate to a given proposition" (Timberlake 1980: 781).This is illustrated in (3) and ( 4). (3) Ved' do ėtogo ja vsegda družestvenno rabotal s moim* / so svoim partnёrom, kem by to ni bylo.'After all, up to that time I had always worked in a friendly way with my.NREFL* / my.REFL partner, whoever he was.' (4) Vsё ėto ne moglo ne radovat' menja, i ja s eščё bol'šim rveniem prinjalsja zanimat'sja s moim / so svoim* partnёrom.
'All this couldn't help but please me, and I began to rehearse with my.NREFL / my.REFL* partner with even more enthusiasm.' In (3), the possessive phrase describes a set of individuals with the property of partner; the NRP should be dispreferred.In (4), the NRP should be chosen because the possessive phrase refers to a unique individual.This understanding of referentiality corresponds to the second component of pragmatic referentiality proposed by Chen (2009Chen ( : 1660)), "the specific reference, characterized by unique individuation, and differentiation from other entities by the same linguistic description."The first component of referentiality in Chen (2009), the presupposition of the actual existence of the referent in the discourse model, is less suitable here as it restricts the referential/non-referential contrast to indefinite expressions and the occurrence of non-referential nominals to opaque (negated and counterfactual) contexts.However, the influence of referentiality on possessive choice is most salient in opaque contexts (Timberlake 1980: 782).
The effect of the animacy of the possessum is well-documented (Timberlake 1980).The NRP was shown to be more frequent with animate than inanimate possessa; the RP equally frequent with both.There is no established explanation for the association of the NRP with animacy; it is suggested that animate possessa might be more inclined to referential use and being the focus of empathy, another influential factor in possessive choice (Perevozchikova forthcoming-2).
To summarize, two types of possessive pronouns can be used for the reflexive function in 1st person contexts in Russian: the RP, specialized in this function, and the NRP, which can be used reflexively, constrained by syntactic, semantic, and pragmatic factors, alongside its core non-reflexive possessive reference function.The referential status of the possessive phrase has been hypothesized to be the major constraining factor; the effect of animacy as a related semantic variable has been established.Since the Russian 1st person NRP is partially similar to the German 1st person possessive and its use is governed by semantic-pragmatic constraints, it is a likely candidate for contact-induced change according to the convergence scenario.The possessive variants are simultaneously involved in ongoing language change in monolingual Russian (and other Slavic languages) and therefore enable addressing the divergence hypothesis.

Research questions and hypotheses
Two questions are addressed in our study.First, do Russian-German bilinguals use the possessive variants equally frequently as speakers of the monolingual variety?Based on the scenarios described above, the NRP could be 1) more frequent in Russian spoken in Germany (convergence hypothesis), 2) equally frequent in both varieties (stability hypothesis), 3) less frequent in bilingual than monolingual Russian (divergence hypothesis).
Second, do Russian-German speakers observe the same linguistic constraints of animacy and referentiality as monolingual speakers?Either 1) Both groups use more NRPs when the possessum is animate and referential compared to inanimate and non-referential (stability hypothesis); 2) only monolinguals show this trend (convergence hypothesis); 3) only bilinguals show this trend (divergence hypothesis); or 4) this occurs in neither group.
However, much like Lealess and Smith (2011) and Poplack and Levey (2010), we expect that both the frequency and conditioning of the possessive choice is subject to substantial individual variation in the bilinguals and possibly also the monolinguals.

Method and data
An experimental approach was chosen over corpus investigation for several reasons.Available bilingual Russian-German corpora such as the RUEG corpus (Wiese et al. 2021) or the Russian Learner Corpus (Rakhilina et al. 2016) only contain data from heritage speakers, i.e., bilinguals who were born in Germany or immigrated as children, whereas our focus is on late bilinguals coming into contact with German after having completely acquired standard Russian.We followed De Bot et al. (1991) in setting the youngest age of L2 onset to 15 years.Even if a suitable corpus did exist, bilingual corpora tend to be small; possessive pronouns with subject reference are not sufficiently frequent in smaller corpora.Additionally, our objective required specific usage contexts and a high degree of comparability between bilingual and monolingual data, which can be best achieved experimentally.
The elicitation technique used was a two-alternative forced choice written sentence completion task.This allows for high statistical power in detecting differences between conditions and relative ease for participants due to binary choice.The sentence completion task, unlike acceptability judgments, is also closer to actual language use; research has shown that forced choice tests reflect the frequency of and linguistic constraints on the variant observed in a quantitative corpus analysis (Arppe and Jarvikivi 2007;Rosenbach 2003).
One major disadvantage of this elicitation format is that it "encourages a high degree of linguistic awareness, a focus on form and the use of metalinguistic knowledge" (Ellis and Roever 2021: 166).Nevertheless, considering that possessives are not selectively and consciously used in Russian, we hoped that the explicit nature of the task would not significantly influence performance.

Materials: two-alternative forced choice sentence completion task
The test was intended to measure the influence of two semantic factors, animacy and referential status of the possessum, on the choice of the possessive pronoun in the reflexive function.Four conditions were constructed exemplifying possible combinations of these factors.Humans and animals were considered animate and all other entities inanimate.Referentiality was operationalized based on the strength of the existential presupposition.Non-referential uses of possessive phrases were prompted in opaque contexts by minimizing the existential presupposition via negation and counterfactual modality as in (6).Referential uses were prompted in non-opaque contexts in past tense and by anaphoric reference to the possessum in subsequent clauses to emphasize that the speaker intends to refer to the entity in question as in ( 5).
( Fourty test items were constructed and divided into two lists.Items were counterbalanced; each participant saw each item in only one condition.Two options to complete the gap, the RP and NRP in a correct morphological form, were presented.Some of the 50 filler items intended to test attention, others were experimental items from an unrelated study.All items were presented in dialogue to increase the naturalness of the task.The test was distributed via the free web-based survey tool Google Forms.

Materials: questionnaire
Speaker-related variables were obtained through a questionnaire.For bilingual speakers, this included questions about personal characteristics (age, sex, duration of residence in Germany), educational level, occupation in the home country and Germany, usage frequency of both languages in different domains, self-evaluation of L1 and L2 proficiency, and linguistic and cultural self-identification.The influence of these variables on possessive choice will be reported in a separate study.

Participants
A total of 96 monolingual Russian speakers residing in Russia and 96 Russian-German bilinguals in Germany were recruited through social networks.The speaker groups are roughly comparable in terms of age, sex, and educational level (Table 1).The bilingual group immigrated to Germany on average at age 27 and had lived in Germany ≈21 years at the time of the study.All bilinguals reported high proficiency in both languages.

Procedure
Participants received a link to the test form with instructions to read the dialogues carefully and intuitively fill the gap with one alternative that best suited the context.The data were analysed in the statistical software R by fitting Possessive pronouns in Russian-German language contact linear mixed effect models.This allowed for including both numerical and categorical predictor variables and controlling for random variance arising from participant and item sampling.A mixed model was constructed to predict possessive type choice in both speaker groups.We set the RP as a reference value because it is more frequently used for the reflexive possessive function.Animacy and referentiality of the possessum and participant group were treated as fixed effects; participant and item as random effects.The model was first constructed with these factors and their interactions.Subsequently, we eliminated the factors and interactions that failed to reach the significance level of 0.05.

Results
Figure 1 plots the overall frequencies of the RP svoj and NRP moj in both data sets.Monolinguals prefer the RP, using the NRP in only 11% of the test items; bilinguals used it around three times more often at 30%. Figure 2 visualizes how often RP and NRP were used with animate and inanimate possessa.The observed pattern is identical for all speakers: the NRP is more frequent when the possessum is animate.Monolinguals used the NRP with inanimate possessa in 8% of test items and animate possessa in 14%.Bilinguals chose the NRP with inanimate possessa in 27% of items and animate possessa in 34%.  Figure 3 plots RP and NRP frequency according to the referential status of the possessum.The NRP is used by all speakers more often with referential than non-referential possessa.This effect seems weak among monolinguals as indicated by the ratio of 10-12% in non-referential compared to referential contexts.Bilinguals used the NRP with non-referential possessa in 27% of examples and referential possessa in 34%.
To test statistical reliability of the observations and whether animacy, referentiality, and group can predict possessive type, we fitted a logistic mixed model.The output of the final model is reproduced in Table 2, showing that all three fixed effects are statistically significant.The coefficient values indicate that Group is more influential than Animacy and Referentiality, i.e., bilinguals use the NRP more often.The effects of animacy and referentiality are smaller but also statistically significant.The NRP is more frequent with animate and referential possessa.The interaction of these two factors with Group was not significant, indicating that Animacy and Referentiality influence both speaker data sets in the same way.
The random effect Item did not significantly improve model fit and was eliminated.Participant was the only significant random effect in the final model with a large effect size.As indicated by Somers' D xy = 0.72 and conditional R 2 = 0.56, the model's total explanatory power, i.e., the fixed and the random effects together, is substantial.However, the part related to the fixed effects alone is very small, with a very low value of marginal R 2 = 0.10.This means that Animacy, Referentiality, and Group explain little variance in the possessive choice without taking into account individual speaker variation.

Discussion
The first research question was whether first-generation Russian-German bilinguals who have been active speakers of both languages for over 20 years use the reflexive and non-reflexive variants of the possessive  Possessive pronouns in Russian-German language contact pronoun equally frequently as monolingual Russian speakers.The results suggest they do not.The RP was chosen more frequently than the NRP by both groups, however, the bilinguals used the NRP three times more often than the monolinguals.The 9:1 ratio between reflexive and non-reflexive possessives in our monolingual data mirrors corpus frequencies (Perevozchikova forthcoming-2).The joint experimental and corpus evidence thus confirm that Russian may be in an advanced stage of change in which having dedicated reflexive pronouns for the reflexive function is almost obligatory (Pekelis 2021).
Our bilingual results suggest that this trend is reversed in Russian spoken in Germany.There are three possible explanations for this finding.First, the increased frequency of the NRP might be due to bilingualism, in line with the convergence hypothesis.Since the German possessive is not specified for reflexivity, it may be perceived as partially similar to the Russian NRP, the form normally reserved for non-reflexive reference but also used reflexively.There is also some degree of phonetic analogy between moj and mein.Guided by partial form and function similarity, bilinguals are likely to expand a more restricted distribution of the NRP towards the less restricted pattern of the L2.This follows the subset-superset transfer principle widely recognized in research on bilingualism and L1 attrition (Gürel 2019: 260).Crucially, as the NRP can also be used in monolingual Russian reflexively, its increasing frequency in the bilingual variety is not ungrammatical, simply unconventional.As discussed by Schmid and Köpke (2017), distributional changes leading to unconventional but not ungrammatical structures are common in language attrition.
Another reason for the increased frequency of the NRP among bilinguals might be simplification and generalization resulting from language processing strategies.To relieve the cognitive burden of managing multiple languages, bilinguals often unconsciously develop simplification strategies by employing generalized patterns common to both languages (Schmid and Köpke 2017).These simplified patterns also tend to be frequent cross-linguistically.In the case of possessives, a dedicated reflexive form for the reflexive function is typologically rare, most languages having one possessive for both functions (Manzelli 1990) and some Slavic languages possibly moving to this type (Pekelis 2021).
Finally, the higher rate of non-reflexive possessives in the bilingual data might reflect bilingual speakers' grammars lagging behind change in monolingual Russian.A steady increase in the use of the reflexive in standard Russian has been reported from the 18th century (Glovinskaja 1996; Pekelis 2021), however, the Russian variety in Germany might preserve an earlier stage in this process, when the non-reflexive was still used more frequently.This, however, remains to be tested in future work.
The second question was whether bilingual speakers observe the same linguistic constraints on possessive variant use.Our findings clearly suggest yes.Even if Russian-German bilinguals use more non-reflexive pronouns than monolinguals overall, they prefer them in the same contexts as monolinguals, namely, when the possessum is animate and referential.This is in line with previous experimental (Gürel 2019;Vulchanova et al. 2022) and variationist studies (Lealess and Smith 2011;Poplack 2008) as well as studies on dialect contact (Meyerhoff and Walker 2007), arguing that linguistic constraints on grammatical variation in language contact tend to remain stable and match those of the monolingual variety.
Although our results suggest that linguistic factors constrain possessive choice in monolingual and bilingual Russian in the same way, their effect was weak.We found a large inter-speaker variability in both groups.Individual speaker-related factors appeared to better explain possessive use than animacy, referentiality and speaker group, which potentially suggests more similarity between individuals overall than between speakers within a given group.If so, this provides evidence for the stability hypothesis and for ongoing change in the possessive variants in Russian.Such intra-and inter-individual variation is expected in language change, as described, e.g., by the variational theory of specialization of Wallenberg (2019).Language change involves gradual and imperfect community coordination in the specialization of variant forms.These processes are manifested in both variable use of competing forms by the same individual and across individuals (Wallenberg 2019).
An interesting qualitative observation from our study is that the effect of referentiality was weaker in monolinguals than bilinguals.Although not supported statistically, this difference could indicate that Russian speakers in Germany still follow the referential constraint on possessive choice, whereas monolinguals are giving it up in favour of the reflexive as the sole form for the reflexive possessive function.It could be that the NRP in monolingual Russian is getting more and more restricted, remaining only in contexts with animate possessa as a focus of empathy.Here we would have a divergence scenario as described by Poplack (2008): minority language speakers separated from the mainstream variety preserve an earlier state lagging behind the change in the monolingual variety.

Conclusion and future directions
This paper aimed to verify the claim about the stability of grammatical variation in a situation of language contact.We focused on the two 1st person possessive pronouns for the reflexive function in monolingual Russian and Russian spoken by first generation immigrant bilinguals in Germany.By means of a forced choice completion task, the usage frequency of the possessive variants and their conditioning by two linguistic constraints, animacy and referentiality of the possessum, was measured.Our results obtained by applying mixed models do not completely support the stability hypothesis, showing that the grammatical variant frequencies change to converge with the contact language, whereas linguistic usage constraints remain stable.
Our study has potential limitations.Elicitation techniques could have affected our results; it is possible that speakers could make at least partially different choices in naturalistic speech.Further research would benefit from using spontaneous speech production data and the comparative variationist methodology, as well as from considering a full set of linguistic constraints on the possessive choice.Individual speaker-related variables require a thorough investigation, which will be our next step.Finally, longitudinal studies of individual bilingual speakers should be conducted to track potential changes in progress.

Table  :
Overview of participants' demographic characteristics.

Table  :
Output of the logistic mixed model predicting possessive type in all participants (N = ).