Reflexives and politeness: possessive pronouns in Russian imperative sentences

Abstract

In Russian, a specialized reflexive possessive pronoun must be used to indicate coreference between the possessor and the subject. However, it has been observed that in the first and second person plural, the reflexivization rule is often violated, and a non-reflexive possessive is used for subject coreference. Existing studies do not distinguish between singular formal and plural informal readings in the second person, assuming no difference between the two contexts. Furthermore, there is limited knowledge regarding the use of possessives in imperative contexts, as previous research has primarily focused on indicatives.

This article aims to investigate the use of possessive pronouns in both formal and informal imperative contexts by addressing two questions. First, how often is a non-reflexive possessive used to indicate coreference with the subject in Russian imperative sentences? Second, what combination of semantic and pragmatic factors allows its use? Apart from the semantic factors that are well documented in the existing literature (referentiality, animacy and alienability of the possessum), we will consider three pragmatic dimensions of politeness that have not been considered so far: form or address (informal $ty$ / formal $Vy$), type of imperative (speaker- / hearer-oriented) and verbal aspect (perfective / imperfective).

The study analyzed the use of possessive pronouns in imperatives in the spoken subcorpus of the Russian National Corpus using generalized linear mixed models and random forests. The findings suggest that the choice of possessive is strongly influenced by the form of address ($ty$ / $Vy$). When addressing the listener with the informal $ty$, a reflexive possessive is almost exclusively used. When addressing the listener with the formal $Vy$, both possessives are used equally often. The choice between them depends on several factors, including the referentiality and alienability of the possessum, the type of imperative, and the verbal aspect. The non-reflexive form is preferred when the possessum is referential, the imperative is speaker-oriented, and, to a lesser extent, when the possessum is alienable and the verb is in the perfective aspect. Additionally, some verbs such as ‘give’ and ‘show’ strongly prefer a non-reflexive possessive.

The results suggest that a reflexive possessive has become the default for subject coreference in Russian second person singular informal contexts. In formal contexts, however, the change is still ongoing, and a non-reflexive possessive is used to express the referential uniqueness of the possessum and to convey negative politeness towards the addressee.

1 Introduction

There are two types of possessive pronouns in Russian, reflexive and non-reflexive. Like personal reflexive pronouns, reflexive possessives (hereafter RPs) express subject coreference. This means that a RP must be used when the adnominal possessor of the object or another non-subject participant has the same referent as the subject. Non-reflexive possessives (hereafter NRPs) must be used for subject disjoint reference, i.e. to indicate that the referent of the possessive is different from that of the subject.

Earlier studies claimed that this complementary distribution is often violated in first person (1P) and second person (2P) contexts, as well as in complex third person (3P) utterances containing secondary predications (Andreevskij, 1973; Bíly, 1981; Klenin, 1974; Padučeva, 1983, 1985, pp. 180‒208; Peškovskij, 1956; Timberlake, 1980). However, recent quantitative studies suggest that the reflexivization rule generally holds in 1P and 2P singular contexts (Pekelis, 2021; Perevozchikova, 2023a, 2023b; Timberlake, 2004, pp. 240‒257; Tiskin, 2019), but less so in the plural, where a NRP has been observed as a frequent means of establishing subject coreference (Pekelis, 2021; Tiskin, 2019).

What existing quantitative studies have not considered is the fact that the morphological 2P plural $vy$ ‘you’ in Russian can also mean a formal / polite singular $Vy$ ‘you.hon’. As the distinction between the two readings is not morphologically annotated in corpora, the use of possessives in formal $Vy$-contexts has not been analyzed separately from the 2P plural. Furthermore, existing quantitative research has focused on indicative contexts, and little is known about the distribution of RPs and NRPs in imperatives.

In previous descriptions of the factors conditioning the choice between a RP and a NRP, 2P imperative contexts have not been discussed in detail either. It has been assumed that the same syntactic, semantic, and pragmatic factors are at play as described for 1P and 3P. However, it has been suggested that in 2P contexts an additional factor of politeness may be important, and that the NRP Vaš sounds more polite than the RP svoj (Padučeva, 1983, p. 196; Pekelis, 2021, p. 87; Yokoyama & Klenin, 1976, p. 266). This suggestion was based on the interpretation of individual examples, and it is not clear to what extent it can be generalized to all 2P contexts.

The present article reports on an empirical investigation of RPs and NRPs in 2P imperative utterances of modern Russian that addresses the research gaps outlined above. The overall aim of the study is to describe the use of a RP and a NRP in imperatives, with special attention to the issue of politeness. Two questions are addressed. First, how often is a RP and a NRP used in 2P imperatives? Second, what combination of semantic and pragmatic factors constrains the use of each possessive?

The semantic factors included in the study are animacy, alienability, and referentiality of the possessum, whose influence on the choice of Russian possessives is well-documented (Andreevskij, 1973; Perevozchikova, 2023a, 2023b; Timberlake, 1980). As for pragmatic factors, we had to come up with our own ideas of what might be important in the context of the 2P imperative, given the lack of relevant studies. We decided to focus on three pragmatic variables that cover potentially relevant dimensions of politeness in 2P imperatives. The first is address form, the distinction between formal/polite $Vy$ vs. informal $ty$. The second is verbal aspect, which is known to influence the politeness of imperatives in Russian (Benacchio, 2010; Lehmann, 2008; Padučeva, 2010). The third factor is the distinction between speaker-oriented and hearer-oriented imperatives, which is considered important in determining the type of speech act and associated values of politeness (Leech, 1983, 2014).

The aim of the study is not only to test how each of these factors individually influences possessive choice, but also to identify which combination of factors best predicts a RP and a NRP. To achieve this, we applied advanced statistical techniques (generalized mixed models and random forests) that allow us to account for the interaction of factors and to assess the weight of each predictor. Unlike previous studies based on fiction or newspaper texts, we used spontaneous conversational data from the spoken subcorpus of the Russian National Corpus (RNC). These data are most representative of the registers in which imperatives naturally occur.

By taking a quantitative approach and using representative data, we aim to contribute to the discussion of reflexive constructions and politeness in Slavic languages. Our findings may also have practical implications for the presentation of the reflexivization rule in textbooks of Russian as a foreign language.

The article is organized as follows. In the next section, we review existing research on RPs and NRPs in Russian, focusing on their frequency (Sect. 2.1) as well as semantic (Sect. 2.2) and pragmatic (Sect. 2.3) factors that constrain their use. The literature review also includes a brief discussion of factors involved in the politeness of Russian imperatives, including verbal aspect, form of address, and speaker- or hearer-oriented type of imperative (Sect. 2.4). The theoretical part ends with a precise formulation of research questions and hypotheses (Sect. 2.5). Section 3 describes the methodology of the study. Section 4 presents the results in several steps. The frequency analysis of individual predictors (Sect. 4.1) is followed by analyses of the interaction of predictors based on mixed effects logistic regression (Sect. 4.2.1) with a detailed analysis of random effects (Sect. 4.2.2). Finally, a random forest analysis is presented (Sect. 4.2.3). The results are discussed in Sect. 5. The paper concludes with a brief summary of the study and possible directions for future research.

2 Theoretical preliminaries

Contemporary approaches describe possessive pronouns as units that limit the denotation of the head noun by establishing a relation based on the linguistic context or world knowledge (Koptjevskaja-Tamm, 2002). The main function of possessive pronouns is to express the referential anchoring of an object (possessum) through the speaker (1P), listener (2P), or the already mentioned entity (3P) (Zifonun, 2017, p. 672).

Some languages make an obligatory distinction between subject-coreferential and subject-disjoint possessive pronouns by having a specific RP that is specialized for coreferential anchoring within a clause. While most of these languages (such as Latin, Albanian, or Scandinavian languages) make the reflexivity distinction in the adpossessive domain only in the 3P, there are languages like Balto-Slavic that mark reflexivity in all persons and numbers. In Russian, a specialized RP svoj must be used if the adnominal possessor of the object or another non-subject participant is coreferential with the subject not only in the 3P but also in the 1P and 2P. A NRP generally cannot be coreferential with the subject of its clause (Rappaport, 1986).

Since the formulation of the reflexivization rule by Peškovskij (1956), several studies have documented examples in the 1P and the 2P where this rule does not hold (Andreevskij, 1973; Bíly, 1981; Klenin, 1974; Padučeva, 1983, 1985; Timberlake, 1980, 1996; Yokoyama & Klenin, 1976). This has led to the opinion that a RP and a NRP are in complementary distribution in the 3P but in free variation in the 1P and 2P. For example, Yokoyama (1999, p. 61) states that “unlike first and second person subjects, which show no clear dominance of one type of possessive over the other, third person subjects in modern literary Russian tend to co-occur more frequently with reflexives”. The obligatoriness of the reflexivization rule in the 3P and its optionality in the 1P and 2P have been also acknowledged in reference grammars (Proekt korpusnogo opisanija russkoj grammatiki, razdel “Mestoimenija”; Timberlake, 2004, pp. 252‒255; Horálek, 1979, p. 355; Isačenko, 1962, p. 436), prescriptive sources (Rozental’, 1997, §168; Rozental‘, 2016, pp. 234‒235), and text books (Nečaeva, 2021, pp. 247‒249; Kirschbaum, 2001, p. 204, Harrison & de Fleming, 2000, p. 66).

Most sources do not differentiate reflexivity in personal and in possessive pronouns, although there are mentions that the reflexivization rule is more obligatory in personal compared to possessive pronouns (Timberlake, 2004, pp. 252, 254; Padučeva, 1983, p. 17). Thus, the section on reflexive pronouns from the Russian corpus grammar (Proekt korpusnogo opisanija russkoj grammatiki) states that reflexivization is often obligatory whenever it is possible. The obligatoriness is illustrated with example 2, where the personal reflexive pronoun sebja must be used to refer to the 3P subject $on$. With pronouns of the 1P, the corpus grammar states that reflexivization is less obligatory and gives example (1) containing the NRP moj coreferential with the subject $ja$.

• (1) On₁ oblil sebja₁ vodoj / *ego₁ vodoj.

  ‘He₁ poured him.refl₁ / *him.nrefl₁ with water.’

• (2) Ja₁ xoču poznakomit’ vas s moej₁ ženoj.

  ‘I₁ want to introduce you to my.nrefl₁ wife.’

The optionality claim has recently been challenged by quantitative studies on large corpora of Russian (Pekelis, 2021; Perevozchikova, 2023a; Tiskin, 2019). They show that optionality is in place in the plural only, whereas in the singular, a RP is a default. Below we will discuss these studies in detail.

No matter whether the free variation of possessive is assumed for all 1P and 2P contexts, or only in the plural, all studies emphasize that optionality does not mean that it does not matter whether to use a RP or a NRP (Honselaar, 1986, p. 236; Padučeva, 1985, p. 181; Timberlake, 1980, p. 780). As Timberlake (1980, p. 780) puts it, whenever the reflexivization becomes optional under certain conditions, it gets constrained by further conditions. Several scholars have discussed examples where a NRP was used instead of an expected RP for subject coreference aiming to uncover motivations underlying the choice (Andreevskij, 1973; Bíly, 1981; Klenin, 1974; Padučeva, 1983, 1985; Timberlake, 2004; Yokoyama & Klenin, 1976). As a result of their efforts, we now have a list of syntactic, semantic, and pragmatic factors that have been suggested relevant for the choice between a RP and a NRP in optional contexts.

Syntactic factors cover in the first place the syntactic distance between the subject and the possessive phrase. It is a complex measure that includes the level of embedding and the degree of detachment of the possessive phrase from the subject, as well as the number and the quality of secondary predications (Bíly, 1981; Timberlake, 2004; Padučeva, 1985; Perevozchikova, 2023a; Rappaport, 1986). We will not discuss syntactic factors in detail, because most utterances in our study represent simple finite clauses making syntactic factors largely irrelevant for our data.

Semantic factors relate to the following properties of the noun modified by the possessive: referential uniqueness, animacy, alienability, and agentivity (Timberlake, 2004; Perevozchikova, 2023a; Padučeva, 1983, 1985). As for pragmatic variables, the point of view or empathy has dominated the discussions (Yokoyama & Klenin, 1976; Yokoyama, 1980, 1999). Unlike semantic and syntactic factors, pragmatic variables have not yet been subject to quantitative studies. Descriptions of how pragmatic factors constrain the possessive choice remain largely speculative due to little quantitative evidence. Additionally, empathy has been described mostly for 1P and 3P utterances, so that we know little about how it operates in the 2P. As semantic and pragmatic factors are of central importance for the present study, we will discuss them in some detail after reviewing quantitative evidence about the distribution of the possessives in modern Russian.

2.1 Frequency of reflexive and non-reflexive possessive pronouns in Russian

Frequency estimations of RPs and NRPs in Russian based on fiction (Andreevskij, 1973, pp. 3‒4; Honselaar, 1986, p. 237; Timberlake, 2004, p. 254) indicate that a RP is preferred over a NRP to a very high degree. Additionally, they point out that a RP is more likely to occur with 2P antecedents than with 1P antecedents. Honselaar (1986, p. 237) states that “first and second person possessives in Russian are preferred over svoj only in case there is a special reason for the speaker to highlight the role of the corresponding first or second person entity”.

Timberlake (2004, p. 254) calculated occurrences of RPs and NRPs in the diary of K. Simonov “Raznye dni vojny” from 1982. Indicative and imperative 2P contexts were treated separately but 2P singular and plural were counted together. Timberlake’s results, presented in the first row of Table 1, reveal that in the 1P, a NRP is very rare in the singular (8%) but makes about a quarter of possessive uses (22%) in the plural. In the 2P, a NRP was used only rarely in the indicative mood (4%) but accounted for almost a half of possessive uses (45%) in the imperative.

Table 1 Percentage of a NRP in existing quantitative accounts

There are three large-scale corpus studies on the distribution of possessives in modern Russian (Pekelis, 2021; Perevozchikova, 2023a; Tiskin, 2019), two of them addressing 2P contexts. Tiskin (2019) assessed the frequency of a RP and a NRP in the Russian National Corpus (RNC) and in the Araneum Russicum Maximum, a large web corpus of modern Russian. He calculated occurrences of 25 most frequent transitive Russian verbs followed by a possessive modifier of an accusative object. His findings reproduced in the second and third rows of Table 1 reveal that a NRP in modern Russian is frequently used in 1P.PL and 2P.PL imperative contexts. In the RNC, a NRP was used in more than 50% of 2P.PL imperatives, and in the Araneum, it occurred less often, in 16% of plural imperatives.

Pekelis (2021) described how frequencies of a RP and a NP in Russian had been changing in the last three centuries. She stated that starting from the 18th century a RP had been gaining ground. In the 19th century, a NRP was still a frequent alternative to a RP in 1P and 2P. In modern Russian, as indicated by her data from the newspaper subcorpus reproduced in the fourth row of Table 1, a RP has been established as a default possessive in the 1P and 2P. The only exception is the 1P.PL, where a NRP is as frequent as a RP. Pekelis (2021) explained it by the fact that 1P.PL is an ideal context for unfolding a semantic contrast between collective vs. distributed possession. A NRP signals collective and generalized possession, whereas a RP implies distributed possession. For example, in 2.1 taken from Pekelis (2021, p. 85), the NRP našix ‘our’ is used to refer to the subject $my$ ‘we’ because the sentence imposes the reading where the 1P subject $my$ ‘we’ includes not only the speaker and associated individuals but also a larger group, in this case the whole nation. The RP svoj is impossible here because it would imply a distributive reading, which is not plausible.

• (3) My poexali spasat’ našix / svoix* brat’ev-armjan.

  ‘We went to save our.nrefl / our.refl* brothers from Armenia.’

The problem with the study of Pekelis (2021) is the fact that her 2P.PL counts do not differentiate between singular formal Vy-uses, plural formal $Vy$-uses, and plural informal $ty$-uses and do not consider 2P imperative contexts. Her conclusion about a RP as a default possessive for subject coreference in the 2P might need revision, especially if we consider the findings of Timberlake (2004) and Tiskin (2019) mentioned above.

This overview shows that there is convincing evidence to state that subject coreference in Russian 1P.SG is expressed overwhelmingly with a RP. In the 1P.PL, both possessives are used with a semantic difference; a NRP expresses collective and generalized possession, whereas a RP expresses distributed possession. The situation in the 2P is less clear. In the 2P.SG indicative and probably also in imperative contexts, a RP dominates. However, there is mixed evidence regarding 2P.PL utterances because existing studies either counted singular formal and plural informal readings together and did not consider imperative utterances. Additionally, available studies are based on written texts and do not include spoken data, which is the most typical environment for 2P utterances. What is missing is an account of the distribution of a RP and a NRP in 2P singular formal contexts separately from 2P plural readings and preferably in spoken data. The study reported here aims to provide such investigation.

2.2 Semantic constraints on usage of reflexive and non-reflexive pronouns

Several studies (Honselaar, 1986; Andreevskij, 1973; Padučeva, 1983, 1985, pp. 204‒208, Timberlake, 1980) have discussed the semantic meanings of a RP and a NRP. According to Honselaar (1986, pp. 238‒239), the RP svoj has an overall meaning of “being somehow related to the prominent entity referred to or implied by the context of the same clause”, while the NRP moj has an overall meaning of ‘being somehow related to the 1P.SG entity’. According to Honselaar (1986, p. 239), the key difference between these meanings is the way they qualify the referent of the head noun. A NRP provides a straightforward qualification to the head noun by relating it to the discourse situation, while a RP “instructs the listener to stay within the narrow context of speaker’s words and to continue the reference to a prominent entity mentioned or implied by the context” (Honselaar, 1986, p. 239). By relating an entity to a concrete 1P or 2P, a NRP conceptualizes a possessum as really existing or at least existing in the form of a definite plan, which contributes to the definiteness of the possessum (Honselaar, 1986, p. 241). A RP presupposes neither the existence nor the definiteness of the possessum.

In a similar vein, Timberlake (1980) described the difference in meaning between a RP and a NRP in terms of referentiality, which he defined based on Donnelan (1966) and Partee (1972). A noun phrase is said to be referential when it is used to refer to a particular person or object whose identity is established independently of a given proposition, and non-referential (=attributive) when it describes qualities or characteristics that can be used to identify a person or object appropriate to a given proposition (Timberlake, 1980, p. 781). For indefinite noun phrases this contrast corresponds to the specific/nonspecific distinction (Haspelmath, 1997, pp. 108‒109); for definite noun phrases it is illustrated by two interpretations of a noun phrase in a sentence like The man who murdered Smith must be insane, the referential interpretation “that specific person” and the non-referential/attributive interpretation “whoever killed Smith” (Donnelan, 1966, p. 257). Timberlake (1980) suggests that a meta-feature of referential uniqueness is the main semantic factor underlying the distribution of RPs and NRPs in Russian. When the noun phrase modified by a possessive is referentially unique, i.e. when it is used to refer directly to its referent, reflexivization is inhibited in Russian and an NRP is often preferred. If a noun phrase modified by a possessive is used to describe properties of an individual as a member of a set, and thus has a reference that is replicated by other noun phrases, it is referentially non-unique and reflexivization is favored (Timberlake, 1980, p. 781).

Referential uniqueness is particularly noticeable in distributive and counterfactual contexts. For instance, consider example 2.2 from Šmelёv (1996, pp. 91‒92), where the possessum žena ‘wife’, modified by the NRP moj, is interpreted referentially to identify a specific existing person. In (4), žena modified by the RP svoj can be interpreted non-referentially as describing any individual that might not even exist in the real world but that has a property of being the wife of the speaker.

• (4) Daže v takoj situacii ja by ne mog udarit‘ moju ženu.

  ‘Even in such situation, I would not be able to hit my.nrefl wife.’

• (5) Daže v takoj situacii ja by ne mog udarit‘ svoju ženu.

  ‘Even in such situation, I would not be able to hit my.refl wife.’

The literature has discussed this referential distinction in terms of discourse-referential versus co-varying interpretation, or strict versus sloppy reading of anaphoric forms. Reflexive pronouns have been observed cross-linguistically to only allow for a sloppy/co-varying interpretation, not a strict reading (Haspelmath, 2023, p. 47). Using a Russian example from Dahl (1973, p. 106) as cited in Haspelmath (2023, p. 49), sentence (5) with the NRP moju ‘my’ suggests a discourse-referential interpretation, whereas example (6) with the RP svoju implies a co-varying reading.

• (6) Tol’ko ja lublju moju ženu.

  ‘Only I love my.nrefl wife.’ (= nobody else loves my wife)

• (7) Tol’ko ja lublju svoju ženu.

  ‘Only I love my.refl wife.’ (= nobody else loves his wife)

Other studies on possessives in Russian have also observed a preference for a discourse-referential interpretation of the noun modified by a NRP (Andreevskij, 1973; Perevozchikova, 2023b), particularly in 1P contexts. However, it is unclear to what extent the referentiality of the head noun determines the choice between a RP and a NRP in 2P utterances.

The second semantic factor that has been shown to influence possessive choice is the animacy of the head noun. It has been argued that the use of a RP is more obligatory with an inanimate head. When the head is animate, both a RP and a NRP can be used (Timberlake, 1980). Quantitative studies by Perevozchikova (2023a, 2023b) and Timberlake (1980) confirmed that a NRP is more frequent with an animate possessum than with an inanimate one. Although the reason for this correlation with animacy is not entirely clear, it has been suggested that it may be a manifestation of a more general principle (Yokoyama, 1999, p. 63), such as the referential uniqueness discussed above (Perevozchikova, 2023a).

Another feature relevant to possessive choice is alienability (Andreevskij, 1973; Yokoyama & Klenin, 1976; Perevozchikova, 2023a). The distinction between alienable and inalienable possession reflects a commonly accepted divide between nouns that denote inherently relational concepts, such as body parts or kinship terms, and therefore imply a possessor (inalienable possessive relation), and nouns that can be freely possessed, such as books, bags, etc. (alienable possessive relation). Yokoyama and Klenin (1976) and Andreevskij (1973) propose that Russian NRPs are mostly used with alienable possessa, whereas RPs can be used with both alienable and inalienable possessa. However, the classification of a noun as alienable or inalienable is not unproblematic. As recently shown (Chappell & McGregor, 1996, p. 9), languages do not treat body parts and kinship terms, which make up the core of the inalienable category, in the same way. Perevozchikova (2023a) demonstrated that Bulgarian kinship terms favor a NRP, while body parts prefer a RP. In addition, other nouns from the personal domain were found to be treated like the core inalienable nouns (Seiler, 1983, p. 13; Heine, 1997, p. 10). What exactly belongs to “the personal domain” depends on the culture and society, meaning that nouns can be classified differently in terms of inalienability (Heine, 1997, pp. 11–12).

Quantitative studies (Perevozchikova, 2023a, 2023b) have shown that animacy and alienability have a predicted influence on possessive choice in Russian, but their effect is small in comparison to other variables.

To sum up the semantic factors, the referential uniqueness of the noun modified by the possessive was claimed to be the main feature underlying the use of RPs and NRPs. The animacy and alienability of the possessum have also been considered relevant. For 1P contexts, research has shown that a NRP is preferred when the possessum is referential, animate, and alienable. It is yet to be determined if these same factors apply to the 2P, which will be investigated in this study.

2.3 Pragmatic factors

Scholarship on reflexive pronouns suggests that empathy, i.e. the speaker’s identification with a person whose point of view he represents (Kuno, 1976, 1987; Kuno & Kaburaki, 1975) is the key pragmatic factor underlying the choice between reflexive and non-reflexive forms in many languages (Kemmer, 1996; Reinhart & Reuland, 1991; Yokoyama, 1975; Yokoyama & Klenin, 1976; Zribi-Hertz, 1989). It is generally agreed that when reflexive and non-reflexive pronouns alternate in the 3P, the reflexive is used when the speaker construes the antecedent as an observer of the content set up in the predication (Cantrall, 1974; Kuno, 1987; Zribi-Hertz, 1989). In other words, the reflexive pronoun expresses the speaker’s empathy with a participant in the clause by adopting an internal point of view associated with that participant. Non-reflexive pronouns are considered to be neutral with regard to empathy and to express an objective perspective on the event. For example, in Russian, the choice between the reflexive personal pronoun sebja ‘self’ and the non-reflexive nego ‘him’ in sentences 2.3 and (8) depends on who evaluates the complexity of the task – the referent of the subject or the speaker. In sentence 2.3, the speaker takes the participant’s perspective and uses the reflexive pronoun sebja. In sentence (8), the speaker evaluates from his or her own point of view and the non-reflexive pronoun is used.

• (8) On vsegda berёtsja za trudnye dlja sebja zadači.

  ‘He always takes on the tasks difficult for himself.’

• (9) On vsegda berёtsja za trudnye dlja nego zadači.

  ‘He always takes on the tasks difficult for him. ’

Scholars of Russian (Padučeva, 1983, p. 20; Yokoyama & Klenin, 1976; Yokoyama, 1999) have observed that in 1P and 2P contexts, the influence of empathy on pronoun choice is opposite to that observed for the 3P. In the 1P and the 2P, it is a NRP that signals the speaker’s identification with the pronoun referent. Yokoyama (1975, p. 79) proposes operationalizing empathy as the distance between the antecedent of a possessive and the speaker. According to Yokoyama, a RP is marked +distance and a NRP as –distance in the 1P and the 2P. In the 1P, the distance relates to the distinction between two different personalities of the speaker: a hear-and-now speaker, on the one hand, and a speaker in other circumstances, on the other (Yokoyama & Klenin, 1976, p. 261).

In the 2P, as pointed by Yokoyama and Klenin (1976, p. 266), distance mainly concerns the relationship between the speaker and addressee, i.e. a NRP generally indicates proximity between the speaker and the addressee, whereas a RP indicates distance. Yokoyama and Klenin (1976, p. 266) illustrate this pragmatic contrast with examples (9) and (10), stating that (9) sounds as “a bit of friendly advice in which the speaker puts himself into the discourse together with the addressee”. On the other hand, they interpret (10) as “a command without any particular interest in how it may be received, with no indication that the speaker is aware of any possible objection to his objective judgment” (Yokoyama & Klenin, 1976, p. 266).

• (10) Ja prošu Vas, vzves’te Vaši slova!

  ‘I beg you, weight your.nrefl words.’

• (11) Ėto, pravda, nevozmožno – slušat’ srazu tri priёmnika. Vyključite svoj, Vas že prosjat.

  ‘This really is impossible – listening to three radios at once. Turn yours.refl off – you are asked!’

The problem with Yokoyama and Klenin’s interpretation of (10), however, is the fact that the context here strongly suggests a distributive interpretation of the noun phrase your radio, which requires the use of a RP. Therefore, it is not clear to what extent the RP svoj in this case actually indicates a distance between the speaker and the addressee. Furthermore, Yokoyama and Klenin’s analysis cannot explain the observation that a NRP sounds impossible in the singular informal variant of examples like (12a), as pointed out by Padučeva (1983, p. 196) and Kolesnikova (2013, p. 6). If a NRP is used to reduce the pragmatic distance between the speaker and the listener addressed with a formal V-pronoun, as suggested by Yokoyama and Klenin (1976), it is not clear why a NRP cannot be used for this purpose in requests to a listener addressed with an informal T-pronoun, as in (12b).

• (12)

  

Several scholars have noticed that a NRP is more appropriate than a RP in the 2P honorific contexts (Padučeva, 1983, p. 196; Pekelis, 2021, p. 87; Kolesnikova, 2013, p. 6; Honselaar, 1986, p. 244). All of them stated that when addressing listeners with the honorific V-pronoun ($Vy$) as in (12a) or (11), a NRP sounds more polite than a RP, especially in the imperative. However, there is no agreement on how and why this interpretation comes about. Padučeva (1983, p. 196) believes that the impoliteness of a RP in examples like (12a) might arise from the connotation of permanent possession in the semantic meaning of svoj. Kolesnikova (2013, p. 196) assigns the politeness effect of the NRP Vaš to “the tautological marking of politeness by doubling of the deictic pronoun”. Honselaar (1986, p. 244) attributes the impoliteness of a RP in examples like (11) to its semantic meaning of referential non-uniqueness. In (11), he believes, the use of svoej as an anchor of wife is impolite as it suggests that she is not a specific, identified person, but rather one of many possible entities.

• (13) Peredajte privet Vašej / svoej žene.

  ‘Give my love to your.nrefl / your.refl wife.’

We see that there are consistent observation in the literature pointing at the fact that the choice of possessive in the 2P contexts might be connected to politeness. A NRP is claimed to sound more polite than a RP, especially in 2P honorific contexts and in imperatives. To our knowledge, however, there have been no empirical research aimed to verify these claims and to clarify why and how this politeness distinction arises. The present study aims to address these gaps by viewing possessives as a linguistic structure that might contribute to the interpretation of an imperative utterance as (im)polite together with other linguistic means, which we will discuss in the next section.

2.4 Polite and impolite imperatives in Russian

Traditionally, the grammatical form of imperative has been associated with direct speech acts such as commands or orders and therefore with impoliteness (Searle, 1975, p. 64). Recent accounts, however, argue that the imperative can express a variety of speech acts, such as offers (e.g., Take a piece of cake), invitations (e.g., Help yourself), or wishes (e.g., Have a good day) (see an overview in Jary & Kissine, 2014). To describe the semantic meaning underlying all these speech acts, Leech (2014, p. 62) introduced the term mand, which refers to representing the state of affairs as unrealized but as desirable. The semantic category of mand, he argues, can express different illocutionary meanings, which cannot be differentiated based solely on the syntactic construction (Leech, 2014, p. 62). For instance, the utterance Take a seat could be an offer, a request, or an order, depending on the continuous rather than absolute context-based judgments about how much freedom the addressee has to comply or not comply (Leech, 2014, p. 63). Similarly, the difference between a command and an offer is based on scaler context-based judgements about “whether the action is envisaged for the benefit of the speaker or of the hearer, which is not always clear “(Leech, 2014, p. 63). Leech (1983, p. 107) assumed that a propositional content of an imperative might be placed on a cost-benefit scale reproduced in 2.4. According to Leech (1983, p. 107),“at some rather indeterminate point on this scale (depending on the context) the relevant value becomes ‘benefit to h’ rather than ‘cost to h’” but in general, there is an increase in hearer’s benefit and in politeness from a) to f).

• (14) Cost-benefit scale (Leech, 1983, p. 107)

  

The benefit factor has been considered important in determining the type of speech act and the associated politeness of imperatives in many other accounts (Hamblin, 1987; Huddleston & Pullum, 2002, pp. 929–931; Quirk et al., 1985, pp. 831–832; Takahashi, 2012, p. 80; Wilson & Sperber, 1988). Recently, de Hoop et al. (2016) have empirically confirmed the influence of the benefit factor on the politeness of imperatives in Dutch. They differentiated between speaker-oriented imperatives, i.e. imperatives that refer to actions that are beneficial to the speaker such as commands or requests, and hearer-oriented imperatives that refer to actions that are in the hearer’s interest such as offers, invitations, or advice. The results showed that hearer-oriented imperatives in Dutch were perceived as positive/polite, whereas speaker-oriented imperatives received a negative/impolite interpretation, which was mitigated by the use of modal particles.

We believe that the distinction between speaker- and hearer-oriented imperatives might be useful in discussing the politeness of imperatives in Russian. Speakers of Russian are known for normatively deploying the imperative form across a wide range of interaction contexts including requests, instructions, offers, and advice (Bolden, 2017; Brehmer, 2000; Larina, 2009; Rykov-Ibsen, 2016). Bolden (2017) demonstrated that the imperative form was chosen in 90% of all Russian requests. Durst-Andersen (2019) reported that 84% of Russian speakers used the imperative form to make an offer.

Studies emphasize that the Russian imperative itself does not have any kind of negative connotations and is a neutral form that is not associated with either politeness or impoliteness (Durst-Andersen, 2019; Larina, 2009). If politeness associations arise with the hearer, they will be linked to other parts of the utterance, such as the address form, verbal aspect, modal particles, intonation, and facial expressions (Durst-Andersen, 2019). It may be necessary to add the choice between a RP and a NRP to this list, considering the observations about the politeness of NRPs in imperatives discussed above. Before turning to the empirical investigation of this issue, we will briefly outline the contribution of the pronominal address form and verbal aspect to the politeness of an imperative utterance.

Regarding address forms, Russian uses the informal/intimate pronoun $ty$ and the formal/distant $Vy$, which is the morphological second-person plural form. This distinction is based on the same principle as in many European languages, where the greater the social and power distance between interlocutors, the more likely the formal $V$-form is used (Brown & Gilman, 1960). In the framework of Brown and Levinson’s politeness theory (1987), the Russian $Vy$ belongs to negative politeness, while the use of the informal $ty$ is a strategy of positive politeness. Further rules regarding pronominal address usage in Russian are outlined in Formanovskaja (2002), Jachnow (1974), and Apresjan (1995). Apresjan (1995, pp. 144–145) identifies six basic situations in which the informal $ty$ is used in Russian. He concludes that a more efficient way to characterize its meaning would be through the concept of the speaker’s private space (ličnaja sfera govorjaščego). This includes the speaker himself and everything that is physically, morally, emotionally, or intellectually close to him. The informal $ty$ is used to include the interlocutor into the speaker’s private space, whereas the formal $Vy$ is used to exclude the interlocutor from the speaker’s private space.

Verbal aspect is another factor that contributes to the interpretation of the Russian imperative form as polite or impolite (Benacchio, 2010; Dicky, 2020; Lehmann, 2008; Padučeva, 2010; Šatunovskij, 2009). According to Benacchio (2010), perfective imperatives express negative politeness/formality because they convey a distant perspective on the action through a metaphorical extension of the core semantic meaning of the perfective aspect. Imperfective imperatives convey a close-up view of an action and tend to occur in less formal contexts. They can sound polite if they request actions that are in the interest of the listener, or they can be perceived as impolite/rude if they request an action in the interest of the speaker (Benacchio, 2010, p. 49). Dicky (2020) following Šatunovskij (2009) argues that (im)polite readings of aspect arise from the category of temporal (in)definiteness. Perfective imperatives express temporal definiteness by communicating “a covert request for the listener to make the choice to perform an action and then to perform it to completion with a desired outcome” (Dicky, 2020, p. 574). As the listener is requested to make the choice to perform the action, perfective imperatives always signal negative politeness (Dicky, 2020, p. 574). Imperfective imperatives generally express temporal indefiniteness. In case of single actions, they indicate that the time for the listener to make the choice whether to perform the action is over (Dicky, 2020, p. 574). Two scenarios are possible here. First, the speaker can be acting on his/her knowledge that the listener has already made the choice to perform the action, which is the case in permissives like (15) taken from Šatunovskij (2009, p. 262) and cited by Dicky (2020, p. 563). In such contexts, the effect of politeness often arises because the imperfective here signals that the speaker knows that the addressee has already made the choice to perform the action and the speaker supports/approves of this choice (Šatunovskij, 2009, pp. 263–264). Second, the speaker can claim the right to make the choice for the listener imposing it on the latter causing either an ‘uncompromising/stern’ or an ‘empathetic/gentle’ interpretation of the imperative (Dicky, 2020, p. 566). An uncompromising variant covers rude imperatives like (15) and (16) from Šatunovskij (2009, p. 270), as well as emergencies, threats, and military orders. An empathetic variant involves the speaker attempting to make a decision for the listener by acting as if s/he knows what is best for the listener (Dicky, 2020, p. 568), as in example (17) from Dicky (2020, p. 568). In such polite suggestions, the imperfective imperative expresses the empathy of the speaker with the addressee.

• (15) Vxodite! Razdevajtes’!

  ‘Come in! Take your coat off!’

• (16) Uxodite! – skazal ja serdito.

  ‘“Go away!” I said angrily’

• (17) Ubirajsja!

  ‘Get lost!’

• (18) Togda davajte vašu sumku, čtoby ne taskat’ eё.

  ‘Then give me your bag, so you won’t have to lug it around.’

It is evident that imperfective imperatives can have a variety of pragmatic interpretations varying from politeness to rudeness. Perfective imperatives, on the other hand, “in and of themselves always communicate a modicum of negative politeness” (Dicky, 2020, p. 574).

In this section, we have seen that the imperative form is not inherently polite or impolite and have outlined three factors (type of imperative, pronominal address, and verbal aspect) that have been suggested to influence the perception of imperative politeness by the listener. Our task in this study is to find out whether and how the choice between a RP and a NRP interacts with these three factors in their influence on the politeness of imperatives.

2.5 Research questions and hypotheses

This study addresses two gaps identified in previous research on the choice between a RP and a NRP in Russian. Firstly, quantitative studies assessing the frequency of both possessives have not made a distinction between 2P formal singular $Vy$ and plural informal $vy$ contexts, and have mostly analyzed indicative utterances. Therefore, it is unclear whether a NRP or a RP is preferred in 2P formal imperative contexts. Second, it has been observed that a NRP sounds more polite than a RP, but it is far from clear how this interpretation arises and how the possessive form interacts with other factors that are also known to contribute to the politeness of an imperative utterance.

The aim of our study is uncover which possessive, a RP or a NRP, is preferred by Russian speakers in 2P imperative contexts and under what conditions. The first question we ask is how frequent is a RP and a NRP in 2P imperatives with $ty$ and with $Vy$ address forms. Based on previous scholarship, we can expect a higher frequency of a NRP with a formal $Vy$ address than with an informal $ty$ address, where a RP should be the default form (Hypothesis 1).

The second question is concerned with the semantic and pragmatic factors that permit the use of a NRP in 2P imperatives. As suggested by previous studies, it is expected that a NRP will be preferred when the possessum is referential, animate, and alienable (Hypothesis 2). Additionally, verbal aspect and imperative type may also influence the possessive choice. A NRP might be preferred with the perfective aspect to reinforce negative politeness conveyed by the perfective, at least with the formal $Vy$ address (Hypothesis 3). Concerning imperative type, we can speculate that a NRP might be more frequent in speaker-oriented imperatives to mitigate their face threatening potential (Hypothesis 4).

Finally, this study aims to investigate not only the individual impact of each semantic and pragmatic predictor, but also their interaction in determining the choice between a RP and a NRP. To achieve this, logistic regression and random forests techniques will be applied.

3 Methodology

In order to test the four hypotheses outlined above, spoken dialogic data was required, as this type of communication is most likely to contain imperatives with both $ty$- and $Vy$-form of address. The spoken subcorpus of the Russian National Corpus (RNC) was chosen as the data source. It contains transcripts of the spoken texts of different types (private speech, public speech, and movie speech tracks) morphologically, semantically, and sociolinguistically annotated with a RNC annotation system. At the time of the search in July 2021, the subcorpus consisted of 4,210 transcripts with a total of 13,399,937 tokens. The largest part of the corpus (52%) contains recordings of public communication, mostly discussions and talks, 38% of the data are transcripts of movies, comedies and dramas. Private conversations make up 10% of the subcorpus. The subcorpus covers the period from 1930 to 2007, but the majority of the data was collected between the 1990s and 2007. Geographically, the subcorpus represents different areas of the European part of Russia. Although the spoken subcorpus of the RNC is considered a representative sample of spoken Russian from the 1930s to the present (Grišina & Savčuk, 2009), it can be questioned to what extent it reflects spontaneous conversational language use, considering that spontaneous spoken conversation is underrepresented in it, while planned speech is overrepresented. In addition, as one reviewer pointed out, the subcorpus contains a relatively large proportion of movie speech tracks, some of which are based on literary works and may therefore not be representative of contemporary spoken Russian. Being aware of these disadvantages, we believe that our choice of the spoken subcorpus of the RNC is justified by the fact that it is the only large, publicly available, morphologically annotated corpus of spoken standard Russian that we know. We also hope that we have partially solved the “time problem” by excluding transcripts of films depicting distant historical periods (e.g., Don Quixote or Sherlock Holmes).

The RNC search interface was used to identify utterances that contained a verb in the imperative form, followed by a possessive pronoun within three words. The search query used was “V & imper & 2p + [1-3] + svoj | vaš”. The query returned 2 839 hits, which included both plural and singular imperative forms with RPs and NRPs. The corpus annotation does not distinguish between simple plural, singular polite, and plural polite readings. Therefore, the plural readings had to be sorted out manually. Additionally, three other types of irrelevant examples were excluded. The first type covers utterances where the possessive phrase and the imperative verb were not within the same clause. The second type includes utterances where the imperative form davajte ‘give us’ was part of a 1P analytical imperative, such as Davajte sosredotočim svoё vnimanie ‘Let us concentrate’. Thirdly, we excluded utterances containing an indicative form that was differentiated from the imperative form by word stress only, as the form provodite ‘spend’ in the interrogative sentence A vot kak Vy provodite svoj dosug? ‘And how do you spend your free time?’

Having excluded irrelevant hits, we ended up with 888 examples, 478 of them (54%) with the $ty$ form of address, and 410 of them (46%) with the polite singular $Vy$ form of address. The resulting data were exported to a spreadsheet, and each example sentence was manually annotated for the factors listed in Table 2. As demonstrated in the table, each factor was annotated with two values. Factors 1 (type of possessive), 2 (form of address), and 4 (verbal aspect) were annotated based on form, and do not require further comments.

Table 2 Linguistic factors and their values investigated in the present study

Coding animacy was also straightforward. Living entities were considered animate and non-living entities inanimate. Regarding alienability, the group of possessa that we classified as inalienable included not only body parts and kinship terms, but also possessed entities that originate from the possessor (such as voice, sweat, and smile), mental states of the possessor (such as mind and fear), and attributes of the possessor (such as name and age). All other nouns that can be freely possessed and that imply a less permanent association between the two entities (such as telephone, dress, and apartment) were considered alienable.

The coding of imperative type and possessum referentiality posed challenges, requiring semantically and pragmatically motivated decisions. The annotation process followed a general procedure. Three linguistically trained student assistants who are Russian native speakers were asked to choose between two specified values. At first, they worked individually and later they discussed problematic sentences to reach a decision together.

The type of imperative was determined based on whether the action requested in the imperative is envisaged for the benefit of the speaker or of the hearer. The student assistants were tasked with coding who benefits more from the action described in the imperative: the speaker or the hearer. For example, in 3, there was a unanimous agreement that the hearer is most likely to benefit from the action of taking care of the feet. Conversely, in (19), the speaker will benefit from the hearer leaving the room, as indicated in the explicit statement of the speaker’s needs: I need to watch the film until the end.

• (19) [Privalov] Razrešite Vas priglasit’ na val’s.

  [Nina] Vy tancuete?

  [Privalov] Otkrovenno govorja, davnen’ko ne tanceval. Tak čto beregite svoi nogi.

  [Pervaja perčatka, film, 1946]

  ‘[Privalov] Let me invite you to for a dance.

  [Nina] Do you dance?

  [Privalov] Frankly speaking, I haven’t danced in a while. So take care of your.refl feet.

  [The first glove, film, 1946]’

• (20) [Saša] Ladno / vsё. Mne nužno dosmotret’ fil’m. Idi v svoju komnatu.

  [Кatja] Kakoj fil’m?

  [Saša] „Tainstvennyj les”. [Razgovor brata s setroj, 2006]

  ‘[Saša] Okay, that’s it. I have to finish watching the movie. Go to your.refl room.

  [Кatja] What movie?

  [Saša] ‘The Mysterious Forest’ [Talk between brother and sister, 2006]’

Referentiality of the noun phrase modified by a possessive pronoun was coded as a binary variable indicating the probability of a referential vs. non-referential/attributive interpretation in a given context. To increase the precision of the referentiality assessment, we examined the extended context rather than just the sentence context. The referentiality of the possessum was evaluated based on the definition of referentiality in Timberlake (1980) outlined in the theoretical section. A noun phrase was considered referential if the coders have agreed that it was used to identify a specific individual or thing in a speech situation, and non-referential if it was used to describe properties of a class of individuals or things which the noun denotes. An intuitive “test” for the non-referential usage was the legitimacy of the parenthetical comment “whoever (s)he / it is.” For example, in (20), there is a situation where a pupil made drawings on the desk. The teacher is scolding the child and is telling her to look at her (girl’s) desk. The usage of desk here is referential as the teacher intends to identify this particular desk in this particular speech situation. In (21), the noun phrase modified by the possessum is non-referential because the speaker, who is a stranger seeing to the addresses seeing them for the first time, is not referring to some specific acquaintances of the addressees, but is rather describing a set of acquaintances from which the listener can choose a particular referent. The noun phrase your acquaintances also passes “the parenthetical comment test”, as it can be logically followed by “whoever they are”.

• (21) [Olga, učitel’] Ja tebja sejčas k direktoru otvedu / i ty ne vyjdeš iz školy / poka za toboj kto-nibud’ iz roditelej ne predёt. Učitelja fiziki čut’ do infarkta ne dovela / portiš’ škol’noe imuščestvo! Posmotri na svoju partu! Čto ty s nej sdelala? А? [Klassnyj čas, 2007]

  ‘[Olga, teacher] I’m taking you to the principal’s office and you won’t leave the school until one of your parents comes to pick you up. I almost gave the physics teacher a heart attack/damaging school property! Look at your.refl desk! What did you do to it? А? [The class hour, film, 2007]’

• (22) [Dama u labirinta] I vot čto ja vam skažu/ milye ledi. Labirint / ėto ne mesto dlja svidanij! Ėto mesto dlja ostryx oščuščenij! Peredajte ėto svoim znakomym. [Troe v lodke ne sčitaja sobaki, к/f, 1979)

  ‘[The lady at the labyrinth] And I’ll tell you what, lovely ladies. The maze is not a place for dating! It’s a place for thrills! Tell that to your acquaintances! Admission is two pence! [Three men in a boat (to say nothing of the dog), film, 1979]’

Additionally to the seven linguistic factors listed in Table 1, we coded two variables to include as random factors in a regression model. The first random effect is the lemma of the imperative verb, motivated by the remark of Tiskin (2019, pp. 6‒7) about the varying frequency of a NRP with different verbs. As a second random effect, we decided to include the lemma of the possessum based on the observation on Czech (Čmejrková, 2011, p. 660) about a NRP to be preferred with particular objects belonging to the private space of the speaker.

The database containing 888 examples annotated based on the nine parameters as just described served as the foundation for the analysis presented in the following section.

4 Findings

This section begins with a graphical exploration of the data, demonstrating how each of the six main factors (address type, imperative type, aspect, referentiality, animacy, and alienability of the possessum) independently influences the frequency of a RP and a NRP in our sample. Then, we will perform a generalized mixed effects modelling with the six main factors and two random effects. After that, a detailed examination of verb lemma as a random variable will be provided. Finally, we will explore how all explanatory factors interact and jointly predict the choice of the possessive form by using the techniques of random forests.

4.1 Individual predictors

For a start, let us explore the distribution of a RP and a NRP according to each of the six main predictor variables visualized in mosaic plots (Figs. 1–6). The cells of Fig. 1 represent proportional frequency of a RP and a NRP with $ty$ and $Vy$ forms of address. It reveals that a RP is used in 94% of $ty$-examples, therefore being a default form to indicate coreference with the imperative subject in informal contexts. The NRP tvoj occurred only in 23 examples, i.e., only in 6% of $ty$-contexts. However, in the $Vy$-contexts, the NRP Vaš was used in 54% of examples, being slightly more frequent than the RP svoj, which occurred in 46% of $Vy$-sentences. A Pearson’s chi-squared test with Yates’ continuity correction confirms that the difference in the distribution of possessive according to the address form is highly significant (X-squared = 264.44, df = 1, p-value < 2.2e-16).

Fig. 1

Frequency of reflexive (r) and non-reflexive (n) according to address form

Figure 2 presents an additional data split according to the type of imperative. The plot shows that speaker-oriented imperatives were more frequent than hearer-oriented imperatives, as indicated by the larger widths of the second and fourth bars compared to the first and third bars. Additionally, a NRP occurred more often in speaker-oriented imperatives than in hearer-oriented imperatives. In $ty$-contexts, a NRP was observed in 8% of speaker-oriented imperatives compared to 2% of hearer-oriented imperatives. A chi-squared test shows that this difference is significant (X-squared = 7.40, df = 1, p-value = .0065). In $Vy$-contexts, a NRP dominated in speaker-oriented imperatives, occurring in 66% of examples, whereas in hearer-oriented imperatives, it occurred in only 23% of examples. The significance of this difference in distribution was confirmed by a chi-squared test (X-squared = 64.78, df = 1, p-value < 0.00001), which means that the imperative type influences the distribution of possessives both in $ty$- and in $Vy$-contexts in the same way, i.e. speaker-oriented imperatives favor a NRP.

Fig. 2

Frequency of reflexive (r) and non-reflexive (n) according to address form and imperative type

Turning to the referentiality of the possessum visualized in Fig. 3, we see that it also seems to influence the frequency of possessives with both address forms. With the $ty$-address, a NRP was used in 7% of examples containing a referential possessum and in only 1% of examples with a non-referential possessum. This difference was significant (X-squared = 10.31, df = 1, p-value = .0013). With the $Vy$-address, a NRP occurred in 69% of examples containing a referential possessum against 20% of sentences with a non-referential possessum. The significance of this difference in distribution was backed up by a chi-squared test (X-squared = 86.70, df = 1, p-value < 0.00001).

Fig. 3

Frequency of reflexive (r) and non-reflexive (n) according to address form and referentiality

Figure 4 presents a mosaic plot that splits the data based on the aspect of the imperative verb. The plot shows that with the $ty$-address, imperfectives were as frequent as perfectives, while with the $Vy$-address, there were twice as many perfectives than imperfectives. The effect of aspect on the type of possessive was opposite for the $ty$- and $Vy$-addresses. With the $ty$-address, a NRP was more frequent with imperfective verbs (7%) than with perfective verbs (3%). However, the statistical analysis showed that the difference was not significant (X-squared = 3.46, p-value = .062614). With the $Vy$-address, a NRP was used in 58% of sentences with perfective verbs and in 44% of examples with imperfectives. A chi-square tests indicated that this difference is significant (X-squared = 7.0716, p-value = .007831).

Fig. 4

Frequency of reflexive (r) and non-reflexive (n) according address form and aspect

Figure 5 visualizes the distribution of possessives with $ty$- and $Vy$-address forms according to the animacy of the possessum. In $ty$-contexts, a NRP occurred more often with inanimate possessa (6%) than with animate possessa (2%). However, this difference turned out statistically insignificant. With $Vy$-forms, a NRP occurred in 53% of examples containing inanimate possessa and in 56% of examples with animate possessa. A chi-squared test confirmed that this small difference is insignificant meaning that animacy does not influences the possessive choice.

Fig. 5

Frequency of reflexive (r) and non-reflexive (n) according to address form and animacy

Finally, Fig. 6 illustrates the frequency of each possessive with alienable and inalienable possessa. In $ty$-contexts, there is no difference between alienable and inalienable possessa, a NRP is used in 5% of examples with both. With the $Vy$-address, however, a NRP occurs in 58% of examples with alienable possessa and in 39% of examples with inalienable possessa. A higher frequency of a NRP with alienable possessa is confirmed by a chi-squared test (X-squared = 9.33, p-value = .002254).

Fig. 6

Frequency of reflexive (r) and non-reflexive (n) according to address form and alienability

Summarizing the individual predictors, we found that a NRP was more frequent with the $Vy$-address compared to the $ty$-address, with speaker- than with hearer-oriented imperatives, and with referential than with non-referential possessa. In $Vy$-contexts, the possessive choice was influenced by verbal aspect and the alienability of the possessum. Animacy appeared irrelevant for the possessive choice with both address forms.

4.2 Interaction of predictor variables

Now that we have examined each predictor individually, we will investigate their interaction. We will begin by presenting a mixed- effects regression model, followed by a discussion of the random effect of verb lemma, and conclude with a random forest analysis.

4.2.1 Regression model

The purpose of the mixed-effects regression model is to determine how the six individual factors described above collectively influence the choice of a possessive and to identify any lexically specific idiosyncrasies related to the verb lemma and the possessum lemma. We used the glmer () function in R (R Core Team, 2023) to fit a generalized linear mixed-effects model. The model aimed to predict the type of possessive (reflexive vs. non-reflexive) from six fixed effects (address form, imperative type, aspect, referentiality, animacy, alienability) and two random effects (verb lemma and possessum lemma). The graphical exploration of the data in the previous section revealed that the influence of aspect, animacy, and alienability differs between $ty$- and $Vy$-contexts. Therefore, we included the interaction of address form with aspect, animacy, and alienability. The reference values for our model are as follows: non-reflexive for the possessive type, hearer-oriented for the imperative type, imperfective for aspect, non-referential for referentiality, animate for animacy, and inalienable for alienability. The model predicts the likelihood of a RP relative to a NRP when the predictors change from the reference value to another value. The final formula we used is as follows:

• (23) Possessive type ∼ address form + imperative type + aspect + referentiality + animacy + alienability + + address form: aspect + address form: animacy + address form: alienability + (1 | verb lemma) + (1 | possessum lemma).

Table 3 summarizes the results of the final regression model. The second column shows the effect size, with positive values indicating a relative preference for a RP and negative values indicating a preference for a NRP. The intercept represents the situation where the address form is $ty$, the aspect of the imperative is imperfective, and the possessum is non-referential, inalienable, and animate. The intercept estimate is positive, indicating an overall preference for a RP in the dataset. The last column of Table 3 shows the p-values, which indicate the significance of the effect of the predictor on the choice of possessive. Three predictors (address form, imperative type, and referentiality of the possessum) significantly influence the likelihood of a RP versus a NRP. The estimates for these three factors indicate a decrease in the likelihood of a RP when the address form is $Vy$ (compared to $ty$), the imperative is speaker-oriented (as opposed to hearer-oriented), and the possessum is referential (vs. non-referential).

Table 3 The output of the logistic regression model predicting reflexive vs. non-reflexive possessive. *** indicate significance < 0.001, ** < 0.01, and * < 0.05

The three other predictors, namely aspect, animacy, and alienability, do not show any significant independent influence. However, the interaction of aspect and alienability with the address form is significant. This indicates that aspect and alienability do have a significant impact on the possessive choice, but their influence differs with the $ty$- and $Vy$-address forms. Based on the graphical inspection of the data in the previous section, we interpret this interaction as showing that the aspect and alienability are significant predictors for the possessive choice only in $Vy$- but not in $ty$-contexts. The negative estimate for the interaction of address form and aspect indicates that in $Vy$-contexts, a RP is less likely with the perfective aspect compared to the imperfective aspect, and with alienable compared to inalienable possessa. As the main effect of animacy and its interaction with the address form are not significant, it can be concluded that animacy is not a significant predictor of the possessive choice in the given data set.

Let us now inspect random effects given in the bottom part of Table 3. Including random effects of verb and possessum lemmas in our model allows the intercept to be adjusted to each verb and possessum lemma separately. The model with such random intercepts assumes that the effects of all predictors are the same across all lemmas, but it allows different lemmas to have different baseline preferences for a RP vs. a NRP. A standard deviation of a random predictor (verb and possessum lemma in our case) specifies how variable lemmas are with respect to their baseline preferences.

In our model, a standard deviation was estimated at 0.97 for the verb lemma and at 0.39 for possessum lemma. It means that the likelihood of the reflexive vs. non-reflexive possessives with individual verbs vary from the average choices with a standard deviation of 0.87 log odds, indicating that particular verb lemmas have different preferences for the possessive type.There is also a sizable random effect of the possessum lemma, as indicated by a standard deviation of almost 0.4 log odds. This means that the baseline preferences for a RP vs. a NRP are not the same for different possessum lemmas. We used the anova () function to assess whether both random effects significantly improve the predictive power of the model. The test showed that excluding the verb lemma leads to a significantly inferior model, while excluding the possessum lemma does not significantly affect the model’s quality. Therefore, the verb. lemma must be included in the final model in addition to the fixed effects.

In summary, the model that best predicts the choice between a RP and a NRP includes the fixed effects of address form, imperative type, verbal aspect, referentiality, and alienability of the possessum, as well as the random effect of verb lemma. We will now assess how well this model fits our data. A RP was observed in 72% of the examples in our data, meaning that a null model that simply guesses a RP every time would be correct in 72% of cases. The regression model must be evaluated against this baseline. It correctly predicts 89% of all cases, with 77% accuracy for NRP uses and 93% accuracy for RP uses, which is a significant improvement over the baseline model.

Other parameters also indicate a good fit for our model. The conditional R-squared value shows that the fixed and random effects in our model account for 74% of the variance in the data. The marginal R-squared value indicates that fixed effects alone account for 67% of the variance in the data. The C-score measure of goodness of fit for the model is 0.92 and. Somers’ Dxy rank is 0.83, which suggests a good model.

We interpret these facts to indicate that the address form, imperative type, verbal aspect, referentiality and alienability of the possessum constrain the use of a RP and a NRP to a considerable degree. A NRP is more likely in imperative utterances when the interlocutor is addressed with the $Vy$-form, when the imperative is speaker-oriented, and when the possessum is referential. Additionally, a NRP with the $Vy$-address is more likely when the verb is perfective and when the possessum is alienable. Although these constrains are valid for all verb lemmas, some verbs have a higher preference for one of the possessives.

4.2.2 The random effect of the verb lemma

The ranef () function from the lme4 package in $R$ was used to determine the estimated deviation of each verb’s average preference for a NRP and a RP from the overall average preference. Table 4 displays ten verb lemmas (out of a total of 320 lemmas), with five indicating a preference for a NRP (on the left side of the table) and five indicating a preference for a RP (on the right side of the table). The higher the estimated deviation, the larger is the difference between the average preferences for a NRP and a RP predicted by the model and the preference shown by a particular verb. Negative values indicate a preference of a verb lemma for a NRP, while positive values indicate a preference of a verb for a RP. Overall, the estimated values for verbs with a NRP are higher than for verbs with a RP, suggesting a stronger tendency for specific verbs to favor a NRP.

Table 4 Verb lemma that show a particular strong preference for a NRP (left) or a RP (right)

Table 4 shows that the imperfective imperative of davat’ ‘give’ strongly prefers a NRP, as illustrated in 4.2.2 for the $Vy$-address, and in (24) for the $ty$-address. Notably, 60% of all examples where a NRP was used with the $ty$-address contain the verb davat’. The perfective lemma dat’ also favors a NRP, but overwhelmingly in $Vy$-forms, as illustrated by (25). Another verb that strongly prefers a NRP in $ty$- and $Vy$-contexts alike is pokazat’ ’show’, as demonstrated in (26) for the $Vy$-address and in (27) for the $ty$-address.

• Sotrudnik biblioteki Zdravstvujte!

  [Studentka] А možno knigi sdat’?

  [Sotrudnik biblioteki] Davajte vaš čitatel’skij bilet!

  [Studentka] Voz’mite. [Razgovor, 2007]

  ‘[Librarian] Hello!

  [Student] Can I check out my books?

  [Librarian] Give me your.nrefl library card!

  [Student] Here you go. [Conversation, 2007]’

• S. Tak / vino s čto nalit’?

  [K.D.] Vik! Davaj tvoj bokal. [Domašnie razgovory, 2000–2003]

  ‘[S] So / where to put the wine in?

  [K.D.] Vic! Give me your.nrefl glass. [Household chat, 2000–2003]’

• Zritel’ Zdravstvujte.

  [Fokusnik] Zdravstvujte / dajte / požalujsta/ vaš galstuk. Snimite. [Karnaval’naja noč‘, k/f, 1956]

  ‘[Audience member] Hello.

  [Magician] Hello/ give/ please/ your.nrefl tie. Take it off. [Carnival Night, film, 1956]’

• Učastnik zasedanija 2 Togda ja predlagaju prervat’ razgovory i … posmotret’ na vsё ėto sobstvennymi glazami.

  [Učastnik zasedanija 3] Nu čto ž / pokažite nam Vašu plёnku / očen’ interesno. [Solaris, к/f, 1972]

  ‘[Session participant 2] Then I suggest we stop talking and ... see it with our own eyes.

  [Session participant 3] Well, show us your.nrefl tape. It’s very interesting. [Solaris, film, 1972]’

• Nataša Моžno ja eё pojdu posmotrju?

  [Evgenij Semёnovič] Da požalujsta / nо…

  [Nataša] Kisen’ka / kisjulja. Pokaži tvoi lapočki. Ty uže vyspalas’? Ja tebja ščas pomoju.

  [Podkidyš, к/f, 1939]

  ‘[Nataša] Can I go see it?

  [Evgenij Semёnovič] Yes please, but...

  [Nataša] Kitty, kitty. Show your.nrefl paws. Have you slept well? I’ll give you a bath.

  [The Foundling, film, 1939]’

• Agaf’ja Lukinična Vot tak ono/ Borisovna / vsegda i polučaetsja. Xočeš’ ljudjam polučše/ a vyxodit zrja.

  [Gruzčik] Polučajte / xozjajuška / vaši fikusy.

  [Aktrisa, к/f, 1942]

  ‘[Agaf’ja Lukinična] That’s how it always turns out, Borisovna. You want to do better for people, but it turns out to be for nothing.

  [Loader] Take your.nrefl ficuses.

  [The Actress, film, 1942]’

Four out of five verbs that prefer a NRP (davat’ ‘give-ipfv’, otdat’ ‘give back-pfv’, pokazat’ ‘show-pfv’, ‘dat’ ‘give-pfv’) seem to be the ones where the action is directed to a state where the listener is no longer the possessor and therefore strongly benefits the speaker. The fifth verb that prefers a NRP is polučat’ ‘get-ipfv’, which is a converse antonym of davat’. An example of an utterance with this verb is given in (28). The verb polučat’ implies an action that does not normally require active participation from the interlocutor, contradicting the overall meaning of the imperative form as an admonition to action. In fact, the imperative form of polučat’ is 40 times less frequent in the RNC than the imperative of davat’. When used in the imperfective aspect, the imperative form of polučat’ conveys a close-up and subjective view of an action, presenting it as something special that deserves attention. A NRP, which implies a referential uniqueness of the possessum, might be preferred with this verb to highlight the unusualness and uniqueness of the event described.

The right half of Table 4 lists five verbs that prefer a RP. Among them is the verb to take back, represented by the perfective form zabrat’, as illustrated in (29) for 2P.SG and its imperfective counterpart zabirat’, as exemplified in (30). Both verbs denote an action directed towards a state where the listener is the possessor and therefore the one who is likely to benefit from the action.

• (30) [Saenko] Ty ne naxal’ničaj / ja te na glotku ne nastupaju! Zabiraj svoё baraxlo!

  [Zagirov] I zaberu! [Vremja, vperёd, к/f, 1965]

  ‘[Saenko] Don’t get cocky, I’m not stepping on your throat! Take your.refl stuff!

  [Zagirov] I will take-ipfv! [Time, forward, film, 1965]’

• (31) [Maksim Perepelica] Po dvadcat’ kopeek za štuku platit’ budem!

  [Javdoxa] О!

  [Maksim Perepelica] A vy eščё derёtes’. Zaberite svoj buket. V drugom meste najdёm i polučše. A za oskorblenija i poboi pered sudom otvetite. [Maksim Perepelica, к/f, 1955]

  ‘[Maksim Perepelica] We’ll pay twenty kopecks a piece!

  [Javdoxa] Oh!

  [Maksim Perepelica] And you’re still fighting. Take-pfv your.refl bouquet. We’ll find a better one elsewhere. And for insults and beatings you’ll answer in court. [Maksim Perepelica, film, 1955]’

Two other verbs favoring a RP are posmotret’ ‘look-pfv’ and idti ‘go-ipfv’, both occurring primarily in constructions with an inalienable possessum. The verb posmotret’ was used mostly in a combination with a body part, such as posmotri($te$) na svoi ruki / nogi / nos / volosy ‘look at your hands / feet / nose / hair’. The verb idti was used in two constructions: 1) go into one’s space (palata ‘ward’, komnata ‘room’, krovat’ ‘bed’) and 2) go to one’s people often referred with proper names (gosti ‘guests’, Elka ‘Elka’, Ljuda ‘Ljuda’). Another verb favoring a RP was the verb poryt’sja ‘go through’, which is strongly associated with an action upon personal belongings such as things, clothes, wallet, and therefore with an inalienable possessive relation.

To sum up, the verb lemma influences the possessive choice. Verbs that favor a NRP express either an action that strongly benefits the speaker, or an action that somehow stands out from the usual and expected. Verbs that show a preference for a RP either convey actions that benefit the listener, or involve an inalienable possessive relation.

4.2.3 Random forests analysis

To assess the relative importance of the predictors for the choice between a RP and a NRP, we applied conditional inference trees and conditional random forests (Breiman, 2001), non-parametric methods that have recently become popular in corpus linguistics (Levshina, 2020). Specifically, we used the conditional variable importance measure from the Cforest () function of the party package in $R$ (Strobl et al., 2009). These techniques achieve the same aim as the logistic regression (predict the choice between a RP and a NRP) but in a different way. The algorithm of random forests works through the data and, by trial and error, tries to establish if an independent variable is a useful predictor. It estimates the likelihood of the value of the response variable (RP vs. NRP) based on a series of binary questions about the values of the predictors.

In the first step, we built a single conditional inference tree for the data with all predictors (address form, imperative type, aspect, referentiality, animacy and alienability of the possessum and verb lemma) included. Figure 7 displays the results, showing which factors contribute to the choice of the possessive and how. The analysis confirms that the address form is the most important predictor. It also reveals that the referentiality of the possessum is a relevant factor for both $ty$- and $Vy$-address forms, whereas the type of imperative is an important factor only for $Vy$-forms. When the address form is $ty$ and the possessum is non-referential (node 11 of Fig. 8), only a RP is used. When the address form is $ty$, and the possessum is referential (node 10), there is a likelihood of around 10% that a NRP will be chosen. When the address form is $Vy$ and the possessum is non-referential, and the imperative is hearer-oriented (node 5), there is a very small likelihood (less than 10%) of a NRP. However, when the imperative is speaker-oriented (node 4), there is a 30% chance of a NRP. When the address form is $Vy$, the possessum is referential, and the imperative is hearer-oriented (node 8), a NRP is used in slightly less than 40% of cases. However, when the imperative is speaker-oriented (node 7), a NRP is preferred to 80%.

Fig. 7

Conditional inference recursive partioning tree for all predictors

Fig. 8

Conditional permutation variable importance for the random forest with all predictors

In the following step, we utilized conditional random forests, an ensemble technique that generates multiple individual trees and averages their predictions. This method typically yields more precise predictions than single tress like the one in Fig. 7 for our data (Levshina, 2020). The output of the conditional random forest analysis is displayed in Fig. 8. What it shows is that the address form is by far the most important predictor of the possessive choice in Russian. The next most important predictor is referentiality of the possessum. Imperative type is the third most influential factor followed by verb lemma and verbal aspect. The effect of alienability is minimal, and animacy does not contribute statistically significant effects, as indicated by the variable importance equal to zero.

In conclusion, the analysis, using conditional inference trees and random forests techniques, confirms that the address form is the most significant predictor of the choice between RP and NRP in 2P imperatives. The second most important predictor is the referentiality of the possessum. Imperative type is the third most important variable, followed by verb lemma. Verbal aspect is a less influential predictor, and the effects of alienability and animacy of the possessum are negligible.

5 Discussion

The first question we addressed in this study was how frequent a RP and a NRP is used in Russian imperative utterances. Based on previous scholarship, we expected a NRP to occur more often in 2P formal $Vy$-contexts than in 2P informal $ty$-contexts. This hypothesis was borne out. When the interlocutor was addressed with the informal $ty$, a NRP was used only in 6% of examples. When the interlocutor was addressed with the formal $Vy$, a NRP was chosen in 54% of cases. A low frequency of a NRP in the informal 2P utterances in our study confirms the results of Timberlake (2004), Tiskin (2019), and Pekelis (2021). The frequency of a NRP in formal 2P contexts in our study was slightly higher than in the estimations of Tiskin (2019) and Timberlake (2004) for 2P.PL imperatives without a distinction between formal and informal readings. This difference may be due to two reasons. First, there might be a general preference in Russian to use a NRP in 2P.SG formal contexts more often than in 2P.PL informal contexts. Unfortunately, we cannot verify this claim, as we did not consider 2P plural readings, leaving it for future studies. Secondly, it may be the case that a NRP is generally more frequent in spoken rather than written language.

This interpretation is supported by the fact that not only was the frequency of NRPs in formal contexts higher in our spoken data than in the written data of Timberlake (2004) and Tiskin (2019), but also the frequency of NRPs in informal singular readings of our spoken data was slightly higher than in the written data of Timberlake (2004), Tiskin (2019), and Pekelis (2021).

The frequency facts suggest that in Russian, the process of establishing a RP as the only form for expressing subject coreference (Večerka, 1979) has almost been completed in 2P.SG informal contexts. In singular 2P and 1P contexts, the possessive system of Russian might have achieved a symmetry to the system of personal pronouns, where the reflexive sebja is the default option for expressing subject coreference in all persons and numbers. However, the plural 1P and 2P contexts and, as we have shown in this study, the singular formal 2P contexts seem to be resisting this diachronic development and remain sites of variation where a NRP is used alongside with a RP.

Why should this variation be preserved exactly in 2P.SG formal contexts and what factors constrain the choice between a RP and a NRP there? This was the second question addressed in the present study. We hypothesized that three semantic factors (referentiality, animacy, and alienability of the possessum), and three pragmatic factors (address form, imperative type, and verbal aspect) condition the choice. The output of a generalized mixed effect model confirmed that referentiality is a significant predictor in both formal and informal contexts. Alienability has a minimal impact in formal contexts, while animacy does not have a significant effect. Referentiality and alienability of the possessum influence the possessive choice in the direction predicted by our hypothesis 2. A NRP is favored when the possessum is alienable and referentially unique.

Regarding pragmatic factors, our regression model showed that the choice between possessives was significantly influenced by address form. Imperative type and verbal aspect only emerged as important predictors in formal contexts. As expected in hypothesis 3, a NRP was more frequent with perfective than with imperfective imperatives. Hypothesis 4 was also confirmed, as a NRP was more frequent with speaker-oriented than with hearer-oriented imperatives.

In addition to the main effects of semantic and pragmatic predictors, we observed a significant random effect of verb lemma. This indicates that while the main effects were consistent across all verbs, individual verbs exhibited varying baseline preferences either for a RP or for a NRP. Further analysis revealed that verbs with a general meaning of give tended to favor a NRP. On the other hand, the verb with the meaning of take, as well as some verbs expressing actions over inalienable possessions, preferred a RP.

Our analysis of the interaction between these seven features revealed that address form is by far the most important factor in the choice between a RP and a NRP. When the hearer is addressed with the informal $ty$, a RP is the default possessive. A NRP can only be used with referential possessa or with some verbs, especially with the imperfective verb davat’ ‘give’. If the hearer is addressed with the formal $Vy$, both RPs and NRPs can be used depending on the combination of the following factors: the referentiality and alienability of the possessum, the type of imperative, verbal aspect, and verb lemma. Among these factors, referentiality of the possessum exerts the strongest influence on the choice, followed by the imperative type and verb lemma. Verbal aspect is a minor factor, and the influence of alienability is negligible. It means that a NRP with $Vy$-forms is preferred when the possessum is referential, when the imperative is speaker-oriented, and, to some extent, when the verb is in the perfective aspect and the possessum is alienable.

A large weight of referentiality both in formal and informal 2P contexts indicates that a RP and a NRP are recruited to convey different semantic meanings. When using a NRP, the speaker construes the possessive relation as highly individuated, referentially unique, and definite. On the other hand, when using a RP, the speaker foregrounds the co-varying, sloppy interpretation of the possessive relation stressing the referential non-uniqueness of the possessum.

In addition, Russian speakers appear to use the distinction between a RP and a NRP to express politeness towards the addressee in 2P formal contexts. Following Brown and Levinson (1987), by politeness we mean a set of redressive actions aimed at minimizing the illocutionary force of speech acts threatening positive or negative face of the addressee. In our case, we suggest that the use of a NRP is a strategy of negative politeness aimed at reducing the imposition of the requested action on the listener. The NRP Vaš having originated from the 2P.PL form, allows the speaker to address the listener indirectly by distributing the responsibility for the requested action among the plurality of addressees. The strategy of avoiding direct reference by implying a plurality of addresses was shown to be responsible for the development of 2P.SG polite pronouns from 2P.PL pronouns in European languages in the first place (Helmbrecht, 2003, p. 195). As argued by Helmbrecht (2003), the avoidance strategy was first applied in face threatening acts and subsequently spread to other speech acts. A connection between an avoidance strategy expressed by a NRP and a face-threatening act is also present in our study. We found that the NRP Vaš is more commonly used in speaker-oriented imperatives, which pose a greater threat to the listener’s negative face than hearer-oriented imperatives, where the requested action is in the hearer’s interest. A NRP might be preferred in speaker-oriented imperatives because the need for a redressive action aimed at saving the negative face of the listener is the strongest in speech acts presenting a larger threat to the listener’s negative face needs (Helmbrecht, 2003). In this respect, it is noteworthy that in both formal and informal contexts of our study, a NRP was preferred with the verbs of giving, which describe an action that tends to benefit the speaker and can be face- threatening for the hearer. Expressing negative politeness, a NRP Vaš is often combined with the perfective aspect, which is another strategy to save the negative face needs of the listener (Dicky, 2020). Both the NRP Vaš and the perfective aspect give the listener more freedom whether to perform the action requested by the imperative, the perfective aspect by requesting the addressee to make the choice to perform the action and a NRP by giving the addressee the choice of not registering a face-threatening act.

While the NRP Vaš is recruited as a means of (negative) politeness, the RP svoj can be used to express impoliteness in combination with other means, as in examples (29) and (30) above. Like Honselaar (1986, p. 244), we suggest that the impoliteness interpretation of the RP may arise by pragmatic inference from the core semantic meaning of the RP, referential non-uniqueness, in contexts where a referentially unique use of the corresponding noun phrase is expected. For example, the context of (30) makes it clear that there is a specific bouquet in the speech situation to which the speaker is referring. By choosing the RP, the speaker invokes the meaning of referential non-uniqueness, which is reinforced by the mention of another bouquet in the following sentence. By violating the expected referentially unique interpretation of the noun phrase, the speaker evokes in the hearer a negative attitude toward the speaker’s behavior, which causes or is presumed to cause offence (Culpeper, 2010, p. 3233). The violation of the expected meaning of referential uniqueness is especially evident with proper names modified by a RP, as in 5, (32), and (33).

• Kol’ka Zapisalis’ / značit?

  [Drankin] Kol’, tak ved’ interesno že!

  [Kol’ka] Interesno? Nu i katites’ k svoemu Serёžen’ke! Hodite pered nim na zadnix lapkax!

  [Drug moj, Kol’ka! к/f, 1961]

  ‘[Kol’ka] Signed up?

  [Drankin] Kolja-voc / it’s interesting, isn’t it?

  [Kol’ka] Interesting? Well, roll away-ipfv to your.refl Sergei.dim! Walk on your hind legs in front of him! [My friend, Kol’ka! film, 1961]’

• Saša Šust I my stadion postroim. Nu požalujsta!

  [Andrej] Da pošli vy naher, ezžajte v svoju Angliju obratno!

  [Garpastum, к/f, 2005]

  ‘[Saša Šust] And we’ll build a stadium. Please!

  [Andrej] Fuck you, go back-ipfv to your.refl England!

  [Garpastum, film, 2005]’

• Kostja Da netu zdes’ Oli / ponimaeš’! Ne živёt ona zdes‘ / god uže! Ne živёt!

  [Žena Kosti] Idite iščite svoju Olju gde-nibud’.

  [Andrej] Mne eё nužno uvidet‘. Mne eё nužno / nužno uvidet‘ sejčas.

  [Žena Kosti] Vot i idite / i… iščite… gde-nibud’.

  [Ženskaja sobstvennost’, к/f, 1999]

  ‘[Kostja] Olja isn’t here, you know! She hasn’t lived here for a year! She doesn’t live here!

  [Kostja’s wife] Go-ipfv find-ipfv your.refl Olja somewhere.

  [Andrej] I need to see her. I need / need to see her now.

  [Kostja’s wife] There you go / and... look... somewhere.

  [Ženskaja sobstvennost’, film, 1999]

In 5 the impoliteness of the imperative utterance is created by several means: by the combination of the particles nu i, by the choice of the verb lexeme (to roll away instead of the neutral to go), by the diminutive form of the proper name, and by the modifying this name with the RP svoj. The use of the RP portrays the person bearing the name as a member of a set rather than as a referentially unique individual, which violates the expectation of referential uniqueness of a proper name and thus contributes to the effect of impoliteness intended by the speaker. In (32) the imperative is preceded by an expletive. In (33) there seems to be no linguistic means other than an RP in the imperative clause that contribute to an impolite interpretation of the imperative. In our data, a RP but not a NRP is used in examples like 5–(33), most often with an imperative of an imperfective verb. It seems that in such examples we have the uncompromising usage of imperfective imperatives discussed in Sect. 2.4. In this usage, “the speaker has already chosen what the listener should do, and is not making the face-saving gesture of giving him/her the opportunity to make the choice to carry out the action”, which makes such imperatives invariably impolite (Dicky, 2020, p. 564). In terms of speaker/addressee orientation, impolite imperatives like 5–(33) could probably be seen as a special case of addressee-oriented imperatives, in which the speaker orders the addressee to perform an action that the speaker knows to be beneficial for the addressee, but at the same time (more or less) undesirable for the speaker. At least such an interpretation is likely in (33), where the addressee Andrej explicitly states his need to see Olja, and the speaker, Kostja’s wife, orders him to perform the action of finding Olja, which should satisfy the addressee’s need. At the same time, the speaker communicates that she does not approve of this action. From our data, it seems that only a RP is used in such imperatives and apparently contributes to their impoliteness by violating the contextual expectation of referential uniqueness.

The final question we want to discuss here is why the NRP tvoj is not used for politeness purposes in the way that the NRP Vaš is. We suggest that the answer lies in the distinction between positive and negative politeness, and, more precisely, in the contrast between (in)directness vs. (in)explicitness of linguistically coded messages as described by Wheeler (1994, p. 158). Positive politeness, according to Brown and Levinson (1987, p. 101), “is redress directed to the addressee’s positive face, his perennial desire that his wants (or the actions/acquisitions/values resulting from them) should be thought as desirable. Redress consist in partially satisfying that desire by communicating that one’s own wants (or some of them) are in some respects similar to the addressee’s wants”. The basis of positive politeness is the feature “presuppose/raise/assert common ground”, the most important manifestation of which is ellipsis. Ellipsis is based on the “the presupposition that the Hearer will contribute actively in interpreting (inexplicit) linguistic messages and that making such a presumption will be positively evaluated socially. That is, by being inexplicit – using ellipsis and contraction, say – a speaker gives attention to the positive face of Hearer by requiring Hearer to infer that Speaker regards Hearer as sufficiently familiar to be able to supply the missing information” (Wheeler, 1994, p. 158). On the other hand, there is negative politeness – “redressive action addressed to the addressee’s negative face: his want to have his freedom of action unhindered and his attention unimpeded. It is the heart of respect behavior, just as Positive Politeness is the kernel of familiar and joking behavior ” (Brown & Levinson, 1987, p. 129). One of its linguistic realizations is conventional indirectness (Brown & Levinson, 1987, p. 130). Wheeler (1994) points out that there is a crucial difference between inexplicitness and indirectness of a linguistic message. An inexplicitly coded message relies on the presumed shared knowledge of the interlocutor, whereas an indirect formulation of a message relies only on the inferential ability of any listener based on the general cooperative principle. This results in different goals of indirectness and inexplicitness. An indirect message aims to give the hearer the (apparent) option of not registering the face-threatening act. The goal of inexplicitness is “to remind Hearer that social distance is so small that there is very little that would count as an FTA” (Wheeler, 1994, p. 158). Applying this distinction to the choice between a RP and a NRP, we suggest that the use of a RP in an imperative utterance directed at someone addressed with an informal T-pronoun is a strategy of positive politeness. The redress in this case is directed at the positive face of the addressee and is based on a higher degree of implicitness of the anaphoric possessive svoj compared to the 2P singular deictic pronoun tvoj. Although the reference of svoj is unambiguous in this context, it still requires the addressee to identify the syntactic subject to which it refers and thus to provide the missing information based on the assumed common ground. The assertion of common ground signals the speaker’s attention to the positive face of the addressee and creates a social proximity that implies that FTAs are unlikely. When possessives are used in an imperative utterance directed at someone addressed with a V-pronoun, a NRP expresses negative politeness based on indirectness. The plural semantics of the NRP gives the addressee an apparent freedom to feel addressed by the imperative or not.

It follows that in Russian a NRP could be a strategy for expressing negative politeness in 2P singular honorific contexts, while a RP could be a means of expressing positive politeness in 2P non-honorific contexts. If we assume that language change originates in positive politeness and that the renewal of negative politeness takes place through obsolete forms of positive politeness (Wheeler, 1994, p. 170), we can explain the choice of possessive in contemporary Russian from a diachronic perspective. A RP as a strategy of positive politeness seems to have pushed a NRP out of use in informal casual styles into the realm of negative politeness, where it is now on its way to entering “the most conventionalized set of linguistic strategies for FTA address” (Brown & Levinson, 1987, p. 130).

6 Conclusion and further directions

We have conducted an empirical study of the choice between a reflexive and a non-reflexive possessive in Russian imperative sentences based on data from the spoken subcorpus of the Russian National Corpus. Our goal was to find out how often both possessives are used, what combination of factors constrains their use, and how influential each factor is. In addition to the semantic variables known from the literature (referentiality, animacy, and alienability of the possessum), we also considered three pragmatic factors: address form, verbal aspect, and imperative type.

We found that the address form is the most important factor in determining the choice between the two possessives in Russian imperative utterances. Almost exclusively a reflexive possessive is used with the informal $ty$-address. With a formal $Vy$-address, there is a 50% chance that a reflexive possessive will be used. A non-reflexive possessive is preferred when the possessum is referentially unique and alienable, the imperative is speaker-oriented, and the verb is perfective.

It seems that the process of establishing the reflexive possessive as the default form for coreference with the subject in Russian is almost complete in 2P informal imperative utterances. In 2P formal imperatives, however, the change is still ongoing, and the use of a reflexive or non-reflexive possessive is determined by the referential uniqueness of the possessive relation and the speaker’s politeness strategy. We argued that the polite interpretation might have arisen as a pragmatic inference from the plurality component in the semantic meaning of the non-reflexive possessive Vaš in face-threatening speech acts. This idea deserves a detailed diachronic investigation showing when and in which contexts the polite meaning of a non-reflexive possessive emerged. Furthermore, experimental studies like de Hoop et al. (2016) are needed to confirm that non-reflexive possessives in imperative contexts are indeed perceived as polite by Russian native speakers. In our study, politeness was inferred from the distributional properties of the possessives, but not measured directly. It would also be interesting to explore the dimension of impoliteness and to investigate in which contexts a reflexive possessive is perceived as impolite. Finally, it is desirable to replicate our study on more recent and more spontaneous conversational data than the spoken subcorpus of the RNC.