Lithuanian reflexive-based impersonals with accusative objects.

This paper surveys Lithuanian impersonal constructions with predicative present passive participles containing non-promoted accusative objects. It is shown that the construction, hitherto considered very rare, is well-attested and productive with one verb class, namely, transitive reflexives. In terms of semantics, transitive reflexives in Lithuanian may be classified as autobenefactives. Autobenefactive reflexives do not exhibit a change in argument structure with respect to their non-reflexive counterparts. In the case of autobenefactives, the morpheme -siattached to the verb adds the meaning that the subject, which mostly has the semantic role of an agent, benefits from the event expressed by the predicate. On the basis of corpus data, we have analysed how widespread impersonal constructions with accusative objects are within the domain of transitive reflexives and which pattern—the accusative or the nominative—is dominant when both are attested. Lastly, we briefly discuss the temporal-aspectual meaning of reflexive-based impersonals as well as the referential properties of implied agents.


Introduction
This article surveys Lithuanian constructions containing a predicative passive participle with a non-promoted accusative object. An example of such a construction is given in (1). For comparison, a canonical promoting passive is exemplified in (2).
(1) Renka-m-a-si ne geriausią, o pigiausią choose-prs.pp-na-rfl not best.acc.sg.m but cheapest.acc.sg.m variantą. (ltTenTen14) alternative(m).acc.sg 'Not the best, but the cheapest alternative is being chosen.' (2) Šiais metais buvo pa-si-rink-t-as Žemaitijos this _ year be.pst3 pvb-rfl-choose-pst.pp-sg.m pln.gen regionas. (ltTenTen14) region.nom.sg 'This year the region of Samogitia was chosen.' (1) and (2) differ in several respects: • agent demotion in (2) is followed by object promotion which is not the case in (1). The promoted object in (2) functions as a canonical subject and agrees with the passive participle in number, gender and case, whereas in (1) the direct object retains the accusative case marking, and the passive participle occurs with the neuter ending -a, which is the default form used in the absence of a proper controller of verbal agreement in a clause. In the following we will refer to the ending -a as 'the non-agreeing form'; • the demoted agent may be expressed as an oblique genitive-marked agent phrase 1 in (2), but not in (1) (cf. Ambrazas et al. 2006, 661); • (1) is limited to indefinite or generic human agency (cf. Geniušienė 2016, 123), whereas in (2) no restrictions on the semantics of the demoted agent are imposed;

Constructions illustrated in
• in (1) present passive participles are almost exclusively used, whereas in (2) both present and past passive participles are possible. Because of these formal and semantic differences we would categorize (1) as a '(subject) impersonal construction' distinct from the canonical agreeing passive. We use the term "impersonal" in the sense of 'lacking an (overt or covert) syntactic subject' (cf. Malchukov & Siewierska 2011, 2). We prefer not to use the term "impersonal passive" because, as shown in Nau, Spraunienė & Žeimantienė (2020), what is traditionally subsumed under the category of "impersonal passive" in Lithuanian (which is roughly defined as any construction with a non-agreeing passive participial form) is indeed a cluster of structurally and semantically different constructions, including generic descriptions, the cumulative-retrospective construction, the evidential construction, the impersonal construction etc. A different approach to passives and impersonals is advocated in Blevins (2003). Blevins argues for a strict division between passives and impersonals, treating them as two different types of verbal diathesis. According to Blevins' approach, the agent in an impersonal is not demoted but merely suppressed, which means that its overt expression is inhibited. An impersonal thus contains a covert agent-subject which is syntactically active. Kibort (2006) shows that the existence of such covert subjects in Polish -no/to constructions can be proved by applying syntactic tests for subjecthood. However, in Lithuanian, the usual syntactic tests for subject status work well with canonical subjects, but they do not give reliable results in the case of putative covert subjects (for details see Spraunienė, Razanovaitė & Jasionytė 2015). As the existence of impersonals in Lithuanian (in Blevins' sense) cannot be verified in a way independent of the theory, we prefer treating the construction exemplified in (1) as a member of the "Passive Family" alongside with other predicative uses of passive participles (see Nau, Spraunienė & Žeimantienė 2020). Thus, in our view, Example (1) represents a subtype of subjectless passives where the agent has been demoted from the subject position but the transitive object retains its syntactic status and case form.
Although the earliest attestation of a Lithuanian impersonal with an accusative object is probably from as long ago as 1573, in the manuscript of the evangelical Postilė of Wolfenbüttel 2 , see example (3), it is reported in the liter-ature that this construction is very rare (cf. LKG 1971, II, 53;Geniušienė 2006;Geniušienė 2016, 121-122;Spraunienė, Razanovaitė & Jasionytė 2015): In Holvoet (2001, 376-377) the retention of the accusative object instead of promoting it to subject is viewed as "a natural shift from a passive towards an impersonal construction", i.e. as an independent development in Lithuanian which is considered a typological parallel to the development of Polish -no/-to impersonals. Danylenko (2005, 158), however, argues that the emergence of the accusative object in the impersonal environment did not occur as a "natural shift", but that it should be attributed to "the Polish adstratum interference throughout the East Lithuanian territory". Interestingly, the use of neuter passive participles with accusative-marked objects was "conspicuously widespread" in Lithuanian press in the beginning of 20 th century (Danylenko 2005, 152). However, such constructions were regarded as calques from Polish and were subsequently banned from the emerging standard language. The famous Lithuanian language reformer Jonas Jablonskis started to correct them, replacing the accusative with the nominative, which he considered indigenous (ibid.). Consequently, Jablonskis' famous grammar from 1922 does not contain a single example of an impersonal construction with an accusative object (ibid.). Thus, sporadic occurrence of impersonals with accusative objects in modern standard Lithuanian may be seen as a result of the efficient purification of the Lithuanian language in the beginning of the 20 th century (though they are by no means considered ill-formed in present day grammars of Lithuanian). Nau, Spraunienė and Žeimantienė (2020) have discovered that the impersonal construction with an accusative object occurs far more frequently with one particular verb class, namely, unprefixed reflexive verbs 3 , such as rinktis 'choose (for oneself)' in (1). These verbs are morphologically defective in that they may only form non-agreeing passive participles, the agreeing form being blocked by the verb-final reflexive marker. In spite of this, they may also be used in promoting passives, for example: (4) Aiškina-m-a-si jų asmenybė. (DLKT) find _ out-prs.pp-na-rfl 3pl.poss identity(f).nom.sg 'Their identity is being investigated.' In (4) the initial object appears in the nominative case, which shows that it has been promoted to subject position. Verbal agreement is not mandatory for subjectful passives in Lithuanian. On the basis of corpus data, Nau, Spraunienė and Žeimantienė (2020, 59-62) investigate the use of two reflexive verbs in impersonal constructions, namely, rinktis 'choose (for oneself)' and aiškintis 'find out (for oneself)'. The investigation shows that these verbs behave very differently with respect to the retention of an accusative object in an impersonal construction: with rinktis the accusative marking is nearly as frequent as the nominative (48.1% and 51.9%, respectively), whereas with aiškintis nominative subjects prevail, non-promoted accusative objects constituting only 5.1% of the data. On the basis of their investigation Nau, Spraunienė and Žeimantienė (2020, 61) draw a tentative conclusion that impersonals with accusative objects are spreading within the domain of reflexive verbs. This fact has not been noticed in the literature on the Lithuanian passive before. Neither Geniušienė's most extensive investigation of the Lithuanian passive (1973, the English edition Geniušienė 2016), nor her subsequent monograph on reflexives (1987) mention that impersonal (passive) forms of reflexive verbs should have a predisposition to retaining accusative objects. Nau, Spraunienė and Žeimantienė (2020, 61) admit that more research is required in order to identify factors that determine the use of accusative versus nominative in such constructions. The present article is intended to fill in this gap. On the basis of what has been said, we have formulated the following research questions: i) How widespread are reflexive-based impersonals with accusative objects? (i.e. Do they extend to other lexemes beyond rinktis and aiškintis?) ii) Which (semantic and formal) classes of reflexives are used in impersonals with accusative objects?
iii) What semantic and formal features are characteristic of reflexive-based impersonals? Our investigation was carried out in the spirit of 'the corpus-driven approach' (Tognini-Bonelli 2001). All our observations are based on corpus evidence. Self-constructed examples are only used in the paper to give a textbook example or to illustrate a difference between two forms in a simplified manner. Our empirical material mainly comes from two sources: The Corpus of Contemporary Lithuanian 4 and the web corpus ltTenTen14 5 . In certain cases, a Google search was done in order to complement the corpus data. We have gathered 464 examples of reflexive-based impersonals with accusative objects (with present passive participles). For details on the methods of data collection and processing see section 3. The cited Lithuanian examples are glossed in accordance with the Salos Glossing Rules (Nau & Arkadiev 2015), avoiding unnecessary segmentation. The rest of the article is structured as follows: section 2 contains a short description of reflexives in Lithuanian mainly focussing on the types of transitive reflexives, in section 3 the method for gathering and the structure of the data are presented, section 4 includes an analysis of the data with respect to the research questions. In section 5 we summarize the main findings and present concluding remarks.

Reflexives in Lithuanian
Reflexives in Lithuanian are traditionally 6 considered derivatives which are formed by adding the reflexive marker -si-(or its allomorph -s) to a non-reflexive base verb (Geniušienė 1987, 356)  The reflexive marker -si-is of pronominal origin; it evolved from an enclitic reflexive pronoun which had lost its ability to occupy the position of a syntactic argument (Holvoet 2020, 11). Although the reflexive marker -si-in Lithuanian has developed various extended uses, including the "potential passive" (or "facilitative") use (Geniušienė 1987;Holvoet 2020, 145f), it is never used to express the passive as in Russian and Scandinavian languages. Constructions with reflexive verbs comprise a heterogeneous set in terms of meaning, productivity and formal properties: while some are clearly lexicalized and belong to the lexicon, others are highly productive and have developed grammatical functions pertaining to the domain of voice 8 . Therefore, the marker -si-may be regarded as a borderline case between inflection and derivation.
There are still traces of the enclitic origin of the reflexive marker in Lithuanian: it may change its position in the verbal stem. If a reflexive verb contains no prefix, the reflexive marker -si-occurs verb finally, as in (5). In reflexive verbs with prefixes, including the verbal negator ne-, the reflexive marker -si-moves to the position between the prefix(-es) and the root, cf. (7). This behaviour shows that -si-is not yet fully grammaticalized as an affix 9 .
(7) [Ir man pasisekė, radau komentarus apie puikią, nuostabią dietologę,] už-si-registravau konsultacijai ir laukiau. (ltTenTen14) pvb-rfl-register.pst1 consultation.dat.sg and wait.pst1 '[And I was lucky to find comments about a nice, wonderful dietitian,] I registered for a consultation and waited.' Prefixes or, more specifically, lexical preverbs 10 , in many cases contribute to the perfective 11 meaning of the verbal lexeme, e.g. užsiregistravau in (7) refers to a completed event in the past. It has to be noted that from the derivational point of view, the prefixed reflexives may be of two types: i) prefixed reflexives that are formed from unprefixed reflexives, e.g., pasikviesti 'invite' ← kviestis + pa-. ii) reflexives with lexicalized prefixes which do not have unprefixed counterparts 12 and therefore can only be analysed as reflexive derivatives of non-reflexive prefixed verbs, e.g., įsigyti 'acquire'← įgyti + -si-. Importantly, the ma-forms of prefixed reflexives are not morphologically defective-they can form agreeing passive constructions (see examples (19-20)).
As most reflexives exhibit a valence decrease with respect to the non-reflexive base verb (this is illustrated by the example pair (5-6)), the majority of reflexives are intransitive. Depending on which argument, direct object, subject or indirect (dative) object, is deleted as a result of valence decrease in reflexive formation, three major syntactic classes of reflexives are distinguished: subjective (e.g. Durys atsidarė 'The door opened'), objective (e.g. Šuo kandžiojasi 'The dog bites') and transitive (dative) reflexives (Geniušienė 1987). Only the latter class will be discussed in detail in the present paper as it may serve as input for impersonals with accusative objects.
Dative transitive reflexives constitute the second largest syntactic class of Lithuanian reflexives. They comprise 28.5% of all reflexive verbs in a list generative framework. For a detailed discussion about the status of the reflexive marker see also Holvoet (2016, 456-460). of reflexives, compiled by Geniušienė, which consists of 3,680 lexemes (the figures are derived from Geniušienė 1987, 73, Table 2). According to Geniušienė (ibid.,[129][130][131][132][133], dative transitive reflexives derive from bivalent nonreflexive verbs with an optional indirect object of Possessor or Beneficiary. A subset of dative transitive reflexives derives from trivalent non-reflexive base verbs with an obligatory indirect object of Recipient. Geniušienė distinguishes between three semantic types of dative transitive reflexives: a) reflexive-possessive, b) reflexive-benefactive, c) reflexive-recipient. In the non-reflexive base verb (8) the referent of the direct object stands in an inalienable (or quasi-inalienable) possession relation to the human referent of the indirect object. In the reflexive verb (9) the inalienable possession relation is established between the Agent and his/her body part (Geniušienė 1987, 130).
Reflexive-benefactive transitive reflexives are labelled so because the subject of the reflexive verb is both an agent and a beneficiary of the event expressed by the verb (ibid., 131), for example: Reflexive-recipient transitive reflexives constitute a small set of verbs denoting putting on and taking off clothes, shoes, etc. The subject of the reflexive is both the agent and the recipient of the predicated situation. An example pair is given in (12-13): (12) Ona už-movė vaikui pirštines. pn.nom pvb-put_on.pst3 child.dat.sg mitten.acc.pl 'Ona put the child's mittens on for him.' (13) Ona už-si-movė pirštines. pn.nom pvb-rfl-put_on.pst3 mitten.acc.pl 'Ona put on her mittens.' Geniušienė (ibid., 134-135) also mentions several cases of non-dative transitive reflexives. These reflexives exhibit no valence change with respect to the non-reflexive base verb. The reflexive marker may be added to a trivalent non-reflexive base verb without deletion of the datival object which plays the semantic role of a recipient, e.g.: (14) Jis niekam savo bėdų 3.nom.sg.m nobody.dat.sg rposs trouble.gen.pl ne-pasakoja. (Geniušienė 1987, 134, our glossing) neg-tell.prs3 'He does not tell anybody of his troubles.' (15) Jis niekam savo bėdų 3.nom.sg.m nobody.dat.sg rposs trouble.gen.pl ne-si-pasakoja. (Geniušienė 1987, 134, our glossing) neg-rfl-tell.prs3 'He does not tell anybody of his troubles.' ('He does not unburden himself to anybody.') There is a subtle meaning difference between (14) and (15): the reflexive variant of the verb refers to a person's need to release the emotional burden by telling about it to someone. Thus the subject-agent of (15) is also a beneficiary of the event predicated by the verb. In other cases the reflexive marker is added to a non-reflexive base verb which may not have an optional indirect object at all. Consider examples (16) and (17): (16) Jonas at-vedė vaiką į pn.nom pvb-bring.pst3 child.acc.sg to mokyklą. (Geniušienė 1987, 135) school.acc.sg 'Jonas (has) brought the child to school.' (17) Jonas at-si-vedė vaiką į pn.nom pvb-rfl-bring.pst3 child.acc.sg to mokyklą. (Geniušienė 1987, 135) school.acc.sg 'Jonas (has) brought the child with him to school.' The meaning difference between (16) and (17) is similar to the difference between the previous example pair. In (16) Jonas brings the child to the school which the child attends (i.e. he does it for the sake of the child), while in (17) Jonas has an interest in or benefits from bringing the child to school (maybe because he works there and does not have a babysitter).
Holvoet (2020,(15)(16)(17)(18)(19)(20)(21)(22)(23) argues that both dative and non-dative transitive reflexives can be subsumed under a broader semantic category of 'autobenefactives'. He draws attention to the fact that the autobenefactive marker -silargely corresponds to 'free datives' (i.e. datives of interest or datives of external possession) but that it never corresponds to an argument of the verb, e.g. the recipient argument of 'give' and 'tell' cannot be represented by the reflexive marker (ibid., 19). Holvoet states that the autobenefactive marker -si-is therefore not a morphological means of marking the "coreferentiality of two valency positions of the verb" (ibid., 20). The autobenefactive -si-does not change the argument structure of the non-reflexive base verb. In this way there is basically no syntactic difference between what Geniušienė calls 'dative' and 'non-dative' transitive reflexives. Both groups share the autobenefactive meaning: that the subject, which usually plays the semantic role of an agent, has an interest in or in some way benefits from the event expressed by the reflexive verb. Korostenskiene (2017, 461) also states that when -si-is added without valency reduction, it "adds the meaning of the subject performing an action for their own benefit", e.g. pasivaikščioti 'go for a walk', pasiskaityti 'read (for oneself)' are autobenefactives with respect to their non-reflexive counterparts pavaikščioti 'walk for a while', paskaityti 'read for a while'. In some autobenefactives the subject is affected by the event in a negative way: (18) Ona su-si-laužė koją. pn.nom pvb-rfl-break.pst3 leg.acc.sg 'Ona broke her leg.' Autobenefactive reflexives are very productive in Lithuanian; however, not all of them are included in dictionaries. Apparently, the autobenefactive marker -si-may be added to most transitive verbs, especially if they have a perfectivizing preverb as in (7).

Data and method
As stated by Geniušienė (2006, 39), Lithuanian allows forming present and past passive participles of nearly all verbs, including reflexives of some classes. In order to investigate the possible spread of (subject) impersonals with accusative objects into the domain of reflexives, we needed a list of transitive reflexives. As impersonals are basically restricted to the present passive participle, our focus was on the non-agreeing forms of the present passive participle (the so-called ma-forms). We adopted the following procedure: first, using the formula Vgpp--npnn-y-p, a search for ma-forms of reflexive verbs was carried out in the (automatically) morphologically annotated Corpus of Contemporary Lithuanian 13 . The search yielded 42,170 occurrences of 963 lexemes 14 . Of these, 150 were identified as transitive reflexive lexemes, see Table 1 below.  Table 1, transitive reflexives constitute 15.6% of all reflexive lexemes occurring in the ma-passive in DLKT. The obtained sample of 150 transitive reflexives was then manually searched through for uses with accusative objects. In order to get more data, we also checked the ma-passives of each transitive reflexive in the sample in the web corpus ltTenTen14. The latter corpus is not morphologically annotated, so a search for morphological forms the same way as it was done in DLKT was not possible. We chose ltTenTen14 because it is the largest Lithuanian corpus available (it is more than twice as big as DLKT); in addition, it is compiled from internet texts which are less strictly edited and therefore more likely to reflect a spreading innovation. The results of the search are presented in Table 2.
Transitive reflexive lexemes used in ma-passives with an accusative object Transitive reflexive lexemes used in ma-passives without an accusative object Total 45 (30%) 105 (70%) 150 (100%) TABLE 2. Transitive reflexive lexemes having at least a single attested case of impersonals with accusative objects in DLKT and ltTenTen14 As is evident from Table 2, only 30% of transitive reflexives in our sample had at least a single occurrence in an impersonal construction with a nonpromoted accusative object. On the basis of the data, we can assume that impersonals have indeed spread into the domain of transitive reflexives but that this spread has not encompassed all transitive reflexives yet.

Analysis
In this section we will examine more closely which reflexives (the prefixed or the unprefixed ones) are more likely to occur in impersonals with accusative objects and which pattern (the accusative or the nominative one) is dominant when both are attested. We will also give a short description of the reflexivebased impersonal construction in terms of formal and semantic properties (word order, semantics of the demoted agent, temporal-aspectual characteristics etc.).

No.
Verbal  Our data shows that the lexical input of impersonals with accusative objects also comprises prefixed reflexives. As mentioned in section 2, the ma-forms of prefixed reflexives are not morphologically defective; they can be used in canonical agreeing passives with nominative subjects, and this is in fact the dominant pattern with most of these verbs. Examples (19) and (20)  Although prefixed transitive reflexives dominate in our sample (118 out of 150), the proportion of prefixed transitive reflexives which have at least a single occurrence of an impersonal with an accusative object in our sample is much smaller than of that of non-prefixed ones, see Table 4.  The data in Table 4 shows that the accusative pattern is more common among unprefixed reflexives than among prefixed reflexives. In addition, some unprefixed reflexives, which do not have many occurrences in the ma-form in both corpora, seem to favour the accusative pattern over the nominative pattern, and we also found additional examples of the accusative pattern though Google searches, as presented in Table 5 below.

Occurrences in impersonals with accusative objects
As can be seen from the data in Table 5, the ma-forms of unprefixed reflexives vilktis 'put on', vežtis 'carry (with oneself)', vytis 'chase (after somebody)' and kviestis 'invite (to one's place)' have either the same or a larger number of attestations with the accusative pattern than with the nominative pattern. Thus, we think that unprefixed reflexives have a greater predisposition to the accusative pattern than the prefixed ones.  In the previous section we saw that many unprefixed reflexives favour the accusative pattern. In this section we present an analysis of 7 reflexives which had the most attested cases of impersonals with accusative objects in our sample, see Table 6. Our goal was to investigate which pattern is preferred when both are attested.
The figures in Table 6 show that three reflexives, namely, prisiminti 'remember', aiškintis 'find out (for oneself)' and skolintis 'borrow (for oneself)', clearly prefer the nominative pattern when used in the ma-form. The verb rinktis 'choose (for oneself)' has more attestations of the use with an accusa-Examples of past passive participles of reflexives with non-promoted accusative objects are cases of the cumulative-retrospective construction, which operates on past passive participles (for details, see Nau, Spraunienė & Žeimantienė 2020, 99).

Concluding remarks
In this paper we investigated reflexive-based impersonals with accusative objects. We found that impersonals with accusative objects have indeed spread within the domain of transitive reflexives which semantically can be characterized as autobenefactives. The subject of an autobenefactive reflexive benefits from the event expressed by the verb. We assume that the spread of impersonals started with unprefixed reflexives which are morphologically defective: they cannot form the agreeing passive forms due to the verb-final reflexive marker. Our data shows that impersonals with accusative objects have also extended to prefixed transitive reflexives which are not morphologically defective and can form agreeing passive participles. The frequency of the accusative vs. nominative pattern is dependent on the verbal lexeme. Some transitive reflexives, e.g. rinktis 'choose (for oneself)', vilktis 'put on', vežtis 'carry (with oneself)', vytis 'chase (after somebody)', kviestis 'invite (to one's place)' and įsimylėti 'fall in love' have either the same or a larger number of attested cases with the accusative pattern than with the nominative pattern. However, the majority of transitive reflexives (especially the prefixed ones) in our sample clearly favour the nominative pattern over the accusative pattern when used in a ma-form. Unprefixed reflexives seem to have a greater predisposition to the accusative pattern than non-prefixed reflexives. More research is needed in order to determine the factors that determine the choice between the nominative and the accusative pattern in reflexive-based impersonals/passives. The meaning of reflexive-based impersonals is mostly habitual generic, in rare cases they refer to an ongoing event (progressive aspect) or a habitual past event. Occasionally, also past passive participles of transitive reflexives may be used with non-promoted accusative objects. Such uses pertain to the cumulative-retrospective construction.
Thus we see that impersonals with accusative objects, which have been nearly ousted from the Lithuanian language because of standardization processes in the beginning of the 20 th century, have again gained access to language use through -ma forms of transitive reflexives.