Modalities of case assignment: The view from Lithuanian.

This paper assesses two competing modalities for the assignment of morphological case. Arguments are provided from Lithuanian against the configurational strategy of Dependent Case (Marantz 1991, Baker 2015) and in favor of case assignment by functional heads (Chomsky 2000, 2001). The first argument comes from a series of Transitive Impersonal constructions in which accusative appears independently, in the absence of a higher, nominative-marked argument, so long as the predicate is two-place and caused, implicating v-Cause as the source of accusative. Further evidence for this analysis comes from the Inferential Evidential, an oblique-subject construction. While the Dependent Case strategy states that nominative automatically shifts to the object if not assigned to the subject, nominative objects are exceedingly rare in the Inferential Evidential, a fact that is entirely consistent with the local, feature-based theory of case advanced in this paper, which relates the appearance of nominative to the Agree relation with Tense.


Theoretical Background on Structural Case
This paper provides a survey of constructions from Lithuanian that bear on the question of case-assigning modality, namely whether case is assigned by configurational rules or by designated functional heads, at least in the case of Lithuanian. 1 The configurational assignment strategy of Dependent Case is given in (1): (1) Dependent Case Theory 2 a. Assign all noun phrases selected by lexical items that assign idiosyncratic case the particular idiosyncratic "lexical" case in question. b. Next, evaluate all noun phrases that did not receive case in the previous step. If there are two distinct NPs in the same local domain such that NP 1 c-commands NP 2 , then value the case feature of NP 2 as accusative (under nominative-accusative alignment). c. Value as nominative the case feature of an NP that has not been assigned lexical or dependent case in the preceding steps.
The constructions analyzed in this paper display two patterns which are unexpected on these configurational rules of Dependent Case. First is the appearance of accusative on the object in the absence of a higher, structurally-case-marked nominal, as in the Transitive Impersonal construction in (2), which challenges the idea of accusative as necessarily the result of case competition (1b): (2) Transitive Impersonal a. Gelia jam kojas nuo šalčio. sting.3.prs him.dat leg.acc.pl from cold.gen.sg 'His legs ache from the cold.' b. Nuo tabako kvapo mane pykina. from tobacco.gen.sg smell.gen.sg me.acc sicken.3.prs 'I am nauseated from the smell of tobacco.' Next we consider nominative on the object of the Inferential Evidential, an oblique-subject construction. The rule in (1c) provides an apparently straightforward account of nominative objects: nominative occurs on the object, as the non-dependent case, in the event that there is no higher, structurally-marked 227 argument in contrast to which object case can be evaluated. An unexpected fact about the Lithuanian Inferential Evidential is that it occurs predominantly with intransitive verbs. As a result, nominative objects, in point of fact, are exceedingly rare in this construction. 3 An example of the Inferential Evidential with an intransitive predicate is given in (3a) and with a (rare) nominative object in (3b): (3) Inferential Evidential a. Iš plačios erdvės matėsi, kad čia from broad.gen.sg.f space.gen.sg see.3.pst.refl that here didelio medžio augta. big.gen.sg.m tree.gen.sg grow.pst.pp.df 'One could see from the broad empty space that a big tree had apparently grown here.' (Spraunienė et al. 2015, 342, from the Corpus of Contemporary Lithuanian) b. Vagies nusikirsta visi kopūstai. thief.gen.sg cut.pst.pp.df all.nom.pl.m cabbage.nom.pl 'Evidently a thief had cut down all the cabbages.' (Geniušienė 2006, 31) I argue on the basis of such constructions against the configurational strategy of Dependent Case. The logic of the argument is to challenge both core predictions of Dependent Case Theory, given in (1b) and (1c), namely that accusative occurs without a case competitor (2) and that nominative fails to occur (productively) in the absence of a case competitor (3b). I evaluate Dependent Case for Lithuanian against the competing theory of structural case assignment by functional heads, namely finite and agreeing T 0 for nominative and a transitive v-head for accusative, as in the standard generative model outlined in (4): (4) Case Assignment by Functional Heads (Chomsky 2000, 121-124) If a functional head F ∊ {T, v} has unvalued phi-features and an NP X has an unvalued case feature (and certain locality conditions hold), then agreement occurs between F and X, resulting in the agreement features of X being assigned to F and the case associated with F (nominative for T 0 , accusative for v 0 ) being assigned to X.
The idea of case assignment by functional heads was initially developed as a system of case by agreement. Agreement with T 0 , for example, values the phi-features on the functional head while simultaneously assigning nominative to its agreement target (normally the subject), as described in (4). In the event that the phi-probe in T 0 fails under default agreement (df), I argue that there is no source for case on an otherwise appropriate agreement target, a condition I take up in the discussion of nominative objects. 4 The discussion proceeds as follows. In Section 2, I review Transitive Impersonal constructions in Lithuanian with an eye toward identifying the source of accusative in relation to a particular v-head (rather than in relation to a particular configuration). Pylkkänen (2008), for example, distinguishes a higher VoiceP as the projection that introduces the external argument from a lower v which introduces causative semantics (see also Harley 2013, Legate 2014, and Alexiadou et al. 2015. I argue that the lower v-Cause is the source of accusative on the object argument, regardless of the properties of the higher Voice head, such that it is possible for accusative to occur in the absence of a higher (nominative) argument so long as the predicate is causative. In Section 3, I review a related challenge to Dependent Case based on the Lithuanian Inferential Evidential. This is an oblique-subject construction, which, according to Dependent Case Theory, should assign nominative to the object NP, since this is the first (and only) structural case deployed in the clause. While the nominative object is indeed attested in the Inferential Evidential, this construction, as mentioned earlier, occurs almost exclusively with intransitive verbs. I refer to the preference for intransitives in this construction as a Transitivity Restriction, which I take to indicate the absence of a productive source for object case (see Lavine 2010Lavine , 2014b, contrary to the predictions of Dependent Case (1c). I conclude in Section 4.

Transitive Impersonals in Lithuanian
Transitive Impersonals are constructions in which structural accusative appears in the absence of an Agent or any nominative argument in the same clause (see, most recently, Lavine and Babby 2019). The Transitive Impersonals given in (2a-b) are repeated below in (5)-(6), along with additional examples in (7)

230
In what follows, I analyze the source of accusative in the Transitive Impersonals in (5)-(8). I argue that Voice is not argument-projecting in this construction-there is no external argument, overt or implicit-thereby barring the application of Dependent Case as the source for accusative on the object argument. I posit instead the lower, v-Cause head as the source of accusative, following Lavine (2016).

Non-Argument-Projecting Voice
The question of whether there is a higher argument (a case competitor) against which accusative is assigned in (5)-(8), in accordance with the Dependent Case Theory, is not trivial, since the oblique causer in the nuo 'from' PP may be construed as underlyingly basic as a higher nominative subject, as in (9)-(10) (see Lavine and Babby 2019, 805-806, fn. 3, for related discussion). I refer to such constructions as "Derived Transitives" to indicate, as I argue below, that the nonvolitional causer is not basic in the subject position and, it follows, is not an argument of Voice. I present evidence below for a principled distinction between genuine external arguments and VP-internal oblique causers. I show that we are dealing with a case of variable causer realization, which applies to a class of Lithuanian two-place, lexically causative verbs that freely project their causer argument either as a nominative subject or as an oblique causer (typically giving "out-of-control" semantics). The appearance of the oblique causer, when merged VP-internally, gives rise to the Transitive Impersonal, a construction in which accusative appears in the absence of an Agent or any nominative argu-ment. If this analysis is on the right track, there is no source for accusative in (5)-(8) on the Dependent Case model.
An initial distinction between genuine external arguments of Voice and VP-internal causers (complement to nuo 'from') is that the former do not occur felicitously in the VP-internal position, as indicated in (11) The intended "pain" interpretation in the examples in (11) fails because the semantic function of the Transitive Impersonal is incompatible with genuine external arguments. It is only with the VP-internal nonvolitional causer that we get the "pain" reading, as a metaphorical extension of the 'stinging' and 'tearing' events in (11) (Seržant 2013, 199). To be sure, while the nuo PP is optional, I assume, in the absence of any overt valency-reducing morphology, that the causer relation is still an argument and that the predicates in (5)-(8) are therefore two-place, but crucially lacking an argument introduced by Voice (i.e., they are dyadic unaccusatives). That is, when not expressed overtly, the causer still remains "unidentified or ambient" (Holvoet 2013, 276; see also Holvoet & Judžentis 2005, 163-168 for a similar account). The idea is that for each of these verbs, there is one sense, with each argument realization expressing a different construal of the event. Following much recent work in the constructionist approach to meaning, multiple interpretations are generated in the syntax on the basis of a single verbal root (see, e.g., Marantz 1997, Borer 2005, Folli and Harley 2005, Alexiadou, Anagnostopoulou, and Schäfer 2006, 2015, Schäfer 2008, and Harley 2013. I take the metaphorical readings for gelti and plėšti to be an artifact of the out-of-control semantics associated with the impersonal (see Babby (1994) on Adversity Impersonals in Russian; and Lavine (2016, 113-114) for additional discussion).
Note that not all Transitive Impersonals allow an alternation with a nominative subject, further indicating that the VP-internal position for the nonvolitional causer is basic and that the subject position in (9)-(10) is derived. In (12)-(13) there is no corresponding nominative subject form, suggesting strongly that the alternation is not necessarily regular, that the nonvolitional causer is not an argument of Voice, and that a nominative subject is not a necessary condition for accusative assignment.  The strongest indication that the oblique causer is basic in its VP-internal position is the fact that Transitive Impersonals do not passivize. This follows from the standard assumption that passivization targets an external argument in Spec,VoiceP. 6 The ungrammatical passives in (14) The inability of Natural Force šalčio 'cold.gen' and the nonvolitional causer to kvapo '[that smell].gen' to appear as passive by-phrases indicates their status as VP-internal causers, and crucially not external arguments. It follows that the non-passive variants of (14)-(15) (i.e., (5)-(6)) display a causative component without (active) Voice (see Alexiadou et al. 2015, ch. 2, for related discussion). It further follows that there is no argument in Spec,VoiceP against which accusative can be assigned as the dependent case. Note that this also suggests that Voice is not the source of accusative on a theory of case assignment by functional heads. In the next section I posit an active v-Cause as the source of accusative. 9 7 It follows that nonvolitional causers are best analyzed as distinct from the all-inclusive Initiator role of Ramchand (2008), Bruening (2013), and Legate (2014). See Lavine, to appear, for further evidence from Polish and Ukrainian against analyzing nonvolitional causers as arguments of Voice.
8 Milena Šereikaitė, p.c., points out that on the non-pain (literal) reading, the passive of perfective sugelti 'sting' is grammatical, as expected, with a genuine external argument, as in (i)

v-Cause as an Accusative Probe
The assignment of accusative in the absence of active Voice implicates the v-head between Voice and lexical VP, namely, v-Cause. That oblique causers occur more generally without Voice (i.e., in the absence of an external argument) is supported by Alexiadou et al. (2015, 47-51), on the basis of data from German, Italian, and English. Internal to Lithuanian, the role of the causative relation (and its syntactic reflex, v-Cause) is illustrated by the failure of accusative to appear in the anticausative alternate of the Transitive Impersonal, as in (16) (Ambrazas et al. 1997, 631) b. *… mane taip lūžta. me.acc so break [anticaus].3.prs The Transitive Impersonal is likewise incompatible with 'by itself' modification, as indicated in (18)-(19). Since 'by itself' modification denies the presence of a causer, as suggested by Alexiadou et al. (2015, 21-22), its incompatibility with the Transitive Impersonal is correctly predicted. 11 (18) Transitive Impersonal (cf. (8) When the cause is removed, as in the anticausative variant of (18), given in (20), 'by itself' modification is grammatical, as expected, and the Theme valtis 'boat' occurs in the nominative: (20) Anticausative (cf. (18)) Valtis pati savaime supo-si. boat(f).nom.sg self.nom.sg.f on.its.own rock.3.pst-refl 'The boat rocked by itself against the waves.' These diagnostics indicate that the Transitive Impersonal is necessarily two-place and caused, even if the causer argument is left unexpressed. This predicts that the sort of independent accusative that I describe here will never occur, in any language, with internally-caused "pure" unaccusatives (in the sense of Levin and Rappaport Hovav 1995).
In section 2.1 it was argued that the Transitive Impersonal does not involve an Agent or Initiator introduced by Voice. It is standardly assumed that the Agent is not part of the lexical entry of verbal roots (Kratzer 1996). Instead, an Agent may or may not be projected by a Voice head, depending on the compatibility of an Agent with the eventuality described. The appearance of accusative case in the absence of an active Voice head indicates that the features of Voice and Cause operate independently in the form of two discrete v-heads, as indicated in the "split vP" in (21), from Lavine and Babby (2019, 810), follow-ing Pylkkänen (2008) (see also Bowers 2002 for an important antecedent to this approach). If an Agent is not projected, as in the case of Transitive Impersonals, Voice is inactive (or absent) in the syntax, as indicated by the shading of the Voice layer.
In the case of lexically-causative verbs, v-Cause is active as an accusative probe, regardless of the setting for Voice, so long as the features [Voice] and [Cause] operate independently. In (21), v-Cause is activated by the presence of oblique nonvolitional causers, which are construed as initiating an out-of-control event in the absence of an active Voice head. Each functional head in the syntax is a predicate of sorts, identified with a particular argument (cf. Bowers 2010). Think of v-Cause as a predicate unsaturated until an argument merges that is capable of independently setting the event in motion. Secondary Instrumentals (Alexiadou et al. 2006) which require human manipulation, like 'fork', would thus fail to saturate the [cause] head. Only causative obliques, expressed overtly or merely understood, saturate the v-Cause predicate, licensing its independent presence in the structure and giving rise to its morphosyntactic reflex of assigning accusative (cf. the licensing of Borer's (2005) Aspect head).
The arrangement of v-heads in (21) applies broadly to similar Transitive Impersonal constructions crosslinguistically. It has recently been argued that this arrangement of v-heads accounts for the Transitive Impersonal in Russian VP # NP-acc and the related Fate Accusative in Icelandic (Lavine 2014a, Lavine and Babby 2019), the impersonal transitive passive in Ukrainian (Bowers 2002, Lavine 2013, and the Japanes adversity causative (Pylkkänen 2008, 89-92). 12

The Active Existential
The Transitive Impersonal in Lithuanian is superficially similar to the Active Existential construction, analyzed by Šereikaitė (2020, and to appear). As is evident in (22)-(23), the Active Existential also occurs with accusative in the absence of a higher nominative argument and, as such, provides additional empirical motivation against a Dependent Case analysis for accusative.  (Ambrazas et al. 1997, 600) The Active Existential construction does not occur with unaccusatives, as illustrated in (24) Intended '[There was a lot of blood in the room.] It feels as if someone fell and died here.' (Šereikaitė, to appear) The glosses in (22)-(23) indicate that the Active Existential projects a thematic Voice head. The incompatibility of the Active Existential with unaccusatives, as in (24), indicates that this Voice head necessarily introduces an external argument. Šereikaitė (2020, to appear) provides a host of syntactic tests indicating that while Voice is thematic in this construction, it does not project an implicit argument. 13 The Active Existential is likewise not a case of 3 rd person pro-drop, which is restricted in its distribution in Lithuanian, in contrast to the freer application of pro-drop for 1 st and 2 nd person pronouns. The Initiator is interpreted as indefinite 'someone' or generic 'some people'. The Voice head of the Active Existential is like that of an active transitive construction in that it is similarly thematic (it introduces an external argument variable) and assigns accusative case to the Theme. The difference is that in the case of the Active Existential the external argument variable is existentially bound in the lexicon, rather than projecting to argument position in Spec,VoiceP (Šereikaitė, to appear, section 5). It follows, on Šereikaitė's analysis, that the projection of an argument in the specifier of Voice, overt or implicit, is not a condition on accusative assignment, so long as Voice is thematic. This is, then, another instance in Lithuanian, independent of the Transitive Impersonal, in which accusative assignment is non-configurational (i.e., not subject to the rules of Dependent Case). 14 13 Tests indicating the lack of an implicit subject include the failure to bind the subject-oriented reflexive savo and the reciprocal vienas kitą 'each other', and the failure to trigger subject-predicate agreement and agreement with the participle of embedded adjuncts, in contrast to the behavior of overt kažkas 'someone'.
14 To be sure, the argumentation in section 2 suggests that the source of accusative is v-Cause and not Voice. A weaker version of the present account, which is compatible with Šereikaitė's analysis, is that active v-Cause is a sufficient condition for accusative assignment, but not a necessary one. This must certainly be the case for stative (non-causative) predicates that appear with accusative:  (Ambrazas et al. 1997, 280) 239 3 A Transitivity Restriction: The Inferential Evidential The strongest argument against Dependent Case Theory challenges both core tenets of the model, (1b) and (1c), repeated below 15 : (1) Dependent Case Theory b. If there are two distinct NPs in the same local domain such that NP 1 c-commands NP 2 , then value the case feature of NP 2 as accusative (under nominative-accusative alignment). c. Value as nominative the case feature of an NP that has not been assigned lexical or dependent case in the preceding steps.
Thus far, we have focused on argumentation against accusative as a dependent case (1b). We now turn to evidence from the Lithuanian Inferential Evidential construction that challenges the condition of Dependent Case in (1c), namely, the status of nominative as the non-dependent case. The Lithuanian Inferential (ma/ta) Evidential is marked formally by a non-agreeing passive participle, the absence of a tense-marking auxiliary, and a genitive-marked subject. The Evidential presents the event as inferred from visual observation (Geniušienė 2006). As previewed earlier, oblique subject constructions are a good testing ground for condition (1c)  (25) Inferential Evidential Čia kiškio gulėta. here rabbit.gen.sg lie.pst.pp.df 'A rabbit must have lain here (as inferred from the marks)' (Geniušienė 2006, 54) (26) Inferential Evidential (= (3a)) Iš plačios erdvės matėsi, kad čia from broad.gen.sg.f space.gen.sg see.3.pst.refl that here didelio medžio augta. big.gen.sg.m tree.gen.sg grow.pst.pp.df 'One could see from the broad empty space that a big tree had apparently grown here.' (Spraunienė et al. 2015, 342,  In earlier work I have taken the genitive marking on the subject to be assigned as an intrinsic lexical property of the ma/ta morpheme itself, that is, as an instance of lexical case. On this approach, ma/ta heads Voice, as an erstwhile marker of the passive, with genitive being associated with the case of the passive by-phrase (see Lavine 2006Lavine , 2010. More recently, the genitive subject has been analyzed as an instance of structural case. A standard feature of structural case is that it does not refer to a fixed theta role. For example, it can be assigned to Agent and Theme subjects alike (cf. (25) and (26)). For this reason (among others), Šereikaitė (2020) and Legate et al. (to appear), treat the genitive subject of the Inferential Evidential as an instance of structural case, assigned by an Evid(ential) head (see also Arkadiev 2018 for a similar conclusion). The precise nature of subject case-lexical vs. structuralis crucial for assessing Dependent Case Theory as it applies to this construction. According to Dependent Case (1c), if the genitive subject is lexically-marked, then nominative automatically shifts to the object, as in standard quirky-subject constructions, which is indeed the case here. Alternatively, if the subject bears structural genitive case, then the object would be expected to bear accusative, dependent on the higher deployment of a structurally-marked case (1b). Finally, any theoretical treatment of the Inferential Evidential must account for the widely reported preference for intransitives in this construction, what I have referred to as a Transitivity Restriction (see fn. 3). There is certainly no sense in which two-place predicates are incompatible with the inferential semantics of the Evidential. The restriction, by hypothesis (and following Lavine 2010), must be syntactic in nature, reducing to a lack of source for object case. So while nominative does occur on the object, as in (27)-(28), the use of the Inferential Evidential with transitive verbs appears to be avoided. So if we were to treat the genitive subject as lexically-case marked, then the use of nominative on the object would follow from Dependent Case (1c), but with no explanation internal to Dependent Case Theory for the Transitivity Restriction. This is a weakness for Dependent Case Theory. Alternatively, if we were to treat the genitive subject as structurally-case marked, then Dependent Case Theory falsely predicts the appearance of accusative on the object (as the lower structurallymarked case). In point of fact, accusative does not appear on the object of the Inferential Evidential (Geniušienė 2006, 38;Šereikaitė 2020, ch. 4). 16 Dependent Case theory therefore fails to capture the distribution of object case in the Inferential Evidential, regardless of how the genitive subject is analyzed.
The fact that nominative is not fully productive on the object fares better on the functional head analysis of case. Nonagreeing (phi-incomplete) T 0 fails to assign structural nominative to the object NP. This failure of the Agree relation, as described in (4), explains why nominative on the object is not readily available in the syntax. Attempts to identify a regular syntactic source for nominative on the object face an overgeneration problem. For example, Šereikaitė (2020) and Legate et al. (to appear) propose that an Evid(ential) head selects a variety of Voice that is featurally specified to assign nominative to the object (rather than the expected accusative). They do not address the "weakness" of the nominative object or the preference for intransitives in this construction. I suggest in Lavine (2010) that nominative has no source for case, that it is syntactically "detached". When it does occur, it is assigned by default, an inherently unstable condition which captures the fact that the nominative object in the Evidential construction is largely obsolete.
Turning to the absence of accusative assignment in the Inferential Evidential, it must be acknowledged that the theory of case assignment by functional head, advanced here, does not fare much better. There is no obvious reason why v-Cause does not assign accusative, thereby freely admitting transitive verbs into this construction. I can only speculate, as I have previously (Lavine 2010), that the failure of accusative to appear in the ma/ta Evidential is a function of its passive etymology. Indeed, as Holvoet notes, while stressing that the evidential does not function as a passive, and in no way alters the predicate's argument alignment, "it must be admitted that there is no rigid line of division, so that one cannot assert that there is no overlap between evidential and passive" (Holvoet 2007, 105). The idea, for present purposes, is that the vestigial passive morpheme ma/ta suppresses accusative (or detransitivizes) by whatever means the passive m/t morpheme suppresses accusative in the canonical passive in the language. I leave the proper analysis of this insight to future research.

Conclusion
Structural case is increasingly analyzed in terms of Dependent Case Theory in generative syntax. This paper has pushed back on this development and has provided evidence for case assignment by designated functional heads. The Transitive Impersonal is offered as a prototype of an "accusative first" construction. The key idea is that the nonvolitional causer in the Transitive Impersonal is merged VP-internally rather than as an argument of Voice (i.e., in Spec,VoiceP). This is demonstrated by the fact that the causer resists passivization and, in some cases, cannot appear as a nominative subject, indicating that its VP-internal position is basic. The absence of a higher argument of Voice means that there is no case competitor against which accusative in the Transitive Impersonal can be assigned as the dependent case. Further, since Voice is non-thematic in this construction, the Voice head is ruled out as the source of accusative. It is proposed instead that the VP-internal nonvolitional causer identifies an active Cause head, independent of Voice, which functions as an accusative case assigner. The Transitive Impersonal is necessarily two-place and caused. The necessity of external causation is demonstrated by the fact that the Transitive Impersonal is incompatible with 'by itself' modification and anticausatives.
Recall that the logic of the argument against Dependent Case demonstrates not only that accusative occurs without a case competitor, but that nominative fails to occur (productively) in the absence of a case competitor. This was demonstrated on the basis of the Lithuanian Inferential Evidential. As we observed above, in the event that the genitive subject of the Evidential is analyzed as an instance of structural case, then the Dependent Case model predicts the appearance of accusative on the object, which does not occur. Alternatively, if the genitive subject is analyzed as an instance of lexical case, then the Dependent Case model predicts the regular appearance of nominative on the object, as the first (and only) structural case deployed. While nominative is attested on the object of the Inferential Evidential, it is rare, if not obsolete, a fact which finds no explanation in Dependent Case Theory. While I have speculated that the non-appearance of accusative on the object may be linked to the fact that the Evidential is historically passive, the failure of nominative to appear on the object follows directly from the non-application of the Agree relation with Tense. Thus, both the unexpected appearance of accusative in the Transitive Impersonal and the equally unexpected "weakness" of the nominative in the Inferential Evidential find a natural explanation only in terms of case assignment by functional heads. gen genitive pvb preverb ins instrumental refl reflexive loc locative sg singular