2.1. Semantic approaches to GN

It has been observed that the distribution of Russian GN is governed by semantic factors, including specificity (Babyonyshev & Brun Reference Babyonyshev and Brun2002), definiteness (Bailyn Reference Bailyn1997) and the semantic type of a noun phrase (Kagan Reference Kagan, Friedman and Gibson2007, Reference Kagan2013, Partee et al. Reference Partee, Borschev, Paducheva, Testelets and Yanovich2011). For instance, the grammatical object of transitives may or may not be marked with GN depending on its definiteness properties (see Kagan Reference Kagan2013: 12). A definite theme favors accusative, as in (5b), with a definite object modified by the demonstrative pronoun and, in (6), with a proper noun. An indefinite theme is compatible with accusative and genitive, as in (5c).

Under negation, genitive is preferred over accusative when the object refers to an abstract entity, like ‘happiness’ in (7b). If the object refers to a concrete noun, such as ‘flower’ in (8b), only accusative has been reported to be available.Footnote ⁸

In equivalent Lithuanian examples, GN is always substituted for an accusative theme object regardless of whether it is definite, referential or indefinite (9).

Similarly, the theme object, be it a concrete entity or an abstract entity, is always genitive when negation is present (10).

It should also be pointed out that Polish GN, which is like Lithuanian GN in many respects, exhibits the same pattern where semantic properties of the object do not play a role as to whether the GN rule can be applied or not (e.g. Błaszczak Reference Błaszczak, Hanneforth and Fanselow2010).

The difference between Russian and Lithuanian is also reflected in the case marking of the theme of unaccusatives. Russian GN is famous for being an unaccusativity test (Pesetsky Reference Pesetsky1982): GN is found with canonical unaccusatives like ‘arrive’ (11b) and passives (12b), but not with unergatives. However, GN is not always obligatory in these environments – sometimes nominative is also possible. See (11a) and (12a).

There is a difference in meaning between the nominative theme and the genitive theme under negation. (11a) has a reading where there is an answer such that it did not arrive, and the same goes for ‘the statement’ in (12a). These refer to a specific answer or a specific statement (an entity of semantic type e), and thus these examples presuppose the existence of a theme. However, when the theme is genitive, it has a reading where no answer and no statement exist, meaning that no existence of a theme is presupposed. On Partee et al.’s (Reference Partee, Borschev, Paducheva, Testelets and Yanovich2011) account, these are properties (of type <e,t>). They propose a semantic analysis – the Property-Type Hypothesis, discussed in Section 6 – to account for this pattern.Footnote ⁹

Whereas a property-type vs. e-type analysis is tenable for Russian, it is not for Lithuanian, in general. GN is ungrammatical in Lithuanian with passives and unaccusatives, as we show in (15)–(16) in Section 2.2 below, regardless of whether the theme is presupposed to exist or not – the theme is always nominative. Thus, this group of predicates differs from Russian unaccusatives. The same goes for the difference between the genitive/accusative alternation in transitive clauses: although it is crucial in Russian whether the object is of type e or <e,t>, it does not affect the outcome in Lithuanian. Recall our examples in (10).

To sum up, while Russian GN can be governed by semantic factors, Lithuanian does not follow the same pattern. Rather, Lithuanian GN applies to accusative theme grammatical objects irrespective of their semantic properties. It is disallowed in constructions with canonical unaccusatives, as opposed to Russian GN. Nevertheless, there is a small subset of unaccusative predicates in Lithuanian, namely locative-existential constructions with verbs of perception and existential predicates, that permit optional GN (see Ambrazas et al. Reference Ambrazas, Geniušienė, Girdenis, Sližienė, Tekorienė, Valeckienė and Valiulytė1997: 667–673, Holvoet Reference Holvoet, Holvoet and Mikulskas2005: 143, Aleksandravičiūtė Reference Aleksandravičiūtė2013). These indeed resemble Russian GN constructions, where genitive DPs are properties of type <e,t>, whereas nominative DPs are of type e. However, these exceptions, discussed in Section 6, do not contradict our claim that Lithuanian GN is a type of structural object case.

2.2. Syntactic approach to GN

Richards (Reference Richards2013) argues, based in part on data from Russian GN, that case is assigned in syntax.Footnote ¹⁰ We agree and, as discussed in Section 5, argue, furthermore, that case is realized late in the derivation, at Vocabulary Insertion at PF (Halle & Marantz Reference Halle, Marantz, Hale and Keyser1993). We demonstrate, nonetheless, that the case replacement analysis proposed by Richards cannot account for Lithuanian GN.

Richards (Reference Richards2013) gives an analysis of case replacement in Lardil and extends it to Russian GN. He splits case into meaningful and meaningless cases; the former category is meaningful at LF, the latter is not. Structural case is meaningless, whereas inherent case, such as instrumental case, is meaningful. In his approach to Russian GN, genitive case is assigned by a negation head to an argument of a transitive verb that originates as an object and has previously been assigned accusative case. After the meaningless structural case, namely accusative, has been assigned to a DP, it is deleted and replaced by genitive. However, GN cannot replace a meaningful case, like instrumental case in (13).

Russian GN also applies to arguments in the passive (see (14b) below) that – without negation – would have been assigned nominative (see also discussion above and example (12)). For Richards (Reference Richards2013), genitive case assignment under negation, like case stacking in Lardil, is subject to timing: genitive is assigned to an argument by negation before it moves to subject position, and nominative case (structural/meaningless case) cannot replace genitive. This analysis can also be extended to unaccusative predicates (see (11b) above).

However, Richards’ approach cannot be extended to Lithuanian. First, GN cannot replace structural nominative, neither in passives (15) nor canonical unaccusatives (16).Footnote ¹¹ Thus, the theme argument that is a grammatical subject is not affected by GN, in contrast to Russian GN. Second, GN in Lithuanian is not sensitive to timing: the passive subject cannot be genitive regardless of whether it is in SpecTP (15a) or in situ (15b), which is also the case for unaccusatives (see (16a) and (16b)). One could have argued for Lithuanian, based on Richards’ analysis, that the DP in (15a) and (16a) moves to subject position before GN is assigned, which could explain the nominative case that is realized on the subject. However, the fact that nominative case is also realized in situ in (15b) and (16b) shows that GN in Lithuanian passives is not affected by the timing of case assignment; GN cannot be realized regardless of whether the DP moves or not.

Furthermore, Lithuanian long-distance GN, as in (17b), may pose a challenge for Richards’ approach.Footnote ¹² In (17a), a predicate ‘teach’ takes an accusative object ‘children’ and an infinitival complement whose main verb ‘paint’ occurs with an accusative object ‘fence’. When the matrix predicate is negated, as in (17b), both the matrix object and the object of the infinitival clause are genitive. This phenomenon is known as long-distance GN (see Arkadiev Reference Arkadiev, Holvoet and Nau2016 and Section 5). Note that it is also possible for the object of the infinitive to bear accusative.

Under an account that places a case feature on NEG, it would have to be able to assign case to two DPs – in this case, to the object of ‘teach’ in the matrix clause and to the object of ‘paint’ in the infinitival clause. That is, the meaningless structural case assigned to both ‘children’ and ‘fence’ would be deleted and subsequently replaced by genitive case assigned by one and the same NEG head. Long-distance GN is not a problem per se for an approach that places a case feature on NEG – this could be accounted for by using, for example, a Multiple Agree approach, as discussed for Polish in Witkoś (Reference Witkoś2008). We, however, will propose a different syntactic approach to Lithuanian GN that successfully derives (15)–(17).

2.3. Interim summary

To summarize, we have reviewed semantic and syntactic approaches that have been used for Russian GN. While the choice of Russian GN can be semantically conditioned when applied to both transitives and unaccusatives, the distribution of Lithuanian GN is not restricted by these semantic factors. We argue that Lithuanian GN is a syntactic phenomenon that should not, however, be analyzed using a case replacement account, such as introduced by Richards (Reference Richards2013) for Russian. We introduce our syntactic approach in Section 5, but now we turn to the difference between GN and other types of genitive case in Lithuanian.

3.1. Non-structural genitive

Some Lithuanian verbs marked with the reflexive -si- take a genitive object. These include stative experiencer-like verbs, such as baimintis ‘be afraid of’, gailėtis ‘be sorry’, gėdytis ‘be ashamed’, saugotis ‘beware of’, as well as verbs like šalintis ‘avoid’ (18) (for a full list, see Ambrazas et al. Reference Ambrazas, Geniušienė, Girdenis, Sližienė, Tekorienė, Valeckienė and Valiulytė1997: 503).

In contrast, Lithuanian GN is not associated with a particular class of verbs. It occurs with both stative (19) and non-stative verbs (20) as long as the predicate can assign structural accusative case (see Section 4 for more examples).

The genitive assigned to an object with experiencer-type verbs in (18) exhibits properties of a non-structural case, whereas GN lacks these properties. The difference between the two is reflected in their behavior with the distributive preposition po ‘each’. DPs with a non-structural case are not compatible with po, whereas structural case DPs are. Po assigns accusative case to its argument, and a PP headed by po can occur in a position where, normally, a structural case is assigned (Anderson Reference Anderson2013, Reference Anderson, Holvoet and Nau2015, E.F. Sigurðsson et al. Reference Sigurðsson, Šereikaitė and Pitteroff2018, Šereikaitė Reference Šereikaitė2020). It can occur as an object (21b) or as a subject of a transitive (22b). In (22b), the preposition takes an accusative complement, and the assignment of nominative is blocked. We therefore take the accusative in (21b) to be assigned by po, but not the verb.Footnote ¹⁴

Anderson (Reference Anderson2013, Reference Anderson, Holvoet and Nau2015) observes that DPs with non-structural case cannot be complements of po. Verbs like tarnauti ‘serve’, which take an object with inherent dative case, are incompatible with this preposition. Neither accusative, which is the case normally assigned by po, nor inherent dative is grammatical in (23) (E.F Sigurðsson et al. Reference Sigurðsson, Šereikaitė and Pitteroff2018, Šereikaitė Reference Šereikaitė2020). Šereikaitė (Reference Šereikaitė2020) hypothesizes that PPs in Lithuanian have a strong phase head which blocks case assignment from outside. At the same time, the assignment of non-structural case, like inherent dative, is obligatory. The PP blocks this case assignment which results in ungrammaticality.

The genitive case of verbs like šalintis ‘avoid’ behaves like inherent case: a DP marked with this case cannot be embedded under po as in (24). The complement of po cannot be accusative either. The genitive case of these predicates requires obligatory assignment just like inherent dative in (23).

In contrast, adding a negation to a structural-case-assigning predicate, which in turn is followed by po, is grammatical.Footnote ¹⁶ This is illustrated with the verb gauti ‘receive’ in (25), whose object becomes genitive under negation. When po is present, it assigns accusative to its complement, and GN is not available (26) (for discussion of GN with PPs, see Section 5.2). The grammaticality of (26), where po assigns accusative under negation, suggests that GN itself behaves like structural case: its failure to be realized on the object does not cause the derivation to crash.Footnote ¹⁷

We have demonstrated that non-structural genitive in Lithuanian differs from GN. Non-structural case is associated with a certain group of predicates, and a DP bearing this case cannot be a complement of po. In contrast, GN does not exhibit properties associated with a non-structural genitive. It can occur with any predicate as long as that predicate assigns structural accusative case to its object, meaning that it is not licensed thematically like a non-structural case. Furthermore, it is compatible with the distributive preposition po, which can occur with the type of arguments that are normally assigned structural case.

3.2. Intensional genitive

Lithuanian has a class of so-called intensional predicates like norėti ‘want’, geisti ‘desire/crave’, laukti ‘wait’, trokšti ‘desire’ and tikėtis ‘hope’. They assign genitive case to their object, as in (27), which is a type known in the Slavic literature as ‘intensional genitive’. The accusative case is not available.

Building on Neidle (Reference Neidle1988), Kagan (Reference Kagan2013) argues that Russian intensional genitive and GN are the same phenomenon. The object of intensional predicates can be assigned genitive case but sometimes it can bear accusative, as in (28). This genitive patterns like Russian GN, which can also be optional with transitive predicates (see Section 2.1). However, Lithuanian intensional genitive cannot be replaced by accusative (27), and thus differs from the Russian genitive in (28).

The intensional genitive and GN are two distinct cases in Lithuanian. Recall that GN in Lithuanian cannot occur in the passive, as in (15) above. In contrast, the intensional genitive can be retained under passivization, as in (29a), and thereby may qualify as a non-structural case (Woolford Reference Woolford2006). It can also advance to nominative (29b), suggesting structural case. Hence, the intensional genitive shows mixed characteristics with respect to its status.Footnote ¹⁸

GN is compatible with the preposition po (recall (26)). The intensional genitive is different. A DP assigned intensional genitive cannot occur as a complement of po, and the accusative is ungrammatical also (30). Intensional genitive is obligatorily assigned, but po blocks the case assignment, which results in ungrammaticality.

To sum up, the intensional genitive is distinct from GN. It is a mixed case that falls between two categories. It must be obligatorily assigned like an inherent case, but it can advance to nominative in the passive (see Šereikaitė Reference Šereikaitė2020), just like a structural case. In contrast, GN does not show this ambiguity.

3.3. Partitive genitive

Yet another type of genitive is partitive genitive, also known as genitive of indefinite quantity (see Ambrazas et al. Reference Ambrazas, Geniušienė, Girdenis, Sližienė, Tekorienė, Valeckienė and Valiulytė1997: 486 and Seržant Reference Seržant, Holvoet and Nau2014 for discussion). This type of genitive, shown in (31), denotes a part or indefinite quantity of something and is usually realized with singular mass nouns and plural count nouns.

Unlike GN, the partitive genitive is incompatible with a definite object, as shown in (32a). Lithuanian lacks definite articles and, therefore, we use a demonstrative šiti ‘these’ in (32) to reinforce a definite interpretation. In (32), a definite object DP is only possible when its case is accusative, as in (32b).

A unified account has been proposed for the partitive genitive and GN in Russian by Pesetsky (Reference Pesetsky1982), in which both cases are realized as genitive by a phonologically null quantifier (also see Pereltsvaig Reference Pereltsvaig1999, Bailyn Reference Bailyn, Arnaudova, Browne, Luisa Rivero and Stojanović2004, Reference Bailyn2012).Footnote ¹⁹ However, Lithuanian GN is not a partitive genitive case. First, the partitive genitive object cannot be definite, as in (32a), whereas a GN object can, as indicated by the availability of the genitive object with the demonstrative in (33).

Second, the partitive genitive cannot be assigned to singular count-noun DPs as in (34a): they are incompatible with an indefinite quantity interpretation like ‘some’. However, GN can occur with singular count-noun objects, as in (34b).

Lastly, the partitive genitive can also surface with passives, as in (35b), as well as with unaccusatives, as in (36b). Note that the theme in both cases occurs in a clause-final position due to (in)definiteness (also see fn. 14 for discussion).

In contrast, GN is incompatible with passives (37–38), or unaccusatives, as in (39) (see also Section 2.2). To ensure that we are testing GN rather than the partitive genitive in (39), we use the demonstrative šie ‘these’, which reinforces a definite interpretation otherwise impossible with the partitive genitive.

Given these differences, we take GN and the partitive genitive to be two distinct cases that deserve separate analyses. We do not attempt to give an analysis of the partitive genitive in Lithuanian.Footnote ²¹ However, we note that due to a number of semantic restrictions, the assignment of partitive genitive does not seem to be related to the assignment of structural case, unlike GN.

3.4. Genitive of evidentials

Lastly, the Lithuanian evidential construction (see Geniušienė Reference Geniušienė, Abraham and Leisiö2006, Lavine Reference Lavine2006, Reference Lavine2010, Spraunienė et al. Reference Spraunienė, Razanovaitė, Jasionytė, Holvoet and Nau2015, Legate et al. Reference Legate, Akkuş, Šereikaitė and Ringe2020, i.a.) presents yet another type of genitive case. This is a non-finite construction with a verb taking a non-agreeing passive morphology. A thematic subject that is typically in the nominative case in an active transitive (40a) appears in the genitive case in the evidential construction (40b). The grammatical object, that would otherwise be in the accusative, is realized in the nominative case.

Legate et al. (Reference Legate, Akkuş, Šereikaitė and Ringe2020) argue that the evidential genitive is a type of structural case assigned by an evidential head to the highest argument in the clause. Genitive of evidentials is realized on the subject of a transitive (40b), unergative (41a), unaccusative (41b) and passive (41c). It is thus associated with a subject position and assigned under A-movement to the highest argument, like structural nominative case.

By contrast, GN does not appear on a thematic subject of unergatives or transitives (42) or a grammatical subject of passives or unaccusatives (37–39). We conclude that GN is not assigned under A-movement, unlike the genitive of evidentials. In other words, it does not target the highest available argument that becomes a subject. Rather, as we argue in Section 4, it tracks structural accusative case.

3.5. Interim summary

We have distinguished between different types of genitives in Lithuanian and demonstrated that GN is different from the rest. An overview is given in Table 1. GN is not a non-structural genitive that is associated with a certain type of a θ-role. Thus it is not assigned thematically. The failure to realize GN, when po is used, does not result in ungrammaticality, unlike what happens with non-structural case. While Lithuanian has a class of mixed cases like the intensional genitive, which bears properties of both structural and non-structural case, the assignment of GN does not show this type of ambiguity. Furthermore, GN differs from the partitive genitive, which has a number of semantic restrictions. GN affects an object that would typically be marked with structural accusative case regardless of its semantic properties (e.g. definiteness). Thus, GN deserves its own analysis. Lastly, GN is not associated with a type of structural case that is assigned under A-movement to the highest argument in the clause as the genitive of evidentials. Even though it is a type of structural case (which has also been argued for Polish GN, see Przepiórkowski Reference Przepiórkowski2000), its assignment is not related to A-movement.

Table 1 Different types of genitives in Lithuanian.

4.1. GN tracks structural accusative case

In this subsection, we demonstrate that GN occurs in various syntactic environments where structural accusative case would typically be assigned.

4.1.1. DPnom-DPacc environment

We start the discussion of GN by reviewing its behavior in DPnom-DPacc environments. A typical example of Lithuanian GN is in transitive clauses with negation where structural accusative case would otherwise surface, as in (43).

Under our account, the direct object is assigned structural case by v in transitive active constructions (for further details, see Section 5). The result is normally accusative, unless the DP is c-commanded by negation; then, genitive is the result. The same pattern can be observed with direct objects of ditransitives (44).

GN does not only apply to direct objects, as the indirect object of the ditransitive verb ‘teach’ is in the genitive under negation.Footnote ²² This suggests that GN does not track a particular θ-role (e.g. a theme, or a particular grammatical function – for example, a direct object). It can affect a goal – an indirect object – as long as that object can be realized in a structural accusative case as in (45) (see Arkadiev Reference Arkadiev, Holvoet and Nau2016 for more examples with ‘teach’).

In order to bear GN, the DP needs to be base-generated below negation. For example, when the verb in the to-infinitive in (46) is negated, the object of that clause, ‘same mistake’, is genitive, as it is base-generated below the negation. The matrix object ‘children’ is accusative, as it is base-generated above the negation.

However, a DP does not need to be under the semantic scope of negation to receive GN. This is reflected in the behavior of the topicalized direct object in (47) which retains its genitive case even when it takes scope over negation, and its surface position is not c-commanded by it.

All in all, the realization of GN is dependent on the syntactic structure, specifically whether or not a grammatical object case is assigned in the structure.

4.1.2. Impersonals

GN is found in impersonals like (48)–(50). Typically, these constructions include an accusative direct object and a subject that is not expressed overtly. Lavine (Reference Lavine, Holvoet and Nau2016) argues that impersonals like (48) have a nonvolitional causer that is not projected in syntax. Šereikaitė (Reference Šereikaitė2021) proposes that constructions like (49) also lack a projected external argument, which is interpreted as an existential ‘someone’. Regardless of whether there is a projected initiator or not, the assignment of accusative case in these constructions is possible. We take these impersonals to have v, which assigns structural case in syntax to a direct object, resulting in accusative case at PF – unless when negation is added, then the result is genitive.

These impersonals are similar to short passives (i.e. without by-phrases): both constructions lack a projected initiator in syntax (Šereikaitė Reference Šereikaitė2022). However, the v in impersonals licenses accusative, while the assignment of accusative in Lithuanian passives is impossible.Footnote ²³ Similarly, GN is possible in the impersonals in (48)–(50), but not in passives (e.g. (38)). The realization of GN does not require the presence of a syntactically projected agent; GN can appear in constructions that lack it as long as they license the assignment of structural object case.

Polish also has an impersonal construction with an accusative theme argument as in (51). Nevertheless, the Polish -no/-to construction, unlike the Lithuanian impersonals above, has been argued to have a projected null subject (Lavine Reference Lavine2005, Legate Reference Legate2014) and thus exhibits active voice characteristics. While the Polish data show us that GN is found in active impersonals with a fully projected agent, the Lithuanian data suggest that GN also occurs in passive-like constructions (i.e. without a projected external argument) as long as they license accusative case.

4.1.3. DPdat-DPacc environment

Lastly, GN appears in dative-accusative constructions. These take a dative subject followed by an object which is assigned structural object case realized with accusative. Crucially, the object becomes genitive when negation is present. Lithuanian pain-verb constructions (see Seržant Reference Seržant2013, Holvoet Reference Holvoet, Holvoet and Nau2016) with a dative possessor and an accusative theme belong to this class. The theme direct object is affected by negation – it is realized in the genitive case (52b).Footnote ²⁴

However, in Polish, verbs like boleć ‘ache’, stać ‘afford’ and kosztowac̀ ‘cost’ take a nominative theme and an accusative possessor. Interestingly, the accusative is either retained on the DP under negation or it becomes genitive (Przepiórkowski Reference Przepiórkowski1999, Błaszczak Reference Błaszczak2001). That is, Polish exhibits optionality in these environments, unlike Lithuanian. It has been argued for Polish that the object of verbs like ‘ache’ is associated with two types of cases: (i) a structural one to which GN can apply and (ii) an inherent one, to which GN does not apply (ibid.).

Returning to Lithuanian dat-acc patterns, an accusative theme is found also in non-finite subordinate clauses with a dative subject (see Ambrazas et al. Reference Ambrazas, Geniušienė, Girdenis, Sližienė, Tekorienė, Valeckienė and Valiulytė1997: 363, Arkadiev Reference Arkadiev, Gast and Diessel2012, Reference Arkadiev2017, Šereikaitė Reference Šereikaitė2020). It is in the genitive under negation (54).

To-infinitive clauses also belong to this class of constructions. The case of arbitrary PRO in Lithuanian is dative as illustrated by the agreement properties of the adjective vienas ‘alone’ in (55a) (see Šereikaitė Reference Šereikaitė2020; for similar facts in Russian, see Landau Reference Landau2008).Footnote ²⁵ The theme object in this configuration is accusative (55a) but realized in the genitive under negation (55b).

To sum up, GN is not sensitive to whether a clause is finite or non-finite. Nor is it sensitive to the case of a subject, be it nominative (see Section 4.1.1) or dative. GN appears on a grammatical object as long as that object can be realized in structural accusative case when negation is absent. In addition to dat-acc constructions, Lithuanian also allows dat-nom constructions. If GN indeed tracks structural object case, then we predict that GN cannot be realized where nominative surfaces (e.g. in dat-nom constructions). We show that this prediction is borne out in the next subsection.

4.2. GN does not track nominative case

Here we discuss environments with structural nominative case that appears on a grammatical subject and a grammatical object. We demonstrate that GN cannot be realized on DPs which would normally be marked with nominative regardless of whether that DP is a subject or an object.

4.2.1. Nominative subjects

A structural-case-marked subject is realized in the nominative case in Lithuanian, whether or not negation is present in the clause. Examples follow with the thematic subject of a transitive verb (56a) and an unergative verb (56b).

The sole arguments of, for example, unaccusatives and passives, that originate as underlying objects, are realized as nominative under negation. In (57a)–(58a), the theme has moved to subject position (SpecTP). If it stays in its original object position, as in (57b)–(58b), it is still nominative. Hence, GN does not track a specific syntactic position. Rather, we argue that GN affects arguments that are assigned structural object case, typically accusative.

Lithuanian like-class verbs (e.g. patikti ‘like’, rūpėti ‘care’) present yet another argument for GN not tracking nominative. These constructions have a non-structural dative experiencer, followed by a nominative theme. The theme is a grammatical subject: it binds the subject-oriented anaphor savo and triggers agreement on the predicate, as in (59).Footnote ²⁶ It cannot bind the anti-subject-oriented anaphor tavo. The theme is nominative in an environment with negation (60).

This is unlike the dat-acc construction in Subsection 4.1.3 where the theme, which typically bears accusative, is realized with GN. The dat-acc construction allows GN, whereas the dat-nom construction does not. The theme in the former bears structural object case, whereas the theme in the latter does not. The juxtaposition of the two constructions is another piece of evidence that GN is sensitive to the type of case the theme bears.

4.2.2. Nominative objects

We now investigate GN in environments with a nominative object. Evidence from evidentials in (61) (see also Section 3.4) demonstrates that GN does not merely track the grammatical function of a DP. In evidentials, the subject is genitive and the theme object is nominative.Footnote ²⁷ The theme is a grammatical object: it binds the anti-subject-oriented pronoun jų instead of the subject-oriented anaphor savo (Lavine Reference Lavine2006). When negation is added, the object is nominative (62).

We could have predicted that GN affects all grammatical objects regardless of the type of case they are marked with. However, the facts from evidentials show that GN does not track a specific grammatical function, such as a grammatical object. Rather, GN affects DPs that would otherwise be realized with accusative case.

4.3. GN does not track non-structural case

We now turn to environments with non-structural case. GN cannot appear on arguments that are marked with a non-structural case – a common property of GN in Slavic languages as well (e.g. Pesetsky Reference Pesetsky1982). This is shown below for direct objects (63) and indirect objects (64) that are assigned non-structural dative.

Quirky dative subjects can also be found in the language. Lack-class verbs (e.g. trūkti ‘lack’, užtekti ‘have enough’) take a dative subject and a genitive theme. As a subject, the dative DP binds the subject-oriented anaphor savo (65a) (see also Šereikaitė Reference Šereikaitė2020). GN cannot affect non-structural case subjects, as shown in (65).

4.4. GN does not track adjuncts

Lastly, we show that GN cannot be realized on adjuncts. Lithuanian measure adjuncts are marked with accusative case (e.g. ‘for x amount of time’); see (66a). Adverbial phrases like ‘every/each day’ also take an accusative temporal DP; see (67a). GN is nonetheless prohibited with these phrases; see (66b)–(67b).

From a morphological perspective, we could have expected GN to apply to all DPs that can be marked with accusative case, including adjuncts, contrary to fact. (66)–(67) suggest that GN does not track a particular morphological case but rather that it is sensitive to syntactic case. In other words, the accusative that appears on adjuncts is not assigned by v, and thus, GN does not apply in this environment.Footnote ^{28 ,} Footnote ²⁹

4.5. Interim summary

We have demonstrated that GN is realized on DPs which (i) are c-commanded by negation and (ii) would otherwise be realized in the accusative case. The availability of GN is not restricted by scope; topicalized objects which are outside the scope of negation still must bear GN. Our main claim is that Lithuanian GN tracks structural object case. We have supported this finding by showing that GN does not track a specific grammatical function since neither a nominative grammatical subject nor an object is affected by GN. Furthermore, the availability of a GN object is not dependent on whether the subject is marked with nominative or dative, or whether the clause is finite or non-finite. In contrast, GN is realized on DPs which are assigned structural accusative case, e.g. the grammatical object of transitives, ditransitives, impersonals and pain-class verbs, and the indirect object of ditransitives. We summarize our findings in Table 2.

Table 2 Summary of what type of case GN tracks.

5.1. Translation and realization of case in the Morphological Component

For our analysis of Lithuanian GN, we adopt H.Á Sigurðsson’s (Reference Sigurðsson2012a, Reference Sigurðsson2012b) notion of case stars. We use them for expository purposes to emphasize a distinction between case in syntax and case at PF.Footnote ³⁰ For him, arguments are A-licensed in syntax. At the Morphological Component (or deep PF, as he refers to it), A-licensing relations are translated accordingly, such that transitive v becomes v* – a v that assigns accusative. Dative is assigned by v*⁺ and genitive by v*⁺⁺. Relations between (eventual) case assigners and DPs realizing case are established in syntax, but ‘case feature values’ (nom, acc, etc.) are determined at PF.

For us, however, v*, v*⁺ and v ^*++ assign case in syntax that is later realized morphologically as accusative, dative and genitive, etc. We refer to case assigned by functional heads like T, v*, etc., as syntactic case. This largely amounts to abstract Case (Vergnaud 1977/Reference Vergnaud, Freidin, Otero and Zubizarreta2008, Chomsky Reference Chomsky1981, Reference Chomsky1995, Legate Reference Legate2008). We assume that when DPs are assigned syntactic case, they are marked accordingly as, for example, DP⁰, DP*, DP*⁺ and DP*⁺⁺. When the derivation is sent to PF, the DPs are still case marked as, for example, DP*; syntactic case then needs to be interpreted at PF. We might expect this to be done when Vocabulary items are inserted. However, we take this translation process of syntactic case at PF to take place earlier in the derivation. That is, at the Morphological Component, of PF, prior to VI, case diacritics on DPs are translated to morphological case features, as in (68).

When, for example, DP* is translated as DPacc, accusative percolates to all case values within the DP (see Norris Reference Norris2014 and E.F. Sigurðsson Reference Sigurðsson2017: §3 for feature percolation accounts). Subsequently, VI uses these values when inserting phonological exponents. We therefore suggest three levels of representation: abstract syntactic case, abstract PF case and its realization at VI. Note that the rules in (68) do not refer to Vocabulary items as we might expect if this took place at VI. In DM, Vocabulary items are inserted after concatenation and pruning, which are part of the linearization process (Embick Reference Embick2010, Reference Embick2015, Ingason Reference Ingason2016). At that point in the derivation, the tree structure is not available, and as a result, the c-command relation is no longer visible. However, we assume that when the syntactic case is translated into abstract PF case, the syntactic structure is still available.Footnote ³¹ As argued in Section 5.2, it is essential for long-distance GN to have access to the negation in the matrix clause when case in the embedded infinitival clause is determined.

Our model makes a clear distinction between case at syntax and morphological case. nom, acc, gen, etc., are morphological labels and reflect how the Morphological Component uses its finite inventory to interpret different syntactic structures. It does not have one morphological case for each syntactic case licensing relation; there is not a one-to-one relation between case in syntax and in morphology. For instance, two different syntactic cases can have the same realization. Lithuanian reflects this. As argued by E.F. Sigurðsson et al. (Reference Sigurðsson, Šereikaitė and Pitteroff2018) and Šereikaitė (Reference Šereikaitė2020), the indirect object of duoti ‘give’ in (69a) is assigned (non-structural) dative case by an Applicative head. Monotransitive verbs like vadovauti ‘manage’ also take a dative object (69b). However, this object bears structural case properties; unlike the dative indirect object of ‘give’, it can, for example, advance to nominative in the passive and is therefore argued to be assigned by v (E.F. Sigurðsson et al. Reference Sigurðsson, Šereikaitė and Pitteroff2018). This suggests a relationship between syntactic cases and their morphological representation, as in (69)–(70), where two distinct syntactic cases, DP^Appl (see (69a)–(70a)) and DP*⁺ (see (69b)–(70b)), have the same morphological outcome, dat. The label DP^Appl reflects here that dative case of indirect arguments is assigned by Appl but not v.

The opposite also exists where a single syntactic case can have two morphological reflections. We propose that structural object case in Lithuanian can have two realizations depending on whether negation is present or not. We argue that Lithuanian GN is not a realization of DP*⁺⁺, which we would normally assume for non-structural (lexical) genitive case (assigned by v*⁺⁺) of the type discussed in Section 3.1, but of DP*. That is, even though the outcome is genitive, GN is not a realization of non-structural case, as in (68d)/(71b), but of structural object case, as illustrated in (71a). Thus, we propose that even though DP* with structural object case is under normal circumstances realized as accusative case, it is realized as genitive under negation in Lithuanian. In other words, GN is a realization of structural object case assigned by v* when it is base-generated below negation.Footnote ³²

In Lithuanian, the syntax of ‘I read a beautiful book’ (72a) is identical to ‘I didn’t read a beautiful book’ (72b) with respect to syntactic case features: when v* probes down and agrees with its object, v* assigns ‘a beautiful book’ structural object case, yielding DP*, whether or not negation is present. For concreteness, we assume that the thematic subject ‘I’ is introduced in the specifier of an agentive Voice (Kratzer Reference Kratzer, Rooryck and Zaring1996, Pylkkänen Reference Pylkkänen2008, Harley Reference Harley2013, Legate Reference Legate2014, i.a.) and moves to SpecTP. It receives its structural subject case from a finite T, resulting in DP⁰. We show the tree structure for these clauses in (73); we add NegP – which we assume to be generated above VoiceP – in parentheses, as it is absent in (72a).Footnote ³³

These syntactic case features are in turn translated into morphological case features at the Morphological Component: DP⁰ as nominative and DP* as either genitive (see (74a)) if that DP is base-generated below negation or accusative (see (74b)), according to the Elsewhere Principle. The translation process uses the syntactic information available at this point to convert the syntactic case features into morphological case features. We show the translation rules for DP* below. This process takes place prior to VI. The abstract PF genitive or accusative case on the object DP in (72) percolates to the nominals within it, such that both ‘beautiful’ and ‘book’ are marked accusative/genitive prior to VI which, lastly, inserts phonological exponents.

It is important to note that the translation process in (74) is not to be interpreted as allomorphy rules where specific heads are realized at VI in a specific way in a certain environment (such as in the vicinity of negation). Allomorphy is usually governed by linear adjacency and cyclic locality (e.g. Embick Reference Embick2010, Reference Embick2015).Footnote ³⁴ However, GN acts on a large scale as it applies to the whole DP, making all elements within the DP be interpreted and eventually realized as genitive case – that is, genitive percolates to all case values within the DP as shown for ‘beautiful book’ in (72b). Furthermore, as discussed in Section 5.2, it is possible to have long-distance GN where the genitive in the embedded infinitival clause can be determined by negation in the matrix clause. Thus, GN does not comply with locality and adjacency constraints common to allomorphy.

After the translation process in (74), Vocabulary items are realized, at VI. We assume an xInfl node which expresses syntactic features of heads from which they are separated at PF (this node is similar to the Agr node in Embick Reference Embick1997, Reference Embick2015; see also the nInfl node in Ingason Reference Ingason2016, which applies to all nominals). In the DP object graži-os knyg-os ‘beautiful book’ in (72b), the feature values of nInfl (i.e. genitive and singular) are realized with the exponent -os in (75), and the same applies to aInfl, which is an adjectival node (Adamson & Šereikaitė Reference Adamson and Šereikaitė2019).

Turning to unaccusatives, we assume the structure in (77). They do not assign structural object case to their theme as they lack v*; the structure contains v, which does not assign case. Unaccusatives also lack agentive semantics, meaning they have no Voice head assigning an external argument θ-role. T assigns structural subject case to the theme, resulting in DP⁰ in syntax, which in turn is translated and realized as nominative, irrespective of the presence or absence of negation.

Our approach therefore explains why a sole argument of unaccusatives (and passives), generated in object position, does not bear genitive case.

Recall from Section 4, that under negation, dat-acc structures have a GN object, whereas dat-nom structures do not, as in (78)–(79) (repeated from (52b) and (60b)). We suggest that dat-acc verb phrases have a v*, whereas dat-nom verb phrases have a v. The former assigns structural object case to its object, resulting in DP*, whereas the object of the latter is assigned case by T, resulting in DP⁰.

Our approach reflects in an interesting way Legate’s (Reference Legate2008) approach to case in languages that have absolutive case as default (ABS=DEF languages). Legate argues for ABS=DEF languages, such as Niuean (see (80)), that morphological absolutive case realizes different syntactic cases, which she refers to as nominative and accusative Cases. The intransitive subject ‘Pita’ in (80a) is morphologically in the absolutive case, but syntactically, according to Legate, in the nominative Case. Likewise, the transitive object ‘tree’ in (80b) is also in the absolutive case morphologically, but its abstract syntactic Case is accusative.

Extending Legate’s account to the current approach, one and the same morphological case, absolutive, realizes both DP⁰ and DP* (see (81a); cf. also (70) above on two syntactic cases translated to dative in Lithuanian). The opposite pattern is also possible; namely, that the same type of structural case can have two distinct morphological realizations, as is the case in Lithuanian GN; see (81b).

Our analysis could potentially also be extended to accommodate Spencer’s (Reference Spencer, Kulikov, Malchukov and de Swart2006) approach to Chuckchee in which syntactic ergative case is sometimes realized as morphological locative case and sometimes as morphological instrumental case.

5.2. Case boundaries

Our approach to Lithuanian GN raises questions regarding phase (Chomsky Reference Chomsky and Kenstowicz2001) and case boundaries. Since GN is a syntax-morphology interface phenomenon, negation must be visible to the DP object at PF, suggesting that the whole VoiceP together with NegP will be sent to PF. For us, the phase is at least as big as NegP, as Neg has to be visible to DP* at PF for it to be translated to genitive.

Furthermore, Lithuanian GN exhibits long-distance dependencies across non-finite clauses which provide important insights into phases and boundaries of case determination: they are quite large as they include embedded infinitival clauses in long-distance GN, as in (17) above, repeated as (82). Note that Polish also allows long-distance GN across non-finite clauses, which also challenges the idea of phases as discussed by Przepiórkowski (Reference Przepiórkowski2000), Błaszczak (Reference Błaszczak2001), Witkoś (Reference Witkoś2008).

The matrix clause object ‘children’ in (82b) is genitive. The object of ‘paint’ in the infinitival clause can also be genitive.Footnote ^{35 ,} Footnote ³⁶ This suggests that when DP* ‘fence’ in the infinitival clause is translated to morphological case, the negation in the matrix is visible to it. Therefore, long-distance GN is also subject to the rule in (74a).

However, long-distance GN cannot apply in wh-infinitives as in (83) with an overt CP element. It also does not apply past overt finite CP boundaries (84).Footnote ³⁷

Assuming that case assignment and its realization is phase-bounded, we take long-distance GN to show that infinitival clauses in Lithuanian without an overt CP layer are not phases (see also Landau Reference Landau2008, who shows that infinitival clauses in Russian do not form a phase boundary). The infinitival clause in (83) and the embedded finite clause in (84) both have a CP element, kur and kad, respectively. Given that long-distance GN cannot be realized, we take these to be phases.

Further indication of overt CPs being the boundaries to GN is seen in full relative vs. reduced relative clauses. In (85a), negation is located outside the full relative clause. As the relative clause has an overt CP layer, it is a phase, and the case of the object in this clause is not affected by the matrix negation. In reduced relatives, there is no overt CP, and the object of this clause can bear GN (85b).Footnote ³⁸

This suggests that full relative clauses constitute phase boundaries, whereas reduced relative clauses do not. Note that the structure of reduced relative clauses in Lithuanian merits further research, but it is outside the scope of this paper.

Finally, Lithuanian has prepositions like į, which assigns accusative to its complement. The accusative complement is not affected by negation, as in (86).

We treat PPs in Lithuanian as phases which prevent the features of the complement from being accessed by functional heads outside PPs (for a PP acting as a phase see, for example, Řezáč Reference Řezáč, Harbour, Adger and Béjar2008 and Šereikaitė Reference Šereikaitė2020).

5.3. Interim summary

We have provided an analysis of GN, arguing that it is a syntax-morphology phenomenon. We proposed that structural object case can be translated into two morphological cases at PF (i.e. either GN or accusative) depending on whether negation is present in the structure or not. This is not case allomorphy, as GN acts on a large scale and applies even to DPs in infinitival clauses. Lastly, instances of long-distance GN show that while finite clauses with an overt CP layer are phases, non-finite clauses lacking an overt C do not constitute phase boundaries.

6.1. Semantically conditioned GN in a locative-existential construction

We begin our discussion of the semantic GN by looking at Russian. Russian has a nominative/genitive alternation in locative-existential constructions (Babby Reference Babby1980, Partee & Borschev Reference Partee, Borschev and Young2004), as in (87). Both (87a) and (87b) state that Petja was not at some concert. However, only in (87a), where Petja is in the nominative, is it possible to proceed with the example and say that, in fact, there was no concert.

Following Babby (Reference Babby1980), Partee & Borschev (Reference Partee, Borschev and Young2004) term the type in (87a) negated declarative sentences (NDS) and (87b) negated existential sentences (NES). To explain the contrast between NDS and NES, Partee & Borschev (Reference Partee, Borschev and Young2004) focus on what they refer to as thing (Petja in the examples above) and location (the concert), one of which is a Perspectival Center (88), presupposed to exist.

The concert in (87b) is the Perspectival Center; thus, the location is presupposed and the existence of the concert cannot be denied. In other words, the NES in (87b) negates the existence of the thing (i.e. Petja) in the presupposed location (i.e. the concert) (see Partee & Borschev Reference Partee, Borschev and Young2004: 218). In contrast, when the thing is the Perspectival Center, as in the NDS in (87a), Petja’s existence is presupposed and the location can be negated (i.e. it is possible to say that there was no concert).

Whether the thing or the location is presupposed affects case marking of the theme: when the thing (Petja) is presupposed, it is nominative, even under negation, whereas when the location is the Perspectival Center, the theme is marked with GN. The case in these situations is conditioned semantically.

A similar construction exists in Lithuanian, which has a class of locative-existential predicates, such as būti ‘be’, egzistuoti ‘exist’, likti ‘remain’, atsirasti ‘appear’. Without negation, they occur with a nominative theme subject and a locative phrase. Nominative is grammatical when negation is present, as in (89a). Surprisingly, genitive can also be used, as in (89b), resulting in a different reading (Ambrazas et al. Reference Ambrazas, Geniušienė, Girdenis, Sližienė, Tekorienė, Valeckienė and Valiulytė1997, Holvoet Reference Holvoet, Holvoet and Mikulskas2005, Aleksandravičiūtė Reference Aleksandravičiūtė2013).Footnote ⁴⁰

As suggested by Holvoet (Reference Holvoet, Holvoet and Mikulskas2005: 144) and Aleksandravičiūtė (Reference Aleksandravičiūtė2013), we can use Partee & Borschev (Reference Partee, Borschev and Young2004)’s system to account for the contrast found in (89a) and (89b). According to Aleksandravičiūtė (Reference Aleksandravičiūtė2013), the subject/thing is the Perspectival Center in (89a). The existence of the subject ‘students’ is presupposed, making the example in (89a) parallel to the Russian NDS in (87a). In both languages, the thing, functioning as a Perspectival Center, is marked with nominative. Aleksandravičiūtė (Reference Aleksandravičiūtė2013) points out that ‘students’ is marked with the genitive case when location is the Perspectival Center, as in (89b). The same case marking in a similar semantic context is found in Russian, as evidenced by the NES in (87b). Unlike the syntactic GN discussed in Section 5, the semantic GN occurs in syntactic environments which lack structural object case. The realization of GN in these constructions seems to be based on the Perspectival Center.

6.2. Semantically conditioned GN with verbs of perception

A similar contrast is seen in another construction in Russian, in which nominative and genitive are both possible under negation but the genitive facilitates a different reading. Partee et al. (Reference Partee, Borschev, Paducheva, Testelets and Yanovich2011) discuss example (90) with a perception verb ‘see’.

When the thing is presupposed (as the Perspectival Center), as in (90a), Maša is of type e and she cannot be seen for some reason, even though she is present. Under such a reading, the DP is nominative. However, the case is genitive in (90b), indicating Maša’s non-existence in an implicit location (i.e. she may not be there at all). It is surprising that the thing in (90b) (and (87b)) is genitive, as Maša (as well as Petja) is the name of an individual, which is type e, and type e arguments in Russian are typically not genitive under negation. Partee et al. (Reference Partee, Borschev, Paducheva, Testelets and Yanovich2011), however, present a Property-Type Hypothesis (91) and argue for a type shift to a property (type <e,t>) which results in a different reading and different case marking.

For (90b), according to Partee et al., the type shift results in the reading ‘being Maša’, compatible with the genitive case. Overall, then, (90) suggests that there is a correlation between case marking and a semantic type.

Lithuanian has a similar perception verb construction with nominative/genitive alternations, depending on the meaning. Perception verbs with the reflexive clitic -si- like matyti-s ‘see-refl’, girdėti-s ‘hear-refl’ or jausti-s ‘feel-refl’ typically take a nominative theme subject (92); the theme cannot be accusative.Footnote ⁴¹

When negation is applied, however, the theme argument of these reflexive predicates occurs either in the nominative or genitive case, as in (93)–(94).

Aleksandravičiūtė (Reference Aleksandravičiūtė2013) argues that there is a semantic difference between the use of the nominative or the genitive in this construction. Under her analysis, nominative presupposes the existence of the argument in question. Only (94), with the argument in the genitive case, but not (93), with nominative case, can have the reading where the girl is not seen because she is not visible, as she has never been a part of the picture (see Aleksandravičiūtė Reference Aleksandravičiūtė2013 for discussion). This pattern is expected, as genitive case is not associated with the ‘existential commitment’ (Aleksandravičiūtė Reference Aleksandravičiūtė2013): there is no evidence for the girl’s presence in the assumed location.

Nevertheless, not only nominative in (93), but also genitive in (94) facilitate the reading where the thing, ‘girl’, is not seen in the picture because, for example, her classmate was standing in front of her. In this respect, Lithuanian differs from Russian, as under the interpretation that presupposes the existence of the thing, the phrase is typically marked with nominative, not genitive. Hence, in certain Lithuanian constructions with negation, there is not a one-to-one correspondence between the case marking and the reading, as opposed to Russian.Footnote ⁴² The observed pattern challenges the Perspectival Center Analysis and merits further research.

6.3. Summary

The overview given in Sections 6.1 and 6.2 shows that even though Lithuanian GN is syntactic for the most part, there is a semantic side that deserves a separate analysis. We have demonstrated that there are two components that distinguish semantic GN from syntactic GN in Lithuanian. First, it is sensitive to semantic factors, unlike syntactic GN. Building on Holvoet (Reference Holvoet, Holvoet and Mikulskas2005) and Aleksandravičiūtė (Reference Aleksandravičiūtė2013), we argued that the (un)availability of GN is related to referentiality and what Partee et al. (Reference Partee, Borschev, Paducheva, Testelets and Yanovich2011) call Perspectival Centers. Second, semantic GN can be realized in syntactic configurations where accusative case is typically not realized – in other words, where syntactic GN does not occur. Semantic GN does not track structural object case, meaning that it deserves a separate analysis from syntactic GN in Lithuanian, which we leave for further research.

Note that the contrast between nominative and genitive discussed for Russian and Lithuanian in Section 6.1 is also found in Polish. Błaszczak (Reference Błaszczak, Hanneforth and Fanselow2010) argues for a syntactic analysis, assuming two structures for the verb ‘be’ with the DP generated in two different locations (see also Holvoet Reference Holvoet, Holvoet and Mikulskas2005 for Lithuanian). However, as she notes, the construction discussed in Section 6.2 is not found in Polish. Her analysis presumably cannot straightforwardly be extended to Lithuanian if the two constructions in Sections 6.1 and 6.2 are to be given one and the same analysis.