Gender Representation and Defaults in Lithuanian

This study examines gender representation and defaults in Lithuanian by investigating the inflection on predicative adjectives (PAs). We provide novel evidence for two types of defaults in the representation of gender, masculine being the unmarked gender, and neuter being the absence of gender. It is demonstrated that neuter PAs appear when the subject lacks gender features accessible for agreement with the PA, which we refer to as non-agreement. In contrast, masculine PAs appear when the PA agrees with a subject bearing an unmarked gender feature. We analyze masculine and feminine as sharing a feature [gend] that originates on n, the locus of gender features (following Lecarme 2002; Lowenstamm 2008; Kramer 2015; 2016), with the default gender – the masculine – bearing only this feature, and the more marked gender – the feminine – having an additional feature [fem]. Neuter corresponds to the absence of these features. By placing gender features on the nominalizing head n, our account explicitly relates the distribution of gender inflection to nominal syntax and agreement.


Introduction
This study examines gender representation and defaults in Lithuanian (Baltic) by investigating the inflection on predicative adjectives (PAs). While it has been suggested before that masculine is a default gender in Lithuanian (Mathiassen 1996;Holvoet & Semėnienė 2006;Bruno 2012), by applying a number of diagnostics associated with default gender (e.g. Corbett 1991;Corbett & Fraser 1999;Kramer 2015) we demonstrate that masculine morphology does not appear on PAs in all environments identified with gender defaults. We show that a single language can in fact have two types of defaults in the representation of gender: masculine is the unmarked gender, whereas neuter, as we argue, corresponds to the the absence of gender features.

Description of Lithuanian gender
Before we proceed to our discussion of default environments, it is first necessary to review descriptive facts about adjectival inflection and gender. Lithuanian has been reported to have three genders: masculine, feminine, and neuter (Ambrazas et al. 1997: 134). However, the neuter is not inherent to any lexical noun in the language; the only inherently neuter elements are the following pronouns (3): 2 (3) Neuter pronouns (adapted from Ambrazas et al. 1997: 181-184) kas 'what'; kažkas 'something'; niekas 'nothing'; šitaĩ 'it, this'; taĩ 'it, this'; víena/kìta 'one/another' for events; vìsa/vìsa taĩ 'everything'; viskas 'everything' While lexical nouns are distributed across two genders, PAs show a three-way contrast (4)-(6). 3 Thus, Lithuanian resembles other languages like Ukrainian in exhibiting more gender categories of PA agreement than there are gender categories of nouns (see Corbett 1980;Corbett & Fraser 1999 for discussion of Ukrainian and other languages).
As can be seen in Table 1, the neuter form in the (i)u class is distinct from all other inflectional forms. In the adjectival class -(i)a, the neuter form and the nominative singular feminine form appear with the same inflectional ending. However, the two show distinct (i) a. Viskas buvo nudažý-t-a. everything.nom.n was paint-pass.pst.ptcp-n 'Everything was painted.' b. Stalai buvo nudažý-t-i. tables.nom.m.pl were paint-pass.pst.ptcp-nom.m.pl 'The tables were painted.' c. Kėdės buvo nudažý-t-os. chairs.nom.f.pl were paint-pass.pst.ptcp-nom.f.pl 'The chairs were painted.' stress patterns, as demonstrated for the minimal pairs in (10)-(11): stress falls on the root in the neuter, and on the inflectional ending in the feminine. It is thus generally possible to distinguish neuter forms. The fact that neuter forms do not alternate for number and case (as shown in Table 1) does not imply that neuter adjectives lack feature distinctions altogether. For example, neuter adjectives can appear in different case positions, which suggests that they can bear a case value. Evidence comes from deadjectival nominals. These forms appear in neuter and can occur in, e.g. accusative object position (12), genitive objects under negation (13), or instrumental complements of prepositions like su 'with' (14). 4 (12) a. Jis mate tik gerus dalykus. He.nom saw only good.acc.m.pl things.acc.m.pl 'He saw good things only.' b. Jis mate gẽra ir blõga. He.nom saw good.n and bad.n 'He saw good and bad.' 4 A reviewer asks if neuter deadjectival nominals can appear as objects of nonstructural case-assigning verbs.
(i-ii) suggests that verbs like tarnauti 'serve,' which take an object with nonstructural dative case, may not be compatible with these nominals. Thus, as the reviewer notes, these elements display similar behavior to indeclinable elements in some Slavic languages (see e.g. Bošković 2006).
he served bad.n and good.n 'He served bad and good.' Having reviewed basic facts of Lithuanian adjectival inflection and gender, we now turn to our discussion on default gender and non-agreement. In the rest of this paper, we generally gloss gender morphology and exclude number and case values on adjectives for ease of exposition, unless otherwise indicated.

Default gender vs. non-agreement forms
In this section, we provide evidence from Lithuanian PAs demonstrating that the language has two types of defaults: gender-default forms and non-agreement forms. We show that PAs agreeing with a subject with default gender appear with masculine morphology, whereas PAs appear with neuter morphology in cases of non-agreement, where the PA cannot agree in gender features with a subject. We use syntactically established tests on defaults from Corbett (1991), Kramer (2015) and others to illustrate this split.

Gender default
Some researchers have suggested that masculine is the default or unmarked gender in Lithuanian (Mathiassen 1996;Holvoet & Semėnienė 2006;Bruno 2012). We provide extensive evidence below demonstrating that masculine indeed functions as the default gender. 5 While pronominal inflection is not the focus of our article, we note that neuter pronouns such as visk-as 'everything' inflects similarly (though not identically) to some masculine nouns (e.g. mišk-as 'forest.m.sg. nom'). We thank an anonymous reviewer for pointing this out. However, we would first like to note that in another line of work, Armoskaite (2011) outlines the possibility that feminine might be a default gender in Lithuanian. She points out that a subclass of nouns that refer to people of either gender, such as dabit-a 'dandy' (15) end with the (nominative singular) suffix -a, which is otherwise typically used with feminine nouns (e.g. ras-a 'dew-f'). One may interpret the availability of feminine declension on these nouns as evidence that feminine is a type of default.
(15) Armoskaite (2011: 120-121) a. Ta moteris yra tikr-àdabita. that.f woman.f is real-f dandy 'That woman is a real dandy.' b. Tas vyras yra tìkr-as dabita. that.m man.m is real-m dandy 'That man is a real dandy.' Even though the -a declension is often associated with the feminine, the noun triggers masculine agreement when it refers to men, as indicated by the masculine adjective 'real' in the example (15b). Nouns like dabita therefore select their declension class independently of gender (Armoskaite 2011: 121). This suggests that declension class and gender are indeed dissociated (see Harris 1991;Alexiadou 2004;Armoskaite 2011;Kučerová 2018;among others). Given this dissociation, we do not treat the evidence from these nouns as indicating that feminine is the default gender.
We provide five pieces of evidence below for masculine being the default gender in Lithuanian.

Mixed-gender groups
Our first two diagnostics pertain to the semantics of gender with animates. Evidence for treating masculine as a default gender comes from mixed-gender groups (see also Holvoet & Semėnienė 2006: 106;Paulauskienė 2007: 72;Bruno 2012: 57). Lithuanian has a class of nouns that can be either masculine or feminine gender depending on the sociocultural gender of the referent, e.g. nouns like atlet-ai 'athletes-m' or atlet-ės 'athletes-f.' When referring to a group consisting of both men and women, the nominal triggers masculine agreement, including on PAs (16). However, when the nominal is feminine, it can only refer to a group of women, and the PA in these cases instead takes a feminine form (17). This supports the unmarked status of the masculine.

Unknown gender
Another piece of evidence concerns unknown gender. Some nominal roots are compatible with either masculine or feminine PAs depending on the gender of the referent. When these roots appear in subjects referring to people of unknown gender, PAs inflect as masculine (e.g. ligon-is 'patient.m' in (18)), which is another indication that masculine is the default gender (Kramer 2009: 17). The feminine DP ligon-ė 'patient-f' can only be used for female referents (19).

Coordination of gender-mismatched DPs
The third diagnostic comes from coordination of gender-mismatched DPs with both animates and inanimates, which shows a "resolving" form of gender agreement on PAs.

Loanwords
We now turn to diagnostics pertaining to the morphological assignment of default gender. Our next piece of evidence for a masculine default comes from gender assignment to loanwords (Corbett 1991: Ch. 4). Bruno (2012: 79) shows that loanwords referring to inanimates are assigned masculine in Lithuanian, lending credence to the idea that masculine serves as a default gender, as exemplified in (24) Building off these observations, we add that inanimate loanword subjects can only trigger masculine and not neuter on PAs, suggesting default assignment is not relativized to the attributive or predicative position.

Onomatopoeia and interjections
While the other tests have probed the representation of gender on PAs, the gender assignment on attributive adjectives also provides important insights. 8 Observe that onomatopoeic sounds can be modified with attributive adjectives, in which case the adjective must be masculine; neither feminine nor neuter is possible (28). 9 7 According to Bruno (2012), loanwords referring to animates, unlike the inanimate expressions, are assigned according to sociocultural gender, as in (i). Furthermore, the assignment of gender may also be influenced by phonological factors, i.e. phonological resemblance to other nouns in the language. One example would be past-a 'pasta-f', which is assigned feminine gender due to its phonological similarity with Lithuanian words belonging to the -a declension (e.g. ras-a 'dew-f'). Our examples presented here are assigned the masculine default gender because they are not conditioned by sociocultural or phonological factors.
(i) pastov-ùs member-is / pastov-ì member-ė usual-m member-m / usual-f member-f 'a usual member' 8 While attributive adjectives are possible with onomatopoeia and interjections, PAs are semantically odd. We therefore test gender inflection of adjectives by using attributives rather than PAs. 9 A reviewer raises the possibility that the attributive adjective takes masculine inflection in examples like (28) because there is an elided noun garsas 'sound' that accompanies this onomatopoeia. Indeed, it is possible to use brrr both with and without an overt noun garsas (i). However, this type of ellipsis is ungrammatical with other phrases in which a noun is modified by another noun. For example, the noun liepų 'linden' is marked with genitive case and it modifies the head noun 'street', but the head noun cannot be elided (ii). Given this contrast, we suggest that the onomatopoeic examples in (28) do not involve elided nouns.

(28)
Adapted from Mathiassen (1996) Mes išgirdome tyl-ų/*-ią/*-ù brrr. we.nom hear.pst.1pl quiet-m/*-f/*-n brrr 'We heard a quiet brrr.' As was illustrated in Section 2, the same type of behavior can be observed with interjections such as oj. Only masculine adjectives can modify the interjection; feminine and neuter adjectives are ungrammatical. 10 (29) Buvo tik tyl-ùs/*-ì/*-ù oj iš mano lūpų ir viskas… be.pst.3 only quiet-m/*-f/*-n intj from my lips and everything 'There was only a quiet oj from my lips and that's it.' (Internet example) 10 Interim summary In this subsection, we have investigated the distribution of gender on PAs in default gender environments. Masculine agreement on PAs is triggered by subjects that are nominals referring to mixed-gender groups, nominals referring to a person of unknown gender, gender-mismatched coordinated nominals, and loanwords. Neither feminine nor neuter forms are available on PAs in these environments, which we take as evidence that masculine is a default gender. In addition, attributive adjectives modifying onomatopoeic sounds and interjections also occur in masculine forms, which we take as further evidence of masculine's default status. 11

Non-agreement
While masculine appears to behave as a default gender, we employ various default diagnostics in this subsection that trigger neuter on PAs. We show that in cases of non-agreement, where a gender feature is not transmitted from a subject to a PA, the adjective appears in the neuter.

Non-structural case
We first discuss instances of non-agreement where the gender features of the subject are not accessible for agreement with a PA. The first piece of evidence comes from grammatical subjects with non-nominative case.
Lithuanian is like Icelandic (on which, see Zaenen et al. 1985;i.a.) in that it has nonnominative grammatical subjects, which bear non-structural cases but otherwise exhibit properties of canonical nominative subjects (Šereikaitė in prep). For example, predicates like trūkti 'to lack', užtekti 'to have enough', stigti 'to be short of', pakakti and 'to suffice' take a dative subject. The dative DP behaves like a subject in that it binds a subject-oriented anaphor savo as in (30) (Legate et al. 2019). The non-nominative subject does not agree with the predicate, and instead the verb displays third-person morphology, which is default morphology in the language (Šereikaitė in prep). We follow H. Sigurðsson (1996), Bobaljik (2008), Preminger (2011, and E.F. Sigurðsson (2017), among others, in taking this to be the result of agreement not obtaining.

(31)
Man trūkst-a/*trūkst-u pinigų. me.dat lack-prs.3/lack-prs.1sg money.gen 'I lack money.' Dative subjects can also occur in constructions with PAs, as in (32)-(33). The dative experiencer here patterns as a non-nominative subject in serving as a binder of the subject-oriented savo. Crucially, in these cases, the PA can only be neuter. We take this to be another instance of non-agreement of the PA, which strikingly takes neuter.

(33)
Man i buvo bais-ù/*-ùs/*-ì net ir savo i kambaryje. me.dat be.pst.3 scared-n/*-m/*-f even and self's room 'I felt scared even in my own room.' In contrast, observe that Lithuanian nominative subjects in constructions similar to (32) trigger agreement on PAs, which is a well-known fact from other languages such as Icelandic. In these cases (34), the PA cannot be neuter; it is either masculine or feminine depending on the gender of the subject.

Pancake Sentences
The second diagnostic is based on so-called "Pancake Sentences". In Scandinavian languages such as Swedish and Norwegian, certain PAs inflect for neuter singular regardless of the features of a subject (Hellan 1986;Enger 2004;Josefsson 2006;Wechsler 2013; cf. also Danon 2012; 2014 on Hebrew ze sentences). For example, in (35), the subject bears common gender and plural number, but the PA is neuter singular. This yields a particular interpretation of the subject; in (35), that is 'eating pancakes is good'.

(35)
Swedish (adapted from Wechsler 2013) Pannkak-or är gott. pancake-comm.pl be.pres good.n.sg 'Situations involving pancakes are good.' (i.e. eating pancakes is good) An analogue to Scandinavian Pancake Sentences can be found in Lithuanian. Certain PAs take neuter even when they occur with masculine or feminine subjects as in (36)-(37). Ambrazas et al. (1997: 137) observe that PAs take neuter when they express the "meaning of a generalized quality", and also "they are never attributes to a noun, and the quality they refer to is never an attribute to a concrete thing". We take one such instance of this to be Pancake Sentences (for other discussion of different environments with neuter, see Ambrazas et al. 1997: 135-137). We further observe that, like Scandinavian Pancake Sentences, neuter PAs in these constructions yield a certain interpretation. These sentences have an eventive interpretation; hence, (36a) means the pancakes 'are healthy to eat' rather than 'are in a healthy state', which is the interpretation with an agreeing PA, as in (36b). The same contrast can be observed in (37).

Infinitival subjects
Now we turn to a case of non-agreement where the subject lacks gender features. Infinitives can appear as subjects of copular sentences (38), in which case they trigger neuter on PAs. We take infinitival subjects to lack gender features, following other authors (Corbett 1991;Corbett & Fraser 1999;Kramer 2015); also see Section 4.3 for discussion.
(38) Ambrazas et al. (1997: 645) Pavargti už tėvynę yra graž-ù/*-ùs/*-ì. suffer.inf for homeland is beautiful-n/*-m/*-f 'To suffer for one's homeland is beautiful.' An anonymous reviewer points out that in these types of sentences, a subject position may potentially be filled with a null expletive, with the to-infinitive clause topicalized. However, the reviewer suggests that this hypothesis is ruled out due to the existence of dative quirky subjects (discussed in Subsection 3.2.1). Crucially in those cases, the subject position is filled by the dative DP rather than a potentially expected null expletive.
Evidence from genitive of negation in Lithuanian suggests that to-infinitive clauses in copular sentences do not behave like A-bar-moved elements, which we would expect if there was a null expletive subject. Lithuanian genitive of negation tracks objects which would otherwise be assigned a structural accusative case, as illustrated in (39a) When the matrix verb is negated, the object of a to-infinitive complement can also be affected by genitive of negation (Arkadiev 2016). Specifically, the object can bear genitive case as well as accusative, as exemplified in (40)-(41). Genitive of negation is retained when the complement is topicalized, thus undergoes A-bar movement, as in (42) The behavior of the to-infinitive clause in copular sentences is parallel to a grammatical subject that is a neuter pronoun, which also cannot appear with genitive case. (44) Viskas/*visko ne-buvo taip leñgv-a. everything.nom.n/everything.gen.n neg-be.3.pst that easy-n 'Everything was not that easy.' We therefore conclude that to-infinitive clauses can be subjects of copular sentences, and thus that the PA is neuter because these clauses lack gender.

Weather-type constructions
Our final case of non-agreement occurs when the subject is not projected, as in weather-type constructions. Weather-type constructions have been suggested to trigger default agreement (e.g. Corbett 1991: 204 on neuter in Russian; Kramer 2015: 143 on feminine in Zayse and Zargulla). In Lithuanian, PAs in weather-type constructions appear with the neuter, with masculine and feminine being ungrammatical (45).
We attribute the non-agreement behavior of PAs in these environments to non-projection of a subject. This is evidenced by a contrast between weather constructions with verbs versus weather constructions with PAs. In (46), weather constructions with verbs such as lyti 'to rain' are grammatical with overt cognate subjects. In contrast, PAs do not allow cognate subjects as in (47). The availability of cognate subjects with weather verbs points to a plausible analysis with weather pro; however, this does not hold for weather-type adjectives, which we take to suggest that no such argument is projected. The non-projection of weather pro in these constructions results in non-agreement. (Note that the language lacks overt expletives).
outside darkness-f is dark-n Intend. 'It is dark (outside).' To conclude, we showed that PAs inflect in the neuter and not in masculine or feminine in non-agreement environments, i.e. when the features of a subject are not visible to a PA. We have distinguished three environments where the non-agreement takes place: i) when the features of the subject are not accessible for agreement, as has been shown with subjects with non-structural case and Pancake Sentences; ii) when the subject lacks gender, as was the case for infinitives; and iii) when the subject is unprojected, as in weather-type constructions.

Summary
In this section, we have demonstrated that Lithuanian presents a split in the distribution of masculine and neuter PAs in environments identified in the literature as being associated with defaults. The masculine PA forms occur in situations where the gender of the subject bears unmarked gender. Specifically, masculine on PAs occurs with subjects which are loanwords, coordinated expressions, nominals referring to a mixed group or referents with unknown gender. In contrast, neuter PA forms occur in situations when gender features of the subject are not transmitted, which we referred to as non-agreement. Given our results, feminine appears to be the marked gender in the language. The results of our diagnostics are summarized in Table 3.
The split behavior we have identified points to the presence of two classes of default diagnostics, which are often conflated in the literature on defaults. In this section, we have provided evidence that one class identifies the unmarked gender feature, whereas the other identifies feature absence.

Analysis
Having examined the split between masculine and neuter PAs in default environments, we now present a morphosyntactic analysis of this split. When PAs agree with a subject that bears unmarked gender, the realization on PAs is masculine. However, when the subject lacks accessible gender features, the realization of the PA is neuter. We model this distinction in terms of feature representation: while masculine and feminine share a feature [gend], neuter corresponds to the absence of gender features. We propose that gender features originate on the nominalizing head n (following Lecarme 2002;Lowenstamm 2008;Kramer 2015;, and claim that in Lithuanian, n must bear gender features; that is, there is no "neuter" n. We extend our analysis to the behavior of neuter pronouns, deadjectival nominalizations, and related phenomena.

Theoretical framework
This subsection consists of two parts: first, we discuss the theoretical background for gender features originating on the nominalizing head n in Section 4.1.1, and then formalize how agreement and non-agreement interact with the realization of PA inflection in Section 4.1.2.

Gender representation and n
We situate our analysis within the Distributed Morphology (DM) framework (Halle & Marantz 1993;Harley & Noyer 1999;Halle 1997;Embick & Marantz 2008; among many others). Following much of the DM literature, we assume categories like noun and adjective are decomposed into a root and a category-defining head (Marantz 1997;2001;Arad 2003;Embick 2010), and that roots must combine with categorizing heads (Embick & Marantz 2008: 6; see also Embick 2015). As sketched in (48), a root combines with n to form a noun or it can combine with a to form an adjective.
(48) a. b. balt-um-a white-n-nom.f.sg 'a white place' 13 Note that the inventory of n heads includes not only a masculine n realized as -∅ but also a feminine n realized as -∅. For example, the the following nouns marked with feminine include a zero nominalizer as in kėd-∅-ė 'chair-n-nom.f.sg' or knyg-∅-a 'book-n-nom.f.sg'. In addition, it is also possible to have overt masculine n heads, e.g. vasar-∅-a 'summer-∅-nom.f.sg' vs. vasar-oj-us vs. 'spring crops-n-nom.m.sg'.
gender of the noun can also be observed through the inflection on agreeing elements such as attributive adjectives. For completeness, we include case, gender, and number information in the glosses here. Lithuanian has a number of other nominalizing suffixes that determine the gender; for discussion and overview of these nominalizers, see Armoskaite 2011; 2014.
For Lithuanian, we propose that the inventory of n heads all bear a gender feature, which we encode as the feature [gend]. Thus the feature [gend] will be included in the representation of both masculine and feminine, since these are the true gender categories in the language. Furthermore, as our results from Section 3 showed, the feminine gender is more marked than masculine. We encode this markedness relationship by analyzing the masculine as only having [gend], while the feminine bears the additional feature [fem], as indicated below in (52). This reflects markedness through feature containment (Bobaljik 2012; also see Harley & Ritter 2002 for a related approach based on feature geometry), where the representation of the more marked gender contains the representation of the unmarked gender. The predictions of this account are explicitly discussed and compared with alternative analyses in Section 4.2 and Section 4.4. Our feature representation of ns is illustrated in (53 While gender features originate on n, they are expressed along with number and case on inflectional suffixes, which we refer to as XInfl (i.e. aInfl, nInfl, etc.). XInfl expresses syntactic features of heads from which they are "dissociated" in the postsyntax (Embick 1997;Embick & Noyer 2007;Kramer 2009;Norris 2014). For example, the noun vakaras consists of a root VAKAR , a null nominalizer that bears [gend] feature, thus masculine, and the nInfl suffix -as, which expresses gender, number, and case features, 14 as in (54a). The noun vakarienė 'supper' is built from the same root, but instead has an overt n suffix -ien that bears [gend][fem], which makes the noun feminine. The example in (54) presents an instance in which a single root can combine with ns of different genders. An anonymous reviewer wonders why it is not the case that any root can combine with ns of any gender. We assume that the selectional relationship between roots and ns is governed by specific semantic and arbitrary licensing conditions (see Kramer 2015: Ch. 3 for discussion). For instance, (54a) is ungrammatical when a feminine n realized as -∅ is used (*[ VAKAR -∅-a] 'evening-n-nom.f.sg'); though this could in principle be built in the morphosyntax, it may not meet the licensing conditions imposed by the interfaces.
Unlike n, gender features are never inherent to adjectives, which yields the following inventory (55).

(55)
Types of ns Types of as

n[fem][gend] a n[gend]
Under our approach, Lithuanian lacks "plain" featureless n; i.e. there is no "neuter" n. This is motivated by the following factors. First, all lexical nouns in the language have gender, being inherently either masculine or feminine, and not neuter. This is reflected in the inventory of ns in (55). Second, attributive adjectives are never neuter (see Section 2). This is consistent with the widely assumed idea that attributive adjectives modify nP.

Agreement vs. non-agreement
We now discuss how gender features are transmitted to PAs. Given that adjectives do not have gender features inherently, we assume that PAs agree with a subject through upward Agree (Baker 2008). 15 Agreement with nominals is with the features of the topmost projection in the nominal domain (Danon 2011;, typically D, which collects features originating internal to the nominal through agreement, including the gender features on n. That D collects gender features is evidenced by inflection on demonstratives, which agree in gender (56). Agreement between a PA and a DP subject is illustrated in (57).
(56) ta vakar-ien-ė that.f supper-n-f 'that supper' Recall that adjectives exhibit a three-way distinction expressed on aInfl: masculine, feminine, and neuter.When an adjective agrees with a nominal bearing [gend][fem], aInfl will exhibit feminine agreement. When the adjective agrees with a nominal bearing [gend], aInfl will exhibit masculine agreement. In contrast, in cases of non-agreement, where the adjective does not have access to features of a nominal to agree with, the adjective will be realized with neuter inflection, which we propose corresponds to the absence of gender features. In the postsyntax, aInfl will be dissociated from a. aInfl will be realized according to the Elsewhere Condition, with neuter forms being the least specified, lacking gender features (cf. Kramer 2015). Masculine and feminine forms, in contrast, will be specified for gender features. This is shown in (58) for the adjective gražu.
Because gender features originate on n, our proposal predicts that adjectives are neuter when they fail to agree with nominals. Hence, non-agreement occurs: i) when a subject lacks n or is unprojected or ii) when agreement between a PA and an argument is disrupted. In contrast, PAs appear in masculine when they agree with a subject that has an unmarked gender feature originating on n. We now discuss how our analysis applies to each of the cases discussed in Section 3.

Analyzing default gender
PAs inflect as masculine in default environments when they agree with DP subjects with unmarked gender. The PA agrees with the topmost projection D, which receives [gend] from n when n is projected. We identify three types of situations when the PA inflects for the unmarked gender (masculine). The first case of unmarked gender inflection on the PA is when the interpretation of the subject does not have a stronger presupposition associated with the feminine gender. Lexical nouns referring to mixed-gender groups (consisting of men and women) and people of unknown gender are derived from a root with n, which bears the feature [gend]. This is consistent with the fact that the PA is inflected in the masculine for both mixed-gender groups (59) and unknown gender (60), as was discussed in Sections 3.1.1 and 3.1.2, respectively.
Atletai yra aukšt-ì/*-os/*-a. athletes.m are tall-m/*-f/*-n 'Athletes (male and female) are tall.' Ligonis yra judr-ùs/*-ì/*-ù. patient.m is restless-m/*-f/*-n 'A patient (unknown gender or male) is restless.' The nominalizing head n with [gend][fem] is not available for either of these cases because when referring to animates, as the feminine feature introduces a stronger presupposition about the sociocultural gender of the referents, namely that they are women. This presupposition is inconsistent with an individual of unknown gender, as it suggests that the referent is known to be a woman. For mixed-gender groups, as is well-known, marked gender presuppositions must apply to each individual in a group. 16 Consequently, feminine is available for these nouns when all of the referents are known to be women (62). 17 In this case, n would bear [gend][fem]. (In this discussion, we are only referring to animate-denoting roots that are compatible with either masculine or feminine. We note that some animate nouns have fixed arbitrary gender, and in those cases, gender presuppositions play no role in the determination of gender of the nominal).
The second situation of PA inflection for unmarked gender is when a subject consists of two coordinated nominals with different genders. As discussed in 3.1.3, coordinated gender-mismatched nominals yield masculine inflection on PAs as in (63). Observe that for completeness we include both singular and plural forms for PAs.

(63)
Stalas ir kėdė buvo purvin-ì/*-as/*-os/*-à/*-a. This follows from our analysis if we assume gender features are syntactically resolved in coordination through set intersection (Wechsler & Zlatić 2003;Börjars & Vincent 2006;Wechsler 2008). In essence, the gender features of the two conjuncts will be percolated up to the coordinate phrase when these features are shared by both conjuncts (see also Anagnostopoulou 2017). Coordination of two feminine nominals will result in feminine inflection on a PA (64)-(65); coordination of masculine nominals will yield masculine by the same principle (not included Kėdė ir lentyna buvo purvin-os/*-ì/*-a. chair.f and shelf.f were dirty-f.pl/*-m.pl/*-n 'The chair and the shelf were dirty.' In our analysis, masculine and feminine share a feature in common, namely [gend]. Therefore, intersecting the features of the two conjuncts yields [gend] as in (66), and this value percolates up to the coordinate phrase. Thus we correctly derive the fact that gender resolution yields masculine and not neuter, as in (63). Observe that if we used an alternative analysis where feminine is [+fem] and masculine is [-fem], we would not be able to derive the coordination facts. This is because the two would have no feature in common, and therefore, their intersection would be empty, thus incorrectly yielding neuter.
Our analysis makes a further prediction for resolution. It is possible to coordinate gendermismatched conjuncts where one is a neuter pronoun. If neuter corresponds to the absence of gender features, we predict that coordinating a neuter pronoun with another nominal, regardless of its gender, will result in neuter inflection on a PA. As we show in (68), this prediction is borne out. 19 The set intersection analysis specifically applies to gender, and does not extend to the resolution of number, which may require set union instead. We thank an anonymous reviewer for pointing this out. 19 One may wonder about the possibility that the PA is neuter in (68) because it is a Pancake Sentence. However, recall that Pancake Sentences require an eventive, generic interpretation (Section 3.2.2). However, this interpretation is not available with predicates like 'dirty' (i), indicating that (68) is not a Pancake Sentence.

(71)
Viskas, kas yra sudėta čia, ir stalas štai ten yra everything.n that be.prs.3 placed here and table.m right there is purvin-a/*-os/*-à/*-ì/*-as. dirty-n/*-f.pl/*-f.sg/*-m.pl/*-m.sg 'Everything placed here and the table over there is dirty.' The last situation is when n is projected and is assigned the unmarked gender rather than the marked gender (feminine). This occurs in cases of loanwords, which we discussed in Section 3.1.4. When a loanword enters the language as a noun, we propose that it enters as a root, which must be nominalized with n, realized as null in (72). Because roots must appear with categorizing heads, loanwords are nominalized with n, being assigned the unmarked masculine by default.

(72)
bin-∅-as 'bin-n-m' [ BIN n [gend] ] n nInfl -as n n -∅ bin Agreement for gender features between the subject and a PA proceeds unhampered, and the PA will receive a value for [gend]. In other words, the neuter is unavailable on PAs because loanword subjects have a projected n bearing [gend]. This feature is visible for agreement with D, and consequently with the PA, resulting in it bearing masculine, as in (73).

(73)
Binas yra žal-ias/-*ià/-*ia. bin.m is green-m/*-f/*-n 'The bin is green.' This analysis of loanwords also applies to onomatopoeia, and interjections, which we discussed in Section 3.1.5. Recall that onomatopoeia, and interjections can be modified with masculine attributive adjectives, as in (74)-(75). They too must appear as roots, and must therefore be nominalized by n to be modified by an attributive adjective. No neuter n is available, so in the absence of sociocultural or phonological conditioning factors for assignment (see fn. 7), the root merges with the n with unmarked [gend]. If neuter were a possible gender on n, we might expect attributive adjectives to inflect as neuter when modifying nouns assigned a default gender such as interjections or onomatopoeia, contrary to fact.
(74) tyl-ùs/*-ì/*-ù oj quiet-m/*-f/*-n intj 'quiet oj' (75) tyl-ùs/*-ì/*-ù brrr quiet-m/*-f/*-n brrr 'quiet brr' Interim summary Our analysis captures the distribution of masculine as a default gender by connecting gender features to the nominalizing head n, and the data support our proposed feature representation for neuter, masculine, and feminine. We now proceed to our discussion of how PAs can fail to have their gender features valued, resulting in neuter inflection.

Analyzing non-agreement
PAs inflect as neuter when they fail to receive features from a subject; in other words, when there is no agreement between the PA and the subject. 21 We now proceed to illustrate three environments where the non-agreement occurs.
The first environment where the PA is neuter is when the gender features of the subject are not accessible for agreement. This happens in situations with non-nominative subjects (76), which we also discussed in Section 3.2.1.

(76)
Man buvo šalt-a/*-as/*-à. me.dat was cold-n/*-m/*-f 'I felt cold.' We suggest that in cases like (76), the dative experiencer enters the derivation with φ-features, but the PA cannot agree with the DP due to the case-discriminating nature of the probe (Preminger 2011;E.F. Sigurðsson 2017). For concreteness, we adopt the idea that case-discrimination is the result of a probe attempting to agree with an argument that has non-nominative case, which causes the probe to abort. This is illustrated in (77), with two slashes indicating non-agreement. The result is that the PA does not value features, and consequently, is realized with neuter inflection. 22 (77) DP dat AP root a // 21 This can be stated in a Failed Agree approach, whereby Agree can fail to value the features on the probe, and this does not cause the derivation to crash, as proposed by Preminger (2011;. 22 An anonymous reviewer asks what triggers a probe if the adjective lacks gender features. We assume that when adjectives enter the derivation, they do carry unvalued φ-features, including u[gend], which triggers the probe's search for a goal. When the probe fails to find a goal with valued [gend], the inflection on the adjective is realized as neuter. Another situation where nominal features are not visible to the PA is in Pancake Sentences (see Section 3.2.2). Observe that the subjects in these instances do project n, as indicated by the agreement of the subject with the attributive adjective in (78).
(78) Trumpos kojinės yra graž-ù/#-ios/*-ūs. short.f socks.f are nice-n/#-f/*-m 'Short socks are nice (to wear).' We propose that the topmost layer of Pancake subjects includes an ZP stacked on top of a DP as in (79). The ZP is responsible for the interpretation of the subject as eventive and generic. Even though the subject has n, we propose that the gender features of Pancake subjects originating on n do not percolate to their topmost projection ZP (cf. Proposing additional structure for these sentences is similar in spirit to Josefsson 2009, 23 which introduces additional covert structure for the subject of Scandinavian Pancake Sentences, and is also similar to Danon's (2012) study, which suggests that the D layer of the subject in Hebrew ze sentences are featurally "defective", yielding non-agreement. The precise nature of ZP is beyond the scope of our paper, and we leave it for further research. The second case where the PA is neuter is when the subject lacks n, and therefore lacks gender, resulting in non-agreement. Infinitival subjects (see Section 3.2.3) present an example of this type of subject. The evidence for the lack of n in infinitival subjects comes from their inability to occur with attributive adjectives, as in (80), which should adjoin to nP.
(81) Jam gręsia stipr-ùs/*-ù per-si-gėr-im-as. he.dat threaten heavy-m/*-n pfv-rfl-drink-n-m 'He (his health) is threatened by heavy over-drinking.' The interpretation of -im/-ym deverbal nominalizations can be generic. We propose that like Pancake Sentences, generic interpretations of -im/-ym nominalizations are derived through the addition of a ZP. Under such an interpretation, the -m nominalization triggers neuter on the PA.

(82)
Stripr-ùs per-si-gėr-im-as yra labai ne-sveik-a/#-as/*-à. strong-m pfv-drink-n-m is very neg-healthy-n/#-m/*-f 'Heavy over-drinking is very unhealthy.' The last environment where the PA is neuter is when the subject is not projected, as in weather-type constructions like (83), which we discussed in Section 3.2.4. Crucially, as suggested in Section 3.2.4, we take these constructions to be different from regular weather constructions with a projected weather pro because constructions with neuter PAs do not allow a nominal argument, whereas constructions with weather verbs like 'rain' do.
(outside) dark-n/*-m/*-f 'It is dark (outside).' We analyze these cases of non-agreement as lacking projected subjects. Consequently, a PA has no argument to agree with, yielding neuter inflection on a PA. (For other potential analyses of weather-type constructions, see also Schäfer 2012; Wood 2017).
(84) AP // Summary In this subsection, we applied our analysis to derive the behavior of neuter as a non-agreement form. We observed that neuter inflection on PAs results from agreement disruption or the non-projection of n. In 4.4, we extend our analysis to other cases of default behavior.

Extensions
We now apply our analysis to other phenomena in the language, suggesting that neuter pronouns and deadjectival nominals lack n, and therefore lack gender. We further discuss how the morphology of neuter adjectives is consistent with our analysis of gender features, which we have represented in terms of containment relations.

Neuter pronouns and their features
Our analysis places inherent gender features solely on n, with neuter being the absence of gender. This correctly predicts that there are no neuter nouns in the language. However, recall from Section 2 that the language does have neuter pronouns, e.g. viskas 'everything.n', which always appear with neuter inflection on PAs was illustrated in (4), repeated here in (85). 24 (85) Viskas yra graž-ù/*-ùs/-*ì. everything.n is beautiful-n/*-m/*-f 'Everything is beautiful.' To account for this, we extend the n hypothesis to pronouns, and propose that in Lithuanian, masculine and feminine pronouns project n, in the spirit of theories that posit complex internal nominal structure in pronouns (e.g. Bjorkman 2017; Panagiotidis 2019 for gender). In contrast, neuter pronouns, as we argue, do not project n, and therefore lack gender. We contend that our analysis is supported by the absence of a number alternation for neuter pronouns. We follow Picallo (2006) and Alexiadou et al. (2011) in taking the projection of n to be required to support the projection of Num. We therefore predict that the language lacks a number distinction between singular and plural for neuter pronouns, unlike the corresponding masculine and feminine. This is borne out, as illustrated by the morphology of demonstrative pronouns in (86), which show a distinction between masculine, feminine and neuter. We take the lack of the distinction between singular and plural forms to suggest that neuter pronouns lack number features in general. It is important to note that the data presented here may constitute evidence that gender and number features in Lithuanian are bundled together, as pointed out by a reviewer; for discussion of this type of bundling approach, see Kramer 2015;2016. (86) a Neuter pronouns can be modified by postnominal adjectives. As (i) shows, these adjectives appear in neuter forms. However, we do not treat these adjectives as attributive. Observe that attributive adjectives can be stacked (ii), while postnominal adjectives cannot (iii) (cf. Larson & Marušič 2004). One plausible analysis is that these adjectives occur in reduced relative clauses (cf. Cinque 2010; 2014); we leave this to future research.
(i) Girdėjau, jog viskas graž-ù yra tik tai, ką pamatome ant scenos. heard, that everything.n beautiful-n yra only that, which see on stage 'I heard that everything beautiful is only what we see on the stage.' https://www.minfo.lt/posts/l-motiejauskaite-pasirinkus-bet-kokia-sporto-saka-reikia-savotiskouzsispyrimo-ir-kad-tai-teiktu-emocine-arba-fizine-nauda (Accessed 11/17/18) (ii) Kiekvien-as brang-us žãl-ias daikt-as yra ant lentynos. every-m expensive-m green-m thing-m is on shelf 'Every expensive green thing is on the shelf.' (iii) *Viskas žãl-ia brang-ù /brang-ù žãl-ia yra ant lentynos. everything-n green-n expensive-n /expensive-n green-n is on shelf 'Everything that is expensive and green is on the shelf.' Based on this distinction, we assume that the structure of pronouns is as sketched in (87). Observe that masculine and feminine personal pronouns project n, and therefore can project Num, while in contrast, neuter pronouns do not project n and therefore cannot project Num. Since some neuter pronouns are also quantified (such as viskas 'everything'), we also include Q in their representation. The last point we would like to make about neuter pronouns concerns case features. We have demonstrated in Section 2 that neuter pronouns like viskas 'everything' and taĩ distinguish different case forms (e.g. visk-ą 'everything-acc.n'). Thus, while these categories lack gender and number, they seem to have case. This is consistent with our analysis, since the D head more generally can bear case features (see Šereikaitė 2019 for evidence of D bearing case features in Lithuanian).

Deadjectival nominals
Another phenomenon to which we can extend our analysis is deadjectival nominals, as in (89). As can be observed, neuter deadjectival nominals can be copular subjects (Ambrazas et al. 1997: 136), in which case they must trigger neuter on a PA. 25 The other possibility is that gender features fail to percolate to the topmost projection of the DP, as was our analysis for Pancake Sentences. The most plausible candidate for preventing percolation is a quantifier, as several of the neuter pronouns are quantificational. Quantifiers, however, do not trigger neuter on a PA when they occur with nouns (i), thus something else would need to be said for their variable blocking behavior.

(89)
Modified from Ambrazas et al. (1997: 136) Sald-ù yra gard-ù/*-ùs/*-ì. sweet-n is delicious-n/*-m/*-f 'Sweet is delicious.' Despite their adjectival morphology, deadjectival nominals share characteristics with nominals in that they appear in argument positions, as mentioned in Section 2. For example, they can occur in accusative object positions, as in (12), repeated in (90a)-(90b). Although they resemble nominals in this respect, we propose that neuter deadjectival nominals lack n, just like infinitives (see Section 4.3), resulting in the neuter inflection seen on the PA. Support for this analysis comes from a comparison between the nominalizations that do involve n and the deadjectival nominals that lack it (for similar distinction see Alexiadou et al. 2012;Alexiadou & Iordǎchioaia 2014). We first note that the deadjectival nominals allow adverbial but not adjectival modification (91a), whereas overtly suffixed nominalizations show the opposite pattern (91b).
(92) a. Saldž-iau yra ger-iau. sweet-comp.n is good-comp.n 'Sweeter is better.' b. *saldž-iau-um-as sweet-comp-n-m 'sweeter-ness' These diagnostics suggest that deadjectival nominals lack internal nominal syntax and instead behave like adjectives. However, as we have pointed out, they are able to appear in argument positions (90a), and thus appear to behave as though they have "external" nominal syntax.
Our analysis draws from a precedent from recent literature showing that certain types of nominalizations are derived without n (Alexiadou et al. 2011;Bošković 2013;Villalba 2013;Alexiadou & Iordǎchioaia 2014;Iordăchioaia 2014;among others). Because these deadjectival nominals lack n, they consequently lack gender. This accounts for both the neuter form of the deadjectival and the fact that they appear with PAs with neuter inflection (89). We assign them the structure in (93), which lacks the nominalizing head n, and therefore lacks gender. We include a DegP to capture the fact that these deadjectivals can appear with comparatives. 26 (93) DP DegP aP root a Deg D As far as case features of these adjectives are concerned, recall that deadjectival nominals can appear in an accusative object position (90). We propose that these deadjectival nominals do indeed receive case values and these values are realized with the same neuter forms. In other words, adjectival neuter inflection is syncretic across case values (see Table (1)).

Morphology of neuter adjectives
Lastly, we relate our analysis to the morphology of neuter adjectives. Recall that our representation of the feminine corresponds to the features [gend] [fem], which contains the representation of the masculine, [gend], while the representation of the neuter lacks gender features. Assuming the Elsewhere Condition, our analysis predicts that it is impossible to syncretize neuter and feminine for aInfl to the exclusion of the masculine (all else being equal). This would be a type of *ABA syncretism (see Bobaljik 2012;Smith et al. 2018; among others).
Recall from Section 2 that adjectives of the iu-class do not exhibit gender syncretism: all three forms have different inflections; examples are provided in Table 4 for the nominative 26 It is reported by Ambrazas et al. (1997) that masculine adjectives can be used in situations similar to the neuter deadjectivals (i). However, they are not always interchangeable (ii) (cf. (89)) and the masculines are in fact restricted in their distribution (ii), tending to have a specific interpretation, possibly suggesting they involve elided nouns or additional nominal structure. We leave these cases to future research. A reviewer notes that the use of masculine adjectives as in (i-ii) is similar to the use of long adjectives in predicative positions in Russian, which have also been analyzed as involving ellipsis or a null noun (e.g. see Bailyn 1994).
sweet-m is delicious-m 'Sweet is delicious.' Table 4: ABC pattern of adjectives.

A B C n m = [gend] f = [gend][fem]
(i)u-class graž-ù graž-ùs graž-ì sald-ù sald-ùs sald-ì (i)a-class žãl-ia žãl-ias žal-ià gẽr-a gẽr-as ger-à singular. In other words, they only exhibit ABC patterns. The adjective class -(i)a distinguishes between neuter and feminine forms with stress, even though the inflectional forms appear to be the same, as shown in Table 4. We take the stress facts to indicate that neuter and feminine forms are in fact distinct, and therefore also exhibit ABC patterns, not ABA. One apparent counterexample comes from a subset of derived adjectives belonging to the (i)a class, which share an inflectional ending and are stressed the same way in the neuter and the nominative feminine singular.
(94) a. dulk-ėt-a dust-a-n 'dusty' b. dulk-ėt-as dust-a-m c. dulk-ėt-a dust-a-f (95) a. raud-ón-a red-a-n 'red' b. raud-ón-as red-a-m c. raud-ón-a red-a-f While the neuter and feminine appear to be identical on the surface, we suggest that this does not run counter to our claims. The adjectives in (94)-(95) contain "strong" stem suffix exponents (cf. Ambrazas et al. 1997: 159-162), e.g. -ėt and -on, which attract stress even in the context of a "strong" suffix like the feminine nominative singular -à. We thus propose that the Vocabulary Items for the feminine and neuter are always distinct. The feminine ending -a includes stress in its Vocabulary Item. However, the stress fails to shift with "strong" stems. As a consequence, these forms overlap with the neuter, whose suffix is not stressed for the (i)a class. 27 (96) Vocabulary Items strong stem non-strong stem -à ↔ [gend][fem] raud-ón-a 'red-a-f' ger-à 'good-f' -as ↔ [gend] raud-ón-as 'red-a-m' gẽr-as 'good-m' -a ↔ elsewhere raud-ón-a 'red-a-n' gẽr-a 'good-n'

Conclusion
The empirical contribution of this paper has been to show that there exists two classes of default diagnostics in the domain of gender representation, which are often conflated in the literature. We have provided evidence that in Lithuanian, one class, masculine, identifies the unmarked gender feature, and the other, neuter, points to the absence of gender features. 27 The failure of these stems to shift stress is not limited to the feminine nominative singular form. In (i-a), we see that for example in feminine locative forms, the inflection is stressed. However, when the stem of the adjective is strong as with raudóna 'red', the inflection is not stressed, and instead, the stem receives the stress, as in (i-b).
(i) a. žal-iosè green-loc.f.sg b. raud-ón-ose red-a-loc.f.sg To capture this split, we offered an explicit and restrictive theoretical DM account of gender representation in Lithuanian, in terms of features, their locus, and how they are transmitted to PAs through agreement. Our study offers a unified "layered" approach to two types of defaults in terms of feature specificity and the Elsewhere Condition, with one, the masculine, being more specific than the other, the neuter. This approach contrasts with some previous studies such as Corbett & Fraser 1999. In this work, there are also two types of defaults: the normal case default and the exceptional case default. The normal case default occurs in cases of underspecification, i.e. "after failing to find a more specific value" (Corbett & Fraser 1999: 71). The exceptional case default, in contrast, occurs in atypical or exceptional circumstances, e.g. where there is no nominal with which to agree. While their approach derives two types of defaults in fundamentally and conceptually different ways, our approach has reduced the difference between defaults purely to feature representation (see Kramer 2015: 141 for a related discussion).
As far as the locus of features is concerned, we provided a novel argument for linking gender features to the nominalizing head (following Lecarme 2002;Lowenstamm 2008;Kramer 2009;among others). Given that the language lacks neuter nouns, we proposed that it lacks an n without gender features and showed that this correctly derives not only the distribution of neuter PAs, but also the properties of deadjectival nominals and neuter pronouns. One alternative would have been to adopt the Distributed Gender Hypothesis (Steriopolo & Wiltschko 2010;Steriopolo 2018), which posits multiple distinct loci of gender features in the nominal domain: in Steriopolo (2018: 4), these loci are at n (semantic/natural gender), and D (discourse/referential gender). For this type of account, neuter features could be located on D but not on n. However, if D can be assigned neuter to all regular nominals, then it is not immediately clear to us how we would derive the ungrammaticality of neuter attributive adjectives. It is also unclear how this type of account would capture the hierarchy effects between masculine and neuter that we have argued for.
Lastly, in terms of agreement, our study further contributes to the growing body of evidence that the inability to agree with an argument does not yield ungrammaticality (cf. Preminger 2014). This is supported by the fact that PAs are neuter in cases of nonagreement (i.e. when the subject lacks gender features accessible for agreement), rather than causing the derivation to crash.
Further questions that remain concern how our representation of gender relates to three-gender languages. Crucially, we are not proposing that these languages lack "neuter" ns. We follow Kramer's (2015) proposal that three-gender languages represent neuter nouns with ns that lack gender features. In these languages, default gender, and non-agreement diagnostics may yield the same results. For example, Icelandic uses neuter inflection on adjectives in different types of default environments. As shown in (97a)-(97b), both onomatopoeia as well as a dative subjects trigger neuter inflection. These environments thus conflate the distinction presented by Lithuanian (cf. (28), and (32) for minimal pairs).
(97) Icelandic a. hátt mjá/voff/búmm loud.n meow/woof/boom 'a loud meow/woof/boom' Gender Default (Onomatopoeia) b. Henni er kalt. her.dat is cold.n 'She feels cold.' Non-agreement (Non-Structural Case) The split between default gender, and non-agreement in Lithuanian affords the opportunity to examine what makes certain syntactic environments "default". Further research should investigate similar gender splits among default environments, especially in languages like Lithuanian that exhibit more gender categories in PA agreement than there are gender categories of nouns. While our system treats defaults, and representation as "layered" for gender, it remains an open question whether other features like case exhibit the same type of behavior, and can be captured under the same type of approach.