Evaluative Suffixes in Russian : The Interaction of Declension Class , Gender , and Animacy *

This is a study of Russian nominalizing evaluative suffixes that form nouns of the -adeclension. Such suffixes are interesting because they can consistently change the animacy, declension class, and grammatical gender of the base to which they attach. However, the resulting nominalizations belong to different grammatical genders that seem to depend on the biological gender of a discourse referent. This work investigates morphosyntactic properties of such evaluative suffixes and proposes an account for the differences in grammatical gender in the framework of Distributed Morphology (Halle and Marantz 1993, Halle 1997, Marantz 1997, among many others), which provides us with formal tools for handling syntactic processes that happen inside a word — in this case, inside evaluative nominalizations. The study contributes to a number of much-debated questions in the current linguistic literature concerning the interaction between grammatical gender and declension class, mixed gender agreement, interpretability of gender features, and default gender.


Grammatical gender 1.2.1 Animate suffixes
The same suffix can form nouns of different grammatical genders: masculine, as in (11); feminine, as in (12); and common gender (MASC or FEM), as in ( 13) and ( 14).It is important to note that when animate suffixes attach to kinship terms, the gender of the base is always preserved, as in ( 11), ( 12). (

Summary
The properties of the nominalizing evaluative suffixes under investigation are summarized in Table 3.They all form nouns of the -a-declension (CLASS II).The majority of the suffixes (excluding -ob, -ot) form animate nouns that belong to different grammatical genders (MASC, FEM, or COMMON).Two vulgar suffixes (-ob, -ot) do not seem to affect animacy (they only attach to inanimate bases and form inanimate nouns) and consistently form nouns of feminine gender.

Analysis
The goals of this research are as follows: first, to investigate the morphosyntactic properties of the nominalizing evaluative suffixes; second, to propose an account for the different grammatical genders of the resulting nominalizations; and third, to account for mixed gender agreement.In §2.1.,I analyze the manner of syntactic attachment of the evaluative suffixes (how do they attach, as a syntactic head or syntactic modifier?).In §2.2., I analyze their place of syntactic attachment (where do they attach in the syntactic tree?).The research is done in the framework of Distributed Morphology (DM) (Halle and Marantz 1993, Halle 1997, Marantz 1997, among many others) which distinguishes between word formation from √roots and from syntactic categories.The central claim of DM is that there is no division between syntax and morphology.The relationships between morphemes are structurally identical to relationships between words.DM contrasts with descriptivist frameworks which view categorization in terms of inflection vs. derivation, but this has been proven problematic with respect to the behaviour of evaluative derivations (Brown and Hippisley 2012, Dressler and Barbaresi 1994, Manova 2004, Scalise 1984, 1988, Stump 1991, 2001, Vinogradov 1972, among others).It has been shown in the literature that the behaviour of evaluative derivations is not wholly inflectional or derivational.In contrast, DM regards inflection and derivation not as primitives, but as derived notions, and thus, this framework can better account for the behaviour of nominalizing evaluative suffixes in Russian.

The manner of syntactic attachment
I propose that the evaluative suffixes under investigation are nominalizing heads, as in ( 17).
The evidence comes from the fact that they can attach to various syntactic categories (adjectives, verbs, nouns) and always form nouns, as shown in ( 18)-( 20

Proposal 2: The feature [CLASS]
I propose that the evaluative suffixes are specified for the feature declension [CLASS II], but they have no gender feature, as shown in (30).4 The reason for this proposal is that the evaluative suffixes consistently form nouns of thea-declension (or CLASS II), as in ( 31)-( 33).
(  34), ( 35) above?Below I argue that the evaluative suffixes under investigation change the declension class of the base and grammatical gender is determined from that declension class in Russian.

Proposal 3: Grammatical gender is determined from declension class in Russian
The gender of the head noun determines agreement patterns, while declension is not relevant for the purpose of agreement.The definitions of declension and gender are given in (37).
(37) DEFINITIONS OF DECLENSION AND GENDER (from Aronoff 1994) a. Declension is a set of lexemes whose members each select the same set of inflectional morphemes.b.Gender is a nominal agreement class.
(38) a. Syntax b.PF: Theme Node Insertion nP nP (from Kramer 2015, 237) If this proposal is correct, we predict that gender features can (but must not) affect declension (Kramer 2015, 233, 237).However, a problem arises, because in Russian, the opposite seems to be true-declension affects gender, but not the reverse (see Aronoff 1994, Corbett 1982, 1991, Fraser and Corbett 1995).This is also supported by the experimental studies of the acquisition of gender in Russian by Tarasenkova (2010).
In Table 4 Corbett 1991, 40).I propose that such nouns are only specified for the feature [GENDER] and not for class, as in (40b) above.
The current proposal is that the evaluative suffixes under investigation are specified only for the feature [CLASS] but they have no gender feature, as in the structure (40c) above.The following question arises: Where does the grammatical gender of evaluative derivations come from?I propose that there are two sources for grammatical gender: (i) gender as a default (determined from [CLASS]), and (ii) gender determined from biological gender (sex) of the referent.I will return to this question later (see §2.6).

Syntactic approaches to account for evaluative derivations
Here I discuss which syntactic approach best accounts for evaluative derivations in Russian: a hierarchical structure approach (Chomsky 2000(Chomsky , 2001) ) or a cyclicity approach (Marantz 2001, Embick 2010, Marvin 2013).To answer this question, I investigate Russian data where an evaluative suffix attaches to a gendered nominal, as in (41).
(41) D The two approaches make completely different predictions.Consider first the hierarchical structure approach.The probe searches downward into its c-command domain for a goal and enters into an Agree relation with the first goal it encounters.In the structure (41) above, D [GEN_] would agree in gender with the lower n [GENDER] because the higher n [CLASS II][ EVAL] has no gender feature, hence it is not a suitable goal.
Consider now the cyclicity approach.n is a phase head that triggers spell-out of its complement.The spelled-out material is not accessible to later operations (Phase Impenetrability Condition, as in Chomsky 1999Chomsky , 2000)).Thus, in the structure (41) above, the higher n [EVAL] triggers spell-out of the lower n [GENDER], which means that D [GEN_] will have no access to the lower n [GENDER].The Russian data in ( 42) and ( 43) show that the cyclicity approach can best account for the data.
In (42c), the hierarchical structure approach predicts that D [GEN_] would agree with the lower n [FEM] because the higher n [EVAL] is not a suitable goal (it has no gender feature).Thus, the resulting evaluative derivation should be feminine.The cyclicity approach makes a different prediction: gender of D [GEN_] has no access to the lower n [FEM].Thus, the gender of the resulting evaluative derivation should be unknown (it could be masculine or feminine, since it denotes a human).The data in (42b) show that this is exactly what we find.
(42) a. kras-ot-a b. kras-ot-ul-ja pretty-NOM-NOM.SG (FEM; CL II) pretty/red-NOM-EVAL-NOM.SG (MASC/FEM; CLASS II) 'beauty' 'pretty person (affect)' In (43c), the hierarchical structure approach predicts that the evaluative derivation should be masculine, while the cyclicity approach again predicts unknown gender, as in (43b).Based on this evidence, I conclude that the Russian data support the cyclic approach; thus, I will assume this approach in this work (see Kramer 2014, 222-25, who reaches the same conclusions for Amharic and Somali).
It was noted above ( §1.2.1) that kinship nouns behave differently -they do not change the gender of the base when an evaluative suffix attaches to them, as in ( 44), ( 45).The question arises: What accounts for the differences between kinship and non-kinship nouns?Bobaljik and Zocca (2011) investigate the behaviour of nominal predicates under ellipsis and show that there are semantic classes of nominals that differ with respect to whether or not the underived masculine forms carry a presupposition of maleness.The data in (46i) show that Russian kinship terms like brat 'brother' carry a presupposition of maleness, while the non-kinship terms like vor 'thief' do not.The data in (46ii) show that when an evaluative suffix is attached, it produces no change in the presupposition of maleness.The question arises as to whether kinship nouns may have a special syntactic feature compared to non-kinship nouns.This question is discussed in the following section.( 46

An interpretable gender features approach
In the Distributed Morphology framework, roots are deprived of any features, including gender (Borer 2005, Acquaviva 2009, Embick and Noyer 2007, Embick 2012, Kramer 2015).Kramer (2015) proposes that gender features are located on n and come in two different types: interpretable, for natural gender, and uninterpretable, for arbitrary gender, as in (47).The "plain" n has no gender feature and the result is gender by morphological default.According to Kramer (2015), interpretable features are legible at LF and can change the interpretation of a linguistic structure (e.g., they can insert a denotation, see Zamparelli 2008, 170).Uninterpretable features are illegible at LF; they do not affect interpretation.Thus, there are no inherent male/female meanings on roots like √mother, √father.Licensing a root in a particular nominal context is what makes it interpreted as male or female (Kramer 2015, 52).For example, in ( 48), the feature i[+FEM] triggers female interpretation and not the meaning of the root.

n i[+FEM] [mother]] = 'female parent'
A problem might arise in this approach which concerns non-nominal derivations with a kinship meaning.An analysis in ( 48) predicts that such derivations (i) either cannot have a male/female interpretation at all (Russian data in (49) contradict this), or (ii) they must be cross-linguistically derived from a nominal that has an interpretable gender feature, as shown in (49d).Further research can show whether (ii) is indeed the case.
(49) a. RUSSIAN VERBS: u-syn-ovit' 'to adopt a son' u-doč'-erit' 'to adopt a daughter' žen-it'sja 'to get married to a woman' vyiti za-muž 'to get married to a man' b.RUSSIAN ADJECTIVES: mat-erinskaja l'ubov' 'mother's love' c.RUSSIAN ADVERBS: razdelit' po-brat-ski 'to divide equally (lit.: like among brothers Kramer's (2015) approach is correct, we expect that in languages with no grammatical gender, (i) either there is no male/female interpretation at all, or (ii) if there is such an interpretation, the interpretable gender features must be present in the syntax.A question arises: What need is there to assume syntactic gender features in languages with no syntactic gender agreement?
If we applied Kramer (2015) to the Russian data in question, the kinship noun brat 'brother' would have the interpretable gender feature i[-fem] (50a), while vor 'thief' would not have this feature (50b), hence these nouns would differ in the syntactic feature [GENDER] and (51c) has no gender feature (morphological default). 5However, the following three problems might arise with the approach in (51).
Second, the feature [CLASS] is not in Kramer's system.Thus, gender as a default, as in (53c), would be unclear in Russian, as default gender can be feminine (CLASS II nouns) or masculine (CLASS I nouns).For example, in Russian CLASS II nouns, when the gender of the referent is unknown (or unimportant), feminine gender agreement is most likely to be used, as in ( 54).This presents an additional piece of evidence for the dependency of Russian grammatical gender on declension class (as discussed in §2.3.2). ( -Tam grjaz-n-ul-ja sid-it.there dirty-ADJ-EVAL-NOM.SG (CLASS II) sit-PRES 5 Thank you very much to Ruth Kramer for a personal discussion of this phenomenon in Russian.
'A dirty person (AFFECT) is sitting there.' -Kak-aja grjaz-n-ul-ja?what-FEM dirty-ADJ-EVAL-NOM.SG 'What (FEM) dirty person (AFFECT)?' In Russian CLASS I nouns, when the gender of the referent is unknown (or unimportant), masculine gender agreement is most likely to be used, as in ( 55).The third potential problem with this approach is undergeneration.Although Kramer's (2015) system accounts for feminine (56a) and masculine (56b) gender agreements with no problem, it cannot account for mixed gender agreement, as in (56c).
(56) a. Èt-a grjaz-n-ul-ja vsë tut this-FEM dirty.person-ADJ-EVAL-NOM.SG everything here za-pačk-al-a.To summarize, if we apply Kramer's (2015) system to Russian evaluative derivations, the system seems to either overgenerate, as in (53), with three homophonous suffixes, or undergenerate, as in (56c), with mixed gender agreement.Therefore, here I do not assume this system; instead, I assume that 'male'/'female' is not a syntactic feature, but a part of the root meaning (Steriopolo and Wiltschko 2010, Kučerová, forthcoming).In other words, I assume no syntactic differences between the nouns 'brother ' and 'thief,' as diagramed in (57).The difference between (57a) and ( 57b) is in the meanings of the roots-the root √brat 'brother' has 'male' as part of its meaning, while the root √vor 'thief' does not (see discussion in §2.4 above).( 57 In the cyclicity approach assumed here, when an EVAL nominal head attaches to a kinship nominal like 'brother' in (58c), it triggers spell-out of its complement (the lower n), and in either approach (Kramer's or the current approach), the meaning 'male' is already accessible at this point.
(  Sauerland (2004) and Matushansky (2013) propose that valuation of context-dependent gender features is driven by the semantic component as presupposition associated with an assignment index (like a pointer to the actual referent).A semantic denotation of masculine/feminine genders is given in (62).A feminine feature associated with the index i will denote a female if the referent is female, as in (62a).A masculine feature with the index i will denote a person if the referent is a person, as in (62b).Thus, masculine gender is compatible with both natural genders.
(62) SEMANTIC DENOTATION OF MASC/FEM (from Kučerová 2017, 27, modelled after Heim 2008) a. [[GEN:fi]]ʷ , ᵍ = λxe.g(i) is a female in w: x b. [[GEN:mi]]ʷ , ᵍ = λxe.g(i) is a person in w: x Although (62) accounts for the Italian data (Kučerová, forthcoming), it does not seem to account for the Russian data, because in the Russian EVAL derivations, feminine gender is compatible with both natural genders, as shown in ( 63), but masculine gender is not, as shown in ( 64).I propose to modify the semantic denotations in (62) to account for Russian nouns of common gender.A feminine feature associated with the index i will denote a person if the referent is a person.A masculine feature with the index i will denote a male if the referent is male.Thus, feminine gender is compatible with both natural genders in Russian common gender nouns.

How can we account for mixed gender agreement?
I propose that a failed AGREE relation -in the sense of Preminger (2009Preminger ( , 2014)), as in (70)can account for mixed gender agreement.
(70) Failed agreement: A descriptive characterization (Preminger 2014: 12) An utterance that is grammatical despite failing to adhere to what is an otherwise obligatory pattern of agreement in the language in question, and for which there is no grammatical variant where agreement surfaces normally, is an instance of failed agreement.
The AGREE operation is obligatory in syntax, modelled in terms of an obligatory operation (71).
(71) FIND(f) (Preminger 2014, 96) Given an unvalued feature f on a head H, look for an XP bearing a valued instance of f and assign that value to H.
According to Preminger (2014, 240-41), agreement can fail if a target with the requisite featural content is absent from the derivation.Crucially, these failures of agreement do not result in ungrammaticality (contra Chomsky 2000Chomsky , 2001)), only in the lack of valuation of the relevant features on the probe.For example, in (72), when the probe D with an unvalued gender feature [GEN_] encounters a goal n that lacks a gender feature, the operation FIND(f) will fail and there is no valuation of the gender feature on D.
(72) FAILED GENDER AGREEMENT D 3 D n [GEN_] 3 n n X [CLASS] [GEN_] In a configuration like (72), where φ-agreement is impossible because of the outright absence of an appropriate target, grammaticality is still possible without φ-agreement (see Preminger 2014: 220).Thus, in the Russian data, we observe two possible outcomes of a failed AGREE relation: (i) Gender is realized as a default (determined from declension [CLASS]); or (ii) Gender is contextually determined via the feature [PERSON] on D, as proposed in Kučerova (forthcoming).If the derivation proceeds cyclically (the approach for which I have argued in §2.4), FIND(f) is triggered upon the merger of an φ-bearing head in each cycle of the derivation (e.g., the ncycle and the D-cycle in the structure (73) above).In the cyclicity approach, both outcomes (i) and (ii) above can be possible, hence mixed gender agreement.For example, when the example in (74a) refers to a male individual, mixed gender agreement can arise, as in (74b).

Conclusions
I have analyzed Russian evaluative suffixes, as in Table 5, in the framework of Distributed Morphology.

Table 3 :
Russian evaluative suffixes, animacy, and gender The evaluative suffixes are not specified for [GENDER] because they form nouns of different genders, as in (34b)-(36b).It is important to note that they can nevertheless change the gender of the base to which they attach.In (34), a masculine noun becomes a common gender noun when the evaluative suffix -in is attached.In (35), a masculine noun becomes feminine when the evaluative suffix -ob is attached.
MASC = MASC a. brat b. brat-ux-a brother.NOM.SG (MASC; CLASS I) brother-EVAL-NOM.SG (MASC; CLASS II) 'brother' 'brother (eval)' It has been proposed before (Arsenijevich 2016, Matushansky 2013) that common gender nouns in Russian are unmarked for gender features.The following question arises: How can we account for a change in the gender of the base, as in (

Table 4 :
Corbett 1982)asses in Russian (modified fromCorbett 1982)If only the gender (not the declension class) of a noun is known, it is impossible to determine its declension class.For example, a feminine noun can belong to CLASS I, CLASS II, or CLASS III; a masculine noun can belong to CLASS I, 39) GRAMMATICAL GENDER CAN BE DETERMINED FROM DECLENSION a. declension CLASS I ⇒ masculine agreement (excluding biological females) b. declension CLASS II ⇒ feminine agreement (excluding biological males) c. declension CLASS III ⇒ feminine agreement d. declension CLASS IV ⇒ neuter agreementIf gender is a syntactic feature and declension class is post-syntactic, we cannot account for the fact that gender can be determined from declension class and not the reverse.Following Kučerová (forthcoming, for Italian), I propose that declension class features and gender features are both present in the syntax, which allows for their interaction.If this is on the right track, we predict the following three combinations in Russian, (40).kenguru 'kangoroo.MASC,' attashe 'attaché.MASC,' and ledi 'lady.FEM.'Such nouns can take different gender agreements, but they do not decline (see CLASS II, or CLASS IV (with just one noun, put' 'way,' in CLASS III); a neuter noun can belong to CLASS IV (with ten neuter nouns in CLASS III).In contrast, grammatical gender can be determined from a combination of declension class and biological gender (seeSteriopolo, forthcoming), as in (39).(CLASSIIIlike vremya 'time' must be specified for the both features [NEUTER] and [CLASS III], as in the structure (40a) above.If only the gender feature [NEUTER] were specified, the declension class would remain unclear (it could be CLASS III or CLASS IV).If only the feature [CLASS III] were specified, grammatical gender would be unclear (it could be feminine or neuter).Russian has a large group of indeclinable nouns, such as pal'to 'coat.NEUT,' ) APPLYING BOBALJIK AND ZOCCA (2011) TO RUSSIAN: KINSHIP NOMINALS .
All feature bundles in (51) contain the category feature n and the semantic feature[EVAL].However, they are different in terms of gender features: (51a) has i[+FEM]; (51b) has i[-FEM], Kučerová (forthcoming)proposes that the φ-feature valuation can be determined from the context and that contextually determined gender (at least in Italian) is assigned on D, as in (61). 6(or [+/-PARTICIPANT] in Nevins' 2007 terms), which is licensed by the syntaxsemantics interface and is associated with an index as part of labelling of the DP.
(65) SEMANTIC DENOTATION OF MASC/FEM (modified for Russian nouns of common gender) a.[[GEN:fi]]ʷ , ᵍ = λxe.g(i) is a person in w: x b. [[GEN:mi]]ʷ , ᵍ = λxe.g(i) is a male in w: xBelow I propose analyses of Russian animate and inanimate derivations.According toKučerová (forthcoming, 31), the contextually driven feature valuation is a last resort operation and it can only apply if the feature is not valued in narrow syntax.Thus, a gender feature on D may be valued via[PERSON]only if the gender feature on D is unvalued at the point of Transfer.But how can we account for mixed gender agreement, as in (69)?This question is discussed below.

Table 5 :
Russian evaluative suffixes (fromSteriopolo 2008, 62)I have argued that the suffixes in question are nominalizing heads, n, that attach above categorized roots (v/a/n), as in (75):The majority of the suffixes (except -ob, -ot) are specified for the feature[ANIMATE].All suffixes are specified for the feature declension[CLASS]but have no gender feature, as in (76) (unlike other nominals that can have both [CLASS] and [GENDER] features).(76)BUNDLES OF FEATURES FOR EVALUATIVE SUFFIXES (i) animate suffixes: n, [EVAL], [ANIMATE], [CLASS] (ii) inanimate suffixes: n, [EVAL], [CLASS] Grammatical gender can either be realized as a default, determined from [CLASS], or it can be contextually determined via D [PERSON], as in (77).This research has contributed to a number of "hot" debates in the current literature concerning the interaction between gender, animacy, and declension class; interpretability of gender features; default gender; and mixed gender agreement.