Gender Assignment in Ukrainian: Language Specific Rules and Universal Principles*

In this paper I present an analysis of gender assignment in Ukrainian, a gender system that has received less attention in scholarly literature than the closely related Russian system. I advance eleven rules of three types, viz. semantic, morphological and morpho-semantic rules. It is shown that these rules give correct predictions, provided that the semantic rules override the morpho-semantic rules, which in turn take precedence over the morphological rules. Furthermore, it is argued that this ranking follows as an automatic consequence of two independently motivated universal principles, viz. Kiparsky’s (1982) Elsewhere Condition and what I shall refer to as the “Core Semantic Override Principle”. Sections 1 through 3 investigate semantic, morphological and morpho-semantic rules, respectively. After a brief discussion of rule interaction in sections 4 and 5, the contribution of the paper is summarized in section 6.


Semantic Rules: Gender and Biological Sex
For the purposes of the present study I adopt Hockett's (1958:231) wellknown definition: "Genders are classes of nouns reflected in the behavior of associated words".Ukrainian has three genders (agreement classes), masculine, feminine and neuter, as witnessed by the sentences in (1), where the nouns xlib 'bread', perepi™ka '(kind of) Ukrainian pastry' and salo 'bacon' take different agreement targets.In the following, we shall be concerned with the rules for assignment of nouns to these three genders.Notice that the animate and inanimate subgenders (cf.Corbett 1991:163ff.) will not be discussed in this paper.
Ukrainian nouns denoting male persons or animals belong to the masculine gender, while nouns denoting female persons or animals are feminine.Thus syn 'son', djadja 'uncle' and ¢erebec' 'stallion' are masculine, while do™ka 'daughter', titka 'aunt' and kobyla 'mare' are feminine.For the sake of explicitness, the rules are given in (2).
(2) Semantic rules: a. Male AE M b.Female AE F This rule format will be applied throughout.The rules are labeled according to the nature of the information to the left of the arrows.Since "male" and "female" are aspects of the meaning of the relevant lexemes, the rule block in (2) is referred to as "semantic".The information to the right of the arrows represents the genders, which I abbreviate as M (masculine), F (feminine) and N (neuter).The arrow stands for an implicational relationship between two pieces of information, in this case biological sex and gender (agreement class).
Relevant for rules (2a-b) is a class of nouns that are traditionally said to have "common gender".Nouns of this type may refer to persons of both sexes, e.g.syrota 'orphan' (for further examples, see Hry£™enko 1997:338).When they refer to males they take masculine agreement, while they consistently combine with feminine agreement targets when they refer to females.Thus a boy may be described as molodoj syrota 'young (masc.)orphan', whereas molodaja syrota 'young (fem.)orphan' is appropriate about a girl.Since the agreement of the so-called common gender nouns depends on the sex of the referent, the rules in (2) correctly assign gender to these nouns.

2.
Morphological Rules: Gender and Inflection Class (Declension) In Ukrainian, there is a close relationship between a noun's gender and its inflection class (declension).I propose six declensions instead of the more traditional analysis with four declensions (cf.e.g.Bilodid 1969 andHry£™enko 1997).Since my analysis is somewhat untraditional, it is useful to start with an explicit definition of inflection class: (3) Inflection class: [A] class of lexemes which share: • a paradigm consisting of a set of "cells", i.e. inflectionally realised morphosyntactic properties or combinations of properties [...], • all the inflectional markers, or exponents, which realise these cells [...] (Carstairs-McCarthy 2000:630).
The paradigms of Ukrainian nouns consist of fourteen cells defined by seven cases in the singular and plural.On the basis of the definition in (3), I shall say that two nouns belong to the same declension if they have the same inflectional endings in the relevant cells.If not, they belong to different declensions.However, as pointed out by Carstairs-McCarthy (2000:632), there is one systematic exception.Even if two nouns combine with different endings, it is customary to relegate them to the same declension if the choice between the endings is predictable on independent grounds.This practice will be adopted here.Thus, zemlja 'earth, land' and voda 'water' belong to the same declension although the former takes the ending -eju in the instrumental singular, while the latter has -oju in this cell of the paradigm.The reason is that the choice of ending is predictable on the basis of the quality of the stem-final consonant.After a palatalized ("soft") consonant -eju is selected, while -oju occurs after non-palatalized ("hard") consonants.
With the definition in (3) in mind, consider now the list of endings given in table 1.The table contains the endings in the singular only, since this is sufficient to establish the number of declensions that are relevant for gender assignment.In order to avoid unnecessary complications, endings combining with stems in palatalized consonants have been omitted in cells with alternations of the -eju/-oju type treated above.All endings are given in phonemic transcription. 2 The Ø symbol represents the absence of an ending (a zero ending).The choice between the alternative endings in column 1 is partly predictable (see e.g.Pugh andPress 1999:70f. andShevelov 1993:958 for overviews).The endings given in column 5 are those of nouns with (oblique) stems in /t/, e.g.telja 'calf' (cf.genitive singular teljat-y).Nouns with (oblique) stems in /n/ (e.g.im'ja 'name', cf.genitive singular imen-i) are inflected somewhat differently.However, since the choice of endings is predictable from the quality of the stem-final consonant, the differences do not form the basis for establishing two declensions.I have only included the endings of the /t/-stems in the table, since these nouns constitute the larger set and even evince some productivity (Shevelov 1993:959).

Table 1: Ukrainian noun inflection: endings in the singular
How many declensions can be established on the basis of table 1?Many cases involve considerable syncretism.For instance, in the locative, the ending -i is found in five of the six columns.Nevertheless, each column represents a declension.Consider first the instrumental, where we have five different endings, viz.-om, -oju, -ju, -am and -Ø.This enables us to establish five declensions.What remains to be shown is whether columns 1 and 4 represent different declensions even though they display the same ending in the instrumental case (and some other cases).At least two arguments suggest that we are dealing with two different declensions.First, columns 1 and 4 involve different endings in the vocative.Secondly, while column 1 involves two or three endings in several cases, no such variation is found in column 4. Since the two columns contain different sets of endings, the definition in (3) above enables us to establish a total of six declensions for Ukrainian.
Against this line of reasoning one might object that some of the differences are predictable on independent grounds if gender is taken into consideration.For instance, nouns with the endings in column 4 are neuter, while nouns with endings from column 1 belong to the masculine gender.Hence, it might be argued, the two columns represent one declension since the differences are predictable from gender.However, Corbett (1982, see also Corbett 1991and Corbett and Fraser 2000) has shown that an approach where declension is predicted on the basis of gender is problematic for Russian, and his argument seems to hold for Ukrainian as well.Consider the relationship between declension and gender, which is spelt out in figure 1. 3 As can be seen from the figure, gender is predictable from declension since a unique path takes us from each declension to one and only one gender.However, if we take gender to be the basis for predicting declension, we get into trouble.For instance, feminine gender is compatible with classes 2 and 3, so if the lexical representation of a noun includes reference to feminine gender, there is no way to infer from this how the noun is inflected.In order to save the gender-to-declension approach, one would have to rely on extensive lexical marking of declension.For instance, one might have to mark all the nouns of class 3 as lexical exceptions.The declension-to-gender approach, on the other hand, does not lead to such problems.In the following discussion, therefore, I shall adopt the latter approach.I shall not conflate any classes even if the differences between them are predictable from gender.Accordingly, I shall assume each of the six columns in table 1 to be separate declensions.

Figure 1: The declension-gender interface
Before we leave table 1, a remark on declension 6 is in order.This declension contains borrowings that are traditionally labeled "indeclinable" and treated as standing outside the declension system.However, as noted by Corbett and Fraser (2000:308), the fact that nouns of this type do not combine with other endings than the zero ending is itself a fact about their inflectional behavior.Since the nouns in question do not fit into patterns 1-5 in table 1, assuming a separate declension for nouns of this type seems justified.
On the basis of figure 1

Morpho-Semantic Rules: Indeclinables and Affective Derivatives
We shall return to the interaction between the morphological rules in (4) and the semantic rules in (2) in sections 4 and 5. Notice, however, that the rules postulated so far are not sufficient to assign the correct gender to all Declension: Ukrainian nouns.Importantly, rule (4f) does not give correct implications for animate nouns in declension 6.Such nouns tend to be masculine unless they refer to females, in which case they are feminine (cf.Pugh and Press 1999:56ff.).Examples include masculines like kakadu 'cockatoo', parvenju 'parvenue', flaminho 'flamingo' and prote¢e 'protégé', and feminines like ledi 'lady'.While the possibility of feminine agreement can be accounted for in terms of rule (2b), we need a rule assigning masculine gender to animate nouns in declension 6.This rule is stated explicitly in (5a): (5) Morpho-semantic rules: a. Declension 6, animate AE M b.Declension 6, language AE F Rule (5b) is intended to capture the fact that indeclinable nouns denoting languages are compatible with feminine agreement, cf.urdu 'Urdu', hindi 'Hindi' and esperanto 'Esperanto'.As noted by Pugh and Press (1999:57), this is presumably related to the fact that the hyperonym mova 'language' is feminine.I label the rules in (5) "morpho-semantic" since they involve a semantic feature in addition to the morphological information about declension.The list of obligatory or optional morpho-semantic rules may certainly be extended.However, a more complete list of such rules will not be attempted here.The main purpose of the present paper is to explore rule interaction, and the rules in (5) are sufficient to illustrate the properties of morpho-semantic rules with regard to this issue.
A group of nouns that need special attention with regard to gender assignment are affective derivatives (diminutives and augmentatives).Let us consider augmentatives in -y£™e, e.g.domy£™e 'house (augm.)',komary£™e 'mosquito (augm.)',borody£™e 'beard (augm.)'.On the face of it, one would expect nouns of this type to be neuter, since they combine with the endings in column 4 in table 1, and thus belong to declension 4.However, in Russian the corresponding augmentatives are reported to behave differently (Corbett 1982, Hippisley 1996, see also Stankiewicz 1968).The Russian suffix -i£™e is transparent in the sense that derived nouns in -i£™e display the same gender as the base noun.Thus, domi£™e is masculine because it is derived from the masculine noun dom 'house', whereas okni£™e is derived from the neuter okno 'window', and hence itself neuter.In other words, the augmentative suffix in Russian does not bear on the gender of the derived noun.In order to test whether this holds for Ukrainian as well, I excerpted all augmentatives in -y£™e from Bevzenko (1985), a one volume backwards dictionary of approximately 134 000 words. 4 The results are given in table 2. The gender predicted by the base noun is represented as shaded cells, and the gender predicted by the augmentative suffix as double frames.The columns marked as M/F, M/N and F/N represent nouns that are reported to vacillate between two genders.

Table 2: Gender assignment in Ukrainian augmentatives in -y£™e
With a total of 95 words the material is too restricted to support strong conclusions.Furthermore, the information in dictionaries may be normative in nature and hence not reflect actual usage.The brief remarks in Hry£™enko (1997:338) suggest that vacillation is widespread.Still the material suggests that Ukrainian differs from Russian in that both the base noun and the augmentative suffix bear on the gender of the derived noun.In some cases, the gender of the derived noun equals the gender of the base noun.This is frequently the case for masculine bases.For inanimate masculine bases, more than half of the relevant nouns are masculine (25 nouns) or vacillate between masculine and neuter (9 nouns).When the base is an animate masculine noun, the gender of the derived noun is always the same 4 I removed all nouns in -y£™e that were not defined as augmentatives in the 11 volume dictionary Slovnyk ukrainskoj movy (Bilodid 1970-80), e.g.horody£™e 'site of ancient settlement'.In order to isolate a potentially interfering factor, I also removed nouns denoting either male or female human beings or animals, e.g.mu¢i™y£™e 'peasant (augm.)'and baby£™e '(old) woman (augm.)',since biological sex may be responsible for the gender of such nouns (cf.rules 2a-b above).
as the gender of the base, with the single exception of borsu™y£™e 'badger (augm.)'which is reported to vacillate between masculine and neuter.However, in many cases the gender of the derived noun appears to be determined by the augmentative suffix.This seems to be the case for most nouns derived from feminine nouns, at least if the base noun is inanimate.As can be seen from the table, all 19 nouns derived from inanimate feminine nouns are reported to have neuter gender.
A detailed description of gender assignment for all affective derivatives in Ukrainian will not be attempted here, as the nouns in -y£™e suffice to illustrate the two principles involved, viz.assignment from the base noun and from the derivational suffix.The important question at this point is how to accommodate these principles in the model proposed in the present study.Assignment by the suffix is unproblematic; nouns of this type obey rule (4d) above, which assigns neuter gender to nouns in declension 4. Assignment by the base noun, on the other hand, is less straightforward.An account in terms of lexical specification of individual nouns is not viable since it would miss the generalization that assignment from the base noun is practically restricted to masculine bases and (almost) obligatory for animate bases belonging to the masculine gender.Consider the morpho-semantic rule in (6) where angled brackets represent morphological structure: This rule assigns masculine gender to declension 4 augmentatives in -y£™e provided that the base has the specifications "masculine" and "animate".It is possible that the rule could be extended to other patterns of affective derivation in other declensions.Furthermore, the animacy requirement in the rule might arguably be removed so that the rule would apply to all masculine bases.As it stands, the rule does not account for the 25 masculine inanimate nouns, so their gender must be specified lexically.However, if the animacy requirement were removed, we would have to specify the gender of the 20 neuter nouns formed from masculine inanimate bases.As there are no clear indications as to which solution is preferable, I have chosen to formulate the rule as restrictively as possible.This decision, however, has no bearing on the issue of rule interaction, to which we turn in section 4.

Rule Interaction: A Hierarchy
Conflicts between rules arise whenever two or more rules predict different genders for a class of nouns.The conflicts are of three types: (7) a. Semantic vs. morphological rules b.Semantic vs. morpho-semantic rules c.Morpho-semantic vs. morphological rules Conflicts of type (7a) arise for common nouns like djadja 'uncle' and male first names like Mykola and Mykyta.Since these nouns denote male human beings, they are relevant for rule (2a), which assigns masculine gender.However, at the same time these nouns are inflected according to the second declension.On the basis of the morphological rule (4b), which assigns feminine gender to nouns of this type, we would therefore expect djadja, Mykola and Mykyta to be feminine.The conflict is resolved in favor of the masculine; nouns denoting males are always masculine regardless of their declension.In this case, therefore, semantic assignment takes precedence over assignment by morphological rules.
Conflicts of type (7b) between semantic and morpho-semantic rules involve nouns denoting female individuals, e.g. common nouns like madam 'madame' and ledi 'lady', as well as female first names like Esfir.These nouns are indeclinable, i.e. they belong to declension 6. Rule (5a) assigns masculine gender to declension 6 nouns denoting animates.However, semantic assignment takes precedence.Nouns denoting females are consistently relegated to the feminine gender.
All the morpho-semantic rules represent special cases that override the more general morphological rules.Thus, rule (5a), which assigns masculine gender to animate nouns in declension 6, e.g.prote¢e 'protégé', takes precedence over the general rule (4f) which relates nouns in declension 6 to the neuter gender.In the same way the morpho-semantic rule (5b) for languages in declension 6 overrides the general rule for this declension, insofar as e.g.hindi 'Hindi' is feminine.The final morpho-semantic rule mentioned above, rule (6), also takes precedence over morphological assignment.Nouns like pavu™y£™e 'spider (augm.)'are masculine due to (6), not neuter as the competing rule (4d) would predict.In all the cases I am aware of, conflicts of type (7c) between morpho-semantic and morphological rules are resolved to the benefit of the former.
The interaction of the rules is summarized in the hierarchy in ( 8), where the sign >> represents "take precedence over": (8) Semantic rules >> Morpho-semantic rules >> Morphological rules

Rule Interaction: Two Principles
The question now arises as to whether the ranking in ( 8) represents an idiosyncratic fact about Ukrainian, or whether it can be derived from universal principles of rule ordering.Opting for the second and theoretically more interesting hypothesis, I shall explore two relevant principles -the Elsewhere Condition of Kiparsky (1982) and what I shall refer to as the "Core Semantic Override Principle".The Elsewhere Condition regulates the order of application of rules of different degrees of specificity.If rule A refers to a proper subset of the nouns referred to by rule B, A takes precedence over B. The Elsewhere Condition enables us to account for rule conflicts of type (7c) where morpho-semantic rules override morphological rules.As the reader will recall, the morpho-semantic rules refer to semantic properties in addition to a declension, while the competing morphological rules only invokes the relevant declension.Thus the nouns to which the morpho-semantic rules apply, constitute subsets of the nouns referred to by the competing morphological rules.For instance, declension 6 nouns denoting animates are a proper subset of declension 6 nouns.Accordingly, the fact that the morphosemantic rules override the morphological rules falls out as an automatic consequence of the Elsewhere Condition.It is not necessary to stipulate the ordering of morpho-semantic rules with regard to morphological rules.
Rule conflicts of the types (7a-b), whereby the semantic rules for biological males and females take precedence over morphological and morpho-semantic rules, represent a somewhat more complex case.No subset relationship holds between biological males (cf.rule 2a) and nouns in the second declension (cf.rule 4b), so the Elsewhere Condition cannot explain why e.g.djadja 'uncle' is masculine as predicted by (2a), and not feminine as predicted by (4b).Nor does the Elsewhere Condition provide an account of the fact that e.g.ledi 'lady' is feminine as predicted by the semantic rule (2b), rather than masculine as suggested by the morphosemantic rule (5a); nouns denoting biological females are not a subset of declension 6 nouns denoting animates.
Since the Elsewhere Condition is not sufficient to account for all the rule conflicts attested in Ukrainian gender assignment, I propose a second universal principle of rule ordering: (9) The Core Semantic Override Principle: Rules referring to biological sex take precedence in gender assignment.
I refer to (9) as the "Core Semantic Override Principle" because biological sex may be considered the semantic core of the category of gender.While the Elsewhere Condition may be called a formal principle of rule ordering since it concerns the logical relationship between rules, the Core Semantic Override Principles is a substantial principle insofar as it favors rules with a particular content.The principle ensures that the semantic rules (2a-b) take precedence since they refer to biological males and females.
There is ample empirical evidence for the claim that the Core Semantic Override Principle is part of universal grammar.According to Dahl (2000:101f.), who has investigated a large language sample including all languages discussed in Corbett (1991), sex is the "major criterion" for the assignment of gender in languages with more than one gender for animates.While Dahl's term "major criterion" may seem opaque, it is clear from his discussion that it implies that sex-based gender assignment tends to take precedence.Notice that the provision "tends to" does not indicate that we are dealing with a mere statistical generalization.Rather, the set of cases where sex-based rules are overridden is limited and well defined.Dahl (2000:103) isolates the following: (10) a. Special morphological rules may take precedence for augmentative and diminutive derivations.b.Special semantic rules may take precedence for nouns denoting young or small animates.c.Special semantic rules may take precedence for certain kinds of animals.d.The "wrong" gender may be used in order to obtain special rhetorical effects ("downgrading" and "upgrading").
German diminutives in -chen and -lein are well known examples of (10a).
As an illustration of special treatment for nouns denoting young or small animates in (10b), Dahl (2000:103) mentions the assignment of neuter gender to unmarried women in certain Polish dialects (see also Corbett 1991:100).As for (10c), in the Australian language Ngangikurrunggurr nouns denoting animals hunted for meat are relegated to a special gender (Dahl 2000:105).Finally, the special effects obtained by the use of s/he about inanimate objects and it about humans in American English serve to illustrate downgrading and upgrading in (10d) (Dahl 2000:105).A detailed discussion of cases of these types is beyond the scope of the present study.Suffice it to say that Dahl's typological evidence strongly suggests that sexbased rules take precedence universally in gender assignment, with the exception of the four well-defined cases in (10).The Core Semantic Override Principle resembles the following observation by Corbett and Fraser: (11) "As is universally the case, the formal gender assignment rules [...] are dominated by the semantic gender assignment rules."(Corbett and Fraser 2000:321) However, (11) represents a stronger claim than the Core Semantic Override Principle, since the former pertains to all semantic rules, while the latter only concerns those semantic rules that invoke biological sex.In languages like Ukrainian, this difference is not critical, since both the semantic rules proposed in this study involve biological sex.However, this is not the case in other languages.For Norwegian, for instance, Trosterud (2001) has proposed 28 semantic gender assignment rules, not all of which refer to biological males or females.At present, it seems fair to say that the interaction of non-biological semantic rules with other rules is not fully understood.Until a better understanding is arrived at, I suggest adopting the more cautious position, i.e. the Core Semantic Override Principle.

Conclusion
Summing up the contribution of the present study, I have proposed the following rule set for the assignment of gender to Ukrainian nouns (the original numbers of the rules in the text are given in square brackets): ( It has been shown that the rules constitute a hierarchy where semantic rules outrank morpho-semantic rules, which in turn override morphological rules.However, this is not an idiosyncratic fact about Ukrainian, but rather falls out as an automatic consequence of two universal principles -the Elsewhere Condition and the Core Semantic Override Principle. , I propose the following gender assignment rules: (4) Morphological rules: a. Declension 1 AE M b.Declension 2 AE F c. Declension 3 AE F d. Declension 4 AE N e.Declension 5 AE N f.Declension 6 AE N G e n d e r o f d e r i v e d n o u n : Gender/animacy of base noun: M