1.1 Genders and declensions in Russian

The majority of Russian nouns are inflected for number and case and are divided into several inflectional classes (declensions) based on their set of inflections. Table 1 presents a widely accepted system with three classes (e.g. Aronoff 1994; Halle 1994; Shvedova 1980). Other approaches have also been proposed,¹ but, since our data do not let us tease them apart, we will not elaborate on this issue. Table 2 shows singular inflections for different declensions (in plural, they have the same inflections in most cases).

As Tables 1 and 2 show, there is a strong correlation between gender and declension, but very often, the gender of the noun cannot be unambiguously determined from its inflection. For example, if a noun ends in -a in nominative singular, it is most likely feminine, but can also be masculine; if it ends in -ju in instrumental singular, it is definitely feminine etc. Adjectives and participles in singular and verbs in past tense singular show gender agreement, which may help to determine the gender of the noun in case of uncertainty. However, masculine and neuter forms coincide in all cases except for nominative and accusative not only in nouns, but also in adjectives and participles, so even agreement information is not always enough.

The correlation between gender and declension is captured in different ways in several major theoretical frameworks. For example, Kramer (2015, 2020) proposes that declension is a syntactic feature situated lower in the tree than the gender feature. This allows gender to influence declension, but not vice versa. On the other hand, Rice (2005) in the framework of Optimality Theory or Corbett & Fraser (2000) in the framework of Network Morphology assume that the declension class influences gender assignment.

²Importantly, Table 1 does not represent several minor groups. Firstly, some animate nouns in the IIa class denoting professions and social roles allow for feminine gender agreement if their referent is female.³ Given that all masculine nouns in the I class are animate, this means that the connection between gender and declension is even stronger for inanimate nouns than for animate ones. Animate nouns have semantic gender, while inanimate nouns do not have any semantic cues and are left with morphophonological cues only.

Secondly, augmentative and diminutive suffixes are not supposed to change the gender of the base noun (e.g. Vinogradov 1947), but as a result many derivates do not readily fit the picture in Table 1. This leads to extensive variation: the base noun gender competes with the gender associated with the inflection.⁴ For example, domina ‘big house’ is listed in (Zalizniak 1987) as a noun with gender variation (M/F): it is derived from a masculine noun dom ‘house’, but inanimate nouns ending in -a are feminine. Thirdly, there is a very small group of nouns with irregular inflections.

Now let us look at indeclinable nouns. Table 3 provides two examples: a declinable noun leto ‘summer’ and an indeclinable noun foto ‘photo’. They have the same final syllable, the same number of syllables and stress pattern. However, in leto the final -o is a nominative singular affix, while in foto it is a part of the root. Indeclinable nouns do not have overt case marking. As these examples show, often it is not immediately clear why a noun is indeclinable from the morphophonological point of view – we will discuss this in more detail in section 2.2.

1.2 The problem of gender markedness

The definition of feature markedness depends on the theoretical framework (see e.g. Haspelmath 2006 for an overview). Structural approaches rely on the so-called representational markedness: a [+a] feature value is more marked than a [−a] value, and no marking is the least marked option. As a result, unmarked gender forms are expected to appear in the structures in which no gender feature is available (e.g. in impersonal sentences).

In Optimality Theory, the unmarked option is the default, which is used when there are no specific requirements to use an alternative option (e.g. C. Rice 2005; K. Rice 2007). Therefore, the unmarked option is expected to be the most frequent and the most productive. Many functional and typological studies hold similar views on markedness.

In spite of these differences, the same feature value is usually selected as unmarked in different approaches. However, this is not true for gender in Russian, which puts it at the centre of an ongoing theoretical debate. In Russian, masculine is the most frequent, while neuter is the least frequent (see Table 1), but is used in impersonal sentences.

Experimental studies also provide conflicting results. Several studies of gender agreement processing found that masculine behaves differently from the two other genders (Akhutina et al. 1999, 2001; Romanova & Gor 2017; Slioussar 2018). Slioussar & Malko (2016) discovered that neuter behaves as unmarked in production and masculine in comprehension in their agreement attraction experiments. Badecker & Kuminiak (2007) who studied gender agreement attraction in production in Slovak, in which the gender system is similar to Russian, also observed that neuter behaved as unmarked.

Structural approaches (e.g. Kramer 2015; Nevins 2011) consider neuter to be unmarked in Russian. For example, Nevins encodes feminine is [+FEM], [−MASC], masculine as [−FEM], [+MASC] and neuter as [−FEM], [−MASC]. For Kramer, feminine is [+FEM], masculine is [−FEM] and neuter corresponds to no gender features. Optimality-Theoretic approaches consider masculine to be unmarked and neuter the most marked (e.g. Rice 2005). Many other authors from different frameworks starting from Jakobson (1960) share this position. Finally, to account for the gender frequency distribution and for the neuter use in impersonal sentences at the same time, Corbett & Fraser (2000) assume masculine to be unmarked on the word level and neuter to be unmarked on the sentence level.

Magomedova & Slioussar (2023) suggest a solution to this paradox. Following various structural approaches, they assume that neuter is representationally unmarked, and conclude that it should be chosen when there are no gender assignment cues. If any cues are present, the ones that point to masculine will tend to win over the others (we will come back to this problem in the discussion section). As a result, both masculine and neuter can be seen as ‘default’ options, but in different circumstances and in different senses.

Magomedova & Slioussar (2023) illustrate the defaultness of masculine by their analysis of diminutive and augmentative nouns and nouns ending in a palatalized consonant in nominative singular. In case of conflicting gender cues, masculine is always significantly more likely to be chosen. As for the defaultness of neuter, they note that it can be observed in impersonal sentences and also in indeclinable nouns. They do not analyse these nouns in their paper, but our study confirms and extends their intuition.

2.1 Semantic and morphophonological factors in gender assignment in the previous studies

In this paper, we refer to semantically motivated gender assignment to indeclinable nouns as semantic analogy. This phenomenon is not well defined: such semantic relations might be similar to hypernymy or to synonymy, in some other cases, they are sporadic conceptual matches. Different terms can be found in literature for this phenomenon: concept association, conceptual link, notional analogy etc. (Murphy 2000, 64–67).

Let us look at semantic and morphophonological factors in more detail. Murphy (2000) found that in contemporary Russian, the impact of these factors may vary for different groups of lexemes. The indeclinables ending in -o and -e are assigned neuter more often than other nouns regardless of their semantic analogy. She has also found that speakers may use masculine as the default gender for some indeclinable nouns: for unknown words, participants chose masculine agreement almost as frequently as neuter, while feminine was rare. This brings us back to the discussion of gender markedness in section 1.2.

Wang (2014) noted the influence of the final -(j)a on feminine gender assignment. Marginal examples of indeclinables ending in -(j)a with feminine agreement are also described in (Mjakilja 2000). Both studies also report semantic analogy effects. Some nouns initially demonstrated gender assignment due to semantic analogy, but later fell under the influence of morphophonological cues. For example, šosse ‘highway’ used to be feminine due to the semantic analogy with doroga ‘road_F’, but later became neuter (Comrie, Stone & Polinsky 1996, 108).

Both morphophonological and semantic analogies are common factors for gender assignment to loanwords cross-linguistically (Corbett 1991, 75–82). For example, a corpus study of Puerto Rican Spanish and Montreal French carried out by Poplack, Pousada & Sankoff (1982) shows that semantic analogy and phonological shape both influence gender assignment to loanwords. However, the size of the effect of each factor can be language-specific as it was found in numerous later works (Rabeno & Repetti 1997; Fuller & Lehnert 2000; Violin-Wigent 2006, among others), including several studies on code-switching in Russian (Leisiö 2001; Pereltsvaig 2004; Chirsheva 2009).

One of the experimental studies on the Vastry platform (https://vastry.ru/, Dobrushina, Staferova & Belokon 2018) showed that speakers are also sensitive to the prescriptive norm in case of gender assignment to indeclinable nouns. The indeclinable word kofe ‘coffee’ is a famous example in Russian where the prescriptive norm (masculine gender) does not coincide with the predominant pattern in real language use (neuter gender). In the experimental study, one group of participants had a reminder that prescriptively kofe is masculine, while the other group did not receive that priming. Participants with priming produced more masculine agreement with other indeclinable nouns as well (like avokado ‘avocado’ and gaspačo ‘gazpacho’) compared to the group without priming.

Another issue that is fundamental for the analysis of semantic factors is the speaker's familiarity with a noun. It plays a minor role for corpus data (people mostly use words they are familiar with), but is hard to control in the experimental settings. Savchuk (2011) conducted a study on the Russian National Corpus (www.ruscorpora.ru), which shows that semantic analogy has a strong impact on some groups of indeclinable nouns. For example, the names of car brands, like Audi ‘Audi’ and Ševrole ‘Chevrolet’, are predominantly used with feminine and masculine agreement, despite being listed in dictionaries as neuter. The analogy with the noun mašina ‘car_F’ suggests feminine agreement, while the analogy with avtomobil’ ‘automobile_M, car_M’ triggers masculine agreement. The study also demonstrated that newly borrowed nouns that are not in the dictionaries yet show much larger variation in gender agreement.

To summarize, previous studies show that many factors are at play: semantic analogy, the default status of neuter and maybe masculine, morphophonological factors. In our corpus study, we collected a large dataset to be able to estimate their relative influence.

2.2 What nouns become indeclinable

The distribution of indeclinable nouns discussed in this section stresses not only the strong connection between genders and declensions, but also the role of the stem-final segment and the relative productivity of different declensions. The pattern in the declension IIa (masculine nouns ending in a consonant) is the most productive, the pattern in the declension I (feminine nouns ending in -(j)a) is fully productive as well. The classes III and especially IIb get new nouns almost exclusively due to several productive suffixes associated with these declensions accidentally, all these suffixes are used to derive abstract nouns.

Russian nouns ending in a consonant in nominative singular are almost always declinable, with the exception of some female first and last names. Let us look at two examples. The last name of the Russian-born linguist Asya Pereltsvaig does not decline because there are no Russian feminine nouns with a stem-final non-palatalized consonant and a zero inflection. The last name of the Russian linguist Natalia Slioussar, spelled as Sljusar’ in Russian, could go to the III declension, but this never happens to such names – apparently, due to the low productivity of this class. Notably, both last names are declinable in masculine, i.e. when they refer to men.

There are examples of loanwords ending in a consonant like xevi-metal ‘heavy metal’ (music genre) that were noted to be indeclinable in the past (Murphy 2000), but now decline normally. All loanwords ending in a non-palatalized consonant go to the IIa declension and become masculine, while loanwords ending in a palatalized or alveo-palatal consonant may be incorporated either in the IIa or in the III declension and become masculine or feminine, respectively. As Savchuk (2011) shows, many of these nouns show gender variation in modern Russian. Importantly, this gender variation always goes hand in hand with declension variation. For example, for the word šampun’ ‘shampoo_M/F’ there are two options in genitive singular: xorošego šampunja ‘good_M.GEN.SG shampoo_GEN.SG’ or xorošej šampuni ‘good_F.GEN.SG shampoo_GEN.SG’. Using a masculine adjective with a III declension form or vice versa is absolutely ungrammatical.

Most Russian nouns ending in -(j)a are declinable. Some first and last names like Derrida are exceptions, although very often, there is variation in real language use. A small number of loanwords do not decline, but mostly monosyllabic ones like fa ‘note fa’ or those ending in a hiatus like fejxoa ‘feijoa’, which stresses the importance of the stem for gender assignment (no native Russian words end in a hiatus). However, there are some exceptions like antraša ‘entrechat’.

All nouns ending in -(j)u are indeclinable, which is not surprising because -ju or -u do not correspond to any nominative singular affix. The same is true for the very few nouns that end in -ė. Nouns ending in -i also do not decline unless -i is reanalysed as a nominative plural affix. However, this usually happens only in colloquial Russian (e.g. dlja xinkalej ‘for khinkali_GEN.PL’ (Georgian dumplings)). Most words ending in -i remain indeclinable, but are often used with plural agreement. Among other things, this means that the speaker does not have to decide on their gender because Russian has gender agreement only in singular. There are also a few indeclinable personal names and toponyms ending in -y (e.g. Gurbanguly (personal name), Janczy ‘the river Yangtze’).

Nouns ending in -o and -e are probably the most interesting. In declinable nouns, these are nominative singular affixes that unambiguously point to neuter (with the exception of some diminutive and augmentative nouns). However, loanwords ending in -o and -e do not go to the IIb declension and remain indeclinable, which is not the case in other Slavic languages. Moreover, some native proper names and toponyms ending in -o, like Kupčino (a district in Saint Petersburg), show variation in declinability. This can be associated with the low productivity of this declension in Russian, although it would be great to offer a more precise theoretical explanation in the future.

3.1 Data collection

In this paper, we present the results of a web-corpus study, with a focus on gender markedness and on idiosyncratic noun groups. We used the LiveJournal subcorpus of the General Internet Corpus of Russian (GICR, http://www.webcorpora.ru/) as we wanted to study variation which is poorly represented in edited texts. The LiveJournal subcorpus consists of blogs and contains 8,720 million words. It has some morphological annotation (morphological ambiguity, including ambiguity arising from case syncretism, is not resolved). We searched for combination of a preposed attributive adjective as an agreement target and an inanimate indeclinable noun as a head.

For the list of heads, we took all indeclinable nouns from the Grammatical Dictionary of the Russian Language (Zaliznjak 1987) and added a few other nouns. Murphy (2000) enumerates over 1,500 indeclinable nouns mentioned in various dictionaries. Firstly, we estimated their frequency subjectively to exclude those that would not be familiar to many Russian speakers (for instance, kavallo, a statue of a horseman in the classical Italian art, or antuka, a type of umbrella popular in the 19th century). We also did not include several highly frequent nouns: kofe ‘coffee’, kafe ‘café’, metro ‘subway’ and taksi ‘taxi’. Using the resulting list, we searched for all available information in the corpus. After that, we excluded some examples that were retrieved by mistake (i.e. did not contain an indeclinable noun with a preposed adjective modifying it) and all nouns with less than two examples of gender agreement.

One instance of gender agreement produced by one author was taken as one observation. If one author produced numerous instances of the same agreement for the same noun, such duplicates were dismissed. However, in 496 cases, one and the same noun was used in different genders by the one and same speaker (in different blog posts or even within one post). These cases were included as separate observations. They show that the gender of indeclinables can vary even in the grammar of a single speaker.

The resulting dataset contained 66,939 data points. We annotated it for gender, number and case. Firstly, many instances were assigned these features automatically based on the inflections of adjectives. After that, other cases were manually disambiguated. Masculine and neuter adjectives are indistinguishable in cases other than nominative and accusative, so these instances were labelled as NM.

There is no gender agreement in plural, but we were interested in the cases of lexicalized plural agreement. Unfortunately, it is not always clear where an indeclinable noun is used as a pluralia tantum and where we deal with a regular plural formation. Eventually, we marked only those cases where we could be almost certain and left them for future research. In the present study, no instances of plural agreement are discussed, except for one example in Table 5 below.

The GICR also provides some metadata, although they are not always reliable.⁸ Most of the blog posts in our dataset were published between 2010 and 2013. The average birth year of the authors in our dataset is 1980. Most of the users who shared their location were in the Russian Federation, mainly in Moscow. There are also users located in Belarus, Ukraine, and Kazakhstan – the countries where there are many native speakers of Russian.

3.2 General statistics

Our dataset contains 66,939 data points. It is not balanced for frequency, as our goal was to collect all the available data. The most frequent noun in our dataset is viski ‘whiskey’ with 4,453 instances, the least frequent noun is napareuli ‘napareuli’ (a sort of wine) with 3 instances.

For the purposes of this study we excluded plural forms and monosyllabic nouns from the analysis. Homonyms were treated as two distinct nouns. This left us with 132 nouns (types) and 32,988 observations (tokens) in the main dataset.

Figure 1 presents gender distribution for the nouns that appeared more than 120 times in our dataset. The absolute majority of nouns show gender variation. For some of them, like xudi ‘hoodie’, reno ‘Renault’ or pežo ‘Peugeot’, all three genders have comparable frequencies.

View Full Size

The distribution of gender agreement for frequent nouns in the dataset

Citation: Acta Linguistica Academica 70, 3; 10.1556/2062.2023.00641

• Download Figure
• Download figure as PowerPoint slide

3.3 Semantic analogy and morphophonological patterns

While labelling root-final segments is a very straightforward task, establishing semantic analogies may be trickier. To the best of our knowledge, studies discussing semantic analogy in gender assignment do not report any issues with the procedure (Poplack, Pousada & Sankoff 1982; Fuller & Lehnert 2000; Violin-Wigent 2006). However, in Russian indeclinables, there might be several or no obvious semantic analogies for a given word.

Therefore, we asked 25 native speakers to provide the most obvious word to describe a given indeclinable noun. Relying on the most popular answers, we introduced the following values for the semantic analogy factor. F, M, N were used for the words with one salient semantic analogy and “?” for the absence of any salient semantic analogy. The M/F label was assigned to the nouns with two salient semantic analogies of M and F gender. These are mostly car brands, their semantic analogies are mašina ‘car_F’ and avtomobil’ ‘car_M’.

The resulting distribution of target nouns by semantic analogy gender is provided in Table 4. Table 4 also shows the distribution of assigned genders: a total number of tokens found, and the percentages of different genders in each category. Masculine is the most frequent as a semantic analogy gender, while neuter is the least frequent. This reflects the general distribution of genders in Russian nouns (see Table 1). However, the absolute majority of indeclinables with a neuter semantic analogy are neuter, while the same cannot be said about masculine and especially about feminine.

Let us compare the distributions of genders assigned in a group of nouns with the same semantic analogy, but with different root-final segments (Table 5). Here, we also included plural instances into the counts. Flamenko, čačača, and sirtaki are dances, and the word tanec ‘dance’ is masculine in Russian. Most instances of flamenko are neuter with zero feminine examples, while čačača has 26% of feminine instances. Sirtaki has no feminine instances, but has 21% of pluralia tantum forms instead, most probably due to the final -i. This may be surprising since sirtaki does not implicate any semantic plurality unlike spaghetti or pince-nez. These examples illustrate the influence of the final vowel on gender assignment.

The role of semantics can be illustrated by gender distribution in homonyms, like safari in Table 5. Safari can denote a nature trip or a popular web-browser. Safari in the first meaning was not labelled as having any salient semantic analogy and shows neuter agreement in the absolute majority of cases. Safari in the second meaning usually agrees in masculine due to the analogy with the word brauzer ‘browser_M’. Some examples of feminine agreement might be explained by the analogy with programma ‘program_F’. Neuter agreement is virtually absent.

Now let us turn to morphophonological factors. Our dataset contains 12 lemmas with the root-final -(j)a, 30 lemmas with final -o, 26 with final -e, 23 with final -(j)u, and 41 with final -i. The distribution of the assigned gender by the final segment is provided in Table 6. Absolute numbers show how many tokens were found, percentages show the share of the respective gender among all nouns with a particular final segment.

Nouns with the final -(j)u are of a particular interest as there is no corresponding declension marker for the nominative case. Thus, (j)u-final roots can be used to examine the suggested defaultness of the neuter gender. Table 7 shows gender assignment in these nouns taking the semantic analogy gender into account. It also includes (j)a-final and i-final nouns that will be discussed below.

Within the (j)u-final subset the group with no obvious semantic analogy is the largest. There are nouns like azu ‘azu’ (a Tatar dish), dežavju ‘deja vu’, interv’ju ‘interview’, kazu ‘kazoo’ (a musical instrument), paspartu ‘passe-partout’, ragu ‘stew, ragout’, randevu ‘rendezvous’, ušu ‘wushu’, fondju ‘fondue’, hokku ‘haiku’, džiu-džitsu ‘jiu-jitsu’. This set of nouns should be the most problematic for the speaker as there are no clear cues for gender assignment. Therefore, the default gender is expected. Indeed, 1,809 of the 2,043 tokens (88.5%) are neuter, 151 are masculine and only 83 are feminine. However, even in the situation of uncertainty we observe some gender variation.

Nouns with the final -(j)a look similar to (j)u-final nouns in terms of the gender distribution and exactly the opposite in terms of the gender assignment cues. As Table 7 shows, this set mostly consists of nouns with contradictory cues. The root-final -(j)a may trigger feminine agreement, but there are almost no words with a feminine semantic analogy. Words with a neuter semantic analogy are also absent (nevertheless, 63% of nouns in this group are assigned neuter).

There are only 12 (j)a-final nouns in our data, namely: amplua ‘character (theatre)’, antraša ‘entrechat’, boa ‘boa’, kinoa ‘quinoa’, krema ‘cream on top of a coffee’, ljulja ‘kebab’, media ‘media’, patua ‘patois’, tanka ‘tanka’, fejxoa ‘feijoa’, fuagra ‘fois-gras’, čačača ‘cha-cha-cha (dance)’. Six of these nouns have hiatus endings, which is very unusual for Russian, and most of them have stress on the root-final vowel. The (j)a-final noun set offers another opportunity to test for the defaultness of neuter in indeclinables – they have contradictory cues for gender assignment, but very often end up being neuter.

-i also does not correspond to any nominative singular affix, but the subset of i-final nouns differs from (j)u-final ones for a number of reasons. Firstly, many i-final nouns show lexicalized plural agreement. Secondly, several nouns with a masculine semantic analogy like viski ‘whiskey’ are much more frequent than the other words in the dataset, which skews the distribution of genders. A decent share of feminine agreement is also due to a particular group of words: five car brand names (audi, lamborgini, micubisi, ferrari, mazerati). As we mentioned earlier, these words have m/f semantic analogy gender. Feminine agreement instances of these nouns constitute 76% of feminine agreement instances in the whole i-final subset. Let us note that neuter agreement is nevertheless highly frequent for the words with feminine and masculine semantic analogy and clearly prevails not only in the n group, but also in the ? group.

Finally, we also looked at stress patterns, which can be considered a purely phonological, rather than a morphophonological factor. As we will show in the next section, stress patterns appear to be significant predictors and improve the predictive power of the model. However, we cannot be completely sure that this significance is actually meaningful. As we have mentioned above, indeclinable nouns are a limited set of words that very often form subgroups with idiosyncratic behaviour, which some of the predictors we used in the models might catch by coincidence. We compared different stress patterns (on the final, prefinal and the second prefinal syllable), but grouping all words with a non-final stress appeared to be more meaningful, so this is what we present in Table 8. Let us add that in declinable nouns, there is no connection between gender and stress patterns.

3.4 Statistical analysis

The statistical analysis was done in the R programming environment (www.r-project.org). We modelled the data with a mixed-effects regression with random intercept by item using the lme4 package (Bates et al. 2015). For post hoc analyses, we used Tukey's tests with a Holm-Bonferroni correction using the glht function from the multcomp package (Bretz, Hothorn & Westfall 2010). We used the same set of independent variables for each model, namely: the root-final segment, the gender of the semantic analogy noun, and stress. We chose neuter as the reference level for semantic analogy gender factor, and -i as a reference for the final segment factor.

Masculine gender assignment. Nouns with a feminine, neuter or unclear semantic analogy are less likely to be assigned masculine than nouns with masculine or mixed semantic analogy (see Table 9). Nouns with a mixed semantic analogy are less likely to be assigned masculine than those with masculine analogy. The root-final vowel and the stress pattern are not significant predictors for masculine gender assignment.

Feminine gender assignment. Nouns with a feminine semantic analogy end up with feminine agreement significantly more often than nouns with any other analogy except for the mixed m/f (see Table 10). As we mentioned above, this group of nouns mostly includes car brands. The feminine semantic analogy (mašina ‘car_F’) is very strong for them, and the other analogy (avtomobil’ ‘automobile_M, car_M’) rather reduces the share of the ‘default’ neuter agreement than undermines the influence of the feminine analogy word. This is probably due to the fact that the word avtomobil’ ‘car_M’ is not as frequent as the word mašina ‘car_F’ (156.9 ipm and 490.4 ipm according to Lyashevskaya & Sharov (2009)). There are no significant differences between masculine, neuter and unclear semantic analogy.

Nouns ending in -(j)a are assigned feminine significantly more often than nouns with any other root-final segment (see Table 11). (j)u-final nouns are significantly more likely to be feminine than nouns with the root-final -e and -o; (j)u-nouns are more likely to be feminine than i-nouns on a tendency level. Nouns with the final stress are significantly less likely to be feminine (β = −1.62, SE = 0.57, P < 0.005) than masculine or neuter.

Neuter gender assignment. As for semantic analogy, every value significantly differs from the others (see Table 12), except for m vs. f and m vs. m/f, which are different on a tendency level.

Considering the root-final vowel, -o is different from -i, -(j)a and -(j)u, but not -e (declinable neuter nouns end in -o or -e). There are no other significant differences (see Table 13). Finally, nouns with the final stress are significantly more likely to be neuter (β = 1.39, SE = 0.42, P < 0.01) than masculine or feminine.

3.5 Discussion. Markedness hierarchy

Data on inanimate indeclinables can complement the discussion on gender defaultness and markedness in the Russian nominal system. Figure 2 shows how often a particular gender is assigned to an indeclinable noun with or without any cues that point to this gender. For example, the root-final -(j)a is a cue to feminine. Figure 2 demonstrates that the share of neuter agreement used in the absence of any cues is considerably larger than the corresponding shares of masculine and feminine agreement.

View Full Size

The distribution of gender agreement depending on the presence or absence of different cues

Citation: Acta Linguistica Academica 70, 3; 10.1556/2062.2023.00641

• Download Figure
• Download figure as PowerPoint slide

Magomedova & Slioussar (2023) argue that neuter is the default gender in Russian, while the special properties of masculine are explained by the fact that it is most frequent and productive and has the most diverse set of nominative singular forms. Therefore, they predict that neuter should be chosen in the absence of any cues. This prediction can be tested only on indeclinables, because in declinable nouns, declension is a very strong cue for gender assignment, and our data support this prediction. In their own study of diminutive and augmentative nouns, Magomedova and Slioussar show that masculine is significantly more likely to be chosen in case of conflicting gender cues. Thus, both masculine and neuter have a special status in the Russian nominal system, but the nature of this status and its manifestations are very different.

Some effects associated with the special status of masculine can also be observed in our study. Unmotivated masculine and feminine agreement is infrequent in our data. In order to compare them, let us form a subset with 672 (5% of all feminine gender agreement instances) and 127 (3% of all feminine gender agreement instances) entries respectively. When nouns have no salient semantic or morphological cues pointing to any gender or when they are assigned an unexpected gender given the morphophonological or semantic factors, they are masculine significantly more often than feminine (according to the two-proportions Z-test, Z-score = 4.2, P < 0.01).

In our sample, there are nouns with decent shares of masculine assignment without any cues to it, like xudi ‘hoodie’ (33% of masculine respectively). Other nouns have smaller shares of unmotivated masculine agreement: for example, 6% of pončo ‘poncho’ instances are masculine, despite its final vowel pointing to neuter. Examples of unmotivated feminine agreement are usually isolated cases, except for the nouns seppuku ‘seppuku’ and xokku ‘haiku’ (10% and 7% respectively).

3.6 Discussion. Approaches to gender assignment and the challenges posed by indeclinable nouns

Different approaches to gender assignment have been proposed, in the frameworks ranging from Distributed Morphology, as in Kramer (2015, 2020), to Relational Morphology, as in Jackendoff & Audring (2020), and Optimality Theory, as in Rice (2005). Ruth Kramer in her paper on gender assignment claims that it is basically a derivation process and gender is assigned by the Merge operation when a bare root or a stem is merged with a nominal head that contains a gender feature. In simplex nous, gender is assigned first on the basis of semantics and then the remainder may be assigned gender based on the declension class.

In the analysis of derived nouns, she assumes lexical decomposition: the speaker already knows how to divide a given word into morphemes as gender assignment is done by the Merge operation (Kramer 2020, 58). She also argues for top-down derivation and doubts the existence of phonological gender assignment. Our data do not support this analysis: not only phonological factors matter, the root-final segment is not separable from the root in any formal way.

The same kind of problem holds for the Relational Morphology approach: although this approach allows analogy as a mechanism for assigning features, there is no formal way to separate the root-final vowel from the root and substitute it with a variable. The ‘similar’ relation can be used instead of the ‘same-same’ relation, but we could not find a formal way to do that. Hence, competition-based approaches are optimal for our data. For example, we can specify ‘ends in -(j)a’ as an OT constraint in a Harmonic Grammar type of approach, which will allow us not only to account for the observed variation, but also to predict the distribution of the genders assigned.