Modeling gender variation in Russian indeclinable nouns

Optimality over structuralism, hierarchical MaxEnt, and degrees of idiosyncrasy

Abstract

In this paper, we provide an analysis of the grammatical gender of 131 inanimate indeclinable Russian nouns based on the data from the General Internet Corpus of the Russian Language. We demonstrate that most nouns show substantial variation, being used in two or even in all three genders: masculine, feminine, and neuter. We identify several factors affecting this, primarily the gender of the semantic analogy noun and the root-final vowel. We argue that these data can be used to compare several major morphological frameworks and conclude that some approaches, namely optimality-theoretic probabilistic ones, are better suited to account for them. We also compare different models within the chosen set of approaches and show that the hierarchical Maximum Entropy (MaxEnt) models are superior to the classical MaxEnt models.