Hungarian speakers use morphological dependencies in inflecting novel forms

Abstract

Theories of morphology must account for lexicalized variation: lexical items that differ unpredictably in their inflection must be memorized individually and differ in their stored representation. When tested on such cases, adult speakers usually follow the Law of Frequency Matching (Hayes et al. 2009), extending gradient phonological patterns from the lexicon. This paper looks at lexicalized variation in the Hungarian possessive: first, I show that a noun’s choice of possessive is partially predicted by its plural form as well as its phonological shape. Then, using a novel nonce word paradigm, I show that Hungarian speakers productively apply this cooccurrence pattern between the plural and possessive. I handle lexicalized variation with diacritic features marking lexical entries and propose that Hungarian speakers have learned a gradient cooccurrence relation between diacritic features indexing their plural and possessive forms, extending the sublexicon model of Gouskova et al. (2015). In this proposal, morphological knowledge is distributed across rules in a generative grammar, individual lexical items indexed for their morphological properties, and pattern-matching grammars storing generalizations over those indexed lexical items.