Speaker
Description
This paper examines the challenges of assigning VerbNet semantic classes to Croatian verb senses, based on data from the Croatian verb lexicon Verbion. The lexicon provides a multi-level description of the 500 most frequent verbs in Croatian, integrating lexical, semantic, syntactic, and morphological information. Verb senses are classified into semantic classes according to WordNet and VerbNet, using English equivalents as an intermediary. Since VerbNet is based on Levin's syntactic-semantic classification of English verbs, applying these classes to Croatian presents certain challenges. The analysis identifies five types of cross-linguistic mismatch. First, differences in lexicalization patterns reflect language-specific conventions governing what semantic components are encoded in the verb itself as opposed to what is expressed syntactically. Second, Croatian reflexive morphology frequently introduces semantic distinctions that have no direct counterpart in VerbNet classes. Third, alternation-based classification cannot always be applied, as such alternations are either restricted or encoded differently in Croatian. Fourth, some VerbNet classes do not adequately capture the syntactic and semantic properties of Croatian verbs. Finally, certain high-frequency verbs are absent from the VerbNet inventory altogether. These findings suggest that VerbNet requires systematic adaptation for Croatian. The study contributes to the development of a Croatian-specific extension while preserving VerbNet’s cross-linguistic comparative potential.