Proceedings of the Annual Meeting of the Cognitive Science Society

Word class conversion refers to the extended use of a wordfrom one grammatical class to another without overt morpho-logical marking. Noun-to-verb conversion, or denominaliza-tion, is one form of word class conversion studied extensivelyin the literature. Previous work has suggested that novel de-nominal verb usages are comprehensible if the listener cancompute the intended meaning based on shared knowledgewith the speaker. However, no existing work has explored thecomputational mechanism under this proposal. We proposea frame-semantic generative model, Noun2Verb, that supportsthe inference and generation of novel denominal verb usagesvia semi-supervised learning. We evaluate this framework ina dataset of denominal verbs drawn from adults and childrenagainst a state-of-the-art model from natural language process-ing. Our results show that Noun2Verb aligns better with humaninterpretation and bridges the gap between machines and hu-mans in lexical innovation.