Word Senses as Clusters of Meaning Modulations: A Computational Model of Polysemy

Most words in natural languages are polysemous; that is, they have related but different meanings contexts. This one-to-many mapping of form to meaning presents a challenge understanding how word learned, represented, and processed. Previous work has focused on solutions which multiple static semantic representations linked single form, fails capture important generalizations about polysemous used; particular, the graded nature senses, flexibility regularity polysemy use. We provide novel view represented processed, focusing is modulated by context. Our theory implemented within recurrent neural network learns distributional information through exposure large representative corpus English. Clusters emerge from model processes individual forms. In keeping with theories semantics, we suggest generalized contexts tokens, emerging as clusters contextually meanings. validate our results against human-annotated gradedness, flexibility, sense individuation, well behavioral findings offline relatedness ratings online sentence processing. The insights into emerges contextual processing both theoretical computational point view.