Verb Interpretation for Basic Action Types: Annotation, Ontology Induction and Creation of Prototypical Scenes

摘要

In the last 20 years dictionaries and lexicographic resources such as WordNet have started to be enriched with multimodal content. Short videos depicting basic actions support the user's need (especially in second language acquisition) to fully understand the range of applicability of verbs. The IMAGACT project has among its results a repository of action verbs ontologically organised around prototypical action scenes in the form of both video recordings and 3D animations. The creation of the IMAGACT ontology, which consists in deriving action types from corpus instances of action verbs, intra and cross linguistically validating them and producing the prototypical scenes thereof, is the preliminary step for the creation of a resouce that users can browse by verb, learning how to match different action prototypes with the correct verbs in the target language. The mapping of IMAGACT types onto WordNet synsets allows for a mutual enrichment of both resources.