IMAGACT: Deriving an Action Ontology from Spoken Corpora

This paper presents the IMAGACT annotation infrastructure which uses both corpus-based and competence-based methods for the simultaneous extraction of a language independent Action ontology from English and Italian spontaneous speech corpora. The infrastructure relies on an innovative methodology based on images of prototypical scenes and will identify high frequency action concepts in everyday life, suitable for the implementation of an open set of languages.