查看文章

The IMAGACT project, which has been funded in Italy with the PAR/FAS program of the Tuscan Region (undertaken by the University of Florence, ILC-CNR, Pisa, and the University of Siena), uses both corpus-based and competence-based methodologies for simultaneous extraction ofa language independent action inventory from spontaneous speech resources of different languages. The IMAGACT framework faces key issues in ontology building. It grounds productive translation relations since it distinguishes the proper usage of verbs from their metaphorical or phraseological extensions; it allows easy identification of types in the variation, it is crosslinguistic in nature, it derives from the actual use oflanguage but it can be freely extended to other languages through competence-based judgments, and it is therefore suitable for filling gaps in lexical resources [1]. The result ofthe annotation work is accessed through the IMAGACT query interface, which is designed to be used to master the verbal lexicon ofa second language during the acquisition process.

The project focuses on high frequency action verbs (approximately 600 lexical entries) of both Italian and English, which represent the basic verbal lexicon of action in the two languages. The purpose of the project is two-fold:(i) to derive information about action verbs adopting a bottom-up approach, from spoken corpora eg the BNC [2] and C-ORAL-ROM [3];(ii) to construct a multilingual ontology of action, anchored to videos. English and Italian action types derived from corpus are mapped onto the same set ofscenes. The validation ofbasic action types so derived is accomplished through a …