Learning Pretopological Spaces for Lexical Taxonomy Acquisition
Résumé
In this paper, we propose a new methodology for semi-supervised acquisition of lexical taxonomies from a list of existing terms. Our approach is based on the theory of pretopology that offers a powerful formalism to model semantic relations and transform a list of terms into a structured term space by combining different discriminant criteria. In order to learn a parameterized pretopological space, we define the Learning Pretopological Spaces strategy based on genetic algorithms. The rare but accurate pieces of knowledge given by an expert (semi-supervision) or automatically extracted with existing linguistic patterns (auto-supervision) are used to parameterize the different features defining the pretopological term space. Then, a structuring algorithm is used to transform the pretopological space into a lexical taxonomy, i.e. a direct acyclic graph. Results over three standard datasets (two from WordNet and one from UMLS) evidence improved performances against existing associative and pattern-based state-of-the-art approaches.