Extraction Semi-Supervisée de Couples d'Antonymes grâce à leur Morphologie

In the framework of meaning representation in Natural Language Processing, we focus on enabling a system to autonomously learn word meanings and semantic relations from user dictionaries, web contents and other lexical resources. For antonymy, as a lexical semantic relation, no resource provides distinctions between complementary, scalar and dual antonymies. In this paper, we present a semi-supervized method to collate such lists, based on operating morphological opposition holding between lexical items. The approach presented here starts from a bootstraped initial list. It is able to augment such lists but also to find out morphological oppositions. We scrutinize the obtained results and discuss the distribution of antonymy types. Didier Schwab, Mathieu Lafourcade et Violaine Prince