AnalogiX : un corpus d’analogies annotées en structure et qualité
Jérémie Roux, Hani Guenoune, Mathieu Lafourcade, Joël Maïzi, Philippe Langlais
Résumé : We describe AnalogiX, a corpus of 5316 analogies between French terms produced by humans and large language models. Unlike existing analogy corpora, this one has the advantage of being annotated with semantic relations between terms as well as a quality indicator for each analogy.We also describe the serious game CompalogiX, which allows players to assess the quality of the proposed analogies. Based on 12865 games played, we show that humans prefer better-constructed analogies from the perspective of intrinsic semantic relations.
Mots clés : analogie,base de connaissances,corpus,grand modèle de langue,similarité,relation,jeu avec un but,jeu sérieux