Évaluer SynLex

Ingrid Falk (1) , Gil Francopoulo (2) , Claire Gardent (2)

1 ATILF - Analyse et Traitement Informatique de la Langue Française
2 TALARIS - Natural Language Processing: representation, inference and semantics

SYNLEX is a syntactic lexicon extracted semi-automatically from the LADL tables. Like the other syntactic lexicons for French which are both available and usable for NLP (LEFFF, DICOVALENCE), it is incomplete and its recall and precision wrt a gold standard are unknown.We present an approach which goes some way towards adressing these shortcomings. The approach draws on methods used for the automatic acquisition of syntactic lexicons. First, a new syntactic lexicon is acquired from an 82 million words corpus. This lexicon is then used to validate and extend SYNLEX. Finally, the recall and precision of the extended version of SYNLEX is computed based on a gold standard extracted from DICOVALENCE.

Mots clés: Syntactic lexicon; Evaluation

Domaines: Informatique et langage [cs.CL]

Ingrid Falk, Gil Francopoulo, Claire Gardent. Évaluer SynLex. Traitement Automatique de la Langue Naturelle - TALN 2007, Jun 2007, Toulouse, France. 10 p. ⟨hal-00162182⟩