Colex: Un outil d’extraction de collocations spécialisées basé sur les fonctions lexicales

We present a method for extracting verb + noun collocations from specialized corpora based on the lexical functions of the Meaning-Text Theory. Lexical functions offer a complete characterization of verbal collocations. They cover the three grammatical relations that obtain between a verb and its arguments and provide high-level semantic descriptors to represent the meanings of the common verbal collocates of terms. Our system uses morphosyntactic patterns to extract verb–argument relations from a parsed corpus of computer texts. We then apply two statistical tests to isolate the true collocations in the list of syntactically extracted combinations. While the better of the two statistical tests identifies collocations with a precision of 71 %, we argue that a semantic filtering of the combinations, based on their ability to be encoded as standard lexical functions, could achieve higher precision.