Modélisation et Traitement Informatique de la Synonymie

Abstract

This paper deals with automatic structuring of semantic values in a dictionary of synonyms. The data we used were first extracted from seven French published dictionnaries of synonyms and then merged to obtain the files we worked on. We explain here why a discrete mathematic representation of synonymic relation is not sufficient to produce a semantic structure (that represents the different meanings of a term but also their overlapping). Then we propose a continuous representation (using data analysis) that ables the machine to produce for each term its semantic values. The system also labels these values with prototypic synonyms and detects synonyms that share different semantic "axes" with the headword. It should be noted that these semantic spaces are obtained automatically for each headword from a homogeneous list of synonyms.