Résumé
In the construction of a thematic representation system of knowledge in Natural Language Processing, it became current to use a thesaurus. Essentially, a thesaurus consists of a set of concepts, often organized in an arborescent structure : it is a fundamental, but relatively fixed tool. But concepts (and language!) are evolving, and this, more and more quickly, particularly in technical fields. We propose a system which makes possible for the notion of concept to evolve by the introduction of the "Basic Notions". Those, necessarily definite on the vector space of the concepts of the thesaurus, constitute another generating system of the space of thematic representation of knowledge. Contrary to the concepts of the thesaurus, the basic notions evolve progressively with the analysis of new texts. We discuss the optimal value of the dimension of the space of representation generated by the basic notions, then of the determination of the acceptions allowing to express them. Lastly, we consider the differentiation between basic notions of general space and those of a specialized field.