In the field of Natural Language Processing, in order to work out a thematic representation system of general knowledge, methods relying on thesaurus have been used for about twenty years. A thesaurus consists of a set of concepts which define a generating system of a vector space modelling general knowledge. These concepts, often organized in a treelike structure, constitute a fundamental, but completely fixed tool. Even if the concepts evolve (we think for example of the technical fields), a thesaurus as for it can evolve only at the time of a particularly heavy process, because it requires the collaboration of human experts. After detailing the characteristics which a generating system of the vector space of knowledge modelling must have, we define the "basic notions". Basic notions, whose construction is initially based on the concepts of a thesaurus, constitute another generating system of this vector space. We then approach the determination of the acceptions expressing the basic notions. Lastly, we clarify how, being freed from the concepts of the thesaurus, the basic notions evolve progressively with the analysis of new texts by an iterative process.