Automatic Aid to Symbolic Cluster Interpretation

This research consists of an attempt to provide algorithms that build abstracts on partition classes; these abstracts are formalized as probabilist symbolic objects and, when necessary, they extract from subsets, bodies of knowledge, in the form of scenarios, tactics or prototypes. An application is developped for the formalization of senegalese fishing tactics. Various problems with relative solutions, such as the unifying or discriminating of description balancing all along the treatment, are illustrated in this context.