L'Interrogation des banques de donneées en langage naturel dans une démarche cognitive complex

Abstract This paper deals with linguistic and mathematical problems which arise in natural language processing of data bases of informations, liable to have a cognitive function. Within the framework of the project AVEROES, questions are asked to a data base providing a description of morphological aspects of roman amphora, in order to analyse and to verify traditional typologies in this archaeological field. To begin with, a language for the formulation of questions is defined; it is planned to associate a question in natural language (LN) with the elements concerned by it, that is, the objects dealt with (the “support” corresponding to a data array copied out from the bank) and the different procedures to be used (the “theme” corresponding to a set of algorithms). Next, it is shown that the matching of algorithms with data arrays, first comes to a problem of choice between representations, between algorithms, and then is likely to lead to uninteresting answers, or even false ones if questions convey an implicit knowledge that is not or cannot be put into words.