From data analysis to neural networks: new prospects for efficient browsing through databases

Querying documentary systems in natural language appears today to be the high road for success. Of course, a high degree of tolerance of these systems to typographical and spelling mistakes, to the use of synonyms, is extremely useful in the man-machine relationship, especially when our goal is to make these systems available to ever wider groups of users. But it appears that the search for a somewhat anthropomor phic dialogue with the computer is premature as long as a certain number of representational problems are not resolved: how to supply the human interlocutor with an overall view of the editorial position(s) adopted by the administrators and indexers of the documentary database, whether implicitly or not? This human interlocutor must have a sure idea of what he might find, of what he will not find. In short, he must know with whom and with what he is dealing! What tools should he be given to find his way all through his search and make it progress at all times in the directions that are most relevant from his point of view?