In this article, we introduce a new method for query generation. This method uses only a logical approach and does not need a statistical process or a natural language processing. The main interest of this new method is the abstraction. In a second part, we discuss a method for learning a text classifier and query generation for this classi.er. The two problems are resolved in a complementary approach using our query generation method and SVM as text classifiers. We use this approach for studying words polysemy. Our method generates queries in order to retrieve documents about a specific sense of the word and in the same time learning the associated text classifier. Our method have good results.