Les bases de connaissances du Web sémantique sont généralement représentées sous
The knowledge base of the semantic Web are often represented by RDF triple repository that form a graph. It requires to use a dedicated language as SPARQL for interrogating them, that is generally not known by non-expert users. Moreover, it also require to know the knowledge base schema. To overcome these problems, the development of question answering systems in natural language is currently growing. In this paper, we propose a new method for the semantic analysis of questions, based on graph transformations, using very general con- straints on the request structure, and that keeps ambiguities that are solved globally at the final step by interrogating the knowledge base. We also developed a method for dealing with lexical gap for identifying databse relations based on WordNet. We obtain good results on relation identification, and the global system show promising results.