This paper presents a mixed approach to spoken language understanding that tries to make best use of the advantages of both statistical and knowledge-based algorithms. Results obtained on ATIS (Air Travel Information System) scenario transferred to Italian language are presented and discussed.