Semantic and pragmatically based re-recognition of spontaneous speech