Uses of higher level knowledge in a speech understanding system: A progress report