This paper presents a system which automatically generates shallow semantic frame structures for conversational speech in unrestricted domains. We argue that such shallow semantic representations can indeed be generated with a minimum amount of linguistic knowledge engineering and without having to explicitly construct a semantic knowledge base.