Comparing spoken dialogue systems in commercial and research background reveals a great discrepancy in state-of-the-art systems. Whereas research systems tend to be very complex, allowing free and flexible dialogues using unrestricted speech, commercial systems are restricted to far less complex solutions such as menu-driven systems based on simple keyword-spotting, mostly doing completely without natural language processing. In our paper we present an approach to speech processing in dialogue systems based on linguistic phrase spotting, that allows for efficient and robust speech understanding in simple application domains. Misunderstandings or under-specification of user requests will thus be detected and complex dialogue strategies for clarification may be activated. In all other cases, the system may operate like a keyword-spotter. The proposed approach therefore is well suited for simple application domains such as information retrieval in commercial dialogue systems.