We describe a domain-independent semantic interpretation architecture suitable for spoken dialogue systems, which uses a decision-list method to effect a transparent combination of rule-based and data-driven approaches. The architecture has been implemented and evaluated in the context of a mediumvocabulary command and control task.