Domain-Level Reasoning for Spoken Dialogue Systems