In this article we report the design principle and the outline of our spoken dialogue translation system from Japanese into Korean. Compared with existing speech translation systems, our prototype system has two major features: 1) Semi-realtime system based on a compact hardware configuration, and 2) Improving dialogue speech recognition rate using a light-weight discourse monitor. Additionally we implemented a user-friendly interface, considering the hotel reservation task environment. The experimental results with evaluation suggested us the feasibility of the system for limited task domains. Based on this experiences, crucial issues for interpreting communications technique will be discussed.