Two-way speech-to-speech (S2S) translation is a spoken language application that integrates multiple technologies including speech recognition, machine translation, text-to-speech synthesis, and dialog management. In recent years, research into S2S systems has resulted in several modeling techniques for improving coverage on broad domains and rapid configuration for new language pairs or domains. This talk will highlight recent advances in S2S area that range from improvements in component technologies to improvements in the end-to-end system for mobile use. I will also present metrics for evaluating the S2S technology, a methodology for determining the impact of different causes of errors, and future directions for research and development.