Understanding Unsegmented User Utterances in Real-Time Spoken Dialogue Systems