Incremental Spoken Dialogue Systems: Tools and Data
Helen Hastie, Oliver Lemon, Nina Dethlefs · 2012
Strict-turn taking models of dialogue do not accurately model human incremental processing, where users can process partial input and plan partial utterances in parallel. We discuss the current state of the art in incremental systems and propose tools and data required for further advances in the field of Incremental Spoken Dialogue Systems. 1 Incremental Spoken Dialogue Systems For Spoken Dialogue Systems (SDS) to be more frequently adopted, advances in the state-of-the-art are necessary to enable highly responsive and conversational systems. Traditionally, the unit of speech has been a whole utterance with strict, rigid turn-taking determined by a voice-activity detector. However, a large body of psycholinguistic literature indicates that human-human interaction is in fact incremental (Tanenhaus and Brown-Schmidt, 2008; Levelt, 1989). Using a whole utterance as the unit of choice makes dialogues longer, unnatural and stilted and ultimately interferes with a user’s ability to focus on their goal (Allen et al., 2001). A new generation of Incremental SDS (ISDS) are being developed that deal with ‘micro-turns ’ (subutterance processing units) resulting in dialogues that are more fluid and responsive. Recent work has shown that processing smaller ‘chunks ’ of input and output can improve the user experience (Aist et