Turn-taking as a design principle for barge-in in Spoken Language Systems

It is widely acknowledged that users of Spoken Language Systems (SLS) want the ability to truncate system prompts by using a barge-in capability (e.g., Basson et al., 1995; Yankelovich et al., 1995). However, little has been published on how barge-in is used or if it adversely affects Automatic Speech Recognition (ASR) and the interface usability. Typically, user requests for barge-in are assumed to be based on the desire to make system interactions faster and therefore more similar to interactions with touch-tone systems. We believe that requests for a barge-in capability are rooted in the notion of discourse as a turn-taking event. Viewed in this way, we believe SLS can be enhanced to develop speech interfaces that are deemed more natural by users, as well as to increase system performance. This study addressed several issues. We found that users new to the system did not need to be informed about the barge-in capability before they attempted barge-in, that they used barge-in during almost half of their interactions with the system, and that they had identifiable patterns of barge-in use consistent with the turn-taking model. Results are presented and consequences for speech interface design as well as algorithm enhancement are discussed.