Problem-Sensitive Response Generation in Human-Robot Dialogs

Abstract

We develop a new mechanism to detect and respond to miscommunications in human-robot dialogs, distinguishing between computer misunderstandings vs. human inexperience.Problem indicators drive an error/help state machine, which augments the dialog state and is used in tailoring response generation.A user study shows that the task success rate and user satisfaction is improved substantially by the two-part miscommunication model.* Now with Lucy Software and