Speech-Plans: Generating Evaluative Responses in Spoken Dialogue

Recent work on evaluation of spoken dialogue systems indicates that better algorithms are needed for the presentation of complex information in speech. Current dialogue systems often rely on presenting sets of options and their attributes sequentially. This places a large memory burden on users, who have to remember complex trade-offs between multiple options and their attributes. To address these problems we build on previous work using multiattribute decision theory to devise speech-planning algorithms that present usertailored summaries, comparisons and recommendations that allow users to focus on critical differences between options and their attributes. We discuss the differences between speech and text planning that result from the particular demands of the speech situation.