How Many Utterances Are Needed to Support Time-Offset Interaction?
Ron Artstein, Anton Leuski, Heather Maio, Tomer Mor-Barak, Carla Gordon, David R. Traum · 2015
A set of several hundred recorded statements by a single speaker is sufficient to address unrestricted questions and sus-tain short conversations on a circumscribed topic. Statements were recorded by Pinchas Gutter, a Holocaust survivor, talk-ing about his personal experiences before, during and after the Holocaust. These statements were delivered to partici-pants in conversation, using a “Wizard of Oz ” system, where live operators select an appropriate reaction to each user utter-ance in real time. Even though participants were completely unconstrained in the questions they could ask, the recorded statements were able to directly address at least 58 % of user questions. The unanswered questions were then analyzed to identify gaps, and additional statements were recorded to fill the gaps. The statements will be put in an automated system using existing language understanding technology, to create the first full working system of time-offset interaction, allow-ing a live conversation with a real human who is not present for the conversation in real time.