Humans are very good at detecting subtle affective changes in a person while speaking with them. A good conversational partner is able to not only detect these changes, but also alter their own way of speaking to suit the needs of their partner. This is especially true in one-on-one tutoring, where the attitude of the student affects the amount of learning that occurs. In this study, I used several methods to develop a model, based on a human tutor, which uses the student's actions (in the form of dialog history, prosody, and utterance timing) and feelings to determine an appropriate response choice. I then asked several participants to receive tutoring from a Wizard-of-Oz spoken dialog system that uses my model to generate acknowledgments and a similar system the randomly generates acknowledgments. I found that while there was no significant difference between the two systems in either the amount of learning or user preference, participants tended to prefer the rule-based system.
展开▼
