A Regression-based Approach to Modeling Addressee Backchannels

During conversations, addressees produce conversational acts—verbal and nonverbal backchannels—that facilitate turn-taking, acknowledge speakership, and communicate common ground without disrupting the speaker’s speech. These acts play a key role in achieving fluent conversations. Therefore, gaining a deeper understanding of how these acts interact with speaker behaviors in shaping conversations might offer key insights into the design of technologies such as computer-mediated communication systems and embodied conversational agents. In this paper, we explore how a regression-based approach might offer such insights into modeling predictive relationships between speaker behaviors and addressee backchannels in a storytelling scenario. Our results reveal speaker eye contact as a significant predictor of verbal, nonverbal, and bimodal backchannels and utterance boundaries as predictors of nonverbal and bimodal backchannels.