International Workshop on Spoken Dialogue Systems, 2016
|
Abstract: Backchannels play an important role in smooth dialogue, especially attentive listening. In this paper, we analyze the morphological patterns (category) of backchannels, and how these relate to linguistic features of the preceding utterance. In particular we consider the type of the previous utterance-end boundary, the linguistic complexity of the previous utterance, and other features. Based on this analysis, we conduct machine learning to create a model to predict a backchannel's morphological pattern from the preceding context. This model outperforms a baseline: its output better matches the actual backchannels made by human counselors, and human listeners rate its output as more natural. |