Perceptual evaluation of backchannel strategies for artificial listeners

Artificial listeners are virtual agents that can listen attentively to a human speaker in a dialog. In this paper, we present two experiments where we investigate the perception of rule-based backchannel strategies for artificial listeners. In both, we collect subjective judgements of humans who observe a video of a speaker together with a corresponding animation of an artificial listener. In the first experiment, we evaluate six rule-based strategies that differ in the types of features (e.g. prosody, gaze) they consider. The ratings are given at the level of a speech turn and can be considered a measure for how human-like the generated listening behavior is perceived. In the second experiment, we systematically investigate the effect of the quantity, type and timing of backchannels within the discourse of the speaker. Additionally, we asked human observers to press a button whenever they thought a generated backchannel occurrence was inappropriate. Both experiments together give insights in the factors, both from an observation and generation point-of-view, that influence the perception of backchannel strategies for artificial listeners.