Turn-taking cue delays in human-robot communication

Fluent communication between a human and a robot relies on the use of effective turn-taking cues. In human speech staying silent after a sequence of utterances is usually accompanied by an explicit turnyielding cue to signal the end of a turn. Here we study the effect of the timing of four turn-yielding cues relative to staying silent. In addition, we examine the relative strength of showing turn-yielding cues individually or in combination. We found that strong, meaningful cues perform considerably better but only if they precede the stay silent cue. The response times to synchronized cues match the stay silent control condition and delayed cues have longer response times. Cue combinations show a winner-take-all effect, as the strongest cue determines the performance of the set. In conclusion, effective human-robot conversation is obtained when the turn-taking cues are clear, meaningful and precede the stay silent cue.