This paper reports on judgement studies regarding the perception of interpersonal stances taken by humans playing the role of a suspect in a police interrogation setting. Our project aims at building believable embodied conversational characters to play the role of suspects in a serious game for learning interrogation strategies. The main question we ask is: do human judges agree on the way they perceive the various aspects of stance taking, such as friendliness and dominance? Four types of stances were acted by eight amateur actors. Short recordings were shown in an online survey to subjects who were asked to describe them using a selection of a number of adjectives. Results of this annotation task are reported in this paper. We explain how we computed the inter-rater agreement with Krippendorff’s alpha statistics using a set theoretical distance metric. Results show that for some of the stance types observers agreed more than for others. Some actors are better than others, but validity (recognizing the intended stance) and inter-rater agreement do not always go hand in hand. We further investigate the effect the expertise of actors has on the perception of the stance that is acted. We compare the fragments from amateur actors to fragments from professional actors taken from popular TV-shows.
- HMI-MI: MULTIMODAL INTERACTIONS
- Body language
- Affect expression
- Interpersonal stance
- Embodied Conversational Agents
- Data reliability