Cross-Cultural Perceptions

While some communicative behavior is universal, other aspects are specific to a subgroup, in much the same way that words have meaning only within a specific language. Different cultures also have different ways of assigning meaning to common behaviors within conversation, such as the amount of space between the participants, the use of gaze and head-gestures such as nodding and head-shakes, and the timing and sequencing of turn-taking. The same non-verbal behaviors often have different meanings within different culture groups, which leads to misunderstanding where behaviors meant from one perspective are interpreted using the meaning system of another, e.g., Arabs stand much closer to each other than East Asians. With virtual humans and animated avatars representing human participation in the virtual world, one cannot simply ignore culture-specific aspects of conversational behavior. For any group of conversing virtual humans and/or avatars, they will either be acting appropriately or inappropriately for a given culture.

In this project, we will build on prior collaboration between the ICT and University of Texas at El Paso (UTEP) on culture-specific conversational behaviors and their applications to virtual humans. The prior work established the soundness of the methodology and that members of different culture groups have significantly different perceptions of the appropriateness of conversational behavior differences of groups of virtual humans engaged in conversation. This work also established several gaps in the required knowledge for building accurate culture-specific models of conversational behavior for virtual humans. In this project, we collected audio/visual (AV) recordings of groups from different cultures engaged in different dialogue activities, including unstructured conversation, storytelling, and collaborative task-related dialogue, with and without an interesting artifact that can be passed around. These recordings are used to serve as the basis for new models of task and culture-specific dialogue behavior, and will also be used as the stimulus for perception studies from subjects in different cultures to help identify similarities and differences in judgments of appropriateness and adherence to norms of behavior for different tasks.
We are annotating the behaviors of interest in the recordings, in order to have a data-set available for training (e.g., setting parameters) and testing of behavioral models, as well as data-mining for features in the recordings that can predict the human judgments of different cultures.