On Manipulating Nonverbal Interaction Style to Increase Anthropomorphic Computer Character Credibility Andrew J. Cowell Pacific Northwest National Laboratory Data & Knowledge Engineering Richland, WA, 99352 andrew@pnl.gov Kay M. Stanney Industrial Engr. & Management Systems University of Central Florida Orlando, FL, 32816 stanney@mail.ucf.edu ABSTRACT This study examined the effectiveness of enhancing human- agent interaction through the use of nonverbal behaviors. A taxonomy is described, which organizes nonverbal behaviors into functional categories and the manner in which they can be embodied (i.e. through gesture, posture, paralanguage, eye contact and facial expression). Prototype computer characters were created according to guidelines extracted from the taxonomy and their efficacy was empirical evaluated. The results indicate that by including trusting nonverbal behaviors, the perceived credibility of a computer character was enhanced, although addition of trusting bodily nonverbal behavior provided little in addition to trusting facial nonverbal behavior. Perhaps more importantly, a character expressing non-trusting nonverbal behaviors was perceived to be the least credible of all characters examined (including a character that expressed no nonverbal behavior). Participants that interacted with this persona perceived the task to be more demanding, made significantly more errors, and rated their interaction less positively and more monotonous than those using trusting personas. They also rated this character to be less likable, accurate, and intelligent. Taken together, the results from this study suggest that there may indeed be benefit to endowing computer characters with nonverbal trusting behaviors, as long as those behaviors are accurately and appropriately portrayed. Such behaviors may lead to a more trusting environment and positive experience for users. Negative character behavior, however, such as non-trusting behavior, may squander the advantages that embodiment brings. Keywords Embodied Conversational Agents; Nonverbal Behavior; Taxonomy of Nonverbal Behavior; Design Guidelines for Agent Embodiment. 1. INTRODUCTION Computer technology is now a crucial element that holds together the social fabric of our lives. An increasing number of people have to interact with computers each day – smart, intelligent people who enjoy a thrilling, satisfying life but can’t set the clock on their VCR or have problems with ATMs. Many seek a more natural means of communicating with their computers, a manner that makes interaction with a computer as easy as interacting with other people, taking advantage of the multimodal nature of human communication. Such interfaces have been designed and implemented and while users should, in theory, gravitate to such anthropomorphic embodiments, quite the contrary has been experienced; users generally have been dissatisfied, frustrated and eventually abandoned their use, Microsoft Bob and the set of Microsoft Office Assistants being famous examples [1]. This suggests a disconnect between factors that make human-human communication engaging and those used by designers to support human-agent interaction. Some researchers [2][3] believe this disconnect may be due to the fact that little definite research has been conducted concerning social aspects of agency. The study takes an in- depth look at the interaction style of computer characters and how embodiment can alter user’s perception of an agent. A taxonomy of nonverbal behavior and specifically of trusting aspects of nonverbal behavior is developed from which design guidelines can be extracted to enhance the interactional style of anthropomorphic computer characters. 2. NONVERBAL BEHAVIORS Several researchers have looked at human-human interactions and studied what kind of strategy evokes a credible response. One commonly accepted method of depicting a credible appearance in human-human interaction, evocatively supported in the literature, is that of nonverbal behavior [4][5][6][7][8]. While many have attempted to define and characterize nonverbal behavior, Ekman’s approach [9], which looked at functions that nonverbal behavior support, may be the most appropriate to interactive design. More specifically, his work culminated in the definition of distinct functional areas, including emblems (i.e., nonverbal behaviors that have a direct verbal equivalent – a wave meaning ‘goodbye’), illustrators (i.e., nonverbal behaviors tied to speech patterns, portraying something analogically about what is being said - an hourglass gesture to indicate a shapely woman), affect displays (i.e.,