This article reports on the benefits and limitations of using "embodied conversational agents" (ECAs) to create hypothetical social situations (called virtual vignettes) for use in assessing interpersonal competency skills of two diverse samples, i.e., male African-American adolescents (n = 125) and male inmates (n = 226).
ECAs are virtual characters rendered on a monitor or screen with whom a viewer conducts a conversation. An ECA application usually incorporates three components: a language processor, a behavior and planning engine, and a visualizer. The language processor accepts spoken, typed, or selected input from the user and maps this input to an underlying semantic representation. The behavior engine accepts semantic content and other input from the user or system and determines ECA behaviors using cognitive, social, linguistic, physiological, planning, and other models. Behaviors in conversational interactions with the virtual person can include an emotional state; actions performed in the virtual environment; gestures, body movements, or facial expressions; and spoken dialog. The virtual vignettes were different for the two groups based on their environmental influences; however, both studies were interested in testing participants' negotiation and conflict-resolution skills in ECA interactions. The findings from the two studies showed differing tendencies for the two samples. The ECA vignettes were sufficiently realistic to elicit behavioral differences among the adolescents, but this was not generally the case for the inmate sample. Still, most users at least accepted ECAs as valid conversational partners. The findings suggest that the ECA technology or the setting in which it is used is not sufficient to engage all users. Its usefulness for assessing individual behaviors may depend on participant characteristics. 8 tables, 4 figures, and 100 references