We present an approach to incorporate interesting and compelling characters in planning-based narrative generation. The approach is based on a computational model that utilizes character actions to portray these as having distinct and well-defined personalities.

As work situations become more complex, virtual reality has proven to be advantageous for the emerging needs in risk-management training. However, building perfectly realistic simulations of the technical systems is not enough to address complex coactivity situations, where human factors play an important part. There is a need for virtual environments that would put trainees into varied ecological situations, inducing knowledge and competencies that would be put into practice in genuine work situations. The V3S project was proposed to build a generic framework for tailor-made virtual environments that can adapt to different application cases, technological configurations, or pedagogical strategies. This framework relies on the integration of multiple explicit models (domain, activity, and risk model). In order to build ecologically valid virtual environments, these models represent not only the prescribed activity, but the situated knowledge of operators about their tasks, including deviations from the procedures. These models are used both to monitor learners' actions, detecting errors and compromises; and also to generate virtual characters' behaviors, subject to erroneous actions. Moreover, dynamic situated feedback allows for progressive learning scenarios, adapting the complexity of the situations to the learner's activity and level. Evaluations have shown a high satisfaction level and encouraging usability measures. In future work, we propose to extend the possibilities of the simulation through the creation and monitoring of adaptive scenarios, adjusting the behavior of virtual characters able to assist or disrupt the user.1

No abstract

This paper presents a tripartite model of narrative in which three different kinds of actions are modeled: story actions, discourse actions, and narration actions. We argue that this model can support particularities of narrative discourse where typical discourse models do not (e.g., ambiguity/change of beliefs), and accounts for the difference in expressivity of the media used to convey the narrative, as opposed to bipartite models.

No abstract