Research in the domain of affective computing is usually concentrated on detection of emotional user behavior. However, less attention is devoted to the question how to enable dialogue systems to overcome problems in the interaction related to emotional user behavior. This talk addresses the latter research question. We present both theoretical considerations and implementation issues of adaptive dialogue management in the NIMITEK prototype spoken dialogue system for supporting users while they solve problems in a graphics system.

The central idea is that the system dynamically refines a dialogue strategy according to the current state of the interaction. We model the state of the interaction as a composite of five interaction features: the state of the task, the user’s command, the focus of attention, the state of the user, and the history of interaction. Dynamical adaption includes three distinct but interrelated decision making processes: When to provide support to the user? What kind of support to provide? How to provide support? These reflect three underlying requirements for a dialogue strategy aimed to support the user. First, support should be timely provided, without relying on the assumption that the user will clearly state a need for support. Second, problems in the interaction may be various (e.g., they may relate to the given task, to the interface language, to the emotional state of the user, etc.) and the user should be provided with useful, sufficient and appropriately emphasized information tailored to a particular problem. And third, the manner of providing support should be tailored to meet the user’s needs, i.e., it should be in accordance with the emotional state of the user.

Finally, we analyze a recorded dialogue between the user and the prototype system that took place during the testing of the system. It illustrates several points of the implemented dialogue strategy: processing of user’s commands, supporting the user, and multilingual working mode.