The interactive scenarios realized in the two prototypes of Virtual Human require an approach that allows humans and virtual characters to interact naturally and flexibly. In this article we present how the autonomous control of the virtual characters and the interpretation of user interactions is realized in the Conversational Dialogue Engine (CDE) framework. For each virtual and real interlocutor one CDE is responsible for dialogue processing. We will introduce the knowledge needed for the CDE-approach and present the modules of a CDE. The real-time requirement resulted in the integrated processing of deliberative and reactive processing, which is needed, e.g., to generate an appropriate nonverbal behavior of virtual characters.