The proposed research aims to intestigate models of dialogue that can support robust and flexible interaction between people and computer systems. When computers are used to teach users about a new topic or technique, users may have difficulty understanding the knowledge and reasoning behind the system's answers. To address this difficulty, this research will specify a computational theory and methods to allow systems to vary the presentation of material dynamically. Specifically, this project will address two important research issues. First, it will provide a specification and representation of the linguistic, intentional, and social information that influence people's understanding of an ongoing dialog and that allow for the detection of misunderstanding, if it should arise. Second, the research will provide methods for integrating and employing these information sources to produce relevant utterances and to identify and resolve communication problems as they arise. Evaluation of the approach will employ two tutoring systems that are under development: one that explains medical concepts and one that teaches general life skills, such as good nutrition and budgeting. Final evaluation of the proposed research will be based on testing with end-user populations, runtime performance, and technical correctness.