Abstract
An important activity in requirements engineering is validation, which is the process of checking whether a model correctly represents a piece of reality and the users' requirements. One technique for supporting validation is explanation generation, which combines paraphrasing of a specification with questionanswer facilities that interactively support a user in exploring a model. In this paper, we propose an architecture and design principles for constructing explanation generation systems for conceptual models. The architecture is partly based on Toulmin's argumentation model, which provides a framework for structuring arguments and Rhetorical Structure Theory. We argue that Toulmin's argumentation model needs support from a discourse theory like RST to create customized explanations to the user. To each type of explanation there exists a set of appropriate RST relations to be used.