Planning Text for Advisory Dialogues: Capturing Intentional and Rhetorical Information

To participate in a dialogue a system must be capable of reasoning about its own previous utterances. Follow-up questions must be interpreted in the context of the ongoing conversation, and the system's previous contributions form part of this context. Furthermore, if a system is to be able to clarify misunderstood explanations or to elaborate on prior explanations, it must understand what it has conveyed in prior explanations. Previous approaches to generating multisentential texts have relied solely on rhetorical structuring techniques. In this paper, we argue that, to handle explanation dialogues successfully, a discourse model must include information about the intended effect of individual parts of the text on the hearer, as well as how the parts relate to one another rhetorically. We present a text planner that records this information and show how the resulting structure is used to respond appropriately to a follow-up question.