DYNAMICALLY GENERATED FOLLOW-UP QUESTIONS

Automatic text generators are at the heart of systems that provide use rs with information. The trick is getting the system to answer follow- up questions as naturally as possible. But even in moderately complex domains, the task of handcrafting explanations using ''canned'' text o r templates is so time-consuming and error-prone that it becomes infea sible. Furthermore, these techniques cannot be extended to let a syste m consider the user's prior knowledge, past problem-solving experience s, or the preceding dialogue. To overcome these limitations, researche rs have focused on automatically synthesizing text directly from under lying knowledge bases. Automatic text-generation systems pose new oppo rtunities-and new problems. Studies of human-human interactions show t hat people often follow up requests for information with more question s. This observation also underscores the need for computer-based infor mation systems to let users ask follow-up questions. This capability i s especially crucial in patient education, for example, where misunder standings could have serious consequences. The ability to handle follo w-up requests in context is essential, even crucial, to applications l ike the patient education system described in this article. The direct ion we've taken presents one alternative to full-fledged natural langu age-understanding and makes it possible to design systems by adopting a pragmatic (and possibly more useful) approach of generating choices for the user. Our initial system evaluations reveal that users are com fortable with the interface as a way to ask follow-up questions.