Beyond Question Answering
Philip R. Cohen* Bolt, Beranek, and Newman, Inc.
C. Raymond Perrault University of Torronto
James F. Allen University of Rochester
Judging from the number of implemented systems, one might conclude that the predominant application of natural-language processing technology is question answering (QA), usually from a highly structured data base. Recent systems have demonstrated enough robustness and coverage in their chosen subsets of natural language that users can accomplish significant work. Although applauding the impressive results as a benchmark for future systems, we claim that interaction with current question-answering systems lacks naturalness, and that the structure of these systems imposes blinders on the development of other applications of natural-language processing. This chapter both supports these claims and proposes a more general architecture for such systems, viewing question-answering as a special case of natural-language dialogue.
We demonstrate, using protocols of actual interactions with a questionanswering system, that users of these systems expect more than just answers to isolated questions. They expect to engage in a conversation whose coherence is manifested in the interdependence of their often unstated plans and goals with those of the system.1 They also expect the system to be able to incorporate its own responses into analyses of their subsequent utterances. Moreover, they maintain these expectations even in the face of strong evidence that the system is not a competent conversationalist. We propose a program of research designed to____________________