Kuniaki Uehara, Koji Ogino
Department of Computer and Systems Engineering, Kobe University
Nada, Kobe 657, Japan
e-mail: uehara@jedi.seg.kobe-u.ac.jp
\baselineskip 5.5mm Dialogue is an interactive process between speaker and listener. The listener needs to be able to ask follow-up questions if he does not understand the speaker's utterance, or wants further elaboration. Answers to such questions must take into account the dialogue context. Furthermore, in the process of generating text, the competent speaker takes into consideration his effect his utterance is likely to have on the listener. In other words, the speaker tries to generate an utterance which is best suited to attain his communicative goal with respect to a particular audience. Unfortunately, current dialogue systems cannot participate in an interactive dialogue with users. In particular, these systems cannot clarify misunderstood utterances, elaborate on previous utterances, or respond to follow-up question in the context of the on-going dialogue. In part, the text-planning components of these systems are limited because their mechanisms are quite simple (called classical planning). However, even the more sophisticated generation techniques employed in computational linguistics are inadequate for responding to follow-up questions. The problem is that these systems view generating responses as a one-shot process. That is, a system is assumed to have one opportunity to produce a response that the user will find satisfactory. This approach requires an enormous amount of detailed knowledge about the user, i.e., user model, in order to produce a suitable context for follow-up questions. However, it will be impossible to build such complete and correct user models. In this research, we will propose case-based text planning approach to dialogue systems in which feedback from the user is an integral part of the text planning process. At the most general level, case-based text planning is a cycle of three steps: Anticipating problems (i.e., misunderstanding, confusion and careless mistake) that are likely to arise in achieving a given set of communicative goals, retrieving from memory the case that comes closest to satisfying the given goals while avoiding these problems, and adapting the plan suggested by this case to meet the current circumstances. If the system receives feedback from the user, it will repeat the above steps to produce a response that the user will find satisfactory. Case-based text planning approach has the following advantages: (1) construction of case base is easier than that of plan operators in an abstract discourse structure model, (2) computational cost will be reduced since sub-optimal solutions can be found out without tracing complex inference which had done previously, (3) as discourse situation changes, dialog system can re-plan responses reactively, which allow various follow-up questions from the user, by revising current text plan case, (4) the system may use information in a user model if it exists, but not require it. We are now developing a natural language consultation system for the UNIX operating system, called ASSIST-R, so as to implement the idea of case-based text planning. Users can ask ASSIST-R how to do things in UNIX, get definitions of UNIX terminology, and get help debugging problems in using commands.
Keywords: dialogue system, communicative goal, case-based planning, user model