真龍主・星音
曖昧さのない対話理解システムの設計
Abstract:This report discussed the current progress in the author's research in dialog
understanding, and plans for the future. A design for a comprehensive disambiguation-based understanding system is presented. First, a report of the status
of the NP plan inference system is given, along with what it can and cannot do.
The current version of the FS-LF system, which translates feature-structures into
logical forms, is presented. Current research on the automatic extraction of illocutionary force from semantic utterances is detailed. Next, the report presents
preliminary results from new research on scheduling parallel processes. After
this, an analysis of the understanding task shows that disambiguation problems,
followed by prediction and prosody problems, are the most common and significant problems in dialog interpretation. Accordingly, it is most important for
ATR to build an understanding system based on disambiguation and prediction.
..The design of a next-generation understanding system that meets these
requirements is presented. The system will perform disambiguation and
prediction by being based partially on trajectories, causal evidential reasoning,
simulation of intentional agents, and reasoning with uncertain non-deterministic
actions. The report next details the research that has been performed for this
design, in trajectories, evidential reasoning, intention theory, and simulating
intentional agents. Finally, a brief summary of the research on Uncertain Non-
Deterministic Action (UNDA) theory is presented, along with some applications
for ATR. Appendicies contain examples of results.