[4Xin2] Poster session 2

[4Xin2-35] Dialogue Breakdown Recovery Strategies Based on Relationship with a Dialogue System

Mai Okada¹, 〇Kazuya Tsubokura¹, Yurie Iribe¹, Norihide Kitaoka² (1.Aichi Prefectural University, 2.Toyohashi University of Technology)

Keywords:Dialogue System, Dialogue Breakdown Recovery, Relationship

In current dialogue systems, dialogue breakdowns, such as making statements that are not true, occur. Since such failures cause users to lose their motivation to interact with the system, it is necessary to repair them appropriately. In this study, we propose a method for repairing dialogue breakdowns by focusing on the relationship between the dialogue system and the user. We collect user utterances in response to dialogue breakdowns and the system's repair sentences in response to those utterances and clarify the differences for each relationship. We assigned dialogue act tags to the collected utterances, and examined a failure recovery method based on the tendency of the tags. When the system fails in each relation, repair sentences are generated according to the user's dialogue acts. We evaluated the repaired utterances of the implemented system and found that the proposed method was the best in three out of six items.