ワークショップ (国内) シチュエーションに合わせたシナリオ誘導とHyperCLOVAを利用した応答生成によるハイブリッド対話システム

吉川 克正, 川本 稔己 (LINE/東京工業大学), 山崎 天, 水本 智也, 小林 滉河, 大萩 雅也, 佐藤 敏紀

第13回 対話システムシンポジウム (第13回 対話システムシンポジウム)


This paper describes our dialogue system submitted to the Situation Track of the Dialogue System Live Competition 5. For this competition, we build a hybrid system that combines a traditional rule-based system with a modern generation-based system. The former is composed of simple extit{if-then} rules according to our pre-build scenario. The latter generates responses by prompt programming approach with HyperCLOVA, a language model based on Transformer. Moreover, our system has several filters to remove the response which does not match the situation or could be ethically incorrect. As a result, our system placed second in the preliminary round. We confirmed that our hybrid approach is useful for the multimodal task of situation-based dialogue generation. On the other hand, robustness and real-time performance in spoken dialogue remain issues. In this paper, we discuss the issues by referring to dialogue logs.

Paper : シチュエーションに合わせたシナリオ誘導とHyperCLOVAを利用した応答生成によるハイブリッド対話システム新しいタブまたはウィンドウで開く (外部サイト)