[3Yin2] Interactive session 1

[3Yin2-39] Proposal of Honorific Style Transfer Task for Supporting Japanese Language Learners

Yuta Matsumoto¹, 〇Yu Hayashizaki¹, Kotaro Kitayama¹, Hiroaki Funayama^1,2, Masato Mita², Kentaro Inui^1,2 (1.Tohoku University, 2.RIKEN)

Keywords:Natural Language Processing, Grammatical Error Correction, Honorific Language, Learning Support

Honorific Language is considered important for smooth communication in social life in Japan, but it is difficult for non-native learners of Japanese to use hororific language properly. However, there has been little discussion about the development of a system that automatically converts ordinary language style, and there is no data to evaluate such a system.
In this study, we propose a new task called honorific conversion, which automatically converts arbitrary ordinary sentences into appropriate honorific sentences while preserving the meaning. We also provided a data annotation scheme and an evaluation dataset to establish this task. Furthermore, we show the validity and feasibility of solving the salutation transformation task as a type of style transformation task by investigating the use of sentence decomposition methods.