Construction of a Corpus of Japanese Honorifics Based on Social Relations

Abstract

In the supervised learning task of NLP, the corpus is very important. So far, most of the large corpora of foreign languages that take interpersonal relations into account have dealt with politeness, formality, emotion, etc. However, in Japanese, the expression of utterances differs depending on the social status of the speakers, especially in the use of honorific expressions. In Japanese, it is very important to take into account such characteristics in order to correctly deal with the differences in meaning in machine translation and dialogue systems. However, so far there is no corpus that deals with honorific expressions based on the speaker's social status. With respect to this, we have constructed a corpus of honorific expressions that includes information on social status relations (KeiCO corpus) based on the system networks in systemic functional linguistics, which deals with the use of language in cultural societies, and confirmed that our corpus can be used for machine learning tasks by its accuracy in a highly practical classification task.