[2F3-GS-10g] AI応用：テキスト処理

[2F3-GS-10g-02] Japanese-English Comparison of Identification Rates of Closed-Ended Questions

〇Atsuya Tsuda¹, Tokuyasu Kakuta¹ (1. Chuo University)

Keywords:Closed-Ended Question, Text Classification, Intermediate Language

As a real practice of legal education, the authors have been coordinating the education program to simulate civil suits on a web system. The program proceeds through the following process; students write in yes/no questions for fact-finding of a sample case; the teaching staffs playing the role of witnesses respond to them. For efficient instruction, there is a demand for automatic detection of sentences that are not yes-no questions. We hypothesized that automatic classification would be easier in English than in Japanese if questions are the same in content, and tested a method that uses English as an intermediate language. In this research, we built an English corpus using the API of DeepL GmbH and compared the discrimination rates of Japanese and English question sentences by machine learning. The results show that the method using English do not have any advantage in the classification accuracy of the test data.