[3K1-OS-2a] OS-2

[3K1-OS-2a-02] ELSI discussions in building the Japanese Tort-case Dataset

〇Hiroaki Yamada¹, Ryutaro Ohara^2,3, Mihoko Sumida³ (1. Tokyo Institute of Technology, 2. Nakamura, Tsunoda and Matsumoto, 3. Hitotsubashi University)

Keywords:Dataset, Legal Judgment Prediction, Natural Language Processing

We constructed the Japanese Tort-case Dataset (JTD) for legal judgment prediction, an important area of AI application research in the legal field. JTD consists of 3,477 civil case judgments annotated by 41 legal experts. There are 7,978 tort-related instances, with 59,697 of their alleged arguments from the involved parties. Since JTD contains legal data, various social issues need to be considered in constructing the dataset, including the appropriateness of computers making judicial predictions and potential biases in the constructed dataset. Moreover, the potential initiative of open data provision of court judgments is still under deliberation in Japan. There needs to be more foundation to share datasets like our JTD, which could relate to the interests of many stakeholders with other researchers. Therefore, we report on the issues considered during our dataset construction process to deepen the discussion on dataset construction and sharing.