Towards Interactivity and Interpretability: A Rationale-based Legal Judgment Prediction Framework

Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 4787 - 4799 December 7-11, 2022 ©2022 Association for Computational Linguistics Towards Interactivity and Interpretability: A Rationale-based Legal Judgment Prediction Framework Yiquan Wu1, Yifei Liu1, Weiming Lu1∗, Yating Zhang2∗ Jun Feng3, Changlong Sun12, Fei Wu1, Kun Kuang1∗ 1Zhejiang University, Hangzhou, China 2Alibaba Group, Hangzhou, China 3State Grid Zhejiang Electric Power Co., LTD, Hangzhou, China {wuyiquan, liuyifei, luwm, kunkuang}@zju.edu.cn, yatingz89@gmail.com changlong.scl@taobao.com, JuneFeng.81@gmail.com, wufei@cs.zju.edu.cn Abstract Legal judgment prediction (LJP) is a funda- mental task in legal AI, which aims to assist the judge to hear the case and determine the judg- ment. The legal judgment usually consists of the law article, charge, and term of penalty. In the real trial scenario, the judge usually makes the decision step-by-step: first concludes the rationale according to the case’s facts and then determines the judgment. Recently, many mod- els have been proposed and made tremendous progress in LJP, but most of them adopt an end- to-end manner that cannot be manually inter- vened by the judge for practical use. Moreover, existing models lack interpretability due to the neglect of rationale in the prediction process. Following the judge’s real trial logic, in this pa- per, we propose a novel Rationale-based Legal Judgment Prediction (RLJP) framework. In the RLJP framework, the LJP process is split into two steps. In the first phase, the model gener- ates the rationales according to the fact descrip- tion. Then it predicts the judgment based on the fact and the generated rationales. Extensive experiments on a real-world dataset show RLJP achieves the best results compared to the state- of-the-art models. Meanwhile, the proposed framework provides good interactivity and in- terpretability which enables practical use. 1 Introduction Deep learning methods, especially the natural lan- guage process (NLP) techniques, have been em- ployed to benefit the legal assistant systems in several aspects, such as controversy focus mining (Duan et al., 2019) and court hearing conversation generation (Ji et al., 2020). As one of the most important components of the judicial process, the task of legal judgment prediction (LJP) has been studied for decades (Kort, 1957; Keown, 1980; Lin et al., 2012; Chalkidis et al., 2019; Ma et al., 2021). LJP is defined to predict the judgment results (e.g., *Corresponding Authors. law article, charge, and term of penalty) of legal cases according to the fact descriptions. In the real trial scenario, as Fig. 1 shows, the judge makes the decision step-by-step: first concludes the rationale according to the case’s facts and then determines the judgment. However, most existing methods focus on designing efficient features following an end-to-end framework (Luo et al., 2017; Zhong et al., 2018) while neglecting the necessity for a system to be interactive and interpretable in real trial scenarios. To improve the practicality and effectiveness of LJP methods in real world applications, we face the following challenges: 1) The prediction process cannot be manually intervened: Legal tasks usu- ally require high accuracy, but current LJP methods are commonly designed …

Content truncated to 500 words