Multi-task Legal Judgement Prediction Combining a Subtask of Seriousness of Charge

Multi-task Legal Judgement Prediction Combining a Subtask of the Seriousness of Charges Zhuopeng Xu, Xia Li*, Yinlin Li, Zihan Wang, Yujie Fanxu and Xiaoyan Lai Guangzhou Key Laboratory of Multilingual Intelligent Processing School of Information Science and Technology Guangdong University of Foreign Studies,Guangzhou,China zhuopengxu@126.com,ksyzformy@126.com,zihanwang0703@126.com, yujiefanxu@126.com,avaxiaoyan@126.com xiali@gdufs.edu.cn Abstract Legal Judgement Prediction has attracted more and more attention in recent years. One of the challenges is how to design a model with better interpretable prediction results. Previous stud- ies have proposed different interpretable models based on the generation of court views and the extraction of charge keywords. Different from previous work, we propose a multi-task legal judgement prediction model which combines a subtask of the seriousness of charges. By in- troducing this subtask, our model can capture the attention weights of different terms of penalty corresponding to the charges and give more attention to the correct terms of penalty in the fact de- scriptions. Meanwhile, our model also incorporates the position of defendant making it capable of giving attention to the contextual information of the defendant. We carry several experiments on the public CAIL2018 dataset. Experimental results show that our model achieves better or comparable performance on three subtasks compared with the baseline models. Moreover, we also analyze the interpretable contribution of our model. 1 Introduction Legal Judgement Prediction (LJP) aims to predict charge, law article and terms of penalty automatically based on the fact descriptions of the criminal cases. It can be used to help the court’s judgement and provide legal guidance and assistance to the public. In recent years, different methods have been proposed to improve the performance of legal judgement prediction task. Some previous studies need to design features manually (Katz et al., 2014; Lin et al., 2012; Liu and Hsieh, 2006; Liu et al., 2015) and some of neural network based models extract features automatically and achieve significant improvements (Liu et al., 2019; Ye et al., 2018; Zhong et al., 2018). However, there are still some challenging problems, including the improvement of the performance and the enhancement of the interpretability of the terms of penalty prediction. For the improvement of the performance in terms of penalty prediction, previous studies use multi- task and joint learning to obtain the sharing information among different subtasks. Zhong et al. (2018) propose a Directed Acyclic Graph structure with topological relations to capture the information attribu- tion among three subtasks, which effectively improve the problem of insufficient fine-grained in LJP. For the enhancement of the interpretability, different solutions are proposed to the problem. Ye et al. (2018) propose a Seq2Seq model to formulate legal judgement prediction task as a natural language generation problem. Their model take fact descriptions and charge labels as input and outputs the court’s view. The outputs are used as an auxiliary information for practical judgement. Liu et al. (2019) propose a multi- task learning model to incorporate charge keywords extracted by TF-IDF and TextRank. Their model has a good interpretability by introduced the keyword information. Although different methods are …

Content truncated to 500 words