Learning to Predict Charges for Criminal Cases with Legal Basis

Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing, pages 2727–2736 Copenhagen, Denmark, September 7–11, 2017. c⃝2017 Association for Computational Linguistics Learning to Predict Charges for Criminal Cases with Legal Basis Bingfeng Luo1, Yansong Feng∗1, Jianbo Xu2, Xiang Zhang2 and Dongyan Zhao1 1Institute of Computer Science and Technology, Peking University, China 2Beijing Institute of Big Data Research, China {bf luo, fengyansong, xujb, xiang.zhang, zhaody}@pku.edu.cn Abstract The charge prediction task is to deter- mine appropriate charges for a given case, which is helpful for legal assistant systems where the user input is fact description. We argue that relevant law articles play an important role in this task, and therefore propose an attention-based neural network method to jointly model the charge pre- diction task and the relevant article extrac- tion task in a unified framework. The ex- perimental results show that, besides pro- viding legal basis, the relevant articles can also clearly improve the charge prediction results, and our full model can effectively predict appropriate charges for cases with different expression styles. 1 Introduction The task of automatic charge prediction is to deter- mine appropriate charges, such as fraud, larceny or homicide, for a case by analyzing its textual fact description. Such techniques are crucial for legal assistant systems, where users could find similar cases or possible penalties by describing a case with their own words. This is helpful for non-legal professionals to get to know the legal basis of their interested cases, e.g., cases they or their friends are involved in, since the massive legal materials and the lack of knowledge of legal jargons make it hard for outsiders to do it on their own. However, predicting appropriate charges based on fact descriptions is not trivial: (1) The differ- ences between two charges can be subtle, for ex- ample, in the context of criminal cases in China, distinguishing intentional homicide from inten- tional injury would require to determine, from the fact description, whether the defendant intended to kill the victim, or just intended to hurt the vic- tim, who, unfortunately died of severe injury. (2) Multiple crimes may be involved in a single case, which means we need to conduct charge predic- tion in the multi-label classification paradigm. (3) Although we can expect an off-the-shelf classifi- cation model to learn to label a case with corre- sponding charges through massive training data, it is always more convincing to make the prediction with its involved law articles explicitly shown to the users, as legal basis to support the prediction. This issue is crucial in countries using the civil law system, e.g., China (except Hong Kong), where judgements are made based on statutory laws only. For example, in Fig. 1, a judgement document in China always includes relevant law articles (in the court view part) to support the decision. Even in countries using the common law system, e.g., the United States (except Louisiana), where the judgement is based mainly on decisions of previ- ous cases, there are still some statutory …

Content truncated to 500 words