Identifying Principals and Accessories in a Complex Case based on the Comprehension of Fact Description

Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 4265–4269 July 5 - 10, 2020. c⃝2020 Association for Computational Linguistics 4265 Identifying Principals and Accessories in a Complex Case based on the Comprehension of Fact Description 1Yakun Hu∗, 2Zhunchen Luo∗, 1Wenhan Chao∗† 1School of Computer Science and Engineering, Beihang University 2Information Research Center of Military PLA Academy of Military Science 1,2Beijing, China 1{huyakun,chaowenhan}@buaa.edu.cn 2zhunchenluo@gmail.com Abstract In this paper, we study the problem of identi- fying the principals and accessories from the fact description with multiple defendants in a criminal case. We treat the fact descriptions as narrative texts and the defendants as roles over the narrative story. We propose to model the defendants with behavioral semantic in- formation and statistical characteristics, then learning the importances of defendants within a learning-to-rank framework. Experimental results on a real-world dataset demonstrate the behavior analysis can effectively model the de- fendants’ impacts in a complex case. 1 Introduction In recent years, much previous work has focused on the building of legal assistant systems with different functions, e.g. searching relevant cases for a given query (Chen et al., 2013), predicting charge labels based on the fact description in a criminal case (Luo et al., 2017; Hu et al., 2018; Zhong et al., 2018), generating the interpretable court views from the fact descriptions (Ye et al., 2018). Though having achieved promising results in this field, most of the work only studies the simple cases with only one defendant. However, there exist lots of com- plex criminal cases in practice, which will involve multiple criminals. In this work, we propose to study the identifi- cation of principals and accessories from the fact description in a criminal case. The principal refers to a criminal who organizes and leads criminal groups to carry out criminal activities or plays a major role in joint crimes. Correspondingly, we re- fer the accessory as the one who plays a secondary or auxiliary role. As the illustration of our task in Fig. 1, given the fact description as well as a list ∗ indicates equal contribution. † Corresponding author. of defendants, we expect to identify the principals and accessories from the defendants. Since the fact descriptions in criminal cases are usually narrative texts which mostly record the criminal events, we treat the defendants in the fact descriptions as the narrative roles and the protago- nists who have the greater impact will be identified as the principles. Narrative comprehension has been studied in NLP for a long time. The tradi- tional method to measure the importances of roles is based on the roles’ dispersion over the story (Karsdorp et al., 2012). It supposes that comparing to less important roles, the roles with bigger im- pact are expected to appear at more places and are more evenly distributed over the story. However, this assumption ignores actions of roles (denoted as behavioral semantic information), which may be a key factor that estimates their impacts in legal- context scenarios. In this paper, we …

Content truncated to 500 words