Everything Has a Cause: Leveraging Causal Inference in Legal Text Analysis

Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 1928–1941 June 6–11, 2021. ©2021 Association for Computational Linguistics 1928 Everything Has a Cause: Leveraging Causal Inference in Legal Text Analysis Xiao Liu1∗, Da Yin2∗, Yansong Feng1,3†, Yuting Wu1 and Dongyan Zhao1,3 1Wangxuan Institute of Computer Technology, Peking University, China 2Computer Science Department, University of California, Los Angeles 3The MOE Key Laboratory of Computational Linguistics, Peking University, China {lxlisa,fengyansong,wyting,zhaody}@pku.edu.cn da.yin@cs.ucla.edu Abstract Causal inference is the process of captur- ing cause-effect relationship among variables. Most existing works focus on dealing with structured data, while mining causal relation- ship among factors from unstructured data, like text, has been less examined, but is of great importance, especially in the legal do- main. In this paper, we propose a novel Graph-based Causal Inference (GCI) frame- work, which builds causal graphs from fact de- scriptions without much human involvement and enables causal inference to facilitate le- gal practitioners to make proper decisions. We evaluate the framework on a challenging simi- lar charge disambiguation task. Experimental results show that GCI can capture the nuance from fact descriptions among multiple confus- ing charges and provide explainable discrimi- nation, especially in few-shot settings. We also observe that the causal knowledge contained in GCI can be effectively injected into powerful neural networks for better performance and in- terpretability. Code and data are available at https://github.com/xxxiaol/GCI/. 1 Introduction Causal inference is the process of exploring how changes on variable T affect another variable Y . Here we call T and Y as treatment and outcome, respectively, and the changes on T are called in- tervention. In other words, the process of drawing a conclusion about whether and how Y changes when intervening on T is called causal inference. Most research in causal inference is devoted to analyzing structured data. Take the research ques- tion how smoking causes lung cancer (Pearl and Mackenzie, 2018) as an example. Smoking, lung cancer, together with distractors like age are ex- tracted from structured data, like electronic health records, and considered as factors. Usually, such * Equal contribution. † Corresponding author. Defendant A rented a car from B on April 22, 2015, and forged an agreement to buy the car from B. On the 26th, A lied that he purchased the car, and used it as a guarantee to borrow money from victim C, defrauding C of 19,000 yuan. Forge Defraud Fraud Credit Card Lie Trust �� Money Figure 1: An example of generated causal graph for the charge fraud. Colored words are matched between the exemplified fact description and the graph. studies properly organize those factors into human- designed structures, e.g., a causal directed acyclic graph (Wright, 1921) with factors {smoking, age, lung cancer} as nodes and causal relations {smok- ing → lung cancer, age → lung cancer} as edges, and perform inference on such structures. Recent works attempt to integrate text infor- mation into causal inference (Egami et al., 2018; Veitch et al., …

Content truncated to 500 words