Leveraging Task Dependency and Contrastive Learning for Case Outcome Classification on European Court of Human Rights Cases

Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics, pages 1103–1111 May 2-6, 2023 ©2023 Association for Computational Linguistics Leveraging Task Dependency and Contrastive Learning for Case Outcome Classification on European Court of Human Rights Cases Santosh T.Y.S.S, Marcel Perez San Blas, Phillip Kemper, Matthias Grabmair School of Computation, Information, and Technology; Technical University of Munich, Germany {santosh.tokala, marcel.perez, phillip.kemper, matthias.grabmair}@tum.de Abstract We report on an experiment in case outcome classification on European Court of Human Rights cases where our model first learns to identify the convention articles allegedly vio- lated by the state from case facts descriptions, and subsequently uses that information to clas- sify whether the court finds a violation of those articles. We assess the dependency between these two tasks at the feature and outcome level. Furthermore, we leverage a hierarchical con- trastive loss to pull together article-specific rep- resentations of cases at the higher level, leading to distinctive article clusters. The cases in each article cluster are further pulled closer based on their outcome, leading to sub-clusters of cases with similar outcomes. Our experiment results demonstrate that, given a static pre-trained en- coder, our models produce a small but consis- tent improvement in classification performance over single-task and joint models without con- trastive loss. 1 Introduction The NLP task of classifying case outcome infor- mation from a textual statement of case facts is generally referred to as Legal Judgement Predic- tion (LJP)(e.g., Aletras et al. 2016; Chalkidis et al. 2019). It has been studied using corpora from dif- ferent jurisdictions, such as the European Court of Human Rights (ECtHR) (Chalkidis et al., 2019, 2021, 2022b; Aletras et al., 2016; Medvedeva et al., 2020; Santosh et al., 2022, 2023), Chinese Criminal Courts (Luo et al., 2017; Zhong et al., 2018; Yue et al., 2021; Zhong et al., 2020; Yang et al., 2019), US Supreme Court (Katz et al., 2017; Kaufman et al., 2019), Indian Supreme Court (Malik et al., 2021; Shaikh et al., 2020), the French court of Cas- sation (¸Sulea et al., 2017), Brazilian courts (Berta- lan and Ruiz, 2020), the Federal Supreme Court of Switzerland (Niklaus et al., 2021), the Turkish Constitutional court (Sert et al., 2021; Mumcuo˘glu et al., 2021), UK courts (Strickson and De La Igle- sia, 2020), German courts (Waltl et al., 2017), the Philippine Supreme court (Virtucio et al., 2018), and the Thailand Supreme Court (Kowsrihawat et al., 2018). In this work, we experiment with classifying case outcomes in the ECtHR A and B benchmark tasks introduced by LexGLUE (Chalkidis et al., 2022b). Task B is to identify the set of articles of the European Convention of Human Rights (ECHR) that the claimant alleges to have been vi- olated, while Task A’s goal is to classify which of the convention’s articles has been deemed violated by the court. The input for both tasks is the case’s fact description that has been extracted from the published judgement document. It should be noted that, despite these tasks …

Content truncated to 500 words