uOttawa at SemEval-2023 Task 6: Deep Learning for Legal Text Understanding

Proceedings of the The 17th International Workshop on Semantic Evaluation (SemEval-2023), pages 580–588 July 13-14, 2023 ©2023 Association for Computational Linguistics uOttawa at SemEval-2023 Task 6: Deep Learning for Legal Text Understanding Intisar Almuslim and Sean Stilwell and Surya Kiran Suresh and Diana Inkpen School of Electrical Engineering and Computer Science University of Ottawa Ottawa, ON, K1N 6N5 {iamul068, sstil051, ssrure044, diana.inkpen}@uottawa.ca Abstract We describe the methods we used for legal text understanding, specifically Task 6 Legal-Eval at SemEval 2023. The outcomes could assist law practitioners and help automate the work- ing process of judicial systems. The shared task defined three main sub-tasks: sub-task A, Rhetorical Roles Prediction (RR); sub-task B, Legal Named Entities Extraction (L-NER); and sub-task C, Court Judgement Prediction with Explanation (CJPE). Our team addressed all three sub-tasks by exploring various Deep Learning (DL) based models. Overall, our team’s approaches achieved promising results on all three sub-tasks, demonstrating the po- tential of deep learning-based models in the judicial domain. 1 Introduction The legal sector generates an overwhelming vol- ume of information from many sources, including law firms, law courts, independent attorneys, and legislators. As a result, tools to help legal profes- sionals manage large volumes of legal documents are becoming increasingly necessary. The LegalEval 2023 shared task (Modi et al., 2023)1 aims to build a legal research community by addressing three specific problems in the legal domain. First, Sub-task A, Rhetorical Roles Predic- tion (RR), involves identifying the rhetorical roles played by various sentences in a given text. Sub- task B, Legal Named Entities Extraction (L-NER), involves identifying distinct named entities that appear in legal texts. Finally, Sub-task C, Court Judgement Prediction with Explanation (CJPE), in- volves predicting the outcome of a given court case, along with an explanation of the reasoning behind the prediction. Our team participated in all three sub-tasks of the shared task. In Sub-task A, we created a hierarchi- cal BiLSTM-CRF model with randomly-initiated 1https://sites.google.com/view/legaleval/home embeddings. In Sub-task B, we created a BERT- based named entity recognition model. In Sub-Task C, we used CaseLawBERT to create judgement pre- diction and explanation models. The rest of this paper is organized as follows: in Section 2, we briefly review the literature back- ground on each sub-task. The detailed descriptions of our approaches to the three sub-tasks are pre- sented in Sections 3, 4, and 5, respectively. Section 6 concludes the paper and comments on future work directions. 2 Related Work Rhetorical role labeling was first automated by (Sar- avanan et al., 2008), where Conditional Random Fields (CRF) developed a generic approach to per- form segmentation using seven rhetorical roles. Ne- jadgholi et al. (2017) developed a method for the identification of factual and nonfactual sentences using fastText. Later, the automatic Machine Learn- ing approaches and rule-based scripts for rhetorical role identification evolved which was compared to (Walker et al., 2019). Kalamkar et al. (2022b) cre- ated a large corpus of RRs and propose transformer- based baseline models for RR prediction. The use of the Bi-LSTM-CRF …

Content truncated to 500 words