NITS_Legal at SemEval-2023 Task 6: Rhetorical Roles Prediction of Indian Legal Documents via Sentence Sequence Labeling Approach

Proceedings of the The 17th International Workshop on Semantic Evaluation (SemEval-2023), pages 751–757 July 13-14, 2023 ©2023 Association for Computational Linguistics NITS_Legal at SemEval-2023 Task 6: Rhetorical Roles Prediction of Indian Legal Documents via Sentence Sequence Labeling Approach Deepali Jain, Malaya Dutta Borah and Anupam Biswas Department of Computer Science and Engineering National Institute of Technology Silchar, India jaindeepali010@gmail.com,{malayaduttaborah,anupam}@cse.nits.ac.in Abstract Legal documents are notorious for their com- plexity and domain-specific language, making them challenging for legal practitioners as well as non-experts to comprehend. To address this issue, the LegalEval 2023 track proposed sev- eral shared tasks, including the task of Rhetori- cal Roles Prediction (Task A). We participated as NITS_Legal team in Task A and conducted exploratory experiments to improve our under- standing of the task. Our results suggest that se- quence context is crucial in performing rhetori- cal roles prediction. Given the lengthy nature of legal documents, we propose a BiLSTM- based sentence sequence labeling approach that uses a local context-incorporated dataset cre- ated from the original dataset. To better repre- sent the sentences during training, we extract le- gal domain-specific sentence embeddings from a Legal BERT model. Our experimental find- ings emphasize the importance of considering local context instead of treating each sentence independently to achieve better performance in this task. Our approach has the potential to improve the accessibility and usability of legal documents. 1 Introduction Legal case documents are typically quite lengthy, often spanning many pages, which can make it time-consuming for legal practitioners and aca- demics to read them in their entirety. In many cases, these professionals may only need to access specific portions of a document, such as the facts of the case or the arguments put forward by the parties involved (Jain et al., 2021d). However, legal case documents are often unstructured and lack clear section headings, unlike research papers or books. This can make it difficult for readers to quickly and efficiently locate the information they need. The lack of structure in legal case documents can be particularly problematic for legal practition- ers who are trying to build a strong legal argument. Without the ability to easily navigate and access relevant information, lawyers may struggle to build a coherent case that is based on sound legal princi- ples and precedents. To address this challenge, researchers and prac- titioners are exploring a range of approaches to structuring legal case documents and making them more accessible to readers. This includes the de- velopment of new tools1 and technologies that can automatically extract key information from legal documents, such as the parties involved, the le- gal issues at stake, and the arguments put forward by each side (Farzindar and Lapalme, 2004; Pol- sley et al., 2016). By leveraging such tools, legal practitioners and academics can more quickly and efficiently access the information they need to build strong legal arguments and advance the field of law. Rhetorical role labeling of sentences is a tech- nique that can help legal practitioners quickly com- prehend the structure and …

Content truncated to 500 words