NITK_LEGAL at SemEval-2023 Task 6: A Hierarchical based system for identification of Rhetorical Roles in legal judgements

Proceedings of the The 17th International Workshop on Semantic Evaluation (SemEval-2023), pages 1154–1160 July 13-14, 2023 ©2023 Association for Computational Linguistics NITK_LEGAL at SemEval-2023 Task 6: A Hierarchical based system for identification of Rhetorical Roles in legal judgements Patchipulusu Sindhu and Diya Gupta and Sanjeevi Meghana and Anand Kumar M Department of Information Technology National Institute of Technology Karnataka, Surathkal {sindhu.191it137, diya.191it215, sanjeevimeghana.191it144, m_anandkumar}@nitk.edu.in Abstract The ability to automatically recognise the rhetorical roles of sentences in a legal case judgement is a crucial challenge to tackle since it can be useful for a number of activities that come later, such as summarising legal judge- ments and doing legal searches. The task is exigent since legal case documents typically lack structure, and their rhetorical roles could be subjective. This paper describes SemEval- 2023 Task 6: LegalEval: Understanding Le- gal Texts, Sub-task A: Rhetorical Roles Pre- diction (RR). We propose a system to auto- matically generate rhetorical roles of all the sentences in a legal case document using Hier- archical Bi-LSTM CRF model and RoBERTa transformer. We also showcase different tech- niques used to manipulate dataset to generate a set of varying embeddings and train the Hierar- chical Bi-LSTM CRF model to achieve better performance. Among all, model trained with the sent2vec embeddings concatenated with the handcrafted features perform better with the mi- cro f1-score of 0.74 on test data. The dataset utilised in our task is available at 1. 1 Introduction In a country with a large population like India, the number of legal cases has been increasing rapidly. People are overloaded by the vast amount of in- formation and documents available online, which is a result of the expansion of the internet and big data (Kemp, 2014). Legal documents are one ex- ample of this type of online information growth in the subject of law. Constitutions, contracts, deeds, orders/judgments/decrees, pleadings, statutes, and wills are some examples of these documents. These documents are very lengthy to read and compre- hend and have a structure that is quite elaborate compared to a regular text. 1https://github.com/Legal-NLP-EkStep/ rhetorical-role-baseline Legal case documents are difficult to process automatically due to their length and unstructured nature If the documents could be divided up into cohesive information pieces, a legal document pro- cessing system would greatly benefit. In a legal case document, Rhetorical role labelling of a sen- tence means identifying the semantic function a sentence of a legal document serves, such as the case’s facts, the parties’ arguments, ruling from the current court, etc. Semantic search, summariza- tion, case law analysis, and other downstream tasks can all benefit from knowing the rhetorical roles that sentences play in a legal case document. How- ever, legal documents lack a clear organizational structure, making it challenging to comprehend the nuanced differences between rhetorical roles. Rhetorical roles(RRs) aid in the extraction of le- gal information, such as cases with similar facts. Comparing various rhetorical role units can also be used to find previous cases that are comparable to the one in …

Content truncated to 500 words