FIT BUT at SemEval-2023 Task 12: Sentiment Without Borders - Multilingual Domain Adaptation for Low-Resource Sentiment Classification

Proceedings of the The 17th International Workshop on Semantic Evaluation (SemEval-2023), pages 143–149 July 13-14, 2023 ©2023 Association for Computational Linguistics LRL_NC at SemEval-2023 Task 6: Sequential Sentence Classification for Legal Documents using Topic Modeling Features Kushagri Tandon, Niladri Chatterjee Department of Mathematics Indian Institute of Technology Delhi Hauz Khas, Delhi-110016, India {kushagri.tandon,niladri.chatterjee}@maths.iitd.ac.in Abstract Natural Language Processing techniques can be leveraged to process legal proceedings for various downstream applications, such as sum- marization of a given judgement, prediction of the judgement for a given legal case, prece- dent search, among others. These applications will benefit from legal judgement documents already segmented into topically coherent units. The current task, namely, Rhetorical Role Pre- diction, aims at categorising each sentence in the sequence of sentences in a judgement document into different labels. The system proposed in this work combines topic mod- eling and RoBERTa to encode sentences in each document. A BiLSTM layer has been utilised to get contextualised sentence repre- sentations. The Rhetorical Role predictions for each sentence in each document are gen- erated by a final CRF layer of the proposed neuro-computing system. This system secured the rank 12 in the official task ranking, achiev- ing the micro-F1 score 0.7980. The code for the proposed systems has been made available at https://github.com/KushagriT/SemEval23_ LegalEval_TeamLRL_NC 1 Introduction Legal domain hosts a wide range of text documents corresponding to different stages of the legal pro- ceedings. Especially in populous countries, such as India, which has a huge and branched judicial system, automation of certain steps in the judicial pipeline can be beneficial to the overall working of the system. Legal documents are generally long and unstructured, and typically consist of jargons specific to the legal domain. Due to presence of various subdomains, such as criminal law, civil law, income-tax law, among others, the systems devel- oped on one domain may not generalise well to other domains. Additionally, the legal domain suf- fers from lack of availability of annotated corpora (Kalamkar et al., 2022). All the above observations make the task truly challenging. Natural Language Processing techniques can be leveraged to process such corpora for various downstream applications, such as judgement sum- marising, judgement outcome prediction, prece- dent search, among others. These applications will benefit from legal judgement documents that are already segmented into topically coherent units. These coherent units are called Rhetorical Roles. This task of semantic segmentation of a judgement document, into Rhetorical Roles, aids in the overall processing of legal documents. The aim of the current task (Modi et al., 2023) is to categorise sentences in a legal case proceeding (judgement document), to different labels. Thus the task effectively boils down to Rhetorical Role Prediction. The dataset (Kalamkar et al., 2022; Malik et al., 2021a) consists of 13 Rhetoric Role labels given in Table 1. This paper proposes a system for Rhetorical Role prediction for legal judgement documents. The proposed approach uses a hierarchical sequence encoder consisting of basic sentence representation and a contextualised sentence representation fol- lowed by a CRF …

Content truncated to 500 words