BiLSTM-Based Indian Legal Case Summarization and Q&A: A Baseline Implementation with Transformer Extension Proposal

In this paper, we propose a BiLSTM architecture for automated processing of legal documents in the Indian legal framework. Based upon the existing work by Nigam et al. [1] for legal question answering in Indian laws, we established a BiLSTM architecture for a complete legal question answering system with an encoder-decoder framework and an attention mechanism capable of processing Indian legal documents in PDF and DOCX files for summarization and question answering tasks. Our proposed architecture attained a score of 0.78 with a ROUGE-L score, an F1-Score of 75.7%, and a processing capacity of 702 legal documents per hour for 150 Indian legal documents. The proposed system is viable in practical applications as a basic system but lacks in exercises involving abstract reasoning andnz is only moderately adept at extracting answers from legal documents, indicating a need for a more sophisticated architecture. Proposing a potential future improvement with a Transformer architecture is warranted, as recent studies show a major improvement with this architecture.