Indian Legal Judgment Summarization using LEGAL-BERT and BiLSTM model with Adaptive Length

The Indian legal system is vast and complex, rapid expansion of legal documentation has created a pressing need for reliable and efficient summarization tools to support legal professionals and researchers. To help reduce the cost and time spent on reading and retrieving critical information from the legal judgment, we introduce an automated summarization technique using deep learning models that helps legal professionals extract key rulings, arguments, and case outcomes quickly and efficiently. We compared two summarization techniques using deep neural networks, specifically LEGAL-BERT and bidirectional long short-term memory (Bi-LSTM) enhanced with an adaptive length mechanism that dynamically determines the optimal summary length based on the complexity and content of each document. We performed our experiment on an Indian Legal Corpus (ILC) dataset and we predict that the BiLSTM approach performs better on ROUGE scores than the LEGAL-BERT model with better recall and stronger fidelity to the original content.