Indian legal text summarization using large language model

This research introduces a novel framework for summarizing legal judgments by integrating two state-of-the-art pretrained models—Google T5-small and Facebook BART—with rhetorical labels to enhance both accuracy and relevance. Each model is fine-tuned to pro- cess distinct rhetorical components, such as facts, arguments, and conclusions, while LoRA adapters ensure efficient fine-tuning with minimal computational over- head. The incorporation of rhetorical labels allows the models to differentiate between various legal roles, ensuring the generated summaries preserve the logical structure and critical reasoning embedded in the text. Our experiments demonstrate that this multi- model, label-guided approach not only outperforms traditional summa- rization methods but also delivers coherent, context aware summaries tailored to the needs of legal practitioners, improving both usability and decision-making efficiency.