Abstract
Legal judgment prediction is essential for enhancing judicial efficiency. Inthis work, we identify that existing large language models (LLMs) underperformin this domain due to challenges in understanding case complexities anddistinguishing between similar charges. To adapt LLMs for effective legaljudgment prediction, we introduce the Ask-Discriminate-Predict (ADAPT)reasoning framework inspired by human judicial reasoning. ADAPT involvesdecomposing case facts, discriminating among potential charges, and predictingthe final judgment. We further enhance LLMs through fine-tuning with multi-tasksynthetic trajectories to improve legal judgment prediction accuracy andefficiency under our ADAPT framework. Extensive experiments conducted on twowidely-used datasets demonstrate the superior performance of our framework inlegal judgment prediction, particularly when dealing with complex and confusingcharges.