LePREC Framework Addresses LLM Limitations in Legal Issue Identification

A novel neuro-symbolic framework named LePREC (Legal Professional-inspired Reasoning Elicitation and Classification) has been introduced to enhance the identification of legal issues through Large Language Models. Researchers compiled a dataset from 769 actual court cases related to the Malaysian Contract Act, utilizing GPT-4o to extract pertinent facts and propose potential legal issues, which were subsequently reviewed by senior legal professionals. Findings indicated that while LLMs produce a variety of issue candidates, their accuracy is lacking, with GPT-4o only achieving 62% precision. This framework integrates neural generation with structured statistical reasoning to bridge this gap. The global challenge of over half the population facing civil justice needs underscores the significance of this study, which was published on arXiv under identifier arXiv:2604.19464v1.