Bridging the Gap in Chinese Legal Conflict Review: A Dataset, Benchmark Tasks, and Framework

Natural language processing has enabled rapid progress in legal artificial intelligence, but automated review of conflicts between superior and subordinate laws remains largely unexplored. Maintaining consistency across legal hierarchies in Chinese legislation is crucial, yet manual review is labor-intensive and error-prone. To address the absence of systematic research in legal conflict review, this study presents three key contributions. (1) we presented the Chinese Legal Conflict Review dataset (LCR-CN), a novel dataset containing 6,995 annotated legislative provisions with conflict types, explanations, and revision suggestions. (2) we proposed three benchmark tasks: (a). Superior Law Retrieval, (b). Conflict Classification, and (c) Conflict Explanation and Revision Generation, to establish a comprehensive evaluation framework for this domain. (3) we fine-tuned multiple language models on LCR-CN and observe a relative accuracy improvement of 72.7% over baseline methods, demonstrating the effectiveness of the dataset and the feasibility of automated legal conflict review. This work introduces the first dedicated dataset (LCR-CN), establishes a standardized task framework, and presents robust baselines for legal conflict review, thereby laying theoretical and methodological foundations for AI-assisted legal conflict resolution.