HCR: Hierarchical Collaborative Reasoning with Interactive Distillation and Swarm Reinforcement for Chinese Spelling Correction

Chinese spelling correction (CSC) remains challenging due to heterogeneous error types and domain-dependent variations. We propose HCR, a hierarchical collaborative reasoning framework that integrates interactive knowledge distillation, swarm reinforcement collaboration, and debate-enhanced arbitration into a unified multi-agent architecture. By specializing agents in orthographic, phonetic, and semantic reasoning and enabling adaptive collaboration, HCR effectively disentangles complex dependencies and refines predictions through iterative consensus. Extensive experiments on three public benchmarks and a real-world medical dataset demonstrate that HCR achieves state-of-the-art performance on both detection and correction tasks and exhibits strong robustness under domain shifts, establishing a solid foundation for advancing interpretable, adaptive, and generalizable collaborative reasoning in CSC.