SSEM:Span-based Semantic Enhancement Method for Chinese Spelling Correction

Current Chinese spelling correction methods usually only consider adjacent word semantics, missing distant word connections. This paper proposes a span-based semantic enhancement method for Chinese spelling correction. The method strengthens the text's overall semantics by a novel span-based semantic enhancement module that captures semantic associations between any two characters. This module not only captures direct neighboring connections but also explores long-distance semantic dependencies. Experiments on the SIGHAN13, SIGHAN14, and SIGHAN15 benchmarks demonstrate the method's effectiveness.