Grammar Correction for Multiple Errors in Chinese Based on Prompt Templates

Grammar error correction (GEC) is a crucial task in the field of Natural Language Processing (NLP). Its objective to automatically detect and rectify grammatical mistakes sentences, which possesses immense application research value. Currently, mainstream grammar-correction methods primarily rely on sequence labeling text generation, are two kinds end-to-end methods. These have shown exemplary performance areas with low density but often fail deliver satisfactory results high-error situations where multiple errors exist single sentence. Consequently, these tend overcorrect correct words, leading high rate false positives. To address this issue, we researched specific characteristics Chinese grammar (CGEC) situations. We proposed method based prompt templates. Firstly, strategy for constructing templates suitable CGEC. This transforms CGEC into masked fill-in-the-blank compatible language model BERT. Secondly, dynamically updating templates, incorporates already corrected template through dynamic updates improve quality. Moreover, used phonetic graphical resemblance knowledge from confusion set as guiding information. By combining BERT’s prediction results, can more accurately select characters, significantly enhancing accuracy model’s results. Our were validated experiments public dataset. The indicate that our achieves higher lower rates scenarios.