The construction of English grammar error correction models based on Seq2Seq model and feedback filtering mechanisms

Abstract The current English grammar error correction model still has low accuracy when dealing with complex sentences and long sentence dependencies. Therefore, this study proposes a Sequence-to-Sequence (Seq2Seq) model that integrates feedback filtering mechanisms. This model combines recurrent neural networks, attention mechanisms, and user feedback optimization mechanisms based on N-grams to improve both accuracy and adaptability in syntax correction tasks. The experimental results demonstrated that the proposed model improved accuracy by 5.69% and the F0.5 value by 1.34% compared to the CAMB model, with particularly notable improvements in correcting common grammar errors such as attributives and prepositions. The average processing time for complex error correction tasks was only 32–55 ms. In conclusion, by combining different networks and mechanisms, the proposed model effectively enhances both accuracy and efficiency of English grammar error correction, thereby providing new solutions for English grammar teaching and automatic translation.