Research on English Grammar Error Correction Algorithms Based on Deep Learning

In traditional English grammar learning scenarios, learners heavily rely on teachers. The scarcity of educational resources means that teachers cannot correct all the grammatical errors made by students, significantly slowing down the learners' progress. Automatic English grammar correction methods can help learners overcome the limitations of educational resources, demonstrating great significance. This paper proposes a deep learning-based English grammar correction algorithm and compares it with several previously reported algorithms, finding that the proposed Transformer structure based on a copy mechanism performs better in correcting grammatical errors. Furthermore, this paper presents a novel method for generating parallel sentence pairs to expand the training corpus and compares it with two other data augmentation methods: random sentence perturbation and reversed grammar error generation. The model is pretrained using training corpora generated by these three methods and later fine-tuned using manually annotated data. Experimental results indicate that the errors generated by the proposed grammar error generation method are closer to the mistakes made by learners, and the model trained on the corpus produced by this method demonstrates better correction performance.