Enhancing Writing Through RoBERTa-Based Transfer Learning

Frequent grammatical errors and loose syntactic structures in English writing training restrict improvement in learners' writing competence. This article proposes a transfer learning strategy based on a Robustly Optimized BERT Pretraining Approach (RoBERTa) to provide accurate analysis and feedback on grammar and syntax. RoBERTa is pre-trained on a large English corpus and then fine-tuned on an English writing dataset with grammatical correctness and syntactic rationality as core objectives. A bidirectional gated recurrent unit (BiGRU) layer is stacked on RoBERTa representations to capture sequence dependencies, followed by a classification layer. During inference, the model outputs quantitative grammar and syntax scores and generates targeted revision suggestions. Experimental results show that the F1 score for verb tense error detection reaches 0.92 and spelling error accuracy reaches 0.97, indicating that the method provides an effective technical path for improving the quality of English writing.