Adaptive Grammar Correction with Error-Type Diagnosis and Learner-Centered Feedback Using Proficiency-Aware Transformer Learning

Grammar error correction systems are widely used to improve learner-written English. However, many existing systems mainly provide corrected sentences and do not explain the reason behind the correction. This limits their usefulness in language learning. This study proposes AGC-LCFNet, an Adaptive Grammar Correction with Learner-Centered Feedback Network. The model corrects grammatical errors, predicts the error type, and generates feedback according to the learner’s proficiency level. AGC-LCFNet uses a shared FLAN-T5 encoder to capture contextual and grammatical features from erroneous sentences. A correction decoder generates the corrected sentence, while a classification head identifies seven error categories, including tense, spelling, punctuation, article, preposition, subject-verb agreement, and sentence structure. A feedback decoder then produces learner-centered explanations. A proficiency-aware gating mechanism adjusts the feedback for beginner, intermediate, and advanced learners. The model was evaluated on BEA-2019, CoNLL-14, JFLEG, and Lang-8 datasets. The results show that AGC-LCFNet achieved the highest F0.5 score across all benchmarks and improved error-type classification performance. The model also produced more useful feedback than template-based and profiling-based baselines. Visual and ablation analysis confirmed that the classification module improves correction accuracy, while the gating mechanism improves feedback adaptation. These findings show that AGC-LCFNet is not only effective for grammar correction but also useful for personalized language learning support.

@article{Sun2026,
  title     = {Adaptive Grammar Correction with Error-Type Diagnosis and Learner-Centered Feedback Using Proficiency-Aware Transformer Learning},
  journal   = {International Journal of Advanced Computer Science and Applications},
  volume    = {17},
  number    = {7},
  year      = {2026},
  publisher = {The Science and Information Organization},
  author    = {Xinxin Sun and Peng Guo},
  doi       = {10.14569/IJACSA.2026.0170762},
  url       = {https://doi.org/10.14569/IJACSA.2026.0170762}
}