Stronger Baselines for Grammatical Error Correction Using a Pretrained Encoder-Decoder Model

Studies on grammatical error correction (GEC) have reported the effectiveness of pretraining a Seq2Seq model with large amount pseudodata. However, this approach requires time-consuming for GEC because size In study, we explore utility bidirectional and auto-regressive transformers (BART) as generic pretrained encoder-decoder GEC. With use GEC, can be eliminated. We find that monolingual multilingual BART models achieve high performance in one results being comparable to current strong English Our implementations are publicly available at GitHub (https://github.com/Katsumata420/generic-pretrained-GEC).