Stronger Baselines for Grammatical Error Correction Using Pretrained Encoder-Decoder Model
share
Grammatical error correction (GEC) literature has reported on the effectiveness of pretraining a Seq2Seq model with a large amount of pseudo data. In this study, we explored two generic pretrained encoder-decoder (Enc-Dec) models, including BART, which reported the state-of-the-art (SOTA) results for several Seq2Seq tasks other than GEC. We found that monolingual and multilingual BART models achieve high performance in GEC, including a competitive result compared with the current SOTA result in English GEC. Our implementations will be publicly available at GitHub.
READ FULL TEXT