MoECE: Mixture of Error Correction Experts

MoECE is a grammatical error correction model built by converting T5-v1.1 models into mixture-of-expert models. MoECE is more computationally efficient than the original T5 models and produces interpretable corrections by identifying the error types for each correction token.

The safety warning is caused by the checkpoints being saved in pickle format. These checkpoints were generated using the Fairseq library and are not directly compatible with the Transformers library. Please refer to the official repository for instructions on how to use the models.

MoECE was introduced in the following paper (PDF):

@inproceedings{qorib-etal-2024-efficient,
    title = "Efficient and Interpretable Grammatical Error Correction with Mixture of Experts",
    author = "Qorib, Muhammad Reza  and
      Aji, Alham Fikri  and
      Ng, Hwee Tou",
    booktitle = "Findings of the 2024 Conference on Empirical Methods in Natural Language Processing",
    month = nov,
    year = "2024",
    address = "Miami",
    publisher = "Association for Computational Linguistics",
}