This repository provides training code of DCN models for Chinese Spelling Check (CSC).
The paper has been accepted in Findings of ACL 2021.
Our code is based on transformers 3.0.
The following command installs all necessary packages:
pip install -r requirements.txt
We test our code using Python 3.6.
The preprocessed training dataset can be downloaded from here(password:hfiw).
To train the DCN model, download the RoBERTa-wwm-ext and copy the model to chinese_roberta_wwm_ext_pytorch, then run:
sh train.sh
The sentence-level experimental results on SIGHAN15 for the default config are as follows:
| model | d-p | d-r | d-f | c-p | c-r | c-f |
|---|---|---|---|---|---|---|
| DCN | 76.84 | 79.64 | 78.21 | 74.74 | 77.45 | 76.07 |
@inproceedings{wang-etal-2021-dynamic,
title = "Dynamic Connected Networks for {C}hinese Spelling Check",
author = "Wang, Baoxin and
Che, Wanxiang and
Wu, Dayong and
Wang, Shijin and
Hu, Guoping and
Liu, Ting",
booktitle = "Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021",
month = aug,
year = "2021",
address = "Online",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2021.findings-acl.216",
doi = "10.18653/v1/2021.findings-acl.216",
pages = "2437--2446",
}