This repository provides the code and data of the work in ACL2021: PLOME: Pre-training with Misspelled Knowledge for Chinese Spelling Correction https://aclanthology.org/2021.acl-long.233.pdf

We have a new work on CSC in ACL 2022: https://github.com/liushulinle/CRASpell

Requirements:

• python3

• tensorflow1.14

• horovod

Instructions:

• Finetune:

  train and evaluation file format: original sentence \t golden sentence

  step1: cd finetune_src ; 
  step2: download the pretrained PLOME model and corpus from https://drive.google.com/file/d/1aip_siFdXynxMz6-2iopWvJqr5jtUu3F/view?usp=sharing ;
  step3: sh start.sh

• Pre-train

  step1: cd pre_train_src ;
  step2: sh gen_train_tfrecords.sh ;
  step3: sh start.sh

Our Pretrained Models:

PLOME model: https://drive.google.com/file/d/1aip_siFdXynxMz6-2iopWvJqr5jtUu3F/view?usp=sharing

cBERT model: https://drive.google.com/file/d/1cqSTpn7r9pnDcvMoM3BbX1X67JsPdZ8_/view?usp=sharing

国内下载地址：

PLOME： https://share.weiyun.com/OREEY0H3

cBERT: https://share.weiyun.com/wXErg7gB