deep-joseon-record-analysis

Official implementation of restoration and translation of joseon historical records. (NAACL 2021)

https://www.aclweb.org/anthology/2021.naacl-main.317/

Datasets and Model

You can download datasets and trained model weights via the following link:

https://tinyurl.com/yj7x8o2w

Run codes

• First, you should preprocess the datasets

  python preprocessing.py 
  

• Then, you can train the model

  bash run_train.sh
  

• After training the model, you can translate the corpus using multi-gpus.

  The translation process have a master server and workers.

  python translate_api.py
  

  (In the other shell) python translation_worker.py
  

• For the details of arguments, please refer codes.

Dependencies

• Python 3.8+ (I recommend to install Anaconda Python.)
• PyTorch 1.8.1
• tensorboard
• sentencepiece
• mpi4py (conda install mpi4py)
• OpenMPI (conda install openmpi)
• ujson
• flask

Contact

If you have any questions on our survey, please contact me via the following e-mail address: rudvlf0413@korea.ac.kr