WLAC-Joint-Training

This repository contains the code for the paper Rethinking Word-Level Auto-Completion in Computer-Aided Translation, EMNLP 2023

Requirements

Create the environment using conda:

conda create -n wlac python=3.6
conda activate wlac
pip install -r requirements.txt

Running the code

1. Prepare the environment variable

export PROJECT_ROOT=path_to_code/WLAC-Joint-Training

Enter the run script directory. For AIOE-BPE model please refer to scripts/aioe_bpe

cd scripts/aioe

Please check all the related files and modify the path to the actual path

2. Preprocess training set

bash ./preprocess_training_set.sh

3. Preprocess validation set

bash ./preprocess_valid_set.sh

4. Run training

For AIOE model:

bash ./train_aioe.sh

For AIOE-Joint model:

bash ./train_aioe_joint.sh

5. Evaluation

bash ./eval.sh

Citation

Please cite as:

@article{chen2023rethinking,
  title={Rethinking Word-Level Auto-Completion in Computer-Aided Translation},
  author={Chen, Xingyu and Liu, Lemao and Huang, Guoping and Zhang, Zhirui and Yang, Mingming and Shi, Shuming and Wang, Rui},
  journal={arXiv preprint arXiv:2310.14523},
  year={2023}
}