VHM

Code for the ACL2022 main conference paper A Variational Hierarchical Model for Neural Cross-Lingual Summarization.

Introduction

In this work, we mainly bulid a variational hierarchical model via conditional variational auto-encoders that introduce a global variable to combine the local ones for translation and summarization at the same time for CLS. The implementation is mainly based on NCLS-Corpora. The dataset we used is from the NCLS and then change the path in 'run_config/*.json' correspondingly.

Requirements

• pytorch >= 1.0
• python 3.6

Usage

Training with the following scripts:

• Build vocab

cd tools 
python -u build_vocab $number < $textfile > $vocab_file

• Start training

python -u train.py -config run_config/train-zhen.json -batch_size $batch_size -kl_annealing_steps $kl_steps -latent_dim $latent_dim

• Start decoding

python -u translate.py -config run_config/decode-zhen.json

Citation

If you find this project helps, please cite our paper :)

@inproceedings{liang-etal-2022-variational,
    title = "A Variational Hierarchical Model for Neural Cross-Lingual Summarization",
    author = "Liang, Yunlong  and
      Meng, Fandong  and
      Zhou, Chulun  and
      Xu, Jinan  and
      Chen, Yufeng  and
      Su, Jinsong  and
      Zhou, Jie",
    booktitle = "Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)",
    month = may,
    year = "2022",
    address = "Dublin, Ireland",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2022.acl-long.148",
    doi = "10.18653/v1/2022.acl-long.148",
    pages = "2088--2099",
    }

Please feel free to open an issue or email me ([email protected]) for questions and suggestions.