ATSum

The code for ACL2020 paper "Attend, Translate and Summarize: An Efficient Method for Neural Cross-Lingual Summarization"

Requirement

• pytorch >= 1.0
• python >= 3.6

Dataset

The datasets used in this paper can be found here.

How to run?

1. You should first obtain the word alignment by the fastalign tool. There is an example in "ATS-NE/beaver/data/savefile/c2e.pkl".

2. Build vocab

cd ATS-NE/tools && python build_vocab $number < $textfile > $vocab_file

3. Build the translation candidates and their index.

cd .. && python -m beaver.data.post_prob

4. Start training

python train.py -config run_config/train.json

5. Start decoding

python translate.py -config run_config/decode.json

Generated summaries and their scores

The summaries generated by our methods as well as their ROUGE score and MoverScore are presented in "resultofpaper" folder. The scripts we use to obtain rouge scores and MoverScore are also included.

Citation

We would appreciate your citation if you find this is beneficial.

@inproceedings{zhu-etal-2020-attend,
    title = "Attend, Translate and Summarize: An Efficient Method for Neural Cross-Lingual Summarization",
    author = "Zhu, Junnan  and
      Zhou, Yu  and
      Zhang, Jiajun  and
      Zong, Chengqing",
    booktitle = "Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics",
    month = jul,
    year = "2020",
    address = "Online",
    publisher = "Association for Computational Linguistics",
    url = "https://www.aclweb.org/anthology/2020.acl-main.121",
    pages = "1309--1321",
}

Contact

If you have any question, please feel free to contact us by sending an email to {junnan.zhu, yzhou, jjzhang, cqzong}@nlpr.ia.ac.cn.

License

This project is licensed under the BSD License - see LICENSE.md for details.

Copyright

The copyright of this code belongs to the authors, and the code is only used for research purposes. Display, reproduction, transmission, distribution or publication of this code is prohibited.