Data2TextWithAuxiliarySupervision

This repository provides the implementation for our paper "Improving Encoder by Auxiliary Supervision Tasks for Table-to-Text Generation" ACL2021 main conference. This code is based on https://github.com/ernestgong/data2text-three-dimensions/.

We provide the scrips (preprocess.sh, train.sh, and translate.sh) to preprocess the dataset, train models, and test. Please refer to these scripts for more details about parameters setting. The ouputs of our model are saved at ./our_results.

Citations

Please kindly cite this work if it helps your research:

@inproceedings{li-etal-2021-improving-encoder,
    title = "Improving Encoder by Auxiliary Supervision Tasks for Table-to-Text Generation",
    author = "Li, Liang  and
      Ma, Can  and
      Yue, Yinliang  and
      Hu, Dayong",
    booktitle = "Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers)",
    month = aug,
    year = "2021",
    address = "Online",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2021.acl-long.466",
    doi = "10.18653/v1/2021.acl-long.466",
    pages = "5979--5989",
}

1. Requirements

All dependencies can be installed via:

pip install -r 3.7.requirements.txt

Note that 3.7.requirements.txt contains unnecessary packages and Python version is 3.7.

2. Dataset

Please refer to https://github.com/ernestgong/data2text-three-dimensions/ and https://github.com/wanghm92/rw_fg to obtain ROTOWIRE and RW-FG datasets. And we provide the necessary config files in ./dataset.

3. Preprocess

The following command will preprocess the data:

bash preprocess.sh

4. Train

The following command will train the model:

bash train.sh

5. Translate

The following command will generate on development or test datasets given trained model:

bash translate.sh

6. Evaluate

To obtain extractive evaluation metrics, please refer to https://github.com/ratishsp/data2text-1 and https://github.com/wanghm92/rw_fg for details.

The following command will compute BLEU score:

perl ref/multi-bleu.perl ref/test.txt < ./our_results/model_pred_rw_test.txt