Mocha-EMNLP22 ☕️

This repository contains data and code for MOCHA in our EMNLP 2022 paper: MOCHA: A Multi-Task Training Approach for Coherent Text Generation from Cognitive Perspective

Processed Datasets

Our processed data can be accessed through the link

• cmv: the processed data for argument generation
• nyt: the processed data for news article writing
• wikiplot: the processed data for story generation

Note: the The New York Times Annotated Corpus is licensed by LDC. If you have the license and want to use the processed data, please contact me.

Requirements

The original code is tested under the following environment:

pytorch==1.7.1
transformers==4.8.2

Code Structure

• finetune_generation_pipeline.py: the code for training and decoding
• run_pipeline.sh: runing script
• eval_utils/evaluation.py: script for automatic evaluations (BLEU, ROUGE, METEOR)

To run the code, you need to specify the model path and data path in run_pipeline.sh, and then run code with the command:

bash run_pipeline.sh

Citation

If you find our work useful, please cite:

@inproceedings{hu-etal-2022-mocha,
    title = "{MOCHA}: A Multi-Task Training Approach for Coherent Text Generation from Cognitive Perspective",
    author = "Hu, Zhe  and
      Chan, Hou Pong  and
      Huang, Lifu",
    booktitle = "Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing",
    month = dec,
    year = "2022",
    address = "Abu Dhabi, United Arab Emirates",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2022.emnlp-main.705",
    pages = "10324--10334",
}

Contact

Zhe Hu (zhehu.94 at gmail.com)