mbartolo/synQA

Tasks

• question-answering

Citation text from the card

@inproceedings{bartolo-etal-2021-improving,
    title = "Improving Question Answering Model Robustness with Synthetic Adversarial Data Generation",
    author = "Bartolo, Max  and
      Thrush, Tristan  and
      Jia, Robin  and
      Riedel, Sebastian  and
      Stenetorp, Pontus  and
      Kiela, Douwe",
    booktitle = "Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing",
    month = nov,
    year = "2021",
    address = "Online and Punta Cana, Dominican Republic",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2021.emnlp-main.696",
    doi = "10.18653/v1/2021.emnlp-main.696",
    pages = "8830--8848",
    abstract = "Despite recent progress, state-of-the-art question answering models remain vulnerable to a variety of adversarial attacks. While dynamic adversarial data collection, in which a human annotator tries to write examples that fool a model-in-the-loop, can improve model robustness, this process is expensive which limits the scale of the collected data. In this work, we are the first to use synthetic adversarial data generation to make question answering models more robust to human adversaries. We develop a data generation pipeline that selects source passages, identifies candidate answers, generates questions, then finally filters or re-labels them to improve quality. Using this approach, we amplify a smaller human-written adversarial dataset to a much larger set of synthetic que

All datasets marked Permissive or share-alike | JSON