@inproceedings{deriu-etal-2022-probing,
    title = "Probing the Robustness of Trained Metrics for Conversational Dialogue Systems",
    author = {Deriu, Jan  and
      Tuggener, Don  and
      Von D{\"a}niken, Pius  and
      Cieliebak, Mark},
    editor = "Muresan, Smaranda  and
      Nakov, Preslav  and
      Villavicencio, Aline",
    booktitle = "Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers)",
    month = may,
    year = "2022",
    address = "Dublin, Ireland",
    publisher = "Association for Computational Linguistics",
    url = "https://preview.aclanthology.org/fix-sig-urls/2022.acl-short.85/",
    doi = "10.18653/v1/2022.acl-short.85",
    pages = "750--761",
    abstract = "This paper introduces an adversarial method to stress-test trained metrics for the evaluation of conversational dialogue systems. The method leverages Reinforcement Learning to find response strategies that elicit optimal scores from the trained metrics. We apply our method to test recently proposed trained metrics. We find that they all are susceptible to giving high scores to responses generated by rather simple and obviously flawed strategies that our method converges on. For instance, simply copying parts of the conversation context to form a response yields competitive scores or even outperforms responses written by humans."
}