@inproceedings{garcia-gilabert-etal-2024-resetox,
    title = "{R}e{S}e{TOX}: Re-learning attention weights for toxicity mitigation in machine translation",
    author = "Garc{\'i}a Gilabert, Javier  and
      Escolano, Carlos  and
      Costa-juss{\`a}, Marta",
    editor = "Scarton, Carolina  and
      Prescott, Charlotte  and
      Bayliss, Chris  and
      Oakley, Chris  and
      Wright, Joanna  and
      Wrigley, Stuart  and
      Song, Xingyi  and
      Gow-Smith, Edward  and
      Bawden, Rachel  and
      S{\'a}nchez-Cartagena, V{\'i}ctor M  and
      Cadwell, Patrick  and
      Lapshinova-Koltunski, Ekaterina  and
      Cabarr{\~a}o, Vera  and
      Chatzitheodorou, Konstantinos  and
      Nurminen, Mary  and
      Kanojia, Diptesh  and
      Moniz, Helena",
    booktitle = "Proceedings of the 25th Annual Conference of the European Association for Machine Translation (Volume 1)",
    month = jun,
    year = "2024",
    address = "Sheffield, UK",
    publisher = "European Association for Machine Translation (EAMT)",
    url = "https://preview.aclanthology.org/jlcl-multiple-ingestion/2024.eamt-1.8/",
    pages = "37--58",
    abstract = "Our proposed method, RESETOX (REdoSEarch if TOXic), addresses the issue ofNeural Machine Translation (NMT) gener-ating translation outputs that contain toxicwords not present in the input. The ob-jective is to mitigate the introduction oftoxic language without the need for re-training. In the case of identified addedtoxicity during the inference process, RE-SETOX dynamically adjusts the key-valueself-attention weights and re-evaluates thebeam search hypotheses. Experimental re-sults demonstrate that RESETOX achievesa remarkable 57{\%} reduction in added tox-icity while maintaining an average trans-lation quality of 99.5{\%} across 164 lan-guages. Our code is available at: https://github.com"
}