{
  "id": "2205.10068",
  "version": "v1",
  "published": "2022-05-20T10:29:46.000Z",
  "updated": "2022-05-20T10:29:46.000Z",
  "title": "Understanding and Mitigating the Uncertainty in Zero-Shot Translation",
  "authors": [
    "Wenxuan Wang",
    "Wenxiang Jiao",
    "Shuo Wang",
    "Zhaopeng Tu",
    "Michael R. Lyu"
  ],
  "comment": "work in progress",
  "categories": [
    "cs.CL",
    "cs.AI",
    "cs.LG"
  ],
  "abstract": "Zero-shot translation is a promising direction for building a comprehensive multilingual neural machine translation (MNMT) system. However, its quality is still not satisfactory due to off-target issues. In this paper, we aim to understand and alleviate the off-target issues from the perspective of uncertainty in zero-shot translation. By carefully examining the translation output and model confidence, we identify two uncertainties that are responsible for the off-target issues, namely, extrinsic data uncertainty and intrinsic model uncertainty. Based on the observations, we propose two light-weight and complementary approaches to denoise the training data for model training, and mask out the vocabulary of the off-target languages in inference. Extensive experiments on both balanced and unbalanced datasets show that our approaches significantly improve the performance of zero-shot translation over strong MNMT baselines. Qualitative analyses provide insights into where our approaches reduce off-target translations",
  "revisions": [
    {
      "version": "v1",
      "updated": "2022-05-20T10:29:46.000Z"
    }
  ],
  "analyses": {
    "keywords": [
      "zero-shot translation",
      "off-target issues",
      "understand",
      "comprehensive multilingual neural machine translation",
      "approaches reduce off-target translations"
    ],
    "note": {
      "typesetting": "TeX",
      "pages": 0,
      "language": "en",
      "license": "arXiv",
      "status": "editable"
    }
  }
}