Understanding and Mitigating the Uncertainty in Zero-Shot Translation
This work addresses quality issues in multilingual neural machine translation systems, which is an incremental improvement for the field of machine translation.
The paper tackled the problem of off-target issues in zero-shot translation by identifying and mitigating two types of uncertainty, resulting in significant performance improvements over strong baselines in experiments on balanced and imbalanced datasets.
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 lightweight and complementary approaches to denoise the training data for model training and explicitly penalize the off-target translations by unlikelihood training during model training. Extensive experiments on both balanced and imbalanced datasets show that our approaches significantly improve the performance of zero-shot translation over strong MNMT baselines.