Adversarial Subword Regularization for Robust Neural Machine Translation
This work addresses robustness issues in machine translation for low-resource and out-of-domain scenarios, representing an incremental improvement over existing subword regularization methods.
The paper tackled the problem of neural machine translation (NMT) models being sensitive to subword segmentation errors by proposing adversarial subword regularization (ADVSR), which uses gradient signals during training to expose diverse segmentations, resulting in improved robustness and performance on low-resource and out-domain datasets.
Exposing diverse subword segmentations to neural machine translation (NMT) models often improves the robustness of machine translation as NMT models can experience various subword candidates. However, the diversification of subword segmentations mostly relies on the pre-trained subword language models from which erroneous segmentations of unseen words are less likely to be sampled. In this paper, we present adversarial subword regularization (ADVSR) to study whether gradient signals during training can be a substitute criterion for exposing diverse subword segmentations. We experimentally show that our model-based adversarial samples effectively encourage NMT models to be less sensitive to segmentation errors and improve the performance of NMT models in low-resource and out-domain datasets.