Doubly-Trained Adversarial Data Augmentation for Neural Machine Translation
This addresses robustness issues in neural machine translation systems, which is an incremental improvement for the machine translation community.
The paper tackles the problem of neural machine translation models being vulnerable to noisy inputs by generating adversarial augmentation samples that attack the model while preserving source-side semantic meaning, resulting in improved model robustness across three language pairs and two evaluation metrics.
Neural Machine Translation (NMT) models are known to suffer from noisy inputs. To make models robust, we generate adversarial augmentation samples that attack the model and preserve the source-side semantic meaning at the same time. To generate such samples, we propose a doubly-trained architecture that pairs two NMT models of opposite translation directions with a joint loss function, which combines the target-side attack and the source-side semantic similarity constraint. The results from our experiments across three different language pairs and two evaluation metrics show that these adversarial samples improve the model robustness.