Faithful Target Attribute Prediction in Neural Machine Translation
This addresses translation errors for users of NMT systems by improving attribute fidelity, but it is incremental as it builds on existing methods for specific attributes.
The paper tackled the problem of uncontrolled attributes like word casing and gender in neural machine translation training data, which cause translation errors, by simultaneously predicting target words and attributes, resulting in translations that more faithfully mirror the training data distribution and enabling data augmentation for uppercased input translation.
The training data used in NMT is rarely controlled with respect to specific attributes, such as word casing or gender, which can cause errors in translations. We argue that predicting the target word and attributes simultaneously is an effective way to ensure that translations are more faithful to the training data distribution with respect to these attributes. Experimental results on two tasks, uppercased input translation and gender prediction, show that this strategy helps mirror the training data distribution in testing. It also facilitates data augmentation on the task of uppercased input translation.