Jam or Cream First? Modeling Ambiguity in Neural Machine Translation with SCONES
This addresses the issue of intrinsic uncertainty in machine translation for practitioners, offering a method to improve translation quality and efficiency, though it is incremental as it builds on existing NMT frameworks.
The authors tackled the problem of modeling ambiguity in neural machine translation by replacing the softmax layer with a multi-label classification layer and a new loss function called SCONES, resulting in consistent BLEU score gains across six translation directions and a 3.9x speedup in inference while matching or improving BLEU scores.
The softmax layer in neural machine translation is designed to model the distribution over mutually exclusive tokens. Machine translation, however, is intrinsically uncertain: the same source sentence can have multiple semantically equivalent translations. Therefore, we propose to replace the softmax activation with a multi-label classification layer that can model ambiguity more effectively. We call our loss function Single-label Contrastive Objective for Non-Exclusive Sequences (SCONES). We show that the multi-label output layer can still be trained on single reference training data using the SCONES loss function. SCONES yields consistent BLEU score gains across six translation directions, particularly for medium-resource language pairs and small beam sizes. By using smaller beam sizes we can speed up inference by a factor of 3.9x and still match or improve the BLEU score obtained using softmax. Furthermore, we demonstrate that SCONES can be used to train NMT models that assign the highest probability to adequate translations, thus mitigating the "beam search curse". Additional experiments on synthetic language pairs with varying levels of uncertainty suggest that the improvements from SCONES can be attributed to better handling of ambiguity.