Reviewed Version (pdf): https://openreview.net/references/pdf?id=Mc83x4evEz
Keywords: resource-rich machine translation, neural machine translation, pre-training, self-supervised learning, joint training
Abstract: Self-supervised pre-training of text representations has been successfully applied to low-resource Neural Machine Translation (NMT). However, it usually fails to achieve notable gains on resource-rich NMT. In this paper, we propose a joint training approach, $F_2$-XEnDec, to combine self-supervised and supervised learning to optimize NMT models. To exploit complementary self-supervised signals for supervised learning, NMT models are trained on examples that are interbred from monolingual and parallel sentences through a new process called crossover encoder-decoder. Experiments on two resource-rich translation bench-marks, WMT’14 English-German and WMT’14 English-French, demonstrate that our approach achieves substantial improvements over a vanilla Transformer and obtains a new state of the art of 46 BLEU on English-French. Results also show that our approach is capable of improving model robustness against input perturbations which is known as a key weakness in contemporary NMT systems.
One-sentence Summary: Train NMT models on resource-rich corpora by exploiting complementary self-supervised learning signals for supervised learning in a joint training framework.
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics