Microsoft Research Asia’s Systems for WMT19

Published:

Please cite:
@inproceedings{xia2019microsoft,
title={Microsoft Research Asia’s Systems for WMT19},
author={Xia, Yingce and Tan, Xu and Tian, Fei and Gao, Fei and He, Di and Chen, Weicong and Fan, Yang and Gong, Linyuan and Leng, Yichong and Luo, Renqian and others},
booktitle={Proceedings of the Fourth Conference on Machine Translation (Volume 2: Shared Task Papers, Day 1)},
pages={424--433},
year={2019}
}

Abstract

We Microsoft Research Asia made submissions to 11 language directions in the WMT19 news translation tasks. We won the first place for 8 of the 11 directions and the second place for the other three. Our basic systems are built on Transformer, back translation and knowledge distillation. We integrate several of our rececent techniques to enhancethe baseline systems: multi-agent dual learning (MADL), masked sequence-to-sequencepre-training (MASS), neural architecture optimization (NAO), and soft contextual data augmentation (SCA).

[PDF]