This paper describes Facebook FAIR's submission to the WMT19 shared news translation task. We participate in two language pairs and four language directions, English <-> German and English <-> Russian. Following our submission from last year, our baseline systems are large BPE-based transformer models trained with the Fairseq sequence modeling toolkit which rely on sampled back-translations. This year we experiment with different bitext data filtering schemes, as well as with adding filtered back-translated data. We also ensemble and fine-tune our models on domain-specific data, then decode using noisy channel model reranking. Our submissions are ranked first in all four directions of the human evaluation campaign. On En->De, our system significantly outperforms other systems as well as human translations. This system improves upon our WMT'18 submission by 4.5 BLEU points.
@article{arxiv.1907.06616,
title = {Facebook FAIR's WMT19 News Translation Task Submission},
author = {Nathan Ng and Kyra Yee and Alexei Baevski and Myle Ott and Michael Auli and Sergey Edunov},
journal= {arXiv preprint arXiv:1907.06616},
year = {2019}
}