An effective method to improve neural machine translation with monolingual
data is to augment the parallel training corpus with back-translations of
target language sentences. This work broadens the understanding of
back-translation and investigates a number of methods to generate synthetic
source sentences. We find that in all but resource poor settings
back-translations obtained via sampling or noised beam outputs are most
effective. Our analysis shows that sampling or noisy synthetic data gives a
much stronger training signal than data generated by beam or greedy search. We
also compare how synthetic data compares to genuine bitext and study various
domain effects. Finally, we scale to hundreds of millions of monolingual
sentences and achieve a new state of the art of 35 BLEU on the WMT'14
English-German test set.
Описание
[1808.09381] Understanding Back-Translation at Scale
%0 Generic
%1 edunov2018understanding
%A Edunov, Sergey
%A Ott, Myle
%A Auli, Michael
%A Grangier, David
%D 2018
%K masterthesis translation
%T Understanding Back-Translation at Scale
%U http://arxiv.org/abs/1808.09381
%X An effective method to improve neural machine translation with monolingual
data is to augment the parallel training corpus with back-translations of
target language sentences. This work broadens the understanding of
back-translation and investigates a number of methods to generate synthetic
source sentences. We find that in all but resource poor settings
back-translations obtained via sampling or noised beam outputs are most
effective. Our analysis shows that sampling or noisy synthetic data gives a
much stronger training signal than data generated by beam or greedy search. We
also compare how synthetic data compares to genuine bitext and study various
domain effects. Finally, we scale to hundreds of millions of monolingual
sentences and achieve a new state of the art of 35 BLEU on the WMT'14
English-German test set.
@misc{edunov2018understanding,
abstract = {An effective method to improve neural machine translation with monolingual
data is to augment the parallel training corpus with back-translations of
target language sentences. This work broadens the understanding of
back-translation and investigates a number of methods to generate synthetic
source sentences. We find that in all but resource poor settings
back-translations obtained via sampling or noised beam outputs are most
effective. Our analysis shows that sampling or noisy synthetic data gives a
much stronger training signal than data generated by beam or greedy search. We
also compare how synthetic data compares to genuine bitext and study various
domain effects. Finally, we scale to hundreds of millions of monolingual
sentences and achieve a new state of the art of 35 BLEU on the WMT'14
English-German test set.},
added-at = {2020-12-20T19:50:29.000+0100},
author = {Edunov, Sergey and Ott, Myle and Auli, Michael and Grangier, David},
biburl = {https://www.bibsonomy.org/bibtex/2af718c966321a4ac768315ee39733d92/festplatte},
description = {[1808.09381] Understanding Back-Translation at Scale},
interhash = {491979113b99930d1adda1730cbc415f},
intrahash = {af718c966321a4ac768315ee39733d92},
keywords = {masterthesis translation},
note = {EMNLP 2018},
timestamp = {2021-02-11T14:20:40.000+0100},
title = {Understanding Back-Translation at Scale},
url = {http://arxiv.org/abs/1808.09381},
year = 2018
}