BriefGPT.xyz
Jun, 2019
标记后翻译
Tagged Back-Translation
HTML
PDF
Isaac Caswell, Ciprian Chelba, David Grangier
TL;DR
提出源端输入带标记的后向翻译方法，以替代传统的加噪声方法，实验证明该方法在英罗马尼亚语翻译中表现优于传统方法，且在英德语翻译中达到同等水平，重新定义了该领域的最佳表现。
Abstract
Recent work in
neural machine translation
(NMT) has shown significant quality gains from
noised-beam decoding
during
back-translation
, a m
→