Improving Neural Machine Translation Models with Monolingual Data

  • 时间: 2016 ACL
  • 作者&机构: School of Informatics, University of Edinburgh

Abstract

  • 目标端monolingual数据在提升NMT流畅度上有着重要的作用。
  • 通过将monoligual训练数据和back-translation的数据进行配对,我们可以将其作为额外训练平行语料库数据,从而提升NMT的翻译性能。

NMT Training with Monolingual Training Data

两种做法:

  • providing monolingual training examples with an empty(dummy) source sentence.
  • back-translation

你可能感兴趣的:(Improving Neural Machine Translation Models with Monolingual Data)