Addressing domain shift in neural machine translation via reinforcement learning

作者：

Highlights：

• REINFORCE-based sentence selection and weighting method for domain adaptation.

• Handling domain shift problem in neural machine translation.

• Different languages encoded into a common script for language model training.

• Maximum likelihood estimation and minimum risk training function are used.

• Proposed method outperforms the existing state-of-the-art approach by ∼2 BLEU points.

摘要

•REINFORCE-based sentence selection and weighting method for domain adaptation.•Handling domain shift problem in neural machine translation.•Different languages encoded into a common script for language model training.•Maximum likelihood estimation and minimum risk training function are used.•Proposed method outperforms the existing state-of-the-art approach by ∼2 BLEU points.

论文关键词：Neural machine translation,Domain adaptation,Low resource languages,Reinforcement learning

论文评审过程：Received 29 October 2021, Revised 15 January 2022, Accepted 27 March 2022, Available online 9 April 2022, Version of Record 19 April 2022.

论文官网地址：https://doi.org/10.1016/j.eswa.2022.117039