arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 1512.02433cs.CL
收藏

神经机器翻译的最小风险训练

Minimum Risk Training for Neural Machine Translation

Baidu Inc.(百度公司);Tsinghua University(清华大学)

Shiqi Shen, Yong Cheng, Zhongjun He, Wei He, Hua Wu, Maosong Sun, Yang Liu

摘要:

我们提出了端到端神经机器翻译的最小风险训练。与传统的最大似然估计不同,最小风险训练能够直接相对于任意评估指标优化模型参数,这些指标不一定是可微的。实验表明,我们的方法在最先进的神经机器翻译系统跨各种语言对的最大似然估计实现了显着的改善。对架构透明,我们的方法可以应用于更多的神经网络,并可能使更多的NLP任务受益。

英文摘要:

We propose minimum risk training for end-to-end neural machine translation. Unlike conventional maximum likelihood estimation, minimum risk training is capable of optimizing model parameters directly with respect to arbitrary evaluation metrics, which are not necessarily differentiable. Experiments show that our approach achieves significant improvements over maximum likelihood estimation on a state-of-the-art neural machine translation system across various languages pairs. Transparent to architectures, our approach can be applied to more neural networks and potentially benefit more NLP tasks.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑