Minimum Risk Training for Neural Machine Translation
About
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.
Shiqi Shen, Yong Cheng, Zhongjun He, Wei He, Hua Wu, Maosong Sun, Yang Liu• 2015
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Machine Translation | WMT En-De 2014 (test) | BLEU27.71 | 379 | |
| Machine Translation | WMT En-Fr 2014 (test) | BLEU31.3 | 237 | |
| Constituency Parsing | Penn Treebank WSJ (section 23 test) | F1 Score95.2 | 55 | |
| Machine Translation (Chinese-to-English) | NIST 2003 (MT-03) | BLEU37.32 | 52 | |
| Machine Translation (Chinese-to-English) | NIST MT-05 2005 | BLEU36.78 | 42 | |
| Machine Translation | NIST MT 06 2006 (test) | BLEU37.22 | 27 | |
| Machine Translation | NIST MT 04 2004 (test) | BLEU0.3941 | 27 | |
| Machine Translation | NIST Zh-En All (test) | BLEU Score37.92 | 10 |
Showing 8 of 8 rows