基于模型不确定性约束的半监督汉缅神经机器翻译(Semi-Supervised Chinese-Myanmar Neural Machine Translation based Model-Uncertainty)
Linqin Wang (王琳钦), Zhengtao Yu (余正涛), Cunli Mao (毛存礼), Chengxiang Gao (高盛祥), Zhibo Man (满志博), Zhenhan Wang (王振晗)
Abstract
基于回译的半监督神经机器翻译方法在低资源神经机器翻译取得了明显的效果,然而,由于汉缅双语资源稀缺、结构差异较大,传统基于Transformer的回译方法中编码端的Self-attention机制不能有效区别回译中产生的伪平行数据的噪声对句子编码的影响,致使译文出现漏译,多译,错译等问题。为此,该文提出基于模型不确定性为约束的半监督汉缅神经机器翻译方法,在Transformer网络中利用基于变分推断的蒙特卡洛Dropout构建模型不确定性注意力机制,获取到能够区分噪声数据的句子向量表征,在此基础上与Self-attention机制得到的句子编码向量进行融合,以此得到句子有效编码表征。实验证明,本文方法相比传统基于Transformer的回译方法在汉语-缅甸语和缅甸语-汉语两个翻译方向BLEU值分别提升了4.01和1.88个点,充分验证了该方法在汉缅神经翻译任务的有效性。- Anthology ID:
- 2021.ccl-1.4
- Volume:
- Proceedings of the 20th Chinese National Conference on Computational Linguistics
- Month:
- August
- Year:
- 2021
- Address:
- Huhhot, China
- Editors:
- Sheng Li (李生), Maosong Sun (孙茂松), Yang Liu (刘洋), Hua Wu (吴华), Kang Liu (刘康), Wanxiang Che (车万翔), Shizhu He (何世柱), Gaoqi Rao (饶高琦)
- Venue:
- CCL
- SIG:
- Publisher:
- Chinese Information Processing Society of China
- Note:
- Pages:
- 35–45
- Language:
- Chinese
- URL:
- https://preview.aclanthology.org/build-pipeline-with-new-library/2021.ccl-1.4/
- DOI:
- Cite (ACL):
- Linqin Wang, Zhengtao Yu, Cunli Mao, Chengxiang Gao, Zhibo Man, and Zhenhan Wang. 2021. 基于模型不确定性约束的半监督汉缅神经机器翻译(Semi-Supervised Chinese-Myanmar Neural Machine Translation based Model-Uncertainty). In Proceedings of the 20th Chinese National Conference on Computational Linguistics, pages 35–45, Huhhot, China. Chinese Information Processing Society of China.
- Cite (Informal):
- 基于模型不确定性约束的半监督汉缅神经机器翻译(Semi-Supervised Chinese-Myanmar Neural Machine Translation based Model-Uncertainty) (Wang et al., CCL 2021)
- PDF:
- https://preview.aclanthology.org/build-pipeline-with-new-library/2021.ccl-1.4.pdf