融入规则信息的神经机器翻译

(苏州大学计算机科学与技术学院,江苏 苏州 215006)

神经机器翻译; 规则信息; 融合

Integrating rule information into neural machine translation
QIN Wenjie,XIONG Deyi*

(School of Computer Science and Technology,Soochow University,Suzhou 215006,China)

DOI: 10.6043/j.issn.0438-0479.201908023

备注

为了将统计机器翻译技术中的规则信息引入到端到端的神经网络模型中,提出了一种将规则信息转化为近似等价的序列信息的方法.在此基础上,提出了两种融入规则信息的神经机器翻译模型,并在基于注意力机制的循环神经网络(RNN)模型上进行了验证.相对于未融入规则信息的基准模型在美国国家标准与技术研究院(NIST)评测集上的评测结果,上述两种模型的双语互译评估(BLEU)值均有所提高.实验表明,将规则等外部知识融入到神经机器翻译系统中是提升模型翻译质量的一种有效途径.

Neural machine translation is currently the most popular research method in the field of machine translation.Introduction of external knowledge into the neural machine translation system has become a research hotspot in this field.To introduce the rule information in statistical machine translation technology into the end-to-end neural network model,we have developed a method to convert the rule information into approximately equivalent sequence information.On this basis,we propose two neural machine translation model frameworks that incorporate rule information,and validate these methods on the attention-based RNN model and find that these methods can achieve a better BLEU point on the NIST evaluation set.Experimental results show that integrating external knowledge such as rules into the neural machine translation system is an effective way to improve the quality of model translation.