HNC基本句类表示式在汉英机器翻译中的应用研究

Application of HNC Basic Sentence Class Expressions in Chinese-English Machine Translation

神经机器翻译发展至今,已经成为目前应用最为广泛的机器翻译方法,其优秀的性能依赖于丰富的语料库和庞大的计算机算力。但通过此方法,在语料库稀缺或者数据算力达到一定规模的情况下,进一步提高翻译质量十分困难。为更好的解决上述问题,实现融入语言学知识的机器翻译是一种重要的方法。本文提出了一种对HNC理论中基本句类语义块表示式进行边界识别,在汉英机器翻译中融入句类知识的方法。首先对属于HNC理论基本句类的句子进行分词和词性标注,根据词性标注的结果进行特征语义块识别和句类分析,然后再由句类表示式对整个句子的语义块表示式进行边界识别。最后设计了一种融入句类知识的汉英神经机器翻译模型,将每个边界划分后的对象单独计算注意力和词向量计算的注意力进行相加,做为输入训练翻译模型。

Up to now,neural machine translation has become the most widely used machine translation method.Its excellent performance depends on rich corpus and huge computer computing power.However,with this method,it is very difficult to further improve the translation quality when the corpus is scarce or the data computing power reaches a certain scale.In order to better solve the above problems,it is an important method to realize machine translation integrating linguistic knowledge.This paper proposes a method to recognize the boundary of the semantic block representation of basic sentence classes in HNC theory and integrate sentence class knowledge into Chinese-English machine translation.Firstly,the sentences belonging to the basic sentence class of HNC theory are segmented and labeled,and the feature semantic block recognition and sentence class analysis are carried out according to the results of part of speech labeling.Then,the boundary of the semantic block representation of the whole sentence is recognized by the sentence class representation.Finally,a Chinese-English neural machine translation model integrating sentence knowledge is designed.The attention calculated by each boundary object separately and the attention calculated by word vector are added as the input training translation model.