基于多粒度与动态词向量的机器翻译关键技术研究  

Research on Key Technologies of machine translation based on multi granularity and dynamic word vector

作　　者：王国英[1] WANG Guoying(The Modern College of Northwest University,Xi’an 710310,China)

机构地区：[1]西北大学现代学院,西安710310

出　　处：《自动化与仪器仪表》2022年第9期181-185,共5页Automation & Instrumentation

基　　金：2021年陕西省新文科研究与改革实践项目《基于在线自建课程的“混合式教学模式”与“混合式一流课程”建设实践探索》;西北大学现代学院校级教育教学改革研究项目(21JG02)的阶段性研究成果。

摘　　要：针对现有机器翻译模型翻译质量低的问题,通过结合多粒度特征融合与基于ELMo模型的动态词向量,并以采用多粒度位置编码和多粒度自注意力机制改进的Transformer模型作为主干网络,构建了一种基于多粒度与动态词向量的机器翻译模型。仿真结果表明,所提模型提高了机器翻译模型性能,在WMT2019Zh-En数据集上和NIST数据集上分别BLEU值分别达到31.53和42.61,相较于单一粒度特征输入和静态词向量嵌入,平均提高了1.1和1.39,具有一定的有效性和优越性。Aiming at the problem of low translation quality of existing machine translation models, a machine translation model based on multi granularity and dynamic word vector is constructed by combining multi granularity feature fusion and dynamic word vector based on Elmo model, and taking the transformer model improved by multi granularity location coding and multi granularity self attention mechanism as the backbone network. The simulation results show that the proposed model improves the performance of the machine translation model. The Bleu values on wmt2019 zh en dataset and NIST dataset reach 31.53 and 42.61 respectively. Compared with single granularity feature input and static word vector embedding, the proposed model improves by 1.1 and 1.39 on average, which has a certain effectiveness and superiority.

关 键 词：多粒度特征 动态词向量 Transformer模型 机器翻译 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]