Research on the Application of Corpus Machine Translation Model Integrating NMT Model and PBSMT Model
李静莹1
扫码查看
点击上方二维码区域，可以放大扫码查看
作者信息
1. 咸阳师范学院,陕西咸阳 712000
折叠
摘要
为了解决传统英语翻译模型存在的效率低、语料库内容覆盖范围较小等问题,研究将神经网络机器翻译(Neural Net-work Machine Translation,NMT)与短语统计机器翻译模型(Phrase Based Statistical Machine Translation,PBSMT)进行融合,用于构建英语翻译模型.研究首先利用NMT构建无监督英语翻译模型,然后将其与PBSMT进行融合,最后利用仿真实验和实际应用来验证翻译模型的性能.结果表明NMT+PBSMT在迭代至895次时,损失值趋于平稳,BLEU评估分数为0.87,同时在数据集 TED2020、UM-Corpus、UNv5.0 中的训练时间和 BLEU 分数分别为 1.97 h、1.72 h、1.85 h 和0.90、0.96、0.93,结果均优于对比翻译模型.这说明研究构建的NMT+PBSMT翻译模型具有较高的有效性和可行性,能够为英语翻译提供有效支持和帮助.
Abstract
In order to solve the problems of low efficiency and limited coverage of corpus content in traditional English translation models,the neural network machine translation method(NMT)and phrase based statistical machine translation model(PBSMT)were fused to construct an English translation model.The study first utilized NMT to construct an unsupervised English translation model,then fused it with PBSMT.Finally,simulation experiments and practical applications were used to verify the performance of the translation model.The results showed that the loss values of NMT+PBSMT tended to stabilize at 895 iterations,with a BLEU eval-uation score of 0.87.At the same time,the training time and BLEU scores in datasets TED2020,UM Corpus,and UNv5.0 were 1.97h,1.72 h,1.85 h,and 0.90,0.96,and 0.93,respectively.The results were better than those of the comparative translation model.This indicates that the NMT+PBSMT translation model constructed in the study has high effectiveness and feasibility,and can provide effective support and assistance for English translation.
维普
万方数据