基于深度学习和上下文感知算法的跨语言翻译方法  

Cross Language Translation Method Based on Deep Learning and Context Aware Algorithms

作　　者：王丹[1] WANG Dan(Department of Public Foreign Language Teaching and Research,Harbin Medical University,Daqing,Heilongjiang 163000,China)

机构地区：[1]哈尔滨医科大学公共外语教研部,黑龙江大庆163000

出　　处：《计算技术与自动化》2025年第1期136-140,共5页Computing Technology and Automation

基　　金：教育部全国基础教育外语教学研究资助项目(JJWYZD2019001)。

摘　　要：针对目前跨语言机器翻译存在性能较低的问题,提出了一种基于深度学习和上下文感知算法的跨语言翻译模型。还提出了一个具有12层的跨语言Transformer编码器-解码器结构,充分学习不同语言表达特征。再提出了一种数据对齐方法,可有效扩充数据样本数量,增加数据多样性和数据量,减少数据和模型之间复杂度的相对差异,缓解过拟合问题。实验阶段,通过有监督、无监督、零样本数据集测试,证明所提模型具备较优性能。实验充分验证了所提模型可有效表达未知语种。与m-Transformer和mRASP等主流模型相比,所提模型BLEU分数较优。实验结果证明了所提出跨语言翻译模型的有效性及实用性,该模型可为跨语言机器翻译领域的发展提供一定借鉴作用。A cross language translation model based on deep learning and context aware algorithms is proposed to address the issue of low performance in current cross language machine translation.A cross language Transformer encoder decoder structure with 12 layers was proposed to fully learn different language expression features.A data alignment method has been proposed,which can effectively expand the number of data samples,increase data diversity and volume,reduce the relative differences in complexity between data and models,and alleviate overfitting problems.During the experimental phase,the proposed model demonstrated superior performance through supervised,unsupervised,and zero sample datasets testing.The experiment fully verified that the proposed model can effectively express unknown languages.Compared with mainstream models such as m-Transformer and mRASP,the proposed model has better BLEU scores.The experimental results demonstrate the effectiveness and practicality of the proposed cross language translation model,which can provide some reference for the development of cross language machine translation.

关 键 词：机器翻译 跨语言模型 深度学习 数据对齐 TRANSFORMER 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]