摘要
针对传统机器翻译准确率不高的问题,提出一个基于预训练语言模型和迁移学习的机器翻译优化模型。首先,利用迁移学习将BERT预训练语言模型迁移至传统Transformer机器翻译模型中,从而快速完成融合模型训练;然后对融合BERT的机器翻译优化模型进行训练直到模型开始收敛;最后,对融合BERT的机器翻译优化模型进行测试。实验结果表明:经过BERT预训练语言模型优化的机器翻译BLEU值一直维持在30以上,且最高达32.92;融合BERT的优化模型对长句的翻译效果平均准确度为91.75%,短句翻译平均准确度为96.05%,比优化前的翻译准确度分别提高了7.88%、5.46%,大幅提高了机器翻译效果,验证了基于语言特征和迁移学习的机器翻译优化模型的可行性与有效性。
A machine translation optimization model based on BERT pre-trained language model and transfer learning is proposed to address the issue of low accuracy in traditional machine translation.Firstly,transfer learning is used to transfer the BERT pre-trained language model to the traditional Transformer machine translation model,thereby quickly completing the fusion model training;Then train the machine translation optimization model fused with BERT until the model begins to converge;Finally,the machine translation optimization model integrating BERT was tested.The experimental results show that the BLEU value of machine translation optimized by BERT pre-trained language model remains above 30,with a maximum of 32.92;The optimization model that integrates BERT has an average accuracy of 91.75%for long sentences and 96.05%for short sentences,respectively,which is 7.88%and 5.46%higher than the translation accuracy before optimization.This greatly improves the machine translation effect and verifies the feasibility and effectiveness of the machine translation optimization model based on language features and transfer learning.
作者
薛俊杰
XUE Junjie(Xi’an Si Yuan University,Xi’an Shaanxi 710038,China)
出处
《自动化与仪器仪表》
2023年第10期183-186,共4页
Automation & Instrumentation
基金
西安思源学院校级项目基金《生态翻译学视角下<陈忠实散文选译>研究》(XASY-A1721)。
关键词
迁移学习
语言特征
中英翻译
机器翻译
BERT预训练语言模型
transfer learning
linguistic features
Chinese-English translation
Machine translation
BERT pre-trained language model
