基于Transformer和视觉特征的多模态机器实时翻译研究  

Multimodal Machine Real time Translation Based on Transformer and Visual Features

作　　者：于英俊 YU Yingjun(Xi’an Fanyi University,Xi’an 710105,China)

机构地区：[1]西安翻译学院,西安710105

出　　处：《自动化与仪器仪表》2024年第10期253-257,共5页Automation & Instrumentation

基　　金：陕西省教育厅科学研究计划项目资助《延安红色旅游翻译与传播路径研究》(23JP053);2022年度陕西高校青年创新团队“中华优秀文化翻译与国际传播创新团队”阶段性成果。

摘　　要：景点导览翻译器对于旅游业发展和跨文化传播具有重要意义。针对传统机器翻译模型无法对未来的语义信息进行预测、在实时翻译中输入资源较为缺乏的问题,研究搭建了基于Transformer和视觉特征的多模态机器实时翻译模型。结果表明,多模态机器翻译模型在Evaluation数据集英语-印地语中的双语评估替补值为42.7,英语-豪萨语中的双语评估替补值为46.9。基于视觉特征的实时机器翻译模型在大多数情况下双语评估替补指标更高。综上所述,所提模型具有较高的翻译准确性,能够对未来上下文进行了有效的预测,模型具有更好的翻译质量,能够有效提高实时机器翻译的质量和性能。研究能够为景点导览翻译器提供一定的技术支持,促进跨文化传播的有效开展。The scenic spot guide translator is of great significance for the development of the tourism industry and cross-cultural communication.In response to the problem that traditional machine translation models cannot predict future semantic information and lack input resources in real-time translation,a multi-modal machine real-time translation model based on Transformer and visual features has been studied and constructed.The results showed that the bilingual evaluation substitute value of the multi-modal machine translation model in the Evaluation dataset was 42.7 in English Hindi,and 46.9 in English Hausa.Real time machine translation models based on visual features tend to have higher substitute indicators for bilingual evaluation in most cases.In summary,the proposed model has high translation accuracy and can effectively predict future contexts.The model has better translation quality and can effectively improve the quality and performance of real-time machine translation.This study can provide certain technical support for scenic spot guide translators and promote effective cross-cultural communication.

关 键 词：机器翻译 视觉特征 多模态 实时翻译 景点导览翻译器 

分 类 号：H059[语言文字—语言学] TP39[自动化与计算机技术—计算机应用技术]