摘要
为实现多语言聊天场景中的机器翻译,提出了一种融合先验知识的机器翻译方法及系统。首先,搭建多聊天系统整体框架,该框架包含应用层、表示层和数据层;其次,为实现多语言聊天,构建BPE+TextCNN的语言检测模型;然后搭建以Transformer翻译模型为基础的Seq2Seq翻译框架,并增加BPE embedding层和先验知识编码器,以此解决传统翻译过度依赖语料库的问题,实现语料库的自学习和训练。结果表明,BPE+TextCNN语言检测模型对语言检测的准确率达98.1%;融合先验知识的机器翻译模型对中-英和英-中互译的BLEU得分达29.75和38.16;系统测试得出,基于机器翻译的多语言聊天系统线上运行良好,用户满意度高达98%。由此表明本研究构建的多语言聊天技术可行,可实现多语种对话翻译。
In order to implement machine translation in multi language chat scenarios,a machine translation method and system integrating prior knowledge were proposed.First,build the overall framework of the multi chat system,which includes the application layer,presentation layer and data layer;Secondly,to achieve multilingual chat,a language detection model based on BPE+TextCNN is constructed;Then,a Seq2Seq translation framework based on the Transformer translation model is built,and a BPE embedding layer and a prior knowledge encoder are added to solve the problem of traditional translation relying excessively on corpora,achieving self-learning and training of corpora.The results show that the BPE+TextCNN language detection model has an accuracy of 98.1%in language detection;The BLEU score of the machine translation model integrating prior knowledge for C-E and E-C translation reached 29.75 and 38.16 respectively;The system test shows that the multilingual chat system based on machine translation works well online,and the user satisfaction is up to 98%.This indicates that the multilingual chat technology constructed in this study is feasible and can achieve multilingual dialogue translation.
作者
杨旭
YANG Xu(Harbin Medical University,Daqing,Heilongjiang 163000,China)
出处
《自动化与仪器仪表》
2023年第6期184-187,共4页
Automation & Instrumentation
