基于自适应寻优控制和多目标学习参数模型的AI人工智能翻译研究  

Research on AI artificial intelligence translation based on adaptive optimization control and multi-objective learning parameter model

作　　者：华琴[1] 赵刚[1] HUA Qin;ZHAO Gang(ShaanXi Institute of Technology,Xi’an 710300,China)

机构地区：[1]陕西国防工业职业技术学院,西安710300

出　　处：《自动化与仪器仪表》2024年第9期33-38,共6页Automation & Instrumentation

基　　金：教育部职业院校外语类专业教学指导委员会2023重点课题(WYJZW-2023TX005)。

摘　　要：在英语翻译软件中,为从句型、时态等多方面提高其人工智能翻译效果,研究在多目标优化的路由分布的基础上,通过网格聚类法,进行特征识别分类,结合多参数融合等方法,进行模糊隶属度函数及多目标学习参数模型的构建,利用多目标优化路由设计,进行相关寻优控制。结果显示,相较于其他方法,研究方法能获得更低的词错误率、更高的重复率、更大的双语评估替补值。在词错误率中,当语句数为627条时,研究算法的词错误率为6.13%,比基于跨语种预训练语言模型的神经机器翻译少23.63%。在重复率中,当样本数为5个时,研究算法、基于多覆盖模型的神经机器翻译的重复率分别为49.24%、36.46%。在双语评估替补值中,研究算法的BLEU值最大,其次为基于跨语种预训练语言模型的神经机器翻译,研究算法的BLEU值为52.36。研究方法能提高翻译软件的准确率、参数估计等多方面性能。To improve the effectiveness of artificial intelligence translation in terms of sentence structure,tense,and other aspects in English translation software,a study was conducted on the basis of multi-objective optimization of routing distribution.Through grid clustering method,feature recognition and classification were carried out,and methods such as multi-parameter fusion were combined to construct fuzzy membership functions and multi-objective learning parameter models.multi-objective optimization of routing design was used for relevant optimization control.The results show that compared to other methods,the research method can achieve lower word error rates,higher repetition rates,and larger bilingual evaluation substitute values.In terms of word error rate,when the number of sentences is 627,the word error rate of the research algorithm is 6.13%,which is 23.63%less than that of neural machine translation based on cross lingual pre trained language models.In terms of repetition rate,when the sample size is 5,the repetition rates of the research algorithm and neural machine translation based on multiple coverage models are 49.24%and 36.46%,respectively.In the bilingual evaluation substitute value,the BLEU value of the research algorithm is the highest,followed by neural machine translation based on cross lingual pre trained language models,with a BLEU value of 52.36.Research methods can improve the accuracy and parameter estimation of translation software in various aspects.

关 键 词：AI人工智能 寻优控制 翻译 多目标 网格聚类 

分 类 号：TP391.2[自动化与计算机技术—计算机应用技术]