深度学习与先验知识结合的英语地名音译技术  

Automatic transliteration technology of English geographical names based on deep learning and prior knowledge

作　　者：王春苗 王继周[2] 毛曦[2] 马维军[2,3] WANG Chunmiao;WANG Jizhou;MAO Xi;MA Weijun(Liaoning Technical University,Fuxin,Liaoning 123000,China;Chinese Academy of Surveying&Mapping,Beijing 100036,China;Capital Normal University,Beijing 100048,China)

机构地区：[1]辽宁工程技术大学测绘与地理科学学院,辽宁阜新123000 [2]中国测绘科学研究院,北京100036 [3]首都师范大学,北京100048

出　　处：《测绘科学》2020年第5期182-188,共7页Science of Surveying and Mapping

基　　金：自然资源部重大项目(12113600000018);科技部重点研发项目(2017-2021)(2017YFB0503600);中国测绘科学研究院基本科研业务经费项目(AR1912)。

摘　　要：针对传统人工地名翻译效率低、且囿于翻译者个人水平差异会导致翻译质量参差不齐,现有机器翻译不能有效解决地名翻译规则的交叉型歧义等问题,该文提出基于深度学习与先验知识相结合的英语地名音译技术,对地名翻译中音标生成、音标优化、音节划分3个核心环节进行研究,提出了基于深度学习的音标生成方法、基于先验知识的音标优化方法和基于双向最大匹配的音节划分方法,解决了机器翻译对地名单词音标识别差、汉字译写不规范的问题。实验结果表明,该地名专名音译技术比传统地名翻译方法时效性高、准确率高、翻译规范。Under the background of "the maintenance and construction of global geographic information resources of China",traditional artificial place name translation is inefficient and limited by the difference of translators’ personal level,which will lead to uneven translation quality. On the other hand,existing machine translation can not effectively solve the cross ambiguity of place name translation rules. This paper puts forward the transliteration technology of English place names based on the combination of deep learning and prior knowledge. It studies three key links in the translation of place names:the generation of phonetic symbols,the optimization of phonetic symbols and the division of syllables. It puts forward the generation method of phonetic symbols based on deep learning,the optimization method of phonetic symbols based on prior knowledge and the division method of syllables based on two-way maximum matching. It solves the problem of machine translation for place names. The problems of poor recognition of phonetic symbols and nonstandard translation and writing of Chinese characters. The experimental results show that the transliteration technology is more efficient,accurate and standardized than the traditional methods.

关 键 词：地名翻译 机器翻译 深度学习 先验知识 双向最大匹配算法 

分 类 号：P208[天文地球—地图制图学与地理信息工程]