基于神经网络的维汉翻译系统实现  

Implementation of Uyghur-Chinese translation system based on neural network

作　　者：张胜刚 艾山.吾买尔[1,2] 吐尔根.依布拉音 买合木提.买买提 米尔夏提.力提甫 ZHANG Shenggang;Hasan Wumaier;Tuergen Yibulayin;Mahmut Maimaiti;Mirxat Litip(School of Information Science and Engineering,Xinjiang University,Urumqi 830046,China;Xinjiang Laboratory of Multi.Language Information Technology,Urumqi 830046,China)

机构地区：[1]新疆大学信息科学与工程学院,新疆乌鲁木齐830046 [2]新疆多语种信息技术重点实验室,新疆乌鲁木齐830046

出　　处：《现代电子技术》2018年第24期157-161,共5页Modern Electronics Technique

基　　金：国家重点基础研究发展计划(2014CB340506);国家自然科学基金(61463048);国家自然科学基金(61331011);国家自然科学基金(61662077);国家自然科学基金(61462083);新疆多语种信息技术实验室开放课题(2016D03023);"自治区青年科技创新人才培养工程"青年博士项目(QN2015BS004)~~

摘　　要：小语种的机器翻译由于开发成本和用户规模等原因一般在开源系统的基础上实现在线服务系统。目前,神经机器翻译提供的源代码以theano编写的居多,但是theano编写的机器翻译因翻译速度较慢而无法满足用户需求。在此以如何实现基于theano的稳定维汉神经网络机器翻译系统为目标开展研究,翻译模型采用ALU神经元的多层双向网络框架,使用django实现了翻译服务接口,选用nginx+uwsgi实现了负载均衡以提高翻译速度。实验结果表明,相比于5个翻译引擎构成的系统,10个翻译引擎的系统翻译速度得到1.3～1.55倍的提高。该文的研究结果对于利用开源系统快速实现能够满足日访问量1 000万次以下的翻译系统具有重要参考价值。The machine translation of minority languages is generally to achieve online service system on the basis of the open source system due to reasons such as development costs and user scale.At present,the source codes provided by neural machine translation are mostly written by the theano,but the machine translation written by the theano cannot meet users′needs due to its slow translation speed.How to realize the stable Uyghur-Chinese neural network machine translation system based on the theano is taken as the research object.The multi-layer bi-directional network framework of the ALU neuron is used as the translation model.The django is used to realize the translation service interface.The nginx+uwsgi is selected to achieve load balancing,so as to improve the translation speed.The experimental results show that,the translation speed of the system constituted by 5 translation engines is 1.3-1.55 times higher than that of the system constituted by 10 translation engines.The re-search results in this paper have an important reference value for using the open source system to rapidly realize the translation system that can meet daily visits of less than 10 million times per day.

关 键 词：小语种 机器翻译 theano 神经网络 开源系统 负载均衡 

分 类 号：TN711-34[电子电信—电路与系统]