摘要
神经网络机器翻译技术模拟人脑神经系统,以深度学习技术为基础,将整个句子作为翻译的基本单元,使得机器翻译的准确率大大提升。谷歌、百度、腾讯三家公司推出的翻译软件都采用了神经网络翻译技术,通过比较可知三者的长句翻译能力突出,译文风格各具特色,但在语言方面都仍面临困境。首先,因语言差异导致的漏译误译;其次,因词义多义引起的选词障碍;第三,因语境因素带来的隐性含义的理解。本文认为只有不断完善深度学习的算法,建立纠错数据库,进行跨学科交流,才能进一步提升机器翻译质量。
Neural Machine Translation(NMT) technology, based on deep learning technology, simulates the human brain nerve system, and takes the whole sentence as the basic unit of translation, which greatly improves the accuracy of machine translation. The translation softwares launched by Google, Baidu, and Tencent all use neural network translation technology. By comparing the three softwares, all of them are excellent in translating long sentence and each has distinctive translation styles. However, they still are faced with troble,such as mistranslations and omission due to language differences, obstacles caused by polysemy, and the difficulty in understanding of implicit meanings caused by contextual factors. To further improve the quality of machine translation, the algorithm of deep learning should be perfected, the error correction database should be established, and interdisciplinary communication should be carried out.
引文
[1]胡壮麟.语言学教程[M].北京:北京大学出版社,2013,4(4):225.
[2]史树敏.机器翻译方法的研究现状[J].内蒙古师范大学学报,2004,33(2):166.
[3]毛颢.若干实用机器翻译软件与网站评介[J].中国科技翻译,2004,17(4):32.
[4]张政.机器翻译难点所在[J].外语研究2005,93(5):61.
[5]荣晶.语义在汉语语法研究中的方法论意义[J].北京师范大学学报(人文社会科学版),2000,106(4):87.