摘要
本文首先简介神经机器翻译,解析其基本原理、主流框架、注意力机制;其次探讨基于注意力机制的神经机器翻译存在的主要问题:集外词翻译、长难句翻译、小语种翻译,并就此问题及最新洞见予以阐述、探究、评析;最后展望基于注意力机制的神经机器翻译的未来发展态势:走向文档级翻译、多语言翻译、多模态翻译及无监督翻译。
First,the mainstream framework and attention mechanism of neural machine translation are introduced.Then,the frontiers and challenges mainly comprising the untranslatability of out-of-vocabulary words,poor quality of long-sentence rendering,and difficulty in minority-language translating,of current attention-based neural machine translation are unveiled via an exploration of its development status.Finally,the future orientations,particularly,of moving to the document-level,multilingual,multimodal and unsupervised translation of attention-based neural machine translation are prospected.
作者
侯强
侯瑞丽
Hou Qiang;Hou Rui-li(College of Foreign Languages,Nankai University,Tianjin 300071,China;Zhongshan College,Inner Mongolia Fengzhou Vocational College,Hohhot 011517,China)
出处
《外语学刊》
CSSCI
北大核心
2021年第5期54-59,共6页
Foreign Language Research
基金
国家社科基金重大项目“双语术语知识库建设与应用研究”(15ZDB102)的阶段性成果。
关键词
神经机器翻译
注意力机制
人工神经网络
循环神经网络
序列到序列模型
编码器—解码器框架
neural machine translation
attention mechanism
artificial neural networks
recurrent neural networks
sequence-to-sequence model
encoder-decoder framework