稀缺资源语言神经网络机器翻译研究综述  

A Survey on Low-resource Neural Machine Translation

作　　者：李洪政 冯冲[1,2] 黄河燕 LI Hong-Zheng;FENG Chong;HUANG He-Yan(School of Computer Science and Technology,Beijing Institute of Technology,Beijing 100081;Beijing Engineering Research Center of High-Volume Language Information Processing and Cloud Computing Applications,Beijing 100081;School of Foreign Languages,Beijing Institute of Technology,Beijing 102488;Key Laboratory of Language,Cognition and Computation Ministry of Industry and Information Technology,Beijing 102488)

机构地区：[1]北京理工大学计算机学院,北京100081 [2]北京市海量语言信息处理与云计算应用工程技术研究中心,北京100081 [3]北京理工大学外国语学院,北京102488 [4]语言工程与认知计算工信部重点实验室,北京102488

出　　处：《自动化学报》2021年第6期1217-1231,共15页Acta Automatica Sinica

基　　金：中国博士后科学基金(2018M640069);国家自然科学基金(61902024,61732005);国家重点研发计划(2018YFC0832104)资助。

摘　　要：作为目前主流翻译方法的神经网络机器翻译已经取得了很大突破,在很多具有丰富数据资源的语言上的翻译质量也不断得到改善,但对于稀缺资源语言的翻译效果却仍然并不理想.稀缺资源语言机器翻译是目前机器翻译领域的重要研究热点之一,近几年来吸引了国内外的广泛关注.本文对稀缺资源语言机器翻译的研究进行比较全面的回顾,首先简要介绍了与稀缺资源语言翻译相关的学术活动和数据集,然后重点梳理了目前主要的研究方法和一些研究结论,总结了每类方法的特点,在此基础上总结了不同方法之间的关系并分析了目前的研究现状.最后,对稀缺资源语言机器翻译未来可能的研究趋势和发展方向进行了展望,并给出了相关建议.As the mainstream approach in the field of machine translation,neural machine translation(NMT)has achieved great improvements on many rich-source languages,but performance of NMT for low-resource languages ae not very good yet.Low-resource NMT has been one of the most popular issues in MT and attracted wide attention around the world in recent years.This paper presents a survey on low-resource NMT research.We first introduce some related academic activities and feasible data sets for the translation,then categorize and summarize several types of approaches mainly used in low-resource NMT in detail,and present their features,as well as relations between them,and describe the current research status.Finally,we propose some advices on possible research trends and directions of this field in the future.

关 键 词：稀缺资源语言 神经网络机器翻译 迁移学习 枢轴语言 回译 多语言翻译 无监督翻译 

分 类 号：TP391.2[自动化与计算机技术—计算机应用技术] TP183[自动化与计算机技术—计算机科学与技术]