机翻字幕质量评估的模型构建研究——以“人人译视界”和“讯飞听见字幕”为例  

The Proposition of Machine-generated Subtitling Quality Assessment Model——Case Studies of“YSJ”and“Iflyrec”

作　　者：雷静[1] 李明明 LEI Jing;LI Ming-ming(School of Foreign Studies,Minzu University of China,Beijing 100081,China)

机构地区：[1]中央民族大学外国语学院,北京100081

出　　处：《大连民族大学学报》2023年第2期176-182,共7页Journal of Dalian Minzu University

基　　金：中央民族大学校级项目(2022QNYL14)。

摘　　要：AI时代神经网络翻译技术的飞速发展催生了一批视频译制平台。以“人人译视界”和“讯飞听见字幕”为研究对象,以动画作品《中华勤学故事》机翻字幕为研究素材,探究国内视频译制平台机翻字幕质量。基于FAR语际字幕评估模型提出了机翻字幕评估模型(FAR 2.0),从功能对等、接受程度、阅读体验三方面对机翻字幕的语义错误、术语翻译错误、译文冗余等问题进行了深入分析。根据研究结果,提出增强语音数据标注积累、加大机器翻译语料训练及强化翻译引擎的意群切分等三点优化建议。With the rapid development of neural machine translation technology in the AI era,a number of video translation platforms have emerged.In order to evaluate the quality of machine-generated subtitling,this paper takes“YSJ”and“Iflyrec”as the research objects,and the machine-generated subtitles of the cartoon“Chinese Stories of Study”as the material when doing the assessing.This study tries to propose a model for assessing machine-generated subtitle(FAR 2.0)based on the FAR model by Jan Pedersen,and analyses semantic error,term error,and redundancy phenomenon from three parameters:functional equivalence,acceptability and readability.This paper gives the following three suggestions for optimization:firstly,enhance the accumulation of speech data annotation;secondly,increase data training on the various corpus;thirdly,enhance the segmentation and combination ability of the translation engine.

关 键 词：机翻字幕 FAR FAR 2.0 人人译视界 讯飞听见字幕 翻译质量评估 

分 类 号：H059[语言文字—语言学]