文章摘要
罗兰,何贤敏,李茂西.句子级别机器译文质量估计研究综述[J].情报工程,2022,8(2):034-050
句子级别机器译文质量估计研究综述
Review of Sentence-level Quality Estimation of Machine Translation
  
DOI:10.3772/j.issn.2095-915X.2022.02.003
中文关键词: 机器翻译;译文质量估计;深度神经网络;预训练语言模型;评价指标
英文关键词: Machine translation; quality estimation; deep neural networks; pre-trained language models; evaluation metrics
基金项目:国家自然科学基金“自动后处理和语法错误校正驱动的译文质量提高方法”(61662031);江西省教育厅研究生创新 基金“基于 Siamese BERT 网络的科技文献 IPC 和 CLC 类目映射研究”(YC2021-S239)。
作者单位
罗兰 1. 江西师范大学计算机信息工程学院 南昌 330022; 
何贤敏 1. 江西师范大学计算机信息工程学院 南昌 330022; 
李茂西 1. 江西师范大学计算机信息工程学院 南昌 330022;2. 江西师范大学管理科学与工程研究中心 南昌 330022 
摘要点击次数: 424
全文下载次数: 305
中文摘要:
      [ 目的 / 意义 ] 近年来,随着估计结果与人工评价的相关性逐步增强,译文质量估计引起了机器翻译研究者们的广泛关注和高度重视,因此对句子级别译文质量估计方法进行归纳整理具有重要意义。[ 方法 / 过程 ] 该文主要是将它们分为基于传统机器学习的方法、基于神经翻译模型的方法和基于预训练语言模型的方法,梳理和对比了这三种译文质量估计方法的代表性工作、交叉工作以及不同方法的发展路线,并介绍了推动译文质量估计研究的相关评测活动和性能评价指标,最后展望句子级别译文质量估计今后的研究方向和发展趋势,并对全文进行总结。[ 结果 / 结论 ] 机器译文质量估计未来的研究方向将会越来越成熟,应用场景也会越来越广泛,有逐步取代传统译文自动评价任务的趋势。
英文摘要:
      [Purpose/Significance] In recent years, with the correlation between estimation results and manual evaluation gradually increasing, translation quality estimation has attracted wide attention and great importance from machine translation researchers, so it is important to summarize and organize the sentence-level translation quality estimation methods. [Method/Process] The paper mainly classifies them into traditional machine learning-based methods, neural translation model-based methods and pre-trained language modelbased methods, compares and contrasts the representative works of these three translation quality estimation methods, the crossover works and the development routes of different methods, and introduces the relevant evaluation activities and performance evaluation indexes that drive the translation quality estimation research, and finally looks forward to the sentence Finally, We look forward to the future research directions and development trends of translation quality estimation at sentence level, and conclude the whole paper. [Results/Conclusion] The future research direction of machine translation quality estimation will become more and more mature, and the application scenarios will become more extensive, with a tendency to gradually replace the traditional translation automatic evaluation tasks.
查看全文   查看/发表评论  下载PDF阅读器
关闭

分享按钮