浅议多种重映射融合的汉英词对齐法  

An Analysis of Chinese-English Word Alignment Method for Multiple Remapping Fusion

作　　者：谢庚全[1] XIE Geng-quan(School of Foreign Languages, Hainan University, Haikou 570228, China)

机构地区：[1]海南大学外国语学院,海南海口570228

出　　处：《海南广播电视大学学报》2019年第2期29-33,共5页Journal of Hainan Radio & TV University

基　　金：2016年海南省自然科学基金项目"基于多预处理机制的多种重映射融合汉英自动词对齐系统研究-以海南旅游文本汉英翻译网上平行语料库创建为例"(编号:20167238)成果之一;2016年海南省哲学社会科学规划课题"海南城市外宣翻译的跨文化文本重构研究"(编号:HNSK(QN)16-134)成果之一

摘　　要：针对自动词对齐工具Giza++只允许源语言到目标语言的一对多映射,并生成了很多不对称的对齐,进而直接影响到词对齐的质量和准确性这一缺陷,文章通过研究发现,基于不同预处理机制的词对齐有着不同的系统上可见优势,相对于采用单一预处理机制,机器学习算法可以从基于多预处理机制的词对齐信息中获益。在此基础上,提出基于多预处理机制的多种重映射融合词对齐方法这一设想,并通过实验验证:通过分词预处理形成尽可能含有正确分词方案的方案集,通过对齐预处理获得尽可能多的可靠对齐点,并通过对齐重映射实现对齐的对称化,随后,将对齐重映射的所有相关特征训练一个对齐融合模型,并将这个对齐融合模型作为监督系统,以显著增加词对齐的准确性。Giza++, an automatic word alignment tool, which only allows one-to-many mapping from source language to target language and generates many asymmetric alignments, will directly affect the quality and accuracy of word alignment. In order to resolve this problem, this paper finds that: firstly, based on different preprocessing mechanisms, the word alignment has different systematic advantages;secondly, compared to the single preprocessing mechanism, the machine learning algorithm can benefit from word alignment information based on multiple preprocessing mechanisms. What’s more, this paper proposes the idea of multiple remapping fusion word alignment method based on multiple pre-processing mechanisms. It is verified by experiments that this program set with correct word segmentation scheme is formed by word segmentation preprocessing and could obtain reliable alignment points as many as possible by alignment preprocessing. Meanwhile, it can achieve symmetry of alignment through alignment remapping. Then, it will train all the relevant features of the alignment remapping with an alignment fusion model, and put this model as the supervision system to significantly increase the accuracy of word alignment.

关 键 词：汉英词对齐 Giza++工具 多预处理机制 重映射融合 

分 类 号：H315.9[语言文字—英语]