TIMSS试题的机器翻译系统构建及其效果

研究以范例为基础的机器翻译技术和英汉双语对应的结构辅助英汉单句语料的翻译。翻译范例是运用一种特殊的结构，此结构包含来源句的剖析树、目标句的字符串、以及目标句和来源句词汇对应关系。将翻译范例建立数据库，以提供来源句作词序交换的依据，最后透过字典翻译，以及利用统计式中英词汇对列和语言模型来选词，产生建议的翻译。研究是以2010年国际数学与科学教育成就趋势调查测验试题为主要翻译的对象，以期提升翻译的一致性和效率。以NIST和BLEU的评比方式，来评估和比较在线翻译系统和本系统所达成的翻译质量。

The Theoretical Basis and Empirical Research on Translation of TIMSS Testing Contents

The paper takes English-Chinese single sentence corpus translation as researching materials which are based on paradigms machine translation and English-Chinese corresponding auxiliary structure. Translation paradigm consists of parse tree, character string target sentence and source sentence. To build translation paradigm＇s database as the basis for exchanging words sequence of source sentences and then produce ideal translation outcome via dictionary and statistical English-Chinese vocabulary. The empirical research takes TIMSS 2010 testing contents as the translation targets with the methods of NIST and BLEU in order to evaluate and compare online translation system and the system we are researching.