基于平行语料的查询翻译词典改进方法研究

Improving the Dictionary of Query Translation Based on Parallel Corpus

面对基于双语词典的跨语言检索查询翻译方法中固有的一对多等翻译模糊问题,已有研究成果存在对于非组合型复合词无法进行准确翻译、双语词典和其他翻译资源联合使用引入较大计算开销等弊端。为建立英汉双向跨语言检索实用性系统,在现有的一部包含若干科技词汇和短语的双语科技词典的基础上,着重研究如何引入平行语料来改进已有的双语词典问题。目标是生成一部基于句对齐平行语料的科技类双语概率词典,为跨语言检索查询翻译消歧提供实时性支持。

Machine readable dictionary （MRD） based translation has been widely used for cross language information retrieval （CLIR）, but it has some shortcomings, especially when one word maps several translation entries. There are two possible drawbacks for the existing studies. One is no correct translation for non-compositional phrases. The other is larger computation consumption when combining with other translation resources. Provided an English-Chinese bilingual dictionary in scientific and technological fields, a novel method to compute translation probability of MRD is proposed. Parallel corpus is foundation for improving the MRD. The new MRD with probability information would be helpful for real-time query translation disambiguation in CLIR.