Multi Lingual Question Answering System Using Word Relationship Expressions

为了实现多语言问答系统，需要多个单语言问答系统作为其组成部分。然而，基于规则的方法对于准备多个单语问题系统来说过于昂贵。因此，本研究采用机器学习的方法来准备它们，而不是基于规则的方法。我们表明，机器学习方法可以有效地以更低的成本准备问答系统，比基于规则的方法具有竞争力的性能。为了提高它们的性能，我们进行了两项研究：一是大型双语词典，以提高关键词翻译性能；二是分析关键词之间的函数关系，以提高答案选择性能。

In order to realize a multilingual question answering system, several monolingual question systems will be required as its components. However, rule based approach is too expensive to prepare several monolingual question systems. Therefore, machine learning approach is employed to prepare them in this study, instead of rule based approach. We showed that machine learning approach is effective to prepare question answering systems in cheaper costs than rule based approaches with competitive performance. In order to improve their performances, two researches were conducted: large scale bilingual dictionary to improve keyword translation performance, and analysis of functional relationship between keywords to improve answer selection performance.