Expected answer type construction using analogical reasoning in a question answering task

In a question answering system (QAS), question analysis component has an important task to determine the expected answer type (EAT) of a given question. Many QAS´s rely their question analysis performance on manually developed patterns, such as in Open Ephyra (OE), one of a state of the art freely available QAS. Recently, there are a number of studies which investigated the influence of statistical relational framework to learn question-answer pairs in particular component of a QAS. In this study, we propose an approach that utilizes the intensity of statistical learning of question-answer pairs as a means to develop EAT patterns. In a question analysis experiment setting by using factoid testing questions from QA@CLEF 2008, our result outperforms the accuracy of manually constructed patterns of OE, with 84.17% against 81.67%.