Automatic question answering for Turkish with pattern parsing

Abstract

In this study we have explained the implementation details of a Automatic Question Answering System for Turkish. Automatic Question Answering (QA) is a sub-field of information retrieval (IR) that deals with finding the answers of submitted free text questions. Automatic QA is important for daily life because, it retrieves the answers or answer sets of submitted questions without wasting time for searching the entire document corpus that include the answers. The important issue in QA systems is to extract the features of each sentence to create their semantic meaning. In this study, documents are processed with pattern matching techniques and features of documents are extracted for question answering system. We propose to categorize the questions automatically and then select the correct answer from corresponding answer set. We have used name entity recognition and pattern matching to categorize the type of questions. Additionally we have used range queries to submit the questions. Instead of using traditional distance metrics we have proposed a new approach to rank the similar documents. Each approach has been evaluated and described in this study.