Bengali Question Classification: Towards Developing QA System

This paper demonstrates the question classification step towards building a question answering system in Bengali. Bengali is an eastern Indo-Aryan language with about 230 million total speakers and one of the most spoken languages in the world. An important first step in developing a question answering system is to classify natural language question properly. In this work, we have studied suitable lexical, syntactic and semantic features to classify the Bengali question. As Bengali question classification is at early stage of development, so for simplicity we have proposed single-layer taxonomy which consists of only nine course-grained classes. We have also studied and categorized the interrogatives in Bengali language. The proposed automated classification work is based on various machine learning techniques. The baseline system based on Naïve Bayes classifier has achieved 80.65% accuracy. We have achieved up to 87.63% accuracy using decision tree classifier.