Bengali Question Answering System for Factoid Questions: A statistical approach

Question answering system in recent days is one of the most trending and interesting topics of research in computational linguistics. Bengali being among the most spoken languages in the world has yet faced difficulties in computational linguistics. This paper demonstrates an attempt to develop a closed domain factoid question answering system for Bengali language. Our proposed system combining multiple sources for answer extraction extracts the answer having the accuracy 66.2% and 56.8% with and without mentioning the object name respectively. The system also hits around 72% documents from which the answer can be extracted. Besides the sub-parts of our system, the question and document classifier provides 90.6% and 75.3% accuracy respectively over five coarse-grained categories.