Learning to classify Marathi questions and identify answer type using machine learning technique

Abstract

One of the budding fields of artificial intelligence is Question Answering (QA). QA is a type of information retrieval in which a set of documents is given, and a QA system attempts to search for the correct answer to the question posed in natural language. Question classification (QC), which is a part of QA system, helps to categorize each question. In QC, the entity type of the answering sentence for a given question in natural language is predicted. QC is a very crucial step in QA system as it helps to take the important decision. For example, QC helps to reduce the possible options of the answer, and thus the answers that match the question class are to be considered. This research takes the first step toward the development of QC system for English-Marathi QA system. This system analyzes the user's question and deduces the expected Answer Type (AType), for which a dataset of 1000 questions from Kaun Banega Crorepati (KBC) was scrapped and manually translated into Marathi. Right now, the result for translation approach for the coarse-grained class is 73.5 percent and the fine-grained class is 47.5 percent, and for the direct approach, it is 56.5 and 30.5 percent for coarse and fine, respectively. Experiments are going on to improve the results.