Automatic question generation for subordinate conjunctions of Marathi

• Published in: 2022 8th International Conference on Advanced Computing and Communication Systems (ICACCS) Vol. 1; pp. 169 - 173
• Main Authors: Patil, Pushpa M., Bhavsar, R.P., Pawar, B.V.
• Format: Conference Proceeding
• Language: English
• Published: IEEE 25.03.2022
• Subjects: Automatic Question generation (AQG); Communication systems; Error analysis; Manuals; Marathi subordinate conjunctive questions; POS; Semantics; Syntactics; Systematics; Tagging
• ISSN: 2575-7288
• DOI: 10.1109/ICACCS54159.2022.9785063

As the internet is growing fast, a lot of multilingual e-content is now easily accessible; this e-content contains information from various domains. This data is being used for a wide range of uses and applications. As systematic evaluation and appraisal methods are evolving, so does the need for a large Question Bank (QB). Manual question creation is a time-consuming and costly process that necessitates domain experts, and there is an increasing movement toward using question banks to generate the question papers for various tests and assessments. As a result, the issue of an automated question generation (AQG) has piqued the interest of the researchers; this problem can be resolved using the Natural Language Processing (NLP) techniques. The reported study here addresses the issues involved in automatic question generation in the context of subordinate conjunctions of Marathi text. We have investigated the utility of conjunctions in the context of AQG at sentence level. Our AQG engine comprises of three stages: question word selection, clause selection and automatic question generation. The central idea of our AQG methodology is to generate the questions at the parts of speech (POS) tagging level only. We have tested our AQG methodology on a test corpus of 100 Marathi complex sentences. To test our AQG methodology, we have used six subordinating conjunctions. Performance of our AQG methodology is measured with manually evaluated syntactic and semantic correctness. We have obtained the precision of 0.83 and a recall of 0.80 on the corpus chosen from standard sixth, seventh and eighth science books. The error analysis is also presented in this paper.