NOIRBETTIK: A reading comprehension based multiple choice question answering dataset in Bangla language

Area of Science:

• Natural Language Processing
• Educational Technology
• Computational Linguistics

Background:

• The COVID-19 pandemic accelerated online education, increasing the need for automated learning and assessment tools.
• Multiple-choice questions (MCQs) are crucial for evaluating comprehension in digital learning environments.
• Existing Bangla language resources for context-based NLP tasks, particularly reading comprehension, are scarce.

Purpose of the Study:

• To introduce NOIRBETTIK, a novel, human-made dataset for reading comprehension-based MCQ answering in Bangla.
• To address the critical gap in high-quality Bangla datasets for NLP and educational applications.
• To facilitate the development of advanced automated systems for Bangla language learning and evaluation.

Main Methods:

• Dataset creation involved sourcing authentic Bangla materials (books, articles, biographies).
• Passages were paired with multiple-choice questions, each having four alternatives.
• The dataset's creation and annotation processes were meticulously documented and compared to existing resources.

Main Results:

• The NOIRBETTIK dataset comprises longer passages and contextually relevant MCQs.
• It is a unique, human-generated resource specifically designed for Bangla reading comprehension tasks.
• The dataset's structure and creation methodology are detailed, ensuring transparency and usability.

Conclusions:

• The release of NOIRBETTIK provides a valuable resource for Bangla NLP research.
• This dataset can significantly enhance reading comprehension systems for Bangla-speaking students.
• NOIRBETTIK is poised to drive innovation in educational technologies for the Bangla language community.