IndicParam

Paper: https://arxiv.org/pdf/2512.00333 IndicParam is a large-scale, graduate-level benchmark dataset designed to evaluate the performance of Large Language Models (LLMs) on low-resource and extremely low-resource Indic languages. The dataset consists of 13,207 multiple-choice questions (MCQs) collected from official UGC-NET language examination papers and their corresponding answer keys. These questions span 11 Indic languages—Nepali, Marathi, Gujarati, Odia, Maithili, Konkani, Santali, Bodo, Dogri, Rajasthani, and Sanskrit along with an additional Sanskrit–English code-mixed subset. Each data instance represents a single MCQ and includes: - A question in the native script of the target language - Four answer options (A–D) - The correct answer label - Metadata such as subject, exam name, and question type The dataset covers a wide range of question formats, including: - Standard multiple-choice questions - Assertion–Reason - List Matching - Fill in the Blanks - Identify Incorrect Statement - Ordering IndicParam is specifically structured to evaluate both: - Language Understanding (LU): linguistic knowledge such as grammar, syntax, and semantics - General Knowledge (GK): domain knowledge including literature, history, and cultural context All questions are preserved in their original scripts (Devanagari, Gujarati, Odia, and Ol Chiki), ensuring authentic evaluation of multilingual capabilities without reliance on transliteration. The dataset is released as a single test split (13,207 samples) and is intended exclusively for evaluation purposes, enabling standardized and reproducible benchmarking of LLMs across diverse Indic languages. Overall, IndicParam provides a comprehensive and challenging evaluation suite for measuring multilingual understanding, cross-lingual generalization, and cultural competence in modern language models.