🇮🇳 INDian Cultural commonsense Inventory with Cross-regional Answers (INDICA)

This is the repository for Common to Whom? Regional Cultural Commonsense and LLM Bias in India.

Github Repository

Authors: Sangmitra Madhusudan, Trush Shashank More, Steph Buongiorno, Renata Dividino, Jad Kabbara and Ali Emami

📄 Paper abstract

Existing cultural commonsense benchmarks treat nations as monolithic, assuming uniform practices within national boundaries. But does cultural commonsense hold uniformly within a nation, or does it vary at the sub-national level? We introduce INDICA, the first benchmark designed to test LLMs' ability to address this question, focusing on India—a nation of 28 states, 8 union territories, and 22 official languages. We collect human-annotated answers from five Indian regions (North, South, East, West, and Central) across 515 questions spanning 8 domains of everyday life, yielding 1,630 region-specific question-answer pairs. Strikingly, only 39.4% of questions elicit agreement across all five regions, demonstrating that cultural commonsense in India is predominantly regional, not national. We evaluate eight state-of-the-art LLMs and find two critical gaps: models achieve only 13.4%–20.9% accuracy on region-specific questions, and they exhibit geographic bias, over-selecting Central and North India as the "default" (selected 30-40% more often than expected) while under-representing East and West. Beyond India, our methodology provides a generalizable framework for evaluating cultural commonsense in any culturally heterogeneous nation, from question design grounded in anthropological taxonomy, to regional data collection, to bias measurement.

Dataset Structure

Files

• dataset.json: Complete dataset (515 questions)

Data Format

{
  "question_id": "q_196",
  "question_text": "Are tinted windows allowed on vehicles, and what types of vehicles most commonly have them?",
  "question_domain": "Traffic and transport behavior",
  "question_subcategory": "Streets and traffic",
  "question_topic": "Understanding Local Traffic Regulations",
  "North": {
    "answer": [
      "In North India, tinted windows are not allowed on vehicles."
    ]
  },
  "South": {
    "answer": [
      "In South India, tinted windows are not allowed on vehicles."
    ]
  },
  "East": {
    "answer": [
      "In East India, tinted windows are not allowed on vehicles."
    ]
  },
  "West": {
    "answer": [
      "In West India, tinted windows are not allowed on vehicles."
    ]
  },
  "Central": {
    "answer": [
      "In Central India, tinted windows are not allowed on vehicles."
    ]
  },
  "pairwise_agreements": {
    "South_North": {"agreement": true},
    "South_East": {"agreement": true},
    "South_West": {"agreement": true},
    "South_Central": {"agreement": true},
    "North_East": {"agreement": true},
    "North_Central": {"agreement": true},
    "East_Central": {"agreement": true},
    "North_West": {"agreement": true},
    "East_West": {"agreement": true},
    "West_Central": {"agreement": true}
  },
  "universal_agreement": true
}

Domains (8)

• Interpersonal Relations
• Festivals and Rituals
• Traffic and transport behavior
• Education
• Clothing and adornment
• Food processing and consumption
• Communication
• Finance

Usage

from datasets import load_dataset
import json

# Load via Hugging Face
dataset = load_dataset("Sangmitra-06/INDICA", data_files="dataset.json")

# Or load directly
with open('full_dataset.json') as f:
    data = json.load(f)

Citation

@misc{madhusudan2026commonwhomregionalcultural,
      title={Common to Whom? Regional Cultural Commonsense and LLM Bias in India}, 
      author={Sangmitra Madhusudan and Trush Shashank More and Steph Buongiorno and Renata Dividino and Jad Kabbara and Ali Emami},
      year={2026},
      eprint={2601.15550},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2601.15550}, 
}