CoolFace
Datasetpublic

Kirtibg/TALES-QA

TALES-QA: Cultural Knowledge Question Bank This repository contains the question bank introduced in our paper “TALES: A Taxonomy and Analysis of Cultural Representations in LLM-generated Stories.” The project webpage can be found here: https://cultural-misrepresentations.github.io/ This project evaluates cultural misrepresentations in LLM-generated stories for diverse Indian cultural identities. As part of this effort, we provide a curated question bank of standalone… See the full description on the dataset page: https://huggingface.co/datasets/Kirtibg/TALES-QA.

sourceHugging Facecc-by-4.0updated 10mo agoView on Hugging Face
0likes70downloads
Dataset Card

TALES-QA: Cultural Knowledge Question Bank

This repository contains the question bank introduced in our paper “TALES: A Taxonomy and Analysis of Cultural Representations in LLM-generated Stories.” The project webpage can be found here: https://cultural-misrepresentations.github.io/

This project evaluates cultural misrepresentations in LLM-generated stories for diverse Indian cultural identities. As part of this effort, we provide a curated question bank of standalone, context-independent questions derived from these misrepresentations. Each question is based on an annotated instance identified by expert annotators with lived experience and native language proficiency across 71 regions of India and 14 languages. The annotations were converted into clear question–answer pairs with accompanying metadata to support evaluation of cultural knowledge in LLMs.

Dataset Structure

The dataset is provided as a CSV file with the following fields:

FieldDescription
IDUnique identifier
DifficultyQuestion difficulty level (e.g., Easy, Medium, Hard)
QuestionQuestion text
AnswerCorrect answer text
TypeQuestion type (e.g., MCQ, One-word, etc.)
LanguageLanguage of the question
Misrepresentation CategoryType of cultural misrepresentation identified
CSI CategoryCulturally specific item being asked through the question

Each row represents a single question-answer pair designed to evaluate cultural knowledge.

Citation

If you find this question bank useful, please cite our work:

@article{bhagat2025tales,
  title={TALES: A Taxonomy and Analysis of Cultural Representations in LLM-generated Stories},
  author={Bhagat, Kirti and Bhatt, Shaily and Velagapudi, Athul and Vashistha, Aditya and Dave, Shachi and Pruthi, Danish},
  journal={arXiv preprint arXiv:2511.21322},
  year={2025}
}