IndicQuest-v2 / README.md
l3cube-pune's picture
Update README.md
bcaa9fb verified
|
Raw
History Blame Contribute Delete
3.51 kB
---
license: cc-by-4.0
task_categories:
- question-answering
language:
- en
- as
- bn
- doi
- gu
- hi
- kn
- kok
- mai
- ml
- mr
- mni
- ne
- or
- pa
- sa
- sd
- ta
- te
- ur
tags:
- indic
- factual-knowledge
- benchmark
- closed-book-qa
size_categories:
- 10K<n<100K
configs:
- config_name: english
data_files: English.csv
- config_name: hindi
data_files: Hindi.csv
- config_name: marathi
data_files: Marathi.csv
- config_name: assamese
data_files: Assamese.csv
- config_name: bengali
data_files: Bengali.csv
- config_name: dogri
data_files: Dogri.csv
- config_name: gujarati
data_files: Gujarati.csv
- config_name: kannada
data_files: Kannada.csv
- config_name: konkani
data_files: Konkani.csv
- config_name: maithili
data_files: Maithili.csv
- config_name: malayalam
data_files: Malayalam.csv
- config_name: manipuri
data_files: Manipuri.csv
- config_name: nepali
data_files: Nepali.csv
- config_name: odia
data_files: Odia.csv
- config_name: punjabi
data_files: Punjabi.csv
- config_name: sanskrit
data_files: Sanskrit.csv
- config_name: sindhi
data_files: Sindhi.csv
- config_name: tamil
data_files: Tamil.csv
- config_name: telugu
data_files: Telugu.csv
- config_name: urdu
data_files: Urdu.csv
---
# IndicQuest v2
A gold-standard multilingual question-answering benchmark for evaluating the India-specific factual knowledge of Large Language Models. 3,471 curriculum-grounded English question–answer pairs across nine domains, translated into 19 Indic languages: 69,420 parallel pairs across 20 languages.
More details can be found in our [paper](http://arxiv.org/abs/2608.15535).
## Dataset structure
One CSV per language, named `<language>.csv` (`english.csv`, `hindi.csv`, `marathi.csv`, …). Every file has the same 3,471 rows in the same order, so the *n*th row of any file is the translation of the *n*th row of `english.csv`.
| Column | Description |
|---|---|
| `question` | The question, in the file's language |
| `answer` | Gold answer, one to five words |
| `Domain` | One of nine domains (see below) |
### Domains
| Domain | Pairs |
|---|---|
| Culture | 534 |
| History | 510 |
| Geography | 499 |
| Law | 494 |
| Political Science | 450 |
| Science | 325 |
| Sports | 273 |
| Art | 233 |
| Commercial Studies | 153 |
| **Total** | **3,471** |
### Languages
English plus Assamese, Bengali, Dogri, Gujarati, Hindi, Kannada, Konkani, Maithili, Malayalam, Marathi, Meitei (Manipuri), Nepali, Odia, Punjabi, Sanskrit, Sindhi, Tamil, Telugu, Urdu.
## Example
| question | answer | Domain |
|---|---|---|
| Which apex body was set up in 1982 to coordinate all institutions involved in the rural financing system? | NABARD | Commercial Studies |
| Which river's water is primarily harnessed by the Ukai project in India? | Tapi river | Geography |
| Which is the oldest football tournament in India, established by Sir Mortimer Durand and held from 1888 in Simla? | Durand Cup | Sports |
## Usage
```python
from datasets import load_dataset
english = load_dataset("l3cube-pune/IndicQuest-v2", data_files="english.csv")
hindi = load_dataset("l3cube-pune/IndicQuest-v2", data_files="hindi.csv")
```
## Citing
```
@article{jain2026l3cube,
title={L3Cube-IndicQuest v2: A Large-Scale Multilingual Benchmark for Evaluating Factual Knowledge of Large Language Models Across Indic Languages},
author={Jain, Rinit and Mahajan, Tirthraj and Joshi, Advait and Joshi, Raviraj},
journal={arXiv preprint arXiv:2608.15535},
year={2026}
}
```