ELR-1000: A Community-Generated Dataset for Endangered Indic Indigenous Languages

ELR-1000 is a rich multimodal dataset documenting traditional recipes in 10 endangered and under-represented languages of India. This community-generated resource contains audio recordings, images, and text descriptions for 1,073 recipes, preserving culinary knowledge and cultural heritage through authentic community participation. Our paper documenting the details can be found here.

Dataset Overview

This dataset captures the rich culinary traditions of indigenous communities across India, documenting recipes that have been passed down through generations. Each recipe entry includes:

• Multimodal Content: Step-by-step images, audio instructions, and detailed text descriptions
• Cultural Context: Stories, memories, and cultural significance associated with each dish
• Practical Information: Ingredients, tools, storage methods, and dietary considerations
• Community Voice: Authentic recordings from community members in their native languages
• Fair Wages: All participants were compensated higher-than living wages for their contributions

Dataset Statistics

Attribute	Value
Total Recipes	1,073
Languages	10 endangered Indic languages
Total Images	11,213
Total Audio	42 hours
Format	Parquet shards (11 files) with embedded media
Modalities	Text, Images (JPG), Audio (WAV)

Languages Included

Assamese, Bodo, Ho, Kaman-Mishmi, Khasi, Khortha, Meitei, Mundari, Sadri, Santhali

Quick Start

from datasets import load_dataset
import matplotlib.pyplot as plt
import os
import IPython.display as ipd

# Load dataset
ds = load_dataset("karya/ELR-1000", split="train", token=os.getenv("hf_token"))
sample = ds[0]

# Display metadata
print(sample['recipe_name'], sample['language'], sample['ingredients'])

# Display recipe texts
print(sample["recipe_steps_text"])

# Display image
if sample['images']:
    plt.imshow(sample['images'][0])
    plt.axis('off')
    plt.show()

# Play audio
if sample['user_audio']:
    audio = sample['user_audio'][0]
    ipd.display(ipd.Audio(data=audio['array'], rate=audio['sampling_rate']))

Dataset Structure

The dataset is stored as Parquet files in the data/ directory. Each Parquet file contains approximately 100 recipes with embedded media (images and audio stored as bytes within the Parquet format).

Data Collection

This dataset was created through community participation, with recipes and recordings contributed by native speakers from various indigenous communities across India. The data collection process prioritized:

• Authenticity: Recipes and recordings sourced directly from community members
• Fair Compensation: All contributors were paid higher-than-living wages for their participation
• Informed Consent: All participants provided explicit consent for their contributions to be shared

Citation

If you use this dataset in your research or projects, please cite the paper:

@inproceedings{joshi2025elr1000,
    title     = {ELR-1000: A Community-Generated Dataset for Endangered Indic Indigenous Languages},
    author    = {Joshi, Neha and Gogoi, Pamir and Mirza, Aasim and Jansari, Aayush and Yadavalli, Aditya and Pandey, Ayushi and Shukla, Arunima and Sudharsan, Deepthi and Bali, Kalika and Seshadri, Vivek},
    booktitle = {Proceedings of the 14th International Joint Conference on Natural Language Processing and the 4th Asia-Pacific Chapter of the Association for Computational Linguistics (IJCNLP-AACL 2025)},
    address   = {Mumbai, India},
    month     = dec,
    year      = {2025},
    note      = {Preprint: arXiv:2512.01077},
    url       = {https://arxiv.org/abs/2512.01077}
  }

Acknowledgments

We gratefully acknowledge the communities and individuals who contributed recipes, recordings, and cultural knowledge to create this dataset. This work is dedicated to preserving and celebrating the rich culinary and linguistic heritage of indigenous communities in India.

License

This work is licensed under a KPL BY-NC-SA-FS 1.0 license. This license allows reusers to distribute, remix, adapt, build upon, and incorporate into software systems, the material in any medium or format for noncommercial purposes only, and only so long as attribution is given to the creator. If you remix, adapt, build upon, or incorporate into software systems, you must license the modified material, including material generated by the software system, under identical terms, and license the software system under the GNU General Public License.

Commercial Use

If you are interested in using this dataset for commercial purposes, please contact us. Profits from commercial licensing will be distributed as royalties to the community members who contributed to this dataset.

Contact

For questions, issues, or contributions, open an issue on the dataset repository or contact us directly.