karya/endangered-recipes-translated-500

Endangered Recipes Translated 500

Endangered Recipes Translated 500 is the next phase of the ELR-1000 research effort. It brings together 500 community-contributed recipes from endangered and under-represented Indic languages, with 50 recipes per language. Alongside the original recipe text, this release includes English translations for recipe names, ingredients, tools, cultural notes, and recipe steps.

The source collection and research context are described in the ELR-1000 paper, "ELR-1000: A Community-Generated Dataset for Endangered Indic Indigenous Languages". This translated release is meant to make the recipes easier to study, compare, and use in language and vision research, while keeping the original language text and community context intact.

Dataset Details

Each recipe entry includes:

• Original recipe text: recipe names, ingredients, tools, stories, cooking context, and preparation steps in the source language.
• English translations: translations for the main recipe fields and matched recipe steps where available.
• Images for cooking steps: images stored with the recipe steps they belong to.
• Cultural notes: stories, occasions, memories, storage practices, health benefits, dietary restrictions, and ways of eating the dish.
• Community-led collection: recipes contributed through the same process as ELR-1000, with informed consent and fair compensation.

Dataset Sources

• New translated dataset: karya/endangered-recipes-translated-500
• Source dataset: karya/ELR-1000
• Source paper: ELR-1000: A Community-Generated Dataset for Endangered Indic Indigenous Languages

License And Attribution

This translated dataset release is published under the Creative Commons Attribution 4.0 International License. You may share and adapt the dataset, including for commercial use, as long as you give appropriate credit.

The original ELR-1000 dataset is a separate release with its own license. We cite ELR-1000 and the ELR-1000 paper here so that users can clearly see where this translated dataset comes from.

Dataset Statistics

Attribute	Value
Total recipes	500
Languages	10
Recipes per language	50
Total images	5,017
Average images per recipe	10.03
Native recipe-step lines	3,559
Translated recipe-step lines	3,559
Top-level translation pairs	10,998
Modalities	Text and images

Languages Included

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

Per-Language Coverage

Language	Recipes	Images	Avg. Images	Sentence Pairs
Assamese	50	618	12.36	1,172
Bodo	50	813	16.26	1,260
Ho	50	500	10.00	1,233
Kaman Mishmi	50	663	13.26	661
Khasi	50	565	11.30	1,029
Khortha	50	443	8.86	1,214
Meitei	50	180	3.60	737
Mundari	50	315	6.30	783
Sadri	50	506	10.12	1,166
Santhali	50	414	8.28	1,123

The dataset contains 420 unique ingredients.

Quick Start

from datasets import load_dataset
import matplotlib.pyplot as plt
import os
from PIL import Image
import io

# This is a gated dataset. Provide a Hugging Face token with access.
ds = load_dataset(
    "karya/endangered-recipes-translated-500",
    split="train",
    token=os.getenv("HF_TOKEN"),
)

sample = ds[0]

print(sample["language"])
print(sample["recipe_name"], "->", sample["recipe_name_translation"])
print(sample["ingredients"])
print(sample["ingredients_translation"])

for step in sample["recipe_steps"][:3]:
    print(step["step_index"], step["native"], "->", step["translation"])

# Display the first available step image.
for step in sample["recipe_steps"]:
    image = step.get("image")
    if image and image.get("bytes"):
        plt.imshow(Image.open(io.BytesIO(image["bytes"])))
        plt.axis("off")
        plt.show()
        break

Relationship To ELR-1000

This dataset builds on the original ELR-1000 collection. ELR-1000 contains 1,073 recipes with native-language text, images, and audio recordings across the same 10 languages. This release selects 500 of those recipes and adds English translations so the material can be used more easily in multilingual and image-text research.

The original ELR-1000 release and this translated release have separate licenses. This translated release is distributed under CC BY 4.0.

Please cite this dataset, the source ELR-1000 dataset, and the ELR-1000 paper when using this translated release.

Data Collection And Translation

The underlying recipes were collected through community participation with native speakers from indigenous and under-represented language communities in India. The collection process focused on authentic recipes, informed consent, and fair compensation. This release adds English translations while preserving the original recipe text.

Intended Uses

This dataset may support research on:

• Recipe understanding with text and images.
• Multilingual NLP and translation research.
• Ingredient, tool, and instruction extraction.
• Food and language research rooted in local communities.
• Dataset development for endangered and under-represented Indic languages.

Limitations

The dataset is a 500-recipe translated subset of ELR-1000 and should not be treated as complete coverage of any language or food tradition. Some fields have more translation coverage than others, and a few recipe-step translations depend on what was available in the source material. Users should also keep in mind that recipes, spellings, and translations can vary across speakers, families, and regions.

Citation

If you use this dataset, please cite the translated dataset:

@dataset{karya2026endangeredrecipestranslated500,
    title     = {Endangered Recipes Translated 500},
    author    = {{Karya}},
    year      = {2026},
    publisher = {Hugging Face},
    url       = {https://huggingface.co/datasets/karya/endangered-recipes-translated-500}
}

Please also cite the source ELR-1000 dataset:

@dataset{karya2025elr1000,
    title     = {ELR-1000: A Community-Generated Dataset for Endangered Indic Indigenous Languages},
    author    = {{Karya}},
    year      = {2025},
    publisher = {Hugging Face},
    url       = {https://huggingface.co/datasets/karya/ELR-1000}
}

And the original ELR-1000 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, images, translations, and cultural knowledge. This work is dedicated to preserving and celebrating the culinary and linguistic heritage of indigenous communities in India.

License

This translated dataset release is licensed under the Creative Commons Attribution 4.0 International License. Please cite this dataset and credit Karya when using, sharing, or adapting it.

Contact

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