nassimjp/pashto_multimodal_v1

Pashto Multimodal v1: A Decade of Pashtun Digital Activism

Dataset Summary

This is a curated, multilingual, and multimodal dataset of ~5,900 high-quality image-text pairs, extracted from the personal X (Twitter) archive of @Pashto_lab. This archive represents over a decade of active digital advocacy (since 2012) by a Pashtun AI researcher and activist based in Japan.

The dataset is a rich, firsthand chronicle of contemporary Pashtun socio-political discourse, created with a clear and consistent focus on:

• ✌️ Peace and anti-war advocacy.
• 🗳️ Democracy and good governance.
• 📚 Education, especially for girls and women.
• 👩 Women's rights and empowerment.
• ⚖️ Human rights and accountability.
• 🎭 Pashtun cultural heritage and language preservation.
• 🏴 Support for the Pashtun Tahafuz Movement (PTM).
• 🚫 Critical perspectives on the Taliban and all forms of extremism.
• A secular, humanist, and culturally-focused worldview.

This is not a random sample; it is a purpose-built thematic archive reflecting a specific, documented voice within the Pashtun diaspora, making it a unique primary source for researchers.

Dataset Structure

Each record has the following schema:

{
  "id": "852812136317308932_0",
  "source": "twitter",
  "image": "images/852812136317308932-C9Uh4mzXoAMPdoQ.jpg",
  "text": "U.S. drops 'Mother of all Bombs' on ISIS tunnel compound in Afghanistan"
}

Field	Type	Description
id	string	A unique identifier for the sample.
source	string	The platform of origin ("twitter" for all records).
image	image	The associated image file.
text	string	The primary text content of the post.

Features and Curation

• Multilingual: The text is in Pashto (primary), Urdu, English, and Japanese.
• Cleaned: Captions and images have been cleaned.
• Deduplicated: Duplicate images and duplicate captions have been removed to ensure quality.
• Source: All data originates from public posts on X (Twitter).

Curator's Perspective and Archive Context

This archive was deliberately created by a Pashtun researcher (@Pashto_lab) with a stated, consistent commitment to the values listed in the summary.

Research Note: This perspective is clearly reflected in the dataset's content. Researchers should recognize this as a valuable, documented viewpoint from a specific community voice. It is ideal for nuanced studies of Pashtun digital activism and diaspora identity but is not a representative sample of all Pashtun opinions.

Statistics

Split	Number of Examples	Dataset Size (Bytes)
Train	5,875	592,099,710
Total	5,875	592,099,710

• Download Size: 677,964,476 bytes

Important Considerations and Content Warning

• Thematic Focus: The content is heavily centered on the themes listed above, which include political activism and human rights.
• Perspective: The dataset presents a specific activist viewpoint. Users should be aware that the text and images may contain strong political opinions and criticism of state actors.
• Content Warning: The dataset contains discussions of violence, conflict, and death. It is intended for research purposes and may not be suitable for all audiences.

Uses and Limitations

Intended Uses

• Research on low-resource languages, especially Pashto.
• Multimodal model training (e.g., image captioning, visual question answering).
• Sentiment analysis and sociopolitical discourse analysis.
• Cultural and historical studies of the Afghanistan-Pakistan region.
• Studies on digital activism, diaspora communities, and online advocacy.

Limitations

• The dataset is not balanced and does not represent all perspectives on the topics covered.
• Text sources are solely from X (Twitter) and may contain noise, slang, or unverified claims.
• Images are provided as-is and may have low resolution or be screenshots.
• The archive reflects the interests and network of a single user (@Pashto_lab).

Licensing and Terms of Use

Please use this dataset responsibly and respect the terms applicable to the original content on X (Twitter).

• License: This dataset is shared under a Creative Commons Attribution-NonCommercial 4.0 International License (CC BY-NC 4.0).
• Research Use: It is intended for non-commercial academic research only.
• Attribution: If you use this dataset, please cite it using the information in the Citation section below.

Citation

If you use this dataset in your research, please cite it as:

@misc{pashto_multimodal_v1,
  author = {Nassim Jan (@Pashto_lab)},
  title = {Pashto Multimodal v1: A Decade of Pashtun Digital Activism (Image-Text Archive)},
  year = {2026},
  publisher = {Hugging Face},
  url = {https://huggingface.co/datasets/nassimjp/pashto_multimodal_v1},
  note = {A curated archive of 5,875 image-text pairs advocating for peace, democracy, human rights, and Pashtun culture.}
}

Acknowledgements

This dataset was curated from the curator's personal public posts on X (formerly Twitter). The curator thanks all researchers who work to advance low-resource languages, multimodal AI, and the understanding of digital activism.