yasalma/tt-structured-content

Dataset Summary

This dataset contains structured textual content in Markdown format extracted from Tatar-language documents, originally in EPUB and PDF formats. The documents include books and other long-form content with rich formatting. The dataset is intended to provide clean, structured, and semantically meaningful content to support natural language processing tasks, content modeling, and research in Tatar language technologies.

The extracted Markdown preserves key structural elements such as:

• Headings and paragraphs
• Images
• Tables
• Table of contents
• Document title
• Footnotes
• Mathematical and chemical formulas (in LaTeX format)

Extraneous elements such as page numbers, headers, and footers have been removed to provide clean and usable text.

Dataset Structure

The dataset is provided in multiple Parquet files with the following columns:

• id: Internal service ID (not relevant for external use)
• publish_year: The year the document was published (may be missing)
• genre: A comma-separated list of genres inferred by Gemini
• text: The structured content in Markdown format

Each row represents one full document.

Supported Tasks and Use Cases

This dataset supports various use cases, including but not limited to:

• Pretraining or fine-tuning language models for the Tatar language
• Building Markdown-aware content models
• Semantic document analysis
• Text segmentation, summarization, or classification
• Cross-lingual NLP (e.g., aligning Tatar with other languages)

Languages

• Tatar (tt, script: Cyrillic)

Source Data

The documents were gathered from publicly available sources and converted using Gemini 2.5. The conversion process aimed to preserve as much of the original document structure and formatting as possible while excluding artifacts such as page numbers and repeated headers/footers.

License

This dataset is released under the MIT License.

You are free to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the dataset, subject to the following condition:

• The original copyright notice and permission notice must be included in all copies or substantial portions of the dataset.

See the full license text here: https://opensource.org/license/mit/

Citation

If you use this dataset in your work, please cite it as: @dataset{tt_structured_content, author = {yasalma}, title = {Structured Tatar Documents in Markdown Format}, year = {2025}, url = {https://huggingface.co/datasets/yasalma/tt-structured-content}, note = {Dataset of structured Markdown content extracted from Tatar-language EPUB and PDF documents.} }

Limitations

• The genre classification is automatically inferred and may not always be accurate.
• Some documents may have minor formatting inconsistencies due to OCR or extraction errors.
• Images and some visual content are referenced in Markdown but not embedded.

Contributions

This dataset was created and maintained by yasalma. Contributions and suggestions are welcome.