Curated by: UzDataLab
**Funded by Initial funding from IT Park Uzbekistan and self-funded efforts; seeking grants from AWS Activate and UNDP AI for Good.
**Shared by UzDataLab, available via Hugging Face community.
Language(s) (NLP): (NLP): Uzbek (uz).
License: apache-2.0

Dataset Sources [optional]

UzDataLab

Uses

Direct Use

This dataset is intended for training and evaluating NLP models, specifically for text generation (e.g., creating Uzbek content) and text classification (e.g., sentiment analysis of Uzbek texts). It is suitable for academic research, low-resource language development, and AI applications in Uzbekistan.

[More Information Needed]

Out-of-Scope Use

The dataset should not be used for commercial purposes without permission, real-time personal data processing, or applications that could perpetuate bias against specific Uzbek dialects or communities without proper validation.

[More Information Needed]

Dataset Structure

The dataset contains 1K to 10K entries, featuring Uzbek text data with annotations. Fields include:

text: Raw Uzbek text (sheva and literary forms). translation: English and Russian translations. category: Tags like "agent", "climate". metadata: Source, date, and quality metrics.

Dataset Creation

The dataset was created to address the scarcity of Uzbek language resources for NLP, aiming to support low-resource language models and promote AI innovation in Central Asia.

Curation Rationale

The dataset was created to address the scarcity of Uzbek language resources for NLP, aiming to support low-resource language models and promote AI innovation in Central Asia.

[More Information Needed]

Source Data

Online chats

Data Collection and Processing

Data was collected from Telegram channels (e.g., Article 365), local news (kun.uz), and community contributions. Processing involved manual transcription, translation using NLLB-200, and quality checks by human reviewers. Tools used: Python (PyPDF2, Pandas), Hugging Face Datasets.

Who are the source data producers?

Data producers include UzDataLab team members and volunteers from Uzbek linguistic communities, representing diverse regions (Tashkent, Samarkand).

Annotations [optional]

Annotations were performed by Team B with guidelines for sentiment (neutral, positive, negative) and entity recognition (locations, domains). Interannotator agreement was 85%, validated by two reviewers.

Annotation process

Who are the annotators?

Annotators are UzDataLab team members (e.g., Asadbek, Muhammadsiddiq) with linguistic expertise.

[More Information Needed]

Personal and Sensitive Information

The dataset contains no personal data. Names and locations are anonymized or generalized (e.g., "Tashkent" as a location, not specific addresses).

Bias, Risks, and Limitations

The dataset may reflect biases from regional dialects or limited sample size. Risks include misuse for biased AI training without proper context.

Recommendations

Users should validate dataset bias with additional Uzbek data and avoid overgeneralization across dialects.

Users should be made aware of the risks, biases and limitations of the dataset. More information needed for further recommendations.

Citation [optional]

@misc{UzDataLab2025, author = {UzDataLab Team}, title = {UzDataLab: Uzbek NLP Dataset}, year = {2025}, url = {https://huggingface.co/datasets/UzDataLab} }

BibTeX:

[More Information Needed]

APA:

UzDataLab Team. (2025). UzDataLab: Uzbek NLP Dataset. Retrieved from

Glossary [optional]

[More Information Needed]

More Information [optional]

Sheva: Regional Uzbek dialect variations. Low-resource: Languages with limited digital data, like Uzbek.

Dataset Card Authors [optional]

Authored by UzDataLab Team (Hayotbek, Jahongir, Ilxom).

Dataset Card Contact

This card reflects the current state of UzDataLab, with plans to expand based on community feedback and additional funding.

Downloads last month: 20