A large-scale part-of-speech tagged corpus for the Tatar language, following Universal Dependencies (UD) conventions. This dataset contains 3.9 million sentences with detailed morphological annotations, including universal part-of-speech tags (UPOS), language-specific tags (XPOS), lemmas, and morphological features.

Dataset Details

Dataset Description

This dataset provides comprehensive part-of-speech annotation for Tatar language texts, covering diverse domains including news, literature, and journalism. It is designed to support the development of NLP tools for Tatar, a low-resource Turkic language.

Curated by: TatarNLPWorld Community
Language(s) (NLP): Tatar (tt)
License: MIT
Annotation scheme: Universal Dependencies (UD) v2

Dataset Sources

Repository: https://huggingface.co/datasets/TatarNLPWorld/tatar-pos
Demo [optional]: [Coming Soon]

Uses

Direct Use

This dataset is intended for:

Training part-of-speech taggers for Tatar
Developing morphological analyzers and disambiguators
Building lemmatization systems
Pre-training language models for Tatar
Cross-lingual learning experiments with other Turkic languages

Out-of-Scope Use

This dataset should not be used for:

Author attribution or stylometry without additional controls
Tasks requiring semantic understanding without additional annotation
Applications where the X tag (unknown) rate of ~22% would cause critical errors

Dataset Structure

Data Fields

Each record in the dataset contains:

Field	Type	Description
`sent_id`	`string`	Unique sentence identifier
`text`	`string`	Original sentence text
`tokens`	`list`	List of token objects

Each token object contains:

Field	Type	Description
`text`	`string`	Word form
`lemma`	`string`	Lemma (base form)
`upos`	`string`	Universal part-of-speech tag
`xpos`	`string`	Language-specific tag
`feats`	`string`	Morphological features (UD format)

Data Statistics

Metric	Value
Total Sentences	3,905,185
Total Tokens	50,467,801
Unique Word Forms	812,687
Unique Lemmas	444,449
Tokens with X tag	10,991,199 (21.78%)

Part-of-Speech Distribution

UPOS Tag	Count	Percentage
PUNCT	11,468,114	22.72%
NOUN	11,241,844	22.28%
X	10,991,199	21.78%
VERB	7,915,684	15.68%
ADJ	2,458,736	4.87%
PRON	1,773,084	3.51%
PART	1,349,735	2.67%
PROPN	995,176	1.97%
CCONJ	743,679	1.47%
ADP	728,426	1.44%
ADV	727,719	1.44%
INTJ	74,405	0.15%

Example Entry

{
  "sent_id": "doc_123_sent_5",
  "text": "Мин китап укыйм.",
  "tokens": [
    {
      "text": "Мин",
      "lemma": "мин",
      "upos": "PRON",
      "xpos": "PRP",
      "feats": "Case=Nom|Number=Sing|Person=1|PronType=Prs"
    },
    {
      "text": "китап",
      "lemma": "китап",
      "upos": "NOUN",
      "xpos": "NN",
      "feats": "Case=Nom|Number=Sing"
    },
    {
      "text": "укыйм.",
      "lemma": "укы",
      "upos": "VERB",
      "xpos": "VBP",
      "feats": "Mood=Ind|Number=Sing|Person=1|Tense=Pres|VerbForm=Fin"
    }
  ]
}

Dataset Creation

Curation Rationale

The dataset was created to address the lack of annotated resources for Tatar language NLP. By providing a large-scale POS-tagged corpus following Universal Dependencies standards, we aim to:

Enable development of modern NLP tools for Tatar
Facilitate cross-lingual research with other Turkic languages
Support low-resource language technology development

Source Data

Data Collection and Processing

Texts were collected from diverse sources including:

News websites (Tatar-inform, Azatliq, etc.)
Literary works and publications
Educational materials
Online media and blogs

Processing pipeline:

Text extraction and cleaning
Sentence segmentation
Tokenization
Manual and automatic annotation
Quality validation

Who are the source data producers?

The original texts were produced by various Tatar language media outlets, authors, and publishers. The annotations were created by the TatarNLPWorld community, including linguists and NLP researchers specializing in Tatar.

Annotations

Annotation process

The annotation followed Universal Dependencies v2 guidelines with adaptations for Tatar-specific phenomena:

Initial automatic annotation using existing tools
Manual correction by trained linguists
Inter-annotator agreement validation
Final quality control checks

Who are the annotators?

The annotations were performed by a team of Tatar language linguists and computational linguists from the TatarNLPWorld community.

Personal and Sensitive Information

The dataset contains publicly available texts from news sources and publications. No personal or sensitive information has been intentionally included. All texts are properly attributed to their original sources.

Bias, Risks, and Limitations

Known Limitations

Domain bias: The corpus may overrepresent certain domains (news, literature) and underrepresent others (social media, technical texts)
X tag rate: ~22% of tokens are marked as X (unknown), indicating annotation challenges or truly ambiguous cases
Temporal coverage: Texts span multiple years but may not equally represent all periods

Recommendations

Users should:

Be aware of the domain distribution when training models
Consider fine-tuning on specific domains if needed
Account for the X tag rate in evaluation metrics
Validate performance on their specific use cases

Citation

BibTeX:

@dataset{tatar_pos_corpus_2026,
    title = {Tatar Universal Dependencies POS Corpus},
    author = {TatarNLPWorld Community},
    year = {2026},
    publisher = {Hugging Face},
    version = {1.0.0},
    url = {https://huggingface.co/datasets/TatarNLPWorld/tatar-pos}
}

APA:

TatarNLPWorld Community. (2026). Tatar Universal Dependencies POS Corpus [Data set]. Hugging Face. https://huggingface.co/datasets/TatarNLPWorld/tatar-pos

Glossary

UPOS: Universal Part-of-Speech tags (e.g., NOUN, VERB, ADJ)
XPOS: Language-specific part-of-speech tags
FEATS: Morphological features in UD format
X tag: Special tag for unknown/unclear cases

More Information

For questions, contributions, or feedback, please open an issue on the Hugging Face repository or contact the TatarNLPWorld community.

Dataset Card Authors

TatarNLPWorld Community

Dataset Card Contact

GitHub: https://github.com/TatarNLPWorld
Hugging Face: https://huggingface.co/TatarNLPWorld

Downloads last month: 26