turkicnlp/generated-ud-tatar

Data Splits

Split	Description
train	Automatically translated and silver-annotated sentences derived from UD training sources
dev	Silver-annotated evaluation data derived from UD test resources
test	Turkic UD parallel corpora - https://github.com/ud-turkic/parallel

---

Dataset Creation

Source Data

This dataset is a derived work based on resources from:

• Universal Dependencies treebanks - https://github.com/UniversalDependencies
• Turkic UD parallel corpora - https://github.com/ud-turkic/parallel
• Turkic language resources collected in the Turkic NLP ecosystem

Training Data Sources

The training split was generated by translating and automatically annotating sentences originating from the following Universal Dependencies treebanks:

• UD_Turkish-IMST (tr_imst-ud-train.conllu)
• UD_Turkish-GB (tr_gb-ud-test.conllu)
• UD_Turkish-FrameNet (tr_framenet-ud-train.conllu)

These datasets were:

1. Translated into Tatar,

2. Processed through the TurkicNLP pipeline,

3. Automatically annotated with UD-compatible morphology and dependency structure.

The resulting annotations are silver-standard.

---

Test Data Sources

The test split is provided by the Turkic UD community:

• ud-turkic/parallel (tt_tuecl-ud-test.conllu)

---

Important Note on Derivation

All data contained in this dataset is a derived work based on previously published UD resources.

Original licensing, attribution, and share-alike requirements of the source datasets are preserved.

Users must cite the original Universal Dependencies treebanks in addition to TurkicNLP when using this dataset.

---

Generation Process

1. Source corpora were selected from publicly available UD resources.
2. Sentences were translated or adapted for Tatar where necessary.
3. The TurkicNLP pipeline automatically produced:
   • tokenization
   • morphological tagging
   • dependency parses
4. Outputs were exported into UD-compatible CoNLL-U format.

Annotations are silver-standard, meaning they are automatically generated and may contain noise.

---

Annotation Type

• Automatic annotation
• Cross-lingual projection
• Rule-based + neural processing
• Silver quality (not manually validated)

---

Intended Uses

Primary Uses

• Training low-resource dependency parsers
• Cross-Turkic transfer experiments
• Benchmarking multilingual models
• Linguistic research on Turkic morphology and syntax

Out-of-Scope Uses

• Linguistic gold-standard evaluation
• High-stakes linguistic analysis without manual verification

---

Known Limitations

• Automatic translation may introduce semantic drift.
• Morphological ambiguity may remain unresolved.
• Annotation quality is lower than manually curated UD treebanks.
• Domain balance depends on source datasets.

Users are encouraged to treat this dataset as silver data.

---

Licensing Information

This dataset contains derived annotations and translations based on datasets from the Universal Dependencies project and related Turkic UD resources.

Original datasets are typically distributed under:

Creative Commons Attribution-ShareAlike 4.0 International (CC BY-SA 4.0) or compatible licenses.

Accordingly, this dataset is released under:

CC BY-SA 4.0

Users must:

• provide attribution to original UD treebanks
• cite Universal Dependencies
• maintain the same license when redistributing derivatives

---

Attribution

Please cite all:

1. Universal Dependencies
2. Turkic UD
3. TurkicNLP

@misc{hakimov2026turkicnlpnlptoolkitturkic,
      title={TurkicNLP: An NLP Toolkit for Turkic Languages}, 
      author={Sherzod Hakimov},
      year={2026},
      eprint={2602.19174},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2602.19174}, 
}

@InProceedings{nivre-et-al:2016:LREC,
  author    = {Nivre, Joakim  and  de Marneffe, Marie-Catherine  and  Ginter, Filip  and  Goldberg, Yoav  and  Hajic, Jan  and  Manning, Christopher D.  and  Ryan, McDonald  and  Petrov, Slav  and  Pyysalo, Sampo  and  Silveira, Natalia  and  Tsarfaty, Reut  and  Zeman, Daniel},
  title     = {Universal Dependencies v1: A Multilingual Treebank Collection},
  booktitle = {Proceedings of the Tenth International Conference on Language Resources and Evaluation (LREC 2016)},
  month     = {may},
  year      = {2016},
  address   = {Portoro{\v{z}}, Slovenia},
  publisher = {European Language Resources Association (ELRA)},
  url       = {https://aclanthology.org/L16-1262},
  pages     = {1659--1666}
}

@inproceedings{akhundjanova-etal-2025-parallel,
    title = "Parallel {U}niversal {D}ependencies Treebanks for {T}urkic Languages",
    author = "Akhundjanova, Arofat  and
      Akkurt, Furkan  and
      Chontaeva, Bermet  and
      Eslami, Soudabeh  and
      Coltekin, Cagri",
    booktitle = "Proceedings of the Eighth Workshop on Universal Dependencies (UDW, SyntaxFest 2025)",
    year = "2025",
    address = "Ljubljana, Slovenia",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2025.udw-1.14/",
    pages = "129--136"
}

---

Dataset Curators

TurkicNLP Project

GitHub: https://github.com/turkic-nlp

---

Ethical Considerations

This dataset is generated from publicly available linguistic data and contains no intentionally collected personal or sensitive information.

---

Contributions

We welcome contributions including:

• improved annotations
• additional Turkic languages
• manual validation
• error reports

Please open an issue or pull request in the TurkicNLP repositories.