clarin-pl/combo-nlp-xlm-roberta-base-latin-ittb-ud2.17

COMBO-NLP Model for Latin

Model Description

This is a Latin-language model based on combo-nlp, an open-source natural language preprocessing system. It performs:

• sentence segmentation (via combo-seg)
• tokenisation (via combo-seg)
• part-of-speech tagging
• morphological analysis
• lemmatisation
• dependency parsing

The Latin model uses FacebookAI/xlm-roberta-base as its base encoder and is trained on UD_Latin-ITTB (UD v2.17).

Evaluation

Evaluation was performed on the UD_Latin-ITTB test split using the standard CoNLL 2018 eval script.

Two evaluation rows are reported:

• Full-text (F1): raw text is segmented by combo-seg, then parsed and compared against gold — measures end-to-end pipeline performance including segmentation quality.
• Aligned accuracy: accuracy on correctly segmented (aligned) tokens — measures parsing quality on tokens that were correctly identified by the segmenter.

Morphosyntactic Tagging

Metric	Tokens	Sentences	Words	UPOS	XPOS	UFeats	AllTags	Lemmas
Full-text (F1)	99.99	95.32	99.99	99.26	97.12	97.86	96.32	99.42
Aligned accuracy	0.00	0.00	0.00	99.26	97.12	97.86	96.32	99.42

Dependency Parsing

Metric	UAS	LAS	CLAS	MLAS	BLEX
Full-text (F1)	92.10	90.50	88.88	85.40	88.49
Aligned accuracy	92.10	90.50	88.80	85.32	88.40

Usage

Install the library from PyPI (assuming you have a virtual environment created):

pip install combo-nlp

The combo-seg segmenter (used to split and tokenise raw text) is installed automatically as a dependency of combo-nlp, so no extra install step is needed.

from combo import COMBO

# Load a pre-trained model with the corresponding combo-seg segmenter
nlp = COMBO("Latin")

# Parse raw text (handles sentence splitting + tokenization)
result = nlp("Vulpes brunnea velox super canem pigrum salit.")

# Inspect results
for sentence in result:
    for token in sentence:
        print(f"{token.form:<15} {token.lemma:<15} {token.upos:<8} head={token.head}  {token.deprel}")

Refer to the combo-nlp documentation for installation and usage instructions:

• https://pypi.org/project/combo-nlp/
• https://pypi.org/project/combo-seg/

License

The training data license: cc-by-nc-sa-3.0 is derived from the Universal Dependencies treebank. For the full license terms, please refer to the LICENSE.txt file in the treebank repository:

• UD_Latin-ITTB LICENSE.txt

Citation

If you use this model, please cite:

Ulewicz, M., Jabłońska, M., Klimaszewski, M., Przybyła, P., Pszenny, Ł., Rybak, P., Wiącek, M., & Wróblewska, A. (2026). COMBO-NLP Models Trained on UD v2.17. Zenodo. https://doi.org/10.5281/zenodo.19650523

@software{combo_nlp_2026,
  author    = {Ulewicz, Michał and Jabłońska, Maja and Klimaszewski, Mateusz and Przybyła, Piotr and Pszenny, Łukasz and Rybak, Piotr and Wiącek, Martyna and Wróblewska, Alina},
  title     = {{COMBO-NLP} Models Trained on {UD} v2.17},
  year      = {2026},
  publisher = {Zenodo},
  doi       = {10.5281/zenodo.19650523},
  url       = {https://doi.org/10.5281/zenodo.19650523}
}

Treebank References

If you use the Latin ITTB treebank data, please also cite:

@article{morpharmo,
  author    = {Gamba, Federica and Zeman, Daniel},
  title     = {Latin Morphology through the Centuries: Ensuring Consistency for Better Language Processing},
  journal   = {Proceedings of the Ancient Language Processing Workshop associated with the 14th International Conference on Recent Advances in Natural Language Processing RANLP 2023},
  year      = {2023},
  address = {Varna, Bulgaria},
url = {https://ufal.mff.cuni.cz/biblio/attachments/2023-gamba-p3787387064232511302.pdf}
}

@article{harmo,
  author    = {Gamba, Federica and Zeman, Daniel},
  title     = {Universalising Latin Universal Dependencies: a harmonisation of Latin treebanks in UD},
  journal   = {Proceedings of the Sixth Workshop on Universal Dependencies (UDW, GURT/SyntaxFest 2023)}},
  year      = {2023},
  address = {Washington, DC, USA},
  url = {https://aclanthology.org/2023.udw-1.2/}
}

@article{lait-ud,
  author    = {Cecchini, Flavio Massimiliano and Passarotti, Marco and Marongiu, Paola and Zeman, Daniel},
  title     = {Challenges in Converting the \emph{Index Thomisticus} treebank into Universal Dependencies},
  journal   = {Proceedings of the Universal Dependencies Workshop 2018 (UDW 2018)},
  year      = {2018},
  address = {Brussels, Belgium}
}

@article{lait,
  author    = {Passarotti, Marco and Dell’Orletta, Felice},
  title     = {Improvements in parsing the index thomisticus treebank. Revision, combination and a feature model for medieval Latin},
  journal   = {Training},
  volume    = {2},
  pages     = {61--024},
  year      = {2010}
}

Resources

• combo-nlp: https://pypi.org/project/combo-nlp/
• combo-seg: https://pypi.org/project/combo-seg/
• UD_Latin-ITTB: https://github.com/UniversalDependencies/UD_Latin-ITTB

Acknowledgements

We gratefully acknowledge Polish high-performance computing infrastructure PLGrid (HPC Center: ACK Cyfronet AGH) for providing computer facilities and support within computational grant no. PLG/2025/018815.