AiLab-IMCS-UL/lv_ud2.18_lv-deberta-base

Latvian morphosyntactic parser

This model is a Latvian joint morphosyntactic parser based on AiLab-IMCS-UL/lv-deberta-base. It predicts UPOS, XPOS, morphological features, lemmas, dependency heads, and dependency relations in a unified multi-task architecture.

The model was trained and evaluated on the Latvian Universal Dependencies Treebank from UD v2.18. In addition to the gold treebank, training used 285k automatically annotated (silver) sentences from the Balanced Corpus of Modern Latvian (LVK2022), labelled by an ensemble of earlier models with analyzer-constrained decoding, and 13.8k annotated multi-word-expression phrases; the silver data is used in a first training stage, the gold treebank alone in the second.

The model does not need the morphological analyzer at inference time. Analyzer-constrained decoding (parser.parse(..., analyzer=True)) is optional and mainly improves XPOS and lemmas.

Results (UD 2.18, test set)

	XPOS	Lemmas	LAS
model alone	95.20	98.62	93.49
with analyzer-constrained decoding	95.31	98.91	93.49

For more details, see the accompanying paper and the GitHub repository: https://github.com/LUMII-AILab/lvnlp.

Citation

@inproceedings{znotins-2026-improving,
  title = {Improving Latvian Morphosyntactic Parsing with Pretrained Encoders and Analyzer-Constrained Decoding},
  author = {Znotins, Arturs},
  booktitle = {Proceedings of the Fifteenth Language Resources and Evaluation Conference (LREC 2026)},
  month = {May},
  year = {2026},
  pages = {11724--11734},
  address = {Palma, Mallorca, Spain},
  publisher = {European Language Resources Association (ELRA)},
  editor = {Piperidis, Stelios and Bel, Núria and van den Heuvel, Henk and Ide, Nancy and Krek, Simon and Toral, Antonio},
  doi = {10.63317/5khpzsaiqrzw},
  url = {https://lrec.elra.info/lrec2026-main-918}
}

Acknowledgements

This work was supported by the EU Recovery and Resilience Facility project Language Technology Initiative (2.3.1.1.i.0/1/22/I/CFLA/002).