Elamur Research⌕
☰
home / papers / 2026 / Language Mixture to Develop Accurate Galician Dependency Parsers: An Exploration of Its Effects

Language Mixture to Develop Accurate Galician Dependency Parsers: An Exploration of Its Effects

Summary

Galician has few manually annotated syntactic treebanks, which limits how accurate its dependency parsers can be. The authors test whether UD annotations from better-resourced languages can fill this gap. They use UDPipe 2 with gold tokenization and compare a baseline that takes no embeddings against variants that take input from Bertinho, BertGL or XML-Roberta. The experiments cover three things: training on PUD-gl versus TreeGal, training on equal-sized PUD samples from languages at different typological distances from Galician, and mixing Galician data with Portuguese (CINTIL) and Spanish (AnCora). Beyond UAS and LAS, they analyse accuracy for each dependency relation.

Parsers trained only on Portuguese or Spanish come close to those trained on Galician. With embeddings, even more distant source languages give usable parsers. Adding cross-lingual data to Galician helps only a little when strong embeddings are present. Parsers also transfer poorly between treebanks. Some relations, notably oblique nominals and adverbial modifiers, stay difficult whatever the training source. The study offers a reproducible recipe for other under-resourced languages, and language mixing is likely most valuable where good embedding models are lacking.

Key points

  • Data setup:
    • For both PUD-gl and TreeGal, each split has 800 training, 50 development and 150 test sentences.
    • The semi-automatically annotated CTG treebank was left out.
    • The typology experiment used 800 PUD sentences from each of Galician, Portuguese, Spanish, French, Italian and German.
  • Mixture partitions:
    • G+PP: 1,600 Galician sentences plus 2,400 CINTIL sentences.
    • G+SS: 1,600 Galician sentences plus 2,400 AnCora sentences.
    • G+PPSS: 1,600 Galician sentences plus 1,200 CINTIL and 1,200 AnCora sentences.
  • Scalability results (Table 6):
    • Without embeddings, LAS goes from 83.5 (Galician only) to as much as 87.3 (G+SS).
    • With the BglB configuration, LAS goes from 89.5 to at most 91.0 (G+PP).
    • In other words, embeddings deliver most of the gain on their own.
  • Error analysis: obl and advmod remain persistent problems that may depend on typology, regardless of the source language.
  • Context and limitations:
    • In UD, Galician has 3 treebanks with about 6k sentences in total, and the default UDPipe 2 model reaches a best LAS of 82.3.
    • The study did no hyperparameter tuning, used only UDPipe 2, and covered a narrow range of genres.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

Fields: Linguistics & languages

Topics: Dependency parsing · Treebanks

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.vardial-1.5

@inproceedings{irastortzaurbieta2026language,
  title = {{Language Mixture to Develop Accurate Galician Dependency Parsers: An Exploration of Its Effects}},
  author = {Irastortza-Urbieta, Xabier and García-Miguel, José M. and Garcia, Marcos},
  booktitle = {Proceedings of the 13th Workshop on NLP for Similar Languages, Varieties and Dialects},
  year = {2026},
  doi = {10.18653/v1/2026.vardial-1.5},
  url = {https://doi.org/10.18653/v1/2026.vardial-1.5}
}

References

28 references, 6 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Pablo Gamallo and Isaac González. 2012. Deppattern: A multilingual dependency parser. In Demo Session of the International Conference on Computational Processing of the Portuguese Language (PROPOR 2012), Coimbra, Portugal. find
  2. Benedikt Adelmann, Wolfgang Menzel, and Heike Zinsmeister. 2021. The impact of word embeddings on neural dependency parsing. In Proceedings of the 17th Conference on Natural Language Processing (KONVENS 2021), pages 1–13, Düsseldorf, Germany. find
  3. KONVENS 2021 Organizers. find
  4. Pablo Gamallo, Pablo Rodríguez, Iria de Dios-Flores, Susana Sotelo, Silvia Paniagua, Daniel Bardanca, José Ramom Pichel, and Marcos Garcia. 2024. Open Generative Large Language Models for Galician. Procesamiento del Lenguaje Natural, 73(0):259–270. find
  5. Marcos Garcia. 2016. Universal dependencies guidelines for the galician-treegal treebank. In Technical Report. LyS Group, Universidade da Coruña. find
  6. Ricardo Carvalho Calero. 1985. O problema ortográfico. Agália: Publicaçom internacional da Associaçom Galega da Lingua, 2:127–134. find
  7. Marcos Garcia. 2021. Exploring the representation of word meanings in context: A case study on homonymy and synonymy. In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long… find
  8. Alexis Conneau, Kartikay Khandelwal, Naman Goyal, Vishrav Chaudhary, Guillaume Wenzek, Francisco Guzmán, Edouard Grave, Myle Ott, Luke Zettlemoyer, and Veselin Stoyanov. 2020. Unsupervised cross-lingual representation learning at scale. In Proceedings of the 58th Annual Meeting of the Association fo… find
  9. Fadli Aulawi Al Ghiffari, Ika Alfina, and Kurniawati Azizah. 2023. Cross-lingual transfer learning for Javanese dependency parsing. In Proceedings of the 13th International Joint Conference on Natural Language Processing and the 3rd Conference of the Asia-Pacific Chapter of the Association for Compu… find
  10. Barbara Plank, Héctor Martínez Alonso, Željko Agić, Danijela Merkler, and Anders Søgaard. 2015. Do dependency parsing metrics correlate with human judgments? In Proceedings of the Nineteenth Conference on Computational Natural Language Learning, pages 315–320, Beijing, China. Association for Comput… find
  11. Goran Glavaš and Ivan Vulić. 2021. Climbing the tower of treebanks: Improving low-resource dependency parsing via hierarchical source selection. In Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021, pages 4878–4888, Online. Association for Computational Linguistics. find
  12. Peng Qi, Yuhao Zhang, Yuhui Zhang, Jason Bolton, and Christopher D. Manning. 2020. Stanza: A python natural language processing toolkit for many human languages. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics: System Demonstrations, pages 101–108, Online.… find our card
  13. Xavier Gómez Guinovart. 2017. Recursos integrados da lingua galega para a investigación lingüística. In Gal-læcia: Estudos de lingüística portuguesa e galega, pages 1037–1048. Universidade de Santiago de Compostela. find
  14. Xulia Sánchez-Rodríguez, Albina Sarymsakova, Laura Castro, and Marcos Garcia. 2024. Increasing manually annotated resources for Galician: the parallel Universal Dependencies treebank. In Proceedings of the 16th International Conference on Computational Processing of Portuguese, pages 587–592, Santia… find
  15. Claudiu Daniel Hromei, Danilo Croce, and Roberto Basili. 2024. U-deppllama: Universal dependency parsing via auto-regressive large language models. Italian Journal of Computational Linguistics, 10(1):21–38. find
  16. Suramya Jadhav, Abhay Shanbhag, Amogh Thakurdesai, Ridhima Sinare, and Raviraj Joshi. 2025. On limitations of LLM as annotator for low resource languages. In Proceedings of the 8th International Conference on Natural Language and Speech Processing (ICNLSP-2025), pages 277–282, Southern Denmark Unive… find
  17. Albina Sarymsakova, Xulia Sánchez Rodríguez, and Marcos García González. 2024. Towards accurate dependency parsing for galician with limited resources. Procesamiento del Lenguaje Natural, 73:247–257. find
  18. Milan Straka. 2018. UDPipe 2.0 prototype at CoNLL 2018 UD shared task. In Proceedings of the CoNLL 2018 Shared Task: Multilingual Parsing from Raw Text to Universal Dependencies, pages 197–207, Brussels, Belgium. Association for Computational Linguistics. find our card
  19. Dan Kondratyuk and Milan Straka. 2019. 75 languages, 1 model: Parsing Universal Dependencies universally. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), pages 2779–279… find our card
  20. Milan Straka and Jana Straková. 2017. Tokenizing, POS tagging, lemmatizing and parsing UD 2.0 with UDPipe. In Proceedings of the CoNLL 2017 Shared Task: Multilingual Parsing from Raw Text to Universal Dependencies, pages 88–99, Vancouver, Canada. Association for Computational Linguistics. find our card
  21. Lucelene Lopes and Thiago Pardo. 2024. Towards portparser - a highly accurate parsing system for Brazilian Portuguese following the Universal Dependencies framework. In Proceedings of the 16th International Conference on Computational Processing of Portuguese - Vol. 1, pages 401–410, Santiago de Com… find
  22. Nasrin Taghizadeh and Heshaam Faili. 2022. Cross-lingual transfer learning for relation extraction using universal dependencies. Computer Speech & Language, 71:101265. find
  23. David Vilares Calvo, Marcos García González, and Carlos Gómez Rodríguez. 2021. Bertinho: Galician bert representations. Procesamiento del Lenguaje Natural, 66:13–26. find
  24. Joakim Nivre, Marie-Catherine de Marneffe, Filip Ginter, Jan Hajič, Christopher D. Manning, Sampo Pyysalo, Sebastian Schuster, Francis Tyers, and Daniel Zeman. 2020. Universal Dependencies v2: An evergrowing multilingual treebank collection. In Proceedings of the Twelfth Language Resources and Eval… find
  25. Daniel Zeman, Martin Popel, Milan Straka, Jan Hajič, Joakim Nivre, Filip Ginter, Juhani Luotolahti, Sampo Pyysalo, Slav Petrov, Martin Potthast, Francis Tyers, Elena Badmaeva, Memduh Gokirmak, Anna Nedoluzhko, Silvie Cinková, Jan Hajič jr., Jaroslava Hlaváčová, Václava Kettnerová, Zdeňka Urešová… find
  26. CoNLL 2017 shared task: Multilingual parsing from raw text to Universal Dependencies. In Proceedings of the CoNLL 2017 Shared Task: Multilingual Parsing from Raw Text to Universal Dependencies, pages 1–19, Vancouver, Canada. Association for Computational Linguistics. find our card
  27. Joakim Nivre, Johan Hall, and Jens Nilsson. 2004. Memory-based dependency parsing. In Proceedings of the Eighth Conference on Computational Natural Language Learning (CoNLL-2004) at HLT-NAACL 2004, pages 49–56, Boston, Massachusetts, USA. Association for Computational Linguistics. find our card
  28. Parsers G P S I F A Meishan Zhang, Gongyao Jiang, Shuang Liu, Jing Chen, and Min Zhang. 2024. LLM-assisted data augmentation for Chinese dialogue-level dependency parsing. Computational Linguistics, 50(3):867–891. find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/urbieta-2026-language-mixture-develop-accurate-galician/