Elamur Research⌕
☰
home / papers / 2026 / Unsupervised Labelling of Mutation Triggers in Welsh

Unsupervised Labelling of Mutation Triggers in Welsh

Summary

Initial consonant mutation in Welsh is difficult for learners and NLP systems. For example, teulu can surface as deulu, nheulu or theulu. Existing tools such as the CyTag2 tagger can mark a token as mutated, but none indicates what caused the change. The authors define a new task, mutation trigger labelling, which assigns each mutated word the lexical, grammatical or syntactic factor responsible. They also propose a taxonomy of trigger types.

No annotated training data exist, so they test two unsupervised baselines on examples drawn from grammar reference books and public corpora:

  • a pipeline that adds a new layer of CG3 Constraint Grammar rules on top of CyTag2's mutation tags
  • multilingual LLMs prompted with a few examples

The study is framed as a feasibility test that sets first baselines. The text supplied here omits the results section, so no scores are reported in this summary. The work matters because it moves Welsh NLP from detecting mutations to explaining them. This could support learner feedback, corpus annotation and explainable grammatical analysis for low-resource languages.

Key points

  • CyTag2 detects mutation by "unmutating" unknown tokens and checking the candidates against its lexicons. For instance, fara is tested as mara and bara, and bara matches as a soft mutation.
  • CG3 rules turn context into trigger labels. In Mae'r ferch bert yn dda, the labels are +FEM-y (feminine noun after the article), +FEM-adj (adjective after a feminine noun) and +RULE-yn (adjective after predicative yn).
  • The evaluation data include examples from Thorne's A Comprehensive Welsh Grammar. These are copyrighted and cannot be redistributed, which limits reproducibility. Related resources discussed include CorCenCC (about 11 million tokens) and the Siarad corpus.
  • The authors acknowledge that coverage of contemporary usage is incomplete and that the data lean towards a prescriptive standard. They position the tool as descriptive rather than normative.
  • Code is available at github.com/ngutierrezrolon/welsh-mutation-trigger-labelling.

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

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.63317/37oxwc9pnyfv

@inproceedings{gutierrezrolon2026unsupervised,
  title = {{Unsupervised Labelling of Mutation Triggers in Welsh}},
  author = {Gutiérrez-Rolón, Nicolás and Alva-Manchego, Fernando},
  booktitle = {Proceedings of the Fifteenth Language Resources and Evaluation Conference},
  year = {2026},
  doi = {10.63317/37oxwc9pnyfv},
  url = {https://doi.org/10.63317/37oxwc9pnyfv}
}

References

24 references, 0 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. G. M. Awbery. 1986. Moves towards a simpler, binary mutation system. In H. Andersen, editor, Sandhi phenomena in the languages of Europe, volume 33 of Trends in Linguistics, Studies and Monographs, pages 161–166. Mouton de Gruyter, Berlin. find
  2. M. J. Ball. 1984. Sociolinguistic aspects of the Welsh mutation system. Ph.D. thesis, University of Wales. Unpublished doctoral dissertation. find
  3. Martin J. Ball and Nicole Müller. 1992. Mutation in Welsh. Routledge, London / New York. find
  4. N. Coupland and M. J. Ball. 1989. Welsh and english in contemporary wales: Sociolinguistic issues. Contemporary Wales, 3:7–40. find
  5. Daniel Cunliffe, Andreas Vlachidis, Daniel Williams, and Douglas Tudhope. 2022. Natural language processing for under-resourced languages: Developing a welsh natural language toolkit. Computer Speech & Language, 72:101311. find
  6. John Davies. 2014. The History of the Welsh Language. University of Wales Press, Cardiff. find
  7. Margaret Deuchar, Peredur Webb-Davies, and Kevin Donnelly. 2018. Building and Using the Siarad Corpus: Bilingual Conversations in Welsh and English, volume 81 of Studies in Corpus Linguistics. John Benjamins. find
  8. Kevin Donnelly. 2013. Kynulliad3: A corpus of 350,000 aligned Welsh and English sentences from the third assembly (2007–2011) of the National Assembly for Wales. http://cymraeg.org.uk/kynulliad3. ↗
  9. Gareth King. 1993. Modern Welsh: A Comprehensive Grammar. Routledge, London. find
  10. Dawn Knight, Steven Morris, Tess Fitzpatrick, Paul Rayson, Irena Spasić, Enlli Môn Thomas, Alex Lovell, Jonathan Morris, Jeremy Evas, Mark Stonelake, Laura Arman, Joshua Davies, Ignatius Ezeani, Steven Neale, Jennifer Needs, Scott Piao, Mair Rees, Gareth Watkins, Lowri Williams, Vignesh Muralidaran,… find
  11. Dawn Knight, Steven Morris, Tony Fitzpatrick, Paul Rayson, Irena Spasić, and Enlli M. Thomas. 2020b. The national corpus of contemporary welsh: Project report \| y corpws cenedlaethol cymraeg cyfoes: Adroddiad y prosiect. ArXiv preprint arXiv:2010.05542. — [arXiv:2010.05542](https://arxiv.org/abs/2010.05542) find
  12. Dawn Knight, Stuart Morris, Bethan Tovey-Walsh, Tom Fitzpatrick, and Laura Anthony. 2020c. Yr amliadur: Frequency lists for contemporary welsh. Technical report, Cardiff University. find
  13. D. Geraint Lewis. 2021. Y Treigladur - A Check-List of Welsh Mutations (Argraffiad Newydd). Y Lolfa. find
  14. Ingo Mittendorf and Louisa Sadler. 2006. A treatment of welsh initial mutation. In Proceedings of the LFG ’06 Conference, pages 343–364, Stanford, CA. CSLI Publications. find
  15. Steve Neale, Kevin Donnelly, Gareth Watkins, and Dawn Knight. 2018. Leveraging lexical resources and constraint grammar for rule-based part-of-speech tagging in welsh. In Proceedings of the 11th Edition of the Language Resources and Evaluation Conference (LREC 2018), volume 3, pages 168–173, Miyazak… find
  16. Office for National Statistics. 2011. 2011 census for wales. https://www.ons.gov.uk/census/2011census. Accessed 13 October 2025. ↗
  17. Milan Straka and Jana Straková. 2017. Udpipe: Trainable pipeline for processing conll-u files performing tokenization, morphosyntactic tagging, lemmatization and dependency parsing. In Proceedings of the CoNLL 2017 Shared Task: Multilingual Parsing from Raw Text to Universal Dependencies, pages 88–9… find
  18. David A. Thorne. 1993. A Comprehensive Welsh Grammar. Blackwell, Oxford. find
  19. David A. Thorne. 1997. Taclo’r Treigladau. Gwasg Gomer, Llandysul. find
  20. Francis M. Tyers. 2009. Rule-based augmentation of training data in Breton-French statistical machine translation. In Proceedings of the 13th Annual Conference of the European Association for Machine Translation, Barcelona, Spain. European Association for Machine Translation. find
  21. Francis M. Tyers and Kevin Donnelly. 2009. Apertium-cy: A collaboratively-developed free rbmt system for welsh to english. The Prague Bulletin of Mathematical Linguistics. find
  22. Welsh Government. 2011. Welsh language measure 2011. https://www.legislation.gov.uk/mwa/2011/1/contents. Accessed 13 October 2025. ↗
  23. Welsh Government. 2017. Cymraeg 2050: A million welsh speakers. https://www.gov.wales/cymraeg-2050-welsh-language-strategy. Accessed 13 October 2025. ↗
  24. Welsh Government. 2018. Welsh language technology action plan. https://www.gov.wales/welsh-language-technology-action-plan. Accessed 13 October 2025. ↗

↑ back to top

This page in Markdown →