Elamur Research⌕
☰
home / papers / 2026 / LLM-Assisted Spanish Dialect Corpus Construction

LLM-Assisted Spanish Dialect Corpus Construction

Summary

Spanish dialect resources usually cover only a few Latin American varieties, and large corpora spanning the whole region are seldom free to reuse. The authors address this with a parallel, multi-dialect corpus of American Spanish that carries pragmatic and domain labels. They first assembled a seed set of 2,000 neutral-Spanish sentences from three sources: a Spanish Wikipedia dump, dialogue lines from Project Gutenberg books, and random or synthetic sentences generated with Gemini and ChatGPT-5. ChatGPT-5 was then prompted zero-shot to rewrite every sentence into country-level dialects. These dialects are organized into five regions (Mexican, Central American, Caribbean, South American, Rioplatense), and a neutral Castilian variety serves as the reference. ChatGPT-5 also assigned pragmatic tags and one domain tag per sentence. For the Wikipedia subset, domains came from Wikipedia's own categories. The raw output was cleaned using scripts, constrained LLM editing and manual review, which removed hallucinations and switches into other languages. The result is 40,000 sentences in JSON and plain text. This provides a reusable method for building annotated dialect data.

Key points

  • Seed data: the seed lexicon has 2,000 sentences: 25% Wikipedia, 25% Project Gutenberg and 50% random/synthetic.
    • The Wikipedia dump is dated 10/01/2025 and was filtered to articles referencing Latin American countries.
    • Gutenberg sentences were chosen by dialogue markers.
  • Final size: the corpus has 40,000 sentences, with 2,000 parallel sentences for each of 20 varieties (American dialects plus neutral Castilian).
  • Inconsistent dialect counts: the paper also mentions 19 dialects in places. Its prompt and its country table do not list the same countries; for example, Peru appears in the prompt but not in the table.
  • Pragmatic labels: a sentence can receive several labels, such as greeting, informal, request, offer, refusal or thanks. For example, ¿Cómo estás? is tagged informal + greeting.
  • Scope: Spain-internal dialects and Equatorial Guinean Spanish are excluded.
  • Evaluation and future work: the provided text reports no quantitative quality evaluation. The authors plan a companion speech corpus.

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 10.63317/3zod9jyperib

@inproceedings{vidal2026llm,
  title = {{LLM-Assisted Spanish Dialect Corpus Construction}},
  author = {VIDAL, Jessica Claribel RAMIREZ and Ouchi, Hiroki and Sakti, Sakriani},
  booktitle = {Proceedings of the SIGUL 2026 Joint Workshop with ELE, EURALI, and DCLRL: Towards Inclusivity and Equality: Language Resources and Technologies for Under-Resourced and Endangered Languages},
  year = {2026},
  doi = {10.63317/3zod9jyperib},
  url = {https://doi.org/10.63317/3zod9jyperib}
}

References

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

  1. Ataman, D., Birch, A., Habash, N., Federico, M., Koehn, P., & Cho, K. (2025). Machine translation in the era of large language models: a survey of historical and emerging problems. Information, 16(9), 723. find
  2. Abdelaziz, A. A. A., Elneima, A. H., & Darwish, K. (2024, May). LLM-based MT data creation: Dialectal to MSA translation shared task. In Proceedings of the 6th Workshop on Open-Source Arabic Corpora and Processing Tools (OSACT) with Shared Tasks on Arabic LLMs Hallucination and Dialect to MSA Machin… find
  3. Abe, K., Matsubayashi, Y., Okazaki, N., & Inui, K. (2018). Multi-dialect neural machine translation and dialectometry. In Proceedings of the 32nd Pacific Asia Conference on Language, Information and Computation. find
  4. Carcelén-Guerrero, A., Posio, P., Kachel, S., & Uclés-Ramada, G. (2025). CoLaGe: Corpus for the Study of Language and Gender in two varieties of spoken Spanish. Corpora, 20(2), 269-285. find
  5. Fuchs, M., & González, P. (2022). Perfect-Perfective variation across Spanish dialects: a parallel-corpus study. Languages, 7(3), 166. find
  6. Hernández-Mena, C. D., Meza-Ruiz, I. V., & Herrera-Camacho, J. A. (2017). Automatic speech recognizers for Mexican Spanish and its open resources. Journal of applied research and technology, 15(3), 259-270. find
  7. Kawasaki, Y. (2026). Digital Linguistic Bias in Spanish: Evidence from Lexical Variation in LLMs. arXiv preprint arXiv:2602.09346. — [arXiv:2602.09346](https://arxiv.org/abs/2602.09346) find
  8. Lu, H., Cheng, G., Luo, L., Zhang, L., Qian, Y., & Zhang, P. (2025, April). Slide: Integrating speech language model with llm for spontaneous spoken dialogue generation. In ICASSP 2025-2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) (pp. 1-5). IEEE. Ma, B., Li,… find
  9. McCallum, L., & Mizumoto, A. (2025). Using Llms for Corpus Linguistics Research: Promise, Disappointment and Progression. Disappointment and Progression. Available at SSRN: https://ssrn.com/abstract=5224441 or http://dx.doi.org/10.2139/ssrn.5224441 — [doi:10.2139/ssrn.5224441](https://doi.org/10.2139/ssrn.5224441) ↗
  10. Molina Martos, Isabel. (2024). Spanish Dialect Classifications. Dialectología. 12. 309-342. 10.1344/Dialectologia2024.2024.10. — [doi:10.1344/dialectologia2024.2024.10](https://doi.org/10.1344/dialectologia2024.2024.10) find
  11. Morin, C., & Larsson, M. M. (2025). A large-scale, unsupervised pipeline for automatic corpus annotation using LLMs: variation and change in the English consider construction. arXiv preprint arXiv:2510.12306. find
  12. Pozhemetskiy, R. (2025). Translation into a low-resource language system: the ability of ChatGPT to create texts in Andalusian dialect of Spanish language (Master's thesis, Itä-Suomen yliopisto). find
  13. Quesada Pacheco, M. Á. (2021). Dialectología histórica del español de América Central. Nivel fonético-fonológico. Revista de Historia de la Lengua Española, (16), 67-100. find
  14. Sakai, Y., Makinae, M., Kamigaito, H., & Watanabe, T. (2024, November). Simultaneous interpretation corpus construction by large language models in distant language pair. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing (pp. 22375-22398). find
  15. Schlumpf, S., & Carreira, S. (2024). Presentación de un corpus para el estudio del español actual. find
  16. Shen, J., Tenenholtz, N., Hall, J. B., Alvarez-Melis, D., & Fusi, N. (2024). Tag-LLM: Repurposing general-purpose LLMs for specialized domains. arXiv preprint arXiv:2402.05140. — [arXiv:2402.05140](https://arxiv.org/abs/2402.05140) find
  17. Weissweiler, L., Köksal, A., & Schütze, H. (2025). Hybrid Human-LLM Corpus Construction and LLM Evaluation for the Caused-Motion Construction. Northern European Journal of Language Technology, 11(1), 27-57. find
  18. Yakhni, S., & Chehab, A. (2025). Fine-tuning llms for low-resource dialect translation: The case of lebanese. arXiv preprint arXiv:2505.00114. — [arXiv:2505.00114](https://arxiv.org/abs/2505.00114) find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/vidal-2026-llm-assisted-spanish-dialect-corpus/