Elamur Research⌕
☰
home / papers / 2026 / Evaluating Data Augmentation Strategies for Training Spanish Misspelling Detection Models

Evaluating Data Augmentation Strategies for Training Spanish Misspelling Detection Models

Paper · 2026 · Manuel Castillo-Sancho, Jordi Porta and Asunción Gómez-Pérez

Summary

Discriminative models for spotting spelling errors need training data with authentic mistakes, and such data is scarce. This paper asks which kind of synthetic error injection best supports Spanish misspelling detection. The authors use CORRSIC, a linguist-annotated corpus of real errors drawn from CREA. They split it 60/40: the larger part supplies error statistics and the smaller part serves as the test set. Three generators are compared:

  • a random character-perturbation model;
  • a keyboard model built on a full Spanish layout, including its Shift, AltGr and diacritic layers;
  • a statistical model whose edit patterns come from Damerau-Levenshtein alignments and are encoded as a weighted finite-state transducer.

The authors also vary how target words are chosen (randomly or by length) and test two augmentation configurations. The statistical generator comes closest to real error distributions. In detection experiments, scores rise as training data grows, while the word-selection choice matters little. The study gives empirical support for grounding synthetic errors in real data when building spell-checkers.

Key points

  • CORRSIC contains 49,676 text segments, 1,696,344 words and 49,958 annotated misspellings. That is a density of 2.95%, covering 30,008 distinct misspelling types.
  • The statistical model achieves the lowest Jensen–Shannon divergence from real errors, at 0.148 nats.
  • The WFST operates over the tropical semiring, with edits weighted by negative log probability. Misspellings are sampled from the k = 1000 shortest paths.
  • The keyboard model handles only substitutions, with probability decaying exponentially as key distance grows. It does not model insertions or deletions.
  • The configurations AUG1 and AUG2 show complementary precision/recall trade-offs. A classifier identifying which generator produced each error reached diagonal shares of about 0.77 (random), 0.85 (keyboard) and 0.79 (statistical).

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

 Institutions

Also on the byline: Real Academia Española.

 Fields, topics and concepts

📚 Shelves 6The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 worksDecision Sciences · 209 worksManagement Science and Operations Research · 108 works

🧭 Topics 1Research subjects and languages the work is about.Evaluation & benchmarks · 1,220 works

Cite

Cite as Manuel Castillo-Sancho, Jordi Porta and Asunción Gómez-Pérez. 2026. Evaluating Data Augmentation Strategies for Training Spanish Misspelling Detection Models. Proceedings of the Third Workshop on Computation and Written Language (CAWL 2026) @ LREC 2026.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.63317/3mw3y4ovzwsz

@inproceedings{castillosancho2026evaluating,
  title = {{Evaluating Data Augmentation Strategies for Training Spanish Misspelling Detection Models}},
  author = {Manuel Castillo-Sancho, Jordi Porta and Asunción Gómez-Pérez},
  booktitle = {Proceedings of the Third Workshop on Computation and Written Language (CAWL 2026) @ LREC 2026},
  year = {2026},
  doi = {10.63317/3mw3y4ovzwsz},
  url = {https://doi.org/10.63317/3mw3y4ovzwsz}
}

References

14 references, 1 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Christopher Bryant, Zheng Yuan, Muhammad Reza Qorib, Hannan Cao, Hwee Tou Ng, and Ted Briscoe. 2023. Grammatical error correction: A survey of the state of the art. Computational Linguistics, 49(3):643–701. find
  2. Manuel Castillo-Sancho. 2025. Modelado de er-ratas para el entrenamiento de un verificador ortográfico en español. Master’s thesis, Universidad Politécnica de Madrid, Madrid, Spain. find
  3. Fred J. Damerau. 1964. A technique for computer detection and correction of spelling errors. Commun. ACM, 7(3):171–176. find
  4. Michael Flor, Yoko Futagi, Melissa Lopez, and Matthew Mulholland. 2015. Patterns of misspellings in L2 and L1 English: a view from the ETS Spelling Corpus. Bergen Language and Linguistics Studies, 6. find
  5. S. Kullback and R. A. Leibler. 1951. On Information and Sufficiency. The Annals of Mathematical Statistics, 22(1):79–86. find
  6. Vladimir I Levenshtein. 1966. Binary codes capable of correcting deletions, insertions and reversals. Soviet Physics Doklady, 10:707. find
  7. Jiahua Lin. 1991. Divergence measures based on the Shannon entropy. IEEE Transactions on Information Theory, 37(1):145–151. \#\# Acknowledgements Jiawei Liu, Min Huang, and Qinghai Miao. 2024. Mitigating spurious correlations in named entity recognition models through counterfactual data augmentation.… find
  8. The authors thank the reviewers for their valuable comments and suggestions, which contributed to improving the quality of the manuscript and to clarifying its scope and limitations. This publication results from work undertaken within the framework of the LEIA (Lengua Española e Inteligencia Artifi… find
  9. Nikita Martynov, Mark Baushenko, Alexander Abramov, and Alena Fenogenova. 2023. Augmentation methods for spelling corruptions. In Proceedings of the International Conference “Dialogue 2023”, volume 2023. find
  10. Mehryar Mohri. 2002. Semiring frameworks and algorithms for shortest-distance problems. J. Autom. Lang. Comb., 7(3):321–350. find
  11. Mehryar Mohri. 2009. Weighted Automata Algorithms, pages 213–254. Springer Berlin Heidelberg, Berlin, Heidelberg. find
  12. Lance A. Ramshaw and Mitchell P. Marcus. 1995. Text Chunking Using Transformation-Based Learning. In Third Workshop on Very Large Corpora, pages 82–94.
  13. Wenqian Ye, Luyang Jiang, Eric Xie, Guangtao Zheng, Yunsheng Ma, Xu Cao, Dongliang Guo, Daiqing Qi, Zeyu He, Yijun Tian, Megan Coffee, Zhe Zeng, Sheng Li, Ting-hao, Huang, Ziran Wang, James M. Rehg, Henry Kautz, and Aidong Zhang. 2025. The clever hans mirage: A comprehensive survey on spurious corre… find
  14. Can Özbey, Hatice Altınok, and Mustafa Umut Demirezen. 2022. A Novel Probabilistic Framework for Modeling Spelling Errors by Term Length and Frequency. In 2022 3rd International Informatics and Software Engineering Conference (IISEC), pages 1–6. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/sancho-2026-evaluating-data-augmentation-strategies-training/