Elamur Research⌕
☰
home / papers / 2026 / CorSpell: Introducing a Semiautomatic Tool for Spelling Normalization in Brazilian Portuguese

CorSpell: Introducing a Semiautomatic Tool for Spelling Normalization in Brazilian Portuguese

Summary

Learner and exam texts contain many non-standard spellings. These variants disrupt tokenization and POS tagging and skew frequency-based measures such as lexical diversity. The authors faced this problem with CorCel, a corpus of roughly 15,000 typed and proofread written responses from Celpe-Bras, Brazil's official proficiency exam for Portuguese as an additional language. Normalizing that many texts by hand would be slow, expensive and inconsistent, so they built CorSpell. It is an open-source web tool that pairs automatic generation of replacement candidates with human validation. Texts are tokenized and checked against Portuguese dictionaries and a whitelist, and a locally hosted LLM proposes context-sensitive corrections. Annotators then accept or edit these suggestions, track their progress and export normalized versions. The authors report shorter annotation time, fewer errors and easier collaboration, but formal benchmarks of the LLM component are still pending. The tool yields a cleaner CorCel for corpus research. The validated annotations will also serve as training data for language models that will later normalize spelling automatically and model Portuguese proficiency.

Key points

  • CorCel holds about 15,000 Celpe-Bras texts drawn from 70,000 image samples. The texts come from four exam editions and are rated on six proficiency levels.
  • The preprocessing pipeline uses spacy-udpipe for tokenization. It combines a common dictionary, an affix dictionary, a customizable whitelist and a self-hosted gemma 3 12B model to flag and correct words.
  • Users can upload zipped batches of .docx and .txt files. Corrections are stored in a database and can be exported. The code is on GitHub (jhssilv/corcel-platform).
  • The authors position CorSpell against SVALA, TEITOK, VARD 2, INCEpTION and the agricultural-domain pipeline by Garcia et al. (2023). In their view, none of these combines Brazilian Portuguese suggestions with an interface suitable for non-programmers.
  • The main annotators are undergraduate Modern Languages students, so the design deliberately favors simplicity over feature-rich platforms.

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 · Machine translation

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.63317/22uey2oj2w49

@inproceedings{schoffen2026corspell,
  title = {{CorSpell: Introducing a Semiautomatic Tool for Spelling Normalization in Brazilian Portuguese}},
  author = {Schoffen, Juliana and Balreira, Dennis Giovani and Stumpf, Elisa Marchioro and Goulart, Larissa and Kuhn, Tanara Zingano and Nunes, Rafael Oleques and Pazzinato, Gabriel Ricci and Hanauer, Isadora Dahmer and Silva, José Henrique de Souza and Divino, Luiza Sarmento and Matte, Marine},
  booktitle = {Proceedings of the Fifteenth Language Resources and Evaluation Conference},
  year = {2026},
  doi = {10.63317/22uey2oj2w49},
  url = {https://doi.org/10.63317/22uey2oj2w49}
}

References

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

  1. Laurence Anthony. 2024. Antconc (version 4.3.1) [computer software]. https://www.laurenceanthony.net/software/AntConc. Tokyo, Japan: Waseda University. ↗
  2. Alistair Baron and Paul Rayson. 2008. Vard2: A tool for dealing with spelling variation in historical corpora. ENEM stands for Exame Nacional do Ensino Médio and is Brazil’s national high school exam, used to assess students’ performance and as the main gateway to higher education. find
  3. Luiza Sarmento Divino. 2021. Índices lexicais de análise para a caracterização dos níveis intermediário e avançado superior no exame celpe-bras: uma pesquisa guiada por corpus. Undergraduate thesis. find
  4. Luiza Sarmento Divino. 2024. Contribuições da linguística de corpus para a definição de níveis de proficiência escrita no exame celpe-bras. Master’s thesis, Universidade Federal do Rio Grande do Sul. find
  5. Luana Q Garcia, Miguel H Chinellato, Helena de M Caseli, and Leandro HM Oliveira. 2023. Pipeline para identificação de erros lexicais e geração de sugestões de correção. In Simpósio Brasileiro de Tecnologia da Informação e da Linguagem Humana (STIL), pages 357–361. SBC. find
  6. Sylviane Granger, Helen Swallow, and Jennifer Thewissen. 2022. The louvain error tagging manual. version 2.0. find
  7. Isadora Dahmer Hanauer. 2022. Influência das inadequações ortográficas em análise de tarefa escrita do celpe-bras guiada por corpus. find
  8. Isadora Dahmer Hanauer. 2023. Caracterização dos níveis intermediário e avançado superior do exame celpe-bras em produções escritas de examinandos no gênero carta/e-mail: contribuições de uma análise guiada por corpus. Undergraduate thesis. find
  9. Isadora Dahmer Hanauer. 2026. Tarefas de compreensão audiovisual para produção escrita no celpe-bras: análise de produções de examinandos guiada por corpus. Master’s thesis, Universidade Federal do Rio Grande do Sul. find
  10. Maarten Janssen. 2016. Teitok: Text-faithful annotated corpora. In Proceedings of the Tenth International Conference on Language Resources and Evaluation (LREC’16), pages 4037–4043. find
  11. Adam Kilgariff, Pavel Ryckly, Pavel Smrz, and David Tugwell. 2004. The sketch engine. i: Williams, g. & s. vessier. In Proceedings of the Eleventh EURALEX International Congress, Lorient, France July 6–10, pages 105–114. find
  12. Jan-Christoph Klie, Michael Bugert, Beto Boullosa, Richard Eckart de Castilho, and Iryna Gurevych. 2018. The INCEpTION platform: Machine-assisted and knowledge-oriented interactive annotation. In Proceedings of the 27th International Conference on Computational Linguistics: System Demonstrations, pages 5–9, Santa Fe, New Mexico. Association for Computational Linguistics. find
  13. Simone Paula Kunrath. 2019. Os descritores gerais e a progressão dos níveis de proficiência do exame CELPE-BRAS. Ph.D. thesis, Universidade Federal do Rio Grande do Sul. find
  14. Lucelene Lopes, Magali Duran, Paulo Fernandes, and Thiago Pardo. 2022. PortiLexicon-UD: a Portuguese lexical resource according to Universal Dependencies model. In Proceedings of the Thirteenth Language Resources and Evaluation Conference, pages 6635–6643, Marseille, France. European Language Resources Association. find
  15. Kaiane Mendel. 2019. Proficiência e autoria na avaliação integrada de leitura e escrita do exame celpe-bras. Master’s thesis, Universidade Federal do Rio Grande do Sul. find
  16. Amanda Michel Raupp. 2024. Características lexicais das produções escritas do exame celpe-bras na tarefa 3 de 2016-2: uma pesquisa guiada por corpus. Undergraduate thesis. find
  17. Lisa Rudebeck and Gunlög Sundberg. 2021. Swell correction annotation guidelines. find
  18. Juliana Schoffen, Elisa Stumpf, Deise Amaral, Luiza Divino, Isadora Hanauer, Isabel Lisboa, Amanda Raupp, and Brenda Xavier. 2024. Compilation and tagging of a corpus with celpe-bras texts. In Proceedings of the 16th International Conference on Computational Processing of Portuguese - Vol. 1, pages 627–632, Santiago de Compostela, Galicia/Spain. Association for Computational Lingustics. find
  19. Juliana Roquele Schoffen, Elisa Marchioro Stumpf, Luiza Sarmento Divino, Isadora Dahmer Hanauer, Deise Amaral, Amanda Michel Raupp, and Brenda de Souza Xavier. 2025. Corcel: A brazilian portuguese <i>corpus</i> of celpe-bras exam written texts. Revista Brasileira de Linguística Aplicada, 25(1):e50034. find
  20. Júlia Luiz Sostruznik da Silveira. 2023. O uso de conjunções em produções escritas no exame celpe-bras: um estudo baseado em corpus. Undergraduate thesis. find
  21. Gabrielle Rodrigues Sirianni. 2020. Entre a certificação e a não certificação no celpe-bras: um estudo sobre os níveis de proficiência na parte escrita do exame. Master’s thesis, Universidade Federal do Rio Grande do Sul. find
  22. Elisa Stumpf, Juliana Schoffen, Luiza Divino, Isadora Hanauer, Amanda Raupp, and Brenda Xavier. 2025. Corpus-driven lexical analyses of corcel: a comparative analysis of preliminary findings of written proficiency in portuguese as an additional language. In Anais do XVI Simpósio Brasileiro de Tecnologia da Informação e da Linguagem Humana, pages 673–681, Porto Alegre, RS, Brasil. SBC. find
  23. Mats Wirén, Arild Matsson, Dan Rosén, and Elena Volodina. 2019. Svala: Annotation of second-language learner text based on mostly automatic alignment of parallel corpora. In CLARIN Annual Conference, Pisa, Italy, 8-10 October, 2018, pages 222–234. Linköping University Electronic Press. find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/papers/schoffen-2026-corspell-introducing-semiautomatic-tool-spelling/