Elamur Research⌕
☰
home / papers / 2026 / Diacritic Restoration for Low-Resource Indigenous Languages: Case Study with Bribri and Cook Islands Māori

Diacritic Restoration for Low-Resource Indigenous Languages: Case Study with Bribri and Cook Islands Māori

Summary

Many Indigenous languages adopted writing only recently, so diacritics are used inconsistently or left out, which hampers automatic processing. This paper addresses diacritic restoration, which means recovering marks in text where they are missing, for two severely under-resourced languages. Bribri is a Chibchan language of Costa Rica whose spelling encodes tone, nasality and vowel laxness. Cook Islands Māori (CIM) is a Polynesian language that marks long vowels with a macron and glottal stops with a saltillo.

The authors benchmark three kinds of systems: character-based translation models, fine-tuned LLMs and zero-shot LLMs. They also track how accuracy grows with training data, compare different resource conditions, and run a small study on correcting inconsistent diacritics.

Fine-tuned character-level LLMs perform best, probably because working on UTF-8 bytes lets them break composite characters into parts. Massively multilingual models fall behind, and zero-shot prompting fails throughout. Dependable output appears at about 10,000 words of training text.

The work responds to explicit community requests for tools. It also analyses errors on tones and glottal stops, and it insists that spelling variation is not a mistake to be eliminated.

Key points

  • Bribri has about 7,000 speakers and at least three orthographies (Constenla et al. 2004, Jara 2018, Margery 2005); the experiments use Constenla's. It marks tone with three diacritics plus a glottal symbol, nasality with an underline and lax vowels with an umlaut. A single word, taî 'much', has been found spelled 14 different ways.
  • CIM has roughly 12,500 speakers in the Cook Islands and about 10,000 in the diaspora. The CIMR orthography (Nicholas 2017) uses the saltillo (U+A78C), which in real text is often replaced by an apostrophe (U+0027), a left quote (U+201C) or an ʻokina (U+02BB).
  • Main findings:
    • Fine-tuned character/byte-level LLMs come out on top.
    • Massively multilingual models underperform with this little data.
    • Zero-shot LLMs perform poorly in every setting.
    • Roughly 10,000 words of data is where reliable restoration begins.
  • Prior work on the related te reo Māori relied on much larger corpora. Cocks et al. (2011) reached 99% accuracy with a Naïve Bayes classifier on 3.8M words, and Nakano (2023) reached 96% with an RNN on 463k words.
  • Ethical stance: the authors frame normalization as the job of tool developers, not language users. They warn against promoting any single spelling variant as the de facto norm.

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 10.63317/4c9samfkamqy

@inproceedings{cotosolano2026diacritic,
  title = {{Diacritic Restoration for Low-Resource Indigenous Languages: Case Study with Bribri and Cook Islands Māori}},
  author = {Coto-Solano, Rolando and Li, Daisy and Ferraz, Manoela Teleginski and Sasse, Olivia and Krupka, Cha and Loaiciga, Sharid and Nicholas, Sally Akevai Tenamu},
  booktitle = {Proceedings of the Fifteenth Language Resources and Evaluation Conference},
  year = {2026},
  doi = {10.63317/4c9samfkamqy},
  url = {https://doi.org/10.63317/4c9samfkamqy}
}

References

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

  1. Milind Agarwal and Antonios Anastasopoulos. 2024. A Concise Survey of OCR for Low-Resource Languages. In Proceedings of the 4th Workshop on Natural Language Processing for Indigenous Languages of the Americas (AmericasNLP 2024), pages 88–102, Mexico City, Mexico. Association for Computational Lingui… find
  2. Sawsan Alqahtani, Ajay Mishra, and Mona Diab. 2020. A Multitask Learning Approach for Diacritic Restoration. arXiv preprint arXiv:2006.04016. find our card
  3. AmericasNLP. 2021. AmericasNLP 2021 Shared Task on Open Machine Translation. https://github.com/AmericasNLP/americasnlp 2021. [Accessed 25-10-2025]. ↗
  4. Carter Anderson, Mien Nguyen, and Rolando Coto-Solano. 2025. Unsupervised, semi-supervised and llm-based morphological segmentation for bribri. In Proceedings of the Fifth Workshop on NLP for Indigenous Languages of the Americas (AmericasNLP), pages 63–76. find
  5. Anthropic. 2025. Introducing Claude Sonnet 4.5. https://www.anthropic.com/news/cla ude-sonnet-4-5. ↗
  6. H Russell Bernard, George N Mbeh, and W Penn Handwerker. 2002. Does marking tone make tone languages easier to read? Human organization, 61(4):339–349. find
  7. Bible Society of South Pacific. 2014. Cook Islands Māori Revised New Testament: Digital Publication. Life.Church / YouVersion. find
  8. Steven Bird. 1999. When marking tone reduces fluency: an orthography experiment in Cameroon. Language and Speech, 42(1):83–115. find
  9. Shobhana L. Chelliah. 2021. Impact of Language Documentation, pages 91–94. Springer International Publishing, Cham. find
  10. John Cocks et al. 2011. A word-based approach for diacritic restoration in Māori. In Proceedings of the Australasian Language Technology Association Workshop 2011, pages 126–130. find
  11. Adolfo Constenla, Feliciano Elizondo, and Francisco Pereira. 2004. Curso Básico de Bribri. Editorial de la Universidad de Costa Rica. find
  12. Rolando Coto-Solano. 2017. Tonal reduction and literacy in Me’ph aa Vátháá. The University of Arizona. find
  13. Rolando Coto-Solano. 2021. Explicit tone transcription improves asr performance in extremely low-resource languages: A case study in bribri. In Proceedings of the first workshop on natural language processing for Indigenous languages of the Americas, pages 173–184. find
  14. Rolando Coto-Solano. 2022. Evaluating word embeddings in extremely under-resourced languages: A case study in bribri. In Proceedings of the 29th International Conference on Computational Linguistics, pages 4455–4467. find
  15. Rolando Coto-Solano, Tai Wan Kim, Alexander Jones, and Sharid Loáiciga. 2024. Multilingual models for asr in chibchan languages. In Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers),… find
  16. Rolando Coto-Solano, Sharid Loáiciga, and Sofía Flores-Solórzano. 2021. Towards universal dependencies for bribri. In Proceedings of the Fifth Workshop on Universal Dependencies (UDW, SyntaxFest 2021), pages 16–29. find
  17. Rolando Coto-Solano, Sally Akevai Nicholas, Samiha Datta, Victoria Quint, Piripi Wills, Emma Ngakuravaru Powell, Liam Koka’ua, Syed Tanveer, and Isaac Feldman. 2022a. Development of automatic speech recognition for the documentation of cook islands māori. In Proceedings of the Thirteenth Language R… find
  18. Rolando Coto-Solano, Sally Akevai Nicholas, Brittany Hoback, and Gregorio Tiburcio Cano. 2022b. Managing data workflows for untrained forced alignment: examples from costa rica, mexico, the cook islands, and vanuatu. The Open Handbook of Linguistic Data Management, 35. find
  19. Rolando Coto-Solano, Sally Akevai Nicholas, and Samantha Wray. 2018. Development of natural language processing tools for cook islands māori. In Proceedings of the Australasian Language Technology Association Workshop 2018, pages 26–33. find
  20. Rolando Coto-Solano and Sofía Flores Solórzano. 2016. Alineación forzada sin entrenamiento para la anotación automática de corpus orales de las lenguas indígenas de costa rica. Kánina, 40(4):175–199. find
  21. Guy De Pauw, Peter W Wagacha, and Gilles-Maurice De Schryver. 2007. Automatic Diacritic Restoration for Resource-Scarce Languages. In International Conference on Text, Speech and Dialogue, pages 170–179. Springer. find
  22. Virginia Eubanks. 2017. Automating inequality: How high-tech tools profile, police, and punish the poor. Picador. find
  23. Ignatius Ezeani, Mark Hepple, and Ikechukwu Onyenwe. 2016. Automatic Restoration of Diacritics for Igbo Language. In International Conference on Text, Speech, and Dialogue, pages 198–205. Springer. find
  24. Isaac Feldman and Rolando Coto-Solano. 2020. Neural machine translation models with back-translation for the extremely low-resource indigenous language bribri. In Proceedings of the 28th International Conference on Computational Linguistics, pages 3965–3976. find
  25. Gabriel H Gilbert, Rolando Coto-Solano, Sally Akevai Nicholas, Lauren Houchens, Sabrina Barton, and Trinity Pryor. 2025. Crosslingual dependency parsing of hawaiian and cook islands māori using universal dependencies. In Proceedings of the 18th International Conference on Parsing Technologies (IWPT… find
  26. Kyle Gorman and Yuval Pinter. 2025. Don’t Touch My Diacritics. In Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 2: Short Papers), pages 285–291, Albuquerque, New Mexico. Association for… find
  27. Kenneth Heafield. 2011. KenLM: Faster and smaller language model queries. In Proceedings of the Sixth Workshop on Statistical Machine Translation, pages 187–197, Edinburgh, Scotland. Association for Computational Linguistics. find our card
  28. Yasser Hifny. 2021. Recent advances in Arabic syntactic diacritics restoration. In ICASSP 2021-2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 7768–7772. IEEE. find
  29. Charles F Hockett. 1967. The quantification of functional load. Word, 23(1-3):300–320. find
  30. Andrej Hucko and Peter Lacko. 2018. Diacritics restoration using deep neural networks. In 2018 World Symposium on Digital Intelligence for Systems and Machines (DISA), pages 195–200. IEEE. find
  31. Instituto de Estudios de las Tradiciones Sagradas de Abia Yala IETSAY. 2001. Narraciones Bribris: Talamanca y Cabagra, II Edición, 2nd edition. Fundación Coordinadora de Pastoral Aborigen, San José, Costa Rica. find
  32. INEC. 2011. X Censo Nacional de Población y VI de Vivienda 2011 - Territorios Indígenas - Principales Indicadores Demográficos y Socioeconómicos. find
  33. Jesin James, Rolando Coto-Solano, Sally Akevai Nicholas, Joshua Zhu, Bovey Yu, Fuki Babasaki, Jenny Tyler Wang, and Nicholas Derby. 2024. Development of community-oriented text-to-speech models for māori ‘avaiki nui (cook islands māori). In Proceedings of the 2024 Joint International Conference on… find
  34. Carla Victoria Jara. 2018. Gramática de la lengua bribri. E-Digital ED. find
  35. Alex Jones, Rolando Coto-Solano, and Guillermo González Campos. 2023. Talamt: Multilingual machine translation for cabécar-bribri-spanish. In Proceedings of the 3rd Workshop on Multi-lingual Representation Learning (MRL), pages 106–117. find
  36. Pratik Joshi, Sebastin Santy, Amar Budhiraja, Kalika Bali, and Monojit Choudhury. 2020. The state and fate of linguistic diversity and inclusion in the NLP world. arXiv preprint arXiv:2004.09095. find our card
  37. McLoddy R Kadyamusuma and Maxwell Kadenge. 2013. The potential effect of non-tone marking on homograph processing in Shona: a preliminary exploration. Per Linguam, 29(2):85–94. find
  38. Jaden Kapali, Keaton Williamson, and Winston Wu. 2025. Statistical and neural methods for Hawaiian orthography modernization. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 35149– 35155, Suzhou, China. Association for Computational Linguistics. find
  39. Sarah Karnes, Rolando Coto-Solano, and Sally Akevai Nicholas. 2023. Towards universal dependencies in cook islands māori. In Proceedings of the Sixth Workshop on the Use of Computational Methods in the Study of Endangered Languages, pages 124–129. find
  40. Jessica Karson and Rolando Coto-Solano. 2024. Morphological tagging in bribri using universal dependency features. In Proceedings of the 4th Workshop on Natural Language Processing for Indigenous Languages of the Americas (AmericasNLP 2024), pages 56–66. find
  41. PJ Kishindo. 1998. On the standardization of Ci-tumbuka and Ciyao orthographies: Some observations. South African Journal of African Languages, 18(4):85–91. find
  42. Philipp Koehn, Hieu Hoang, Alexandra Birch, Chris Callison-Burch, Marcello Federico, Nicola Bertoldi, Brooke Cowan, Wade Shen, Christine Moran, Richard Zens, Chris Dyer, Ondřej Bojar, Alexandra Constantin, and Evan Herbst. 2007. Moses: Open Source Toolkit for Statistical Machine Translation. In Proc… find
  43. Ettien N Koffi. 2014. Towards an Optimal Representation of Tones in the Orthographies of African Languages. Linguistic Portfolios, 3(1):12. find
  44. Taku Kudo and John Richardson. 2018. SentencePiece: A simple and language independent subword tokenizer and detokenizer for neural text processing. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing: System Demonstrations, pages 66– 71, Brussels, Belgium. Assoc… find our card
  45. Olli Kuparinen, Aleksandra Miletić, and Yves Scherrer. 2023. Dialect-to-Standard Normalization: A Large-Scale Multilingual Evaluation. In Findings of the Association for Computational Linguistics: EMNLP 2023, pages 13814–13828, Singapore. Association for Computational Linguistics. find
  46. Phuong Le-Hong. 2021. Diacritics generation and application in hate speech detection on Vietnamese social networks. Knowledge-Based Systems, 233:107504. find
  47. Wesley Y Leonard. 2023. Refusing “endangered languages” narratives. Daedalus, 152(3):69–83. find
  48. Nikola Ljubešić, Tomaž Erjavec, and Darja Fišer. 2016. Corpus-Based Diacritic Restoration for South Slavic Languages. In Proceedings of the Tenth International Conference on Language Resources and Evaluation (LREC’16), pages 3612– 3616, Portorož, Slovenia. European Language Resources Association (EL… find
  49. Hieu-Thi Luong and Hai-Quan Vu. 2016. A non-expert Kaldi recipe for Vietnamese speech recognition system. In Proceedings of the Third International Workshop on Worldwide Language Service Infrastructure and Second Workshop on Open Infrastructures and Analysis Frameworks for Human Language Technologie… find
  50. Enrique Margery. 2005. Diccionario fraseológico bribri-español español-bribri, second edition. Editorial de la Universidad de Costa Rica. find
  51. Abir Masmoudi, Salima Mdhaffar, Rahma Sellami, and Lamia Hadrich Belguith. 2019. Automatic diacritics restoration for Tunisian dialect. ACM Transactions on Asian and Low-Resource Language Information Processing (TALLIP), 18(3):1– 18. find
  52. Jim Mathias and Thomas L Kennedy. 1980. Computers, Language Reform, and Lexicography in China: A Report. Washington State University Press. find
  53. Ministry of Finance and Economic Management, Government of the Cook Islands. 2021. find
  54. Census 2021: Key findings. https://www.mfem.gov.ck/statistics/census-and-surveys/census/267-census-2021. ↗
  55. Lilo Moessner. 2017. Standardization. The History of English, 4:167–187. find
  56. Alberto Muñoz-Ortiz, Verena Blaschke, and Barbara Plank. 2025. Evaluating Pixel Language Models on Non-Standardized Languages. In Proceedings of the 31st International Conference on Computational Linguistics, pages 6412–6419, Abu Dhabi, UAE. Association for Computational Linguistics. find
  57. Masato Nakano. 2023. Using Neural Network Models for Diacritics Restoration in Māori. Master’s thesis, Georgetown University. find
  58. Jakub Náplava, Milan Straka, Pavel Straňák, and Jan Hajič. 2018. Diacritics Restoration Using Neural Networks. In Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018), Miyazaki, Japan. European Language Resources Association (ELRA). find
  59. Cao Hong Nga, Nguyen Khai Thinh, Pao-Chi Chang, and Jia-Ching Wang. 2019. Deep Learning based Vietnamese Diacritics Restoration. In 2019 IEEE international symposium on multimedia (ISM), pages 331–3313. IEEE. find
  60. Sally Akevai Nicholas. 2012. Te Vairanga Tuatua o te Te Reo M=aori o te Pae Tonga: Cook Islands M=aori (Southern dialects). Collection SN1 at catalog.paradisec.org.au [Open Access]. Collector: Sally Akevai Nicholas. find
  61. Sally Akevai Nicholas. 2017. Ko te Karāma o te Reo Māori o te Pae Tonga o Te Kuki Airani: A Grammar of Southern Cook Islands Māori. Unpublished PhD thesis, University of Auckland, Auckland N.Z. http://hdl.handle.net/2292/32929 [Accessed 2017-07-14]. ↗
  62. Sally Akevai Nicholas and Rolando Coto-Solano. 2019. Glottal variation, teacher training and language revitalization in the cook islands. In Proceedings of the 19th International Congress of Phonetic Sciences, University of Melbourne, Australia, pages 3602–3606. find
  63. Sally Akevai Te Namu Nicholas. 2018. Language contexts: Te Reo Māori o te Pae Tonga o te Kuki Airani also known as Southern Cook Islands Māori. Language Documentation and Description, 15:64. find
  64. Attila Novák and Borbála Siklósi. 2015. Automatic diacritics restoration for Hungarian. In Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing, pages 2286–2291. find
  65. OpenAI. 2025. ChatGPT (GPT-5) [Large language model]. https://chat.openai.com/. Accessed: 2025-10-25. ↗
  66. Zeynep Ozer, Ilyas Ozer, and Oguz Findik. 2018. Diacritic restoration of Turkish tweets with word2vec. Engineering Science and Technology, an International Journal, 21(6):1120–1127. find
  67. Eva Pettersson, Beáta Megyesi, and Joakim Nivre. 2013. Normalisation of historical text using context-sensitive weighted Levenshtein distance and compound splitting. In Proceedings of the 19th Nordic conference of computational linguistics (Nodalida 2013), pages 163–179. find our card
  68. Elena Llamas Pombo. 2012. Spanish. In Susan Baddeley and Anja Voeste, editors, Orthographies in Early Modern Europe, pages 15–62. De Gruyter Mouton. find
  69. Akaiti Ponga. 2013. Genealogy book of Akaiti Ponga. Retrieved from https://familyse arch.org/search/catalog/2017778?av ailability=Family%20History%20Libr ary accessed January 16, 2013. International Genealogical Index. Index. FamilySearch. . Index based upon data collected by the Genealogical Societ… find
  70. Vineel Pratap, Qiantong Xu, Anuroop Sriram, Gabriel Synnaeve, and Ronan Collobert. 2020. MLS: A large-scale multilingual dataset for speech research. arXiv:2012.03411. arXiv preprint Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, and Peter… find our card
  71. Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, and Peter J. Liu. 2020b. Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer. Journal of Machine Learning Research, 21(140):1–67. find
  72. David Roberts. 2009. Visual crowding and the tone orthography of African languages. Written Language & Literacy, 12(1):140–155. find
  73. David Roberts, SR Borgwaldt, and T Joyce. 2013. A tone orthography typology. Typology of writing systems, 51:85. find
  74. Wael Salloum and Nizar Habash. 2011. Dialectal to standard Arabic paraphrasing to improve Arabic-English statistical machine translation. In Proceedings of the first workshop on algorithms and resources for modelling of dialects and language varieties, pages 10–21. find
  75. Carlos Sánchez Avendaño. 2013. Lenguas en peligro en Costa Rica: vitalidad, documentación y descripción. Revista Káñina, 37(1):219–250. find
  76. Floria Edaena Saynes-Vazquez. 2002. Zapotec language shift and reversal in Juchitan, Mexico. Ph.D. thesis, The University of Arizona. find
  77. Yves Scherrer and Nikola Ljubešić. 2016. Automatic normalisation of the Swiss German Archi-Mob corpus using character-level machine translation. In Proceedings of the 13th Conference on Natural Language Processing (KONVENS 2016), pages 248–255. find
  78. Jiatong Shi, Jonathan D. Amith, Xuankai Chang, Siddharth Dalmia, Brian Yan, and Shinji Watanabe. 2021. Highland Puebla Nahuatl Speech Translation Corpus for Endangered Language Documentation. In Proceedings of the First Workshop on Natural Language Processing for Indigenous Languages of the Americas… find
  79. Yao Shi, Hui Bu, Xin Xu, Shaoji Zhang, and Ming Li. 2020. Aishell-3: A multi-speaker mandarin tts corpus and the baselines. arXiv preprint arXiv:2010.11567. — [arXiv:2010.11567](https://arxiv.org/abs/2010.11567) find
  80. Sofía Flores Solórzano and Rolando Coto-Solano. 2017. Comparison of two forced alignments systems for aligning bribri speech. CLEI Electronic Journal, 20(1):2–1. find
  81. Lukas Stankevičius, Mantas Lukoševičius, Jurgita Kapočiūtė-Dzikienė, Monika Briedienė, and Tomas Krilavičius. 2022. Correcting diacritics and typos with a ByT5 transformer model. Applied Sciences, 12(5):2636. find
  82. NLLB Team, Marta R. Costa-jussà, James Cross, Onur Çelebi, Maha Elbayad, Kenneth Heafield, Kevin Heffernan, Elahe Kalbassi, Janice Lam, Daniel Licht, Jean Maillard, Anna Sun, Skyler Wang, Guillaume Wenzek, Al Youngblood, Bapi Akula, Loic Barrault, Gabriel Mejia Gonzalez, Prangthip Hansanti, John Hof… find
  83. Nick Thieberger. 2014. PARADISEC: Building methods for preserving ethnographic fieldwork recordings and providing long term access. Australasian Sound Archive, The, 39:43–53. find
  84. Quang-Linh Tran, Gia-Huy Lam, Van-Binh Duong, and Trong-Hop Do. 2021. A Study on Diacritic Restoration Problem in Vietnamese Text using Deep Learning based Models. In 2021 IEEE International Conference on Communication, Networks and Satellite (COMNETSAT), pages 306– 310. IEEE. find
  85. Maximilian Viatori and Gloria Ushigua. 2007. Speaking Sovereignty: Indigenous Languages and Self-Determination. Wicazo Sa Review, 22(2):7–21. find
  86. Linting Xue, Aditya Barua, Noah Constant, Rami Al-Rfou, Sharan Narang, Mihir Kale, Adam Roberts, and Colin Raffel. 2022. ByT5: Towards a Token-Free Future with Pre-trained Byte-to-Byte Models. Transactions of the Association for Computational Linguistics, 10:291–306. find
  87. Linting Xue, Noah Constant, Adam Roberts, Mihir Kale, Rami Al-Rfou, Aditya Siddhant, Aditya Barua, and Colin Raffel. 2021. mT5: A Massively Multilingual Pre-trained Text-to-Text Transformer. In Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Ling… find
  88. Shiyue Zhang, Ben Frey, and Mohit Bansal. 2022. How can NLP Help Revitalize Endangered Languages? A Case Study and Roadmap for the Cherokee Language. find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/papers/solano-2026-diacritic-restoration-low-resource-indigenous/