Examples

Instead of prescribing rules which cannot possibly fit the wide heterogeneity of community goals, this wiki works with concrete examples: tools, collaborations, and approaches, each with its upsides and downsides. Through these examples, we also attempt to put terms into a concrete context. We hope these examples can be a resource for communities, researchers, and practitioners to have more productive conversations about this topic.

All

Name	Genre	Stance	Themes	Summary
AmericasNLP Shared Tasks	Case	Promising	Capacity-building, Data sovereignty	A recurring research competition at the workshop on NLP for Indigenous Languages of the Americas.
Chatbots With Human Likeness	Case	Mixed	Consent and likeness, Representation and misrepresentation	Systems that give a tool a real person's face, voice, or words can be powerful and personal, but also raise important questions.
Choctaw Nation and Oracle	Case	Promising	Data sovereignty, Digitization and legacy materials, Capacity-building	A Nation-led collaboration with a commercial vendor to build a Choctaw translation model, with the language data kept under Choctaw control.
Community-Trained Transcription	Case	Promising	Capacity-building, Digitization and legacy materials	A short workshop taught language workers to build their own speech-to-text systems.
Image Generation For Language Learning	Case	Mixed	Representation and misrepresentation, Consent and likeness, Data quality and review	Using AI image generation to make visuals for language-learning materials, like flashcards and children's books.
OpenAI and Indigenous Languages	Case	Cautionary	Data sovereignty, Consent and likeness	OpenAI's Whisper was trained on Māori and Hawaiian recordings scraped without community consent.
Putting Data Online	Case	Mixed	Data sovereignty, Digitization and legacy materials	Putting community language materials online, weighing access against protection and consent.
Scots Wikipedia	Case	Cautionary	Data quality and review, Representation and misrepresentation	A non-speaker wrote a huge amount of inaccurate Scots-language Wikipedia content, producing a flawed resource that later fed AI training.
Glosbe	Tool	Mixed	Data quality and review, Data sovereignty	A large crowd-sourced, well-documented, multilingual dictionary that can be a valuable or risky source of data for language technology.
LLM-RBMT	Tool	Promising	Data sovereignty, Data quality and review	A translation system that constrains a language model to a hand-built grammar via constrained decoding, so every output is guaranteed grammatical by construction.
Mukurtu CMS	Tool	Mixed	Data sovereignty, Sustainability and succession, Digitization and legacy materials	A free, open-source content management system built for Indigenous communities.
Ojibwe Chat	Tool	Cautionary	Data quality and review, Representation and misrepresentation	A public Ojibwe translation chatbot that confidently returns incorrect output.
Transkribus	Tool	Mixed	Digitization and legacy materials, Data sovereignty	A widely used handwritten- and printed-text recognition platform run by a European cooperative that is strong on user ownership and EU data protection, but its cloud service is permitted to use uploaded material to improve its models.

Tools

Name	Technology	Stance	Summary
Glosbe	Dictionary	Mixed	A large crowd-sourced, well-documented, multilingual dictionary that can be a valuable or risky source of data for language technology.
LLM-RBMT	MT, LLM	Promising	A translation system that constrains a language model to a hand-built grammar via constrained decoding, so every output is guaranteed grammatical by construction.
Mukurtu CMS	Archive/CMS	Mixed	A free, open-source content management system built for Indigenous communities.
Ojibwe Chat	MT, LLM	Cautionary	A public Ojibwe translation chatbot that confidently returns incorrect output.
Transkribus	OCR	Mixed	A widely used handwritten- and printed-text recognition platform run by a European cooperative that is strong on user ownership and EU data protection, but its cloud service is permitted to use uploaded material to improve its models.

Cases & patterns

Name	Themes	Stance	Summary
AmericasNLP Shared Tasks	Capacity-building, Data sovereignty	Promising	A recurring research competition at the workshop on NLP for Indigenous Languages of the Americas.
Chatbots With Human Likeness	Consent and likeness, Representation and misrepresentation	Mixed	Systems that give a tool a real person's face, voice, or words can be powerful and personal, but also raise important questions.
Choctaw Nation and Oracle	Data sovereignty, Digitization and legacy materials, Capacity-building	Promising	A Nation-led collaboration with a commercial vendor to build a Choctaw translation model, with the language data kept under Choctaw control.
Community-Trained Transcription	Capacity-building, Digitization and legacy materials	Promising	A short workshop taught language workers to build their own speech-to-text systems.
Image Generation For Language Learning	Representation and misrepresentation, Consent and likeness, Data quality and review	Mixed	Using AI image generation to make visuals for language-learning materials, like flashcards and children's books.
OpenAI and Indigenous Languages	Data sovereignty, Consent and likeness	Cautionary	OpenAI's Whisper was trained on Māori and Hawaiian recordings scraped without community consent.
Putting Data Online	Data sovereignty, Digitization and legacy materials	Mixed	Putting community language materials online, weighing access against protection and consent.
Scots Wikipedia	Data quality and review, Representation and misrepresentation	Cautionary	A non-speaker wrote a huge amount of inaccurate Scots-language Wikipedia content, producing a flawed resource that later fed AI training.

Cautionary tales

Name	Genre	Themes	Summary
Ojibwe Chat	Tool	Data quality and review, Representation and misrepresentation	A public Ojibwe translation chatbot that confidently returns incorrect output.
OpenAI and Indigenous Languages	Case	Data sovereignty, Consent and likeness	OpenAI's Whisper was trained on Māori and Hawaiian recordings scraped without community consent.
Scots Wikipedia	Case	Data quality and review, Representation and misrepresentation	A non-speaker wrote a huge amount of inaccurate Scots-language Wikipedia content, producing a flawed resource that later fed AI training.

Created Jun 22, 2026 · Updated Jun 29, 2026