Language Toolkit
A language toolkit your nation owns outright
A plan for saving, teaching, and using your language, with AI as a helper. It follows one rule: your community decides what is shared, what is taught, and what stays sacred. Run it on your own computers, or let us help you set it up.
Two ways to use this plan
Run it yourself
Your own tech staff and language program set it up and run it on your community’s computers. Everything here is free for you to use and change.
Build it with us
Our Native-led team sets it up where you choose, teaches your language staff how to run it, and hands you the keys. You keep everything. We keep nothing.
Structure
Four pillars on one foundation
A tribal language program will not upload recordings of elders to any tool until it knows three things. Who owns the files. Who can hear them. Where they go. So control of your language is not one of four pillars. It is the ground all four stand on.
Preserve
Recordings, written words, documents, and songs, saved and sorted the way your community wants.
Teach
Lessons, practice, and a tutor that only uses words your speakers have approved.
Create
Stories, songs, and devices that put the language to use in daily life.
Empower
Controls for the people in charge of your language, plus training so your community runs the tools itself.
Protect
Your nation’s own locked vault. Rules on every item. A wall between sacred words and any AI. Speakers approve everything. A record of who looked at what. And the right to leave and take everything with you.
Context
What “sacred” means in practice
To protect sacred words, we first have to name the kinds of limits communities actually put on their words. Every nation is different, but a few patterns come up again and again. A few systems built by and for Indigenous people already handle these well. We borrow their ideas. The vault stays yours.
Mukurtu
Free archive software. Built at Washington State University together with Indigenous communities.
Said MOOK-oo-too. It is a Warumungu word for a bag that keeps things safe. Every recording, photo, or document in Mukurtu carries rules about who may see it and when. Tribal libraries and language programs across the United States use it today.
Traditional Knowledge Labels
From Local Contexts, a project started by legal scholar Jane Anderson and Mukurtu’s Kim Christen.
Small digital tags a community puts on its own recordings, photos, or texts to say how they may be used. Examples: TK Seasonal, TK Women General, TK Secret/Sacred. Museums, archives, and universities recognize them, so your nation’s rules travel with the file.
FirstVoices
An online language archive run by the First Peoples’ Cultural Council in British Columbia, Canada.
Dozens of First Nations keep their dictionaries, recordings, and lessons there. Each community’s own staff decide what is public and what is for members only. It is the best working example of a language site run by the community itself.
The levels below are a starting point, not a fixed list. The people in charge of your language (we call them the language authority) set their own, and the system has to let you add levels no outsider would think of.
| Level | Who can see it | Examples | Can AI use it? |
|---|---|---|---|
| PUBLIC | Anyone | Greetings, numbers, place names your nation chooses to share, outreach material | Yes, if your language authority says so |
| COMMUNITY | Enrolled members and approved learners | Most everyday words, family stories, classroom recordings | Yes, but only by AI that belongs to your nation |
| ROLE | One group only: women, men, a clan, a society, initiated people, or elders | Teachings for women or men only, society songs, clan knowledge | Only for that group, and usually not at all |
| SEASONAL | Community members, but only during the right season | Stories told only in winter, material tied to the ceremony calendar | Same rule as its base level, but only in season |
| SACRED | Only the people your language authority names | Ceremony language, prayers, names of people who have passed where that is the custom, and anything else the authority says | Never. Locked with a key the AI never has. |
The trust question is real
Communities have watched outside groups claim they own recordings made with the community’s own elders. People had to fight for the right to their own words. This plan is built the opposite way. Ownership, the keys, and the exit are in your nation’s hands from day one. Not a promise. Built in.
A quieter risk with AI
An AI can make up words that sound right but are wrong. For a language with a few hundred fluent speakers, that does real harm, because learners can’t tell the difference. So the wall works both ways. Sacred words never go in. And nothing comes out as part of the record until a fluent speaker approves it.
Principles
The principles this toolkit follows
Neither set of principles below came from a tribe in the United States, and most people in Indian Country have never heard of them. They still matter, for two reasons. They are the closest thing to a shared rulebook for Indigenous data, so funders, universities, and tech companies are starting to recognize them. And they say what any tribal council would say if asked who should control the language: the people it belongs to.
If your nation already has its own data rules or a research review board, those come first. These principles fill the gaps. Each one below is matched to something you can actually see in the toolkit.
Collective Benefit · Authority to Control · Responsibility · Ethics
The CARE Principles
Written in 2019 by the Global Indigenous Data Alliance. That is a group of Indigenous researchers and organizations from the United States, Canada, Australia, and New Zealand, including the Native Nations Institute at the University of Arizona. CARE was the Indigenous answer to the push for “open data.” It says data about Indigenous people should serve them and be controlled by them, not simply handed to everyone.
Ownership · Control · Access · Possession
OCAP®
Created by First Nations in Canada in 1998 and kept by the First Nations Information Governance Centre. It started when First Nations refused to let the government run a health survey about them without them. Today any researcher working with First Nations data is expected to follow it. Only the trademark is Canadian. The four ideas fit a tribal nation in the United States just as well.
Collective Benefit
CARE · Global Indigenous Data Alliance
Every tool serves your nation’s own language goals. Your data is never mixed with another nation’s. No company profits from it. Speakers are paid and credited.
Authority to Control
CARE
Your nation names who is in charge of the language: the language department, a council, or elders. That group sets every rule, approves every entry, and can shut off outside access.
Responsibility
CARE
Whoever runs the system says openly what it does with your data, keeps a record of who looked at what, and lets an outside expert check the rules.
Ethics
CARE
Nothing an AI makes goes in the record until a fluent speaker approves it. Every entry shows who said it and who approved it. No made-up words.
Ownership
OCAP® · First Nations Information Governance Centre
Your nation owns the recordings, the written words, and any AI trained on them. A partner gets no rights beyond what the agreement allows for doing the work.
Control
OCAP®
The rules are built into the storage itself, not just the screen. No app and no AI can get around them.
Access
OCAP®
Your nation can see and download all of its data at any time, in standard file formats, without asking permission.
Possession
OCAP®
Your nation’s vault is separate from everyone else’s. It can live on tribal computers or in a cloud location you choose, and your nation holds the keys.
Traditional Knowledge Labels fit on top of this. Your language authority can put the standard labels (TK Secret/Sacred, TK Seasonal, TK Women General, TK Community Voice, and others) on any item. If a file ever leaves the vault, your nation’s rules go with it.
How it works
Where the wall between your words and AI sits
The whole design fits in one drawing. Your words live in your nation’s vault, sorted into zones. A gate checks the rules before any AI can read anything. Whatever the AI makes comes back as a draft for a speaker to approve before it goes in the record.
Your nation’s vault
Its own storage and its own keys. One per nation. Kept wherever your nation chooses.
The rules gate
Built into the storage itself. Checks the level, the group, the season, and whether AI is allowed before letting anything through.
- •Rules live in the storage, not just on the screen
- •The sacred zone is locked so AI can’t read it, no matter what
- •Every look at a file is written down for your language authority
AI helpers
Turning speech into text, tutoring, story drafts, cleaning up old audio. They only use approved words, and they show where each word came from.
- •Speech AI trained on your own recordings, owned by your nation
- •Outside AI companies used only with a written promise not to keep or learn from your data
- •Option to run all AI on your own computers so nothing leaves your community
If your own team builds it, these choices matter most
- •
One nation, one vault. Give each nation its own separate database and file storage. Set the access rules inside the database itself. Running it on tribal computers is the surest way to keep possession.
- •
Two sets of keys. The everyday keys open the public and community zones. The sacred zone has its own key that the AI and any outside partner never get. Your nation holds it and shares it only with people the language authority names.
- •
The approved word list is the source of truth. Tutors and story tools only pull from approved entries and show which entry each word came from. If an answer can’t be traced to one, it gets flagged, not shown.
- •
Works without internet. Many reservations have weak internet. Recording, reviewing, and lessons must work offline and catch up later, with the same rules checked when they sync.
- •
Let people inspect the rules engine. The part that decides who can see what should be open to anyone your nation asks to check it. That is a matter of trust as much as code.
Trust
How trust gets earned, in order
In Indian Country, trust comes from relationships, and it comes in steps. It depends on who vouches for you, what you do before you ask for anything, and whether leaving is truly possible. None of these steps are software.
- 1
Agreement first, software second.
Start with a Language Data Sovereignty Agreement, adopted by your council. It names who is in charge of the language, where the data lives, who owns any AI trained on it, and how to delete, download, or shut off access. Nothing gets installed until it is signed.
- 2
Start with what is already public.
The first things in the vault are what your nation already shares: outreach words, published dictionaries, classroom lessons. Nothing restricted goes in until the people in charge have watched the system work.
- 3
Relationships, not sales calls.
Setup happens in person, through people the community already knows. If Codetalkers is involved, our Native-led team comes to you.
- 4
Show, don’t promise.
Every setup gets a plain public page saying what is stored, where, which AI is used, and who has access. The rules engine is open to inspection. An outside expert checks it before any restricted words go in.
- 5
Pay speakers, credit elders.
People who record, review, or approve are paid, and their names appear on each entry in the way the community thinks is right.
- 6
Make leaving easy.
The download-everything button and the shut-off switch are on your language authority’s dashboard from day one. A nation that knows it can leave is a nation that can stay.
Building it
What to build, in what order
Protect and Preserve come first, together, because nothing else can exist without a vault to draw from. Teach and Create each build on the approved words the first phase produces. Empower is not a phase. The dashboard for the people in charge grows with every release.
Phase 1
Protect + Preserve
The vault and the rules gate. Upload recordings and documents, tag each one with its rules, turn speech into text with an AI your nation owns, send every transcript to a speaker for approval, and download everything. The dashboard for the people in charge ships here, with roles, rules, and the record of who looked at what.
Ships: A working archive with the public and community levels live. The sacred level turns on only after an outside expert checks the rules.
Phase 2
Teach
Lessons and a chat tutor that only use approved words and show where each word came from. Pronunciation practice using your nation’s own speech AI. A phone app that works without internet.
Ships: A learner app for your language program’s classrooms.
Phase 3
Create
Story drafting with elder review, song tools, and connections to home devices, all going through the same gate and review line.
Ships: The creative tool your community asks for first, not the one anyone guesses at.
Bring this to your language program
Whether you want to run it entirely on your own computers or have our Native-led team set it up with you, the first step is the same: a talk with the people in charge of your language about what stays sacred.
The template is a starting point for your language leaders and your tribe’s lawyer. It is not legal advice.
Built on
- CARE Principles for Indigenous Data GovernanceGlobal Indigenous Data AllianceThe four principles this toolkit follows, with the full text and the people behind it.
- The First Nations principles of OCAP®First Nations Information Governance CentreOwnership, Control, Access, Possession, explained by the First Nations group that keeps them.
- Traditional Knowledge LabelsLocal ContextsTags a community puts on its own material to say how it may be used. Free to use.
- Mukurtu CMSWashington State UniversityFree archive software with access rules built in. Many US tribal libraries already use it.
- FirstVoicesFirst Peoples’ Cultural CouncilA working example of a language archive run by the community itself, used by dozens of First Nations.