Two Indigenous Languages Disappear Every Month and the Oral Knowledge Recorded Only in Those Languages Disappears With Them

The library that burns one room at a time

Every two weeks, on average, a language that has existed for hundreds or thousands of years is spoken by its last fluent elder. When that person dies, the language does not simply lose a speaker. It loses everything encoded in it. Oral histories that were never written down. Traditional knowledge about plants, weather patterns, ecological relationships, and sustainable practices accumulated across generations of observation. Ceremonial language with no equivalent in any other tongue. A way of understanding the world that developed independently over millennia and that, once gone, cannot be reconstructed.

The United Nations declared 2022 to 2032 the International Decade of Indigenous Languages. The declaration followed years of calls from the Permanent Forum on Indigenous Issues for urgent, coordinated action. The Permanent Forum's documentation of the extinction rate — two languages per month on average — represents the floor of the problem. More pessimistic but realistic projections suggest 90 to 95 percent of the world's languages will be extinct or seriously endangered by the end of this century. Humanity may have only 300 to 600 oral languages left that are currently unthreatened.

The asymmetry that explains everything

Frontiers in Communication research published in 2025 documents a statistic that explains the structural problem. 97% of the world's population communicates in just 4% of the world's languages. The remaining 3% of the world's population speaks and communicates in the other 96% of languages.

This means that every endangered language community is by definition a small community — small enough that its economic and political leverage is limited, small enough that market-based solutions do not naturally address its needs, and small enough that even well-intentioned institutional programmes frequently reach it imperfectly or not at all. The communities whose languages are most at risk are exactly the communities with the least capacity to advocate for the resources that preservation requires.

In the United States, this looks like 245 indigenous languages, 65 of which are already extinct and 75 of which are near extinction with only a few elder speakers remaining. The US Administration for Native Americans provides grants for language revitalisation, and notes explicitly that this is a matter of urgency requiring more than the current funding levels support. Federal acknowledgment that funding is inadequate sits alongside the data showing languages continue to disappear.

The AI dimension that makes it more urgent

The technology that could most help with endangered language preservation — AI-assisted transcription, speech recognition, and machine translation — is the same technology that is simultaneously accelerating the dominance of the languages that already have the most speakers.

Research on automatic speech recognition for indigenous languages documents that AI-based natural language processing solutions exist for only a few dozen of the world's 7,000 languages. Building AI tools for an endangered language requires training data, which requires prior documentation, which requires resources the communities most affected typically lack. The languages with the most speakers attract the most AI development investment, which makes tools for those languages better, which makes those languages more useful in digital environments, which disadvantages speakers of other languages further.

The digital divide identified in a March 2026 systematic review of 79 preservation studies is not simply about internet access. It is about the entire digital ecosystem of language support — the AI assistants, the translation tools, the speech recognition systems — that make participating in the digital world easier and more productive in dominant languages than in endangered ones. A young person in an endangered language community who wants to build a career in a digital economy finds that their heritage language is not supported by the tools they use every day. The practical pressures toward dominant language use compound the cultural pressures that have historically driven language shift.

What the recording captures and what it cannot

The most common preservation strategy is audio and video documentation — recording elder speakers, capturing oral histories, archiving ceremonial language and traditional knowledge before the last fluent speakers are gone. This is necessary and valuable. It is also insufficient as a standalone strategy.

A Nature npj Heritage Science systematic review published in March 2026 identifies what researchers call digital fossilisation as the risk of documentation without community engagement: archives that preserve the form of a language — the sounds, the grammar, the vocabulary — without preserving the living community that gives it meaning. A language that exists only in an archive is not a living language. It cannot be passed to children in the ordinary course of family life, cannot adapt to new experiences, cannot fulfill the social functions that language serves. Documentation extends the window for revitalisation but does not substitute for it.

Revitalisation requires communities. Communities require resources. The communities most urgently in need of revitalisation support are, by the structural logic of endangered language demographics, the communities with the least access to those resources.

The peer-reviewed extinction timeline

Researchers from the Université de Montréal, Friedrich-Schiller-Universität Jena, and the Max Planck Institute published a peer-reviewed probabilistic projection for 27 indigenous languages in Canada in February 2025. The methodology used Canadian census data and demographic modelling, not advocacy assumptions.

The model found that speaker numbers could decline by more than 90% in 16 of the 27 languages over the period 2001 to 2101. Dormancy risks — the probability of a language having no remaining speakers within a given period — could surpass 50% among five languages. These are languages that are currently spoken, by real communities, with living elder speakers who carry the knowledge encoded in them. The trajectory the model describes is not speculative. It is based on the demographic reality of who speaks these languages today and what the age distribution of those speakers implies about the future.

The research conclusion is calibrated rather than alarmist: the pace of extinction is slower than the most pessimistic projections had suggested, but still alarming. Slower than feared and faster than acceptable. The window is not closed. It is closing.

Proof signals

United Nations International Decade of Indigenous Languages. The UN proclaimed 2022 to 2032 the International Decade of Indigenous Languages after repeatedly calling for urgent action through the Permanent Forum on Indigenous Issues. The formal UN decade declaration is itself evidence of institutional recognition that the extinction rate has reached a threshold requiring coordinated global response. More pessimistic but realistic estimates claim that 90 to 95 percent of the world's languages will become extinct or seriously endangered by the end of this century. Humanity may have only 300 to 600 oral languages left that are unthreatened by the end of this century.

Royal Society Open Science peer-reviewed research February 2025. Researchers from the Université de Montréal, Friedrich-Schiller-Universität Jena, and the Max Planck Institute published a probabilistic population projection for 27 Indigenous languages in Canada. The model found speaker numbers could decline by more than 90% in 16 languages over the period 2001 to 2101 and that dormancy risks could surpass 50% among five languages. This is peer-reviewed quantification of the extinction trajectory, not advocacy, the numbers come from Canadian census data and demographic modelling.

Applied Sciences research on AI and indigenous languages. Peer-reviewed research on automatic speech recognition for indigenous languages documented that AI-based natural language processing solutions are only available for a few dozen languages out of 7,000. The research notes that globalisation and technological advances in AI have further accelerated the hazard to minority languages because AI solutions are built only for dominant languages. When a community's language is not supported by AI tools, it cannot benefit from AI-assisted documentation, translation, or education, the tools that could help preserve the language are not available for the languages most at risk.

Nature npj Heritage Science March 2026. A systematic review of 79 studies on digital safeguarding of intangible cultural heritage published in March 2026 found that effective preservation requires synergising digital methods with community participation. Technology-focused but community-shallow projects risk what the researchers call digital fossilisation, the creation of archives that preserve the form of a language without the living community that gives it meaning. The review identifies the digital divide as a key challenge: the communities most in need of digital preservation tools are often the communities with least access to the technology required to use them.

Frontiers in Communication 2025. Research published in Frontiers in Communication noted that 97% of the world's population communicates in just 4% of the world's languages. The remaining 3% of the world's population, the speakers of 96% of languages, represents every endangered language community. The math means that the people with the most cultural diversity and the most at risk of permanent linguistic loss are a very small global minority with proportionally less political and economic leverage to advocate for preservation resources.

What to actually do about it

Existing attempts fall short in specific ways:

• National government language programmes: Many governments fund indigenous language preservation through grant programmes. The Native American Languages Act and the Esther Martinez Act in the United States, the Administration for Native Americans grants, and equivalent programmes in Canada, Australia, and New Zealand provide some funding. The funding levels are consistently described as inadequate relative to the scale of the extinction threat and the cost of comprehensive documentation, digitisation, and revitalisation work. Government programmes also typically require communities to navigate grant application processes that are burdensome relative to the capacity of small communities.
• University linguistics departments: Academic linguists document endangered languages as part of research programmes. The documentation produced serves academic purposes and is typically housed in institutional archives with limited community access. The community whose language is being documented often does not have meaningful control over how the documentation is used, stored, or accessed. The relationship between academic documentation and community-led revitalisation is frequently described in the research literature as insufficient and sometimes counterproductive.
• Existing AI language tools: AI-based natural language processing, speech recognition, and machine translation exist as mature technologies for dominant languages. They are available for a few dozen languages out of 7,000. Building AI tools for an endangered language requires sufficient training data, which requires significant prior documentation effort, which requires resources the communities most affected typically lack. AI accelerates the development of language tools for already well-resourced languages and cannot currently be readily applied to languages with limited existing digital documentation.
• Community-led digital archives: Community-controlled digital archives have been identified as the most effective approach for endangered language preservation, combining technical infrastructure with community ownership of content and access decisions. Building and maintaining these archives requires sustained funding, technical expertise, and community capacity that most endangered language communities do not have access to. The communities that most need community-controlled archives are the ones least positioned to build and maintain them independently.
• Apps and language learning platforms: Duolingo and similar platforms have developed courses for some endangered languages including Hawaiian, Navajo, and Māori with significant community support. This is genuine progress for specific languages with engaged communities and institutional backing. The approach does not scale to the 40% of languages that are threatened, most of which lack the community infrastructure, digital presence, and platform relationship required to develop a mainstream language learning course. The languages most at risk are not the ones that have made it onto Duolingo.

Before going further, it is worth pressure-testing the idea against these questions:

1. Could AI-assisted transcription and speech recognition, trained on even small amounts of existing audio documentation for an endangered language, produce tools useful enough for community documentation efforts to meaningfully extend the window before the last fluent speakers are gone?
2. The communities most affected by language extinction are often the communities with least access to technology and funding. What does an equitable distribution of preservation resources look like, and who has the mandate and the capacity to provide it at the scale the extinction rate requires?
3. Digital fossilisation, archives that preserve form without preserving the living community, is identified as a real risk of technology-led preservation. What does community-led preservation look like when the community itself is very small and the technology required is very expensive?
4. Duolingo has built courses for some endangered languages with significant community backing and institutional support. Could a similar model, a platform that makes endangered language learning accessible while routing revenue back to preservation efforts, scale to languages without existing institutional relationships?
5. The oral knowledge encoded in endangered languages includes traditional ecological knowledge that is increasingly recognised as valuable for biodiversity conservation and climate adaptation. Does this create a practical conservation funding argument for language preservation that goes beyond cultural value and could access a larger funding pool?