The Crisis
Half of all human languages will be silent by 2100
Language loss accelerates alongside urbanisation, colonisation, and digital monoculture. English and Mandarin dominate digital spaces — over 95% of internet content exists in fewer than 12 languages. The remaining 6,988 are largely invisible online, leaving their speakers without digital representation or preservation infrastructure.
UNESCO's Atlas of World Languages in Danger currently tracks 2,500+ threatened languages. Critically endangered languages — those with fewer than 100 fluent speakers, most of them elderly — are disappearing before documentation can begin. Linguists estimate that 4,000 languages remain completely undocumented.
What dies with a language
- Medicinal plant knowledge encoded in native terminology — estimated thousands of undocumented remedies
- Ecological knowledge: species names, seasonal patterns, landscape features with no equivalents
- Kinship structures and social organisation models unavailable in dominant languages
- Unique phonological systems that reveal how the human vocal tract can be used
- Historical oral record stretching back thousands of years before writing
- Mathematical and spatial concepts embedded in grammatical structure — some languages encode cardinal directions rather than relative position
Successful revitalisation cases
- Hebrew — only language in history revived from no native speakers to a national language (2 million speakers today)
- Welsh — declined to 19% Welsh-speaking in 1981; statutory protection + education policy reversed it; now growing
- Māori — near-extinction reversed via immersion schools (kura kaupapa); plateau reached, now expanding
- Basque — survived despite centuries of suppression; EU minority language protections stabilised it
Research Findings
Documentation race against time
The Endangered Archives Programme (British Library) has funded 300+ archive projects in 70 countries, digitising recordings, manuscripts, and oral traditions. But the pace of documentation lags the pace of extinction by an estimated factor of 10.
AI and language preservation
- Large language models trained on endangered languages can generate new content and teach non-native speakers — but require minimum corpus size
- Automatic speech recognition now viable for languages with as few as 10 hours of recorded speech (Meta MMS project, 2023)
- AI risks: homogenisation pressure as speakers shift to AI-capable languages; training data biased toward written, Western languages
- SIL International has documented 7,000+ languages; 3,000 have audio recordings; 1,000 have written grammars
Primary drivers of extinction
- Economic pressure: dominant languages control employment, education, and government services
- Colonial legacy: suppression policies in Australia, Americas, and Africa created generational breaks
- Urbanisation: rural communities relocate, mixed-language environments favour dominant tongue
- Digital exclusion: no keyboard, no spellcheck, no autocomplete = no digital presence = no next generation
- Education systems: most countries provide schooling only in national or colonial languages