Of the world's roughly 7,000 spoken languages, half will be silent by 2100. A field guide to the half that is dying, the linguists trying to record them, and the communities trying to bring them back.
A language dies, on average, every two weeks. Within a century, between 50 and 90 percent of human linguistic diversity will be gone. The catastrophe is well-documented and largely unstoppable.
Michael Krauss's 1992 paper in Language, "The world's languages in crisis," named the scale of the loss for the first time. Krauss estimated that of the roughly 6,000 languages then spoken, 90 percent were "moribund or doomed" — no longer being learned by children. The paper hit linguistics like a fire alarm. Subsequent surveys (UNESCO Atlas, Ethnologue, ELP) have refined the number but not the conclusion.
The cause is rarely sudden. A language dies because its speakers, generation by generation, raise their children in a different one — usually a regional or national language with more economic and educational utility. The shift is rational at the household level and devastating at the species level.
This deck covers what a language is when it is dying, who has noticed, what documentation has captured, and which revitalisations have actually worked. Hebrew, Hawaiian, Welsh, Māori — the bright spots. And the long, sober list of what we have lost since the question began to be asked.
UNESCO's Atlas of the World's Languages in Danger uses a five-step scale, codified in 2003:
Vulnerable. Most children speak it, but its use is restricted to certain domains (e.g. the home).
Definitely endangered. Children no longer learn the language as mother tongue at home.
Severely endangered. Spoken by grandparents and older generations; the parent generation may understand but does not speak it to children.
Critically endangered. The youngest speakers are grandparents and older, and they speak the language partially and infrequently.
Extinct. No speakers remain.
The hinge is "intergenerational transmission." Once children stop being raised in a language, the clock starts. The remaining fluent speakers will live ~40-60 more years, and after that the language exists only in recordings.
Joshua Fishman's Reversing Language Shift (1991) introduced the eight-stage Graded Intergenerational Disruption Scale (GIDS) — a more granular tool used by working revitalisation programs. Fishman's stage 6 ("intergenerational informal use in family, neighborhood, and community") is the threshold below which any revitalisation must reach to be self-sustaining.
Michael Krauss, then director of the Alaska Native Language Center, published "The world's languages in crisis" in Language 68(1) in March 1992. The paper opened with a calculation: of roughly 6,000 known languages, perhaps 600 (10%) had a "good chance" of surviving to 2100, perhaps 3,000 were "definitely safe," and the remaining 3,000 were already moribund. Subsequent recountings have found Krauss's framing roughly correct.
Krauss compared the loss to mass biological extinction, then went further: "Surely, just as the extinction of any animal species diminishes our world, so does the extinction of any language." The analogy stuck. The "biolinguistic diversity" framing — language loss as a parallel crisis to species loss — descends from this paper.
Krauss also estimated that of the ~210 indigenous languages of Alaska and the United States, only ~20 were still being learned by children. He had spent his career documenting Eyak, the language of a single village on the Copper River; in 1992 it had two fluent speakers. The last, Marie Smith Jones, died in 2008. Krauss attended her funeral.
The paper's larger effect was institutional. Linguistics began to reorganise around documentation. The Endangered Languages Project, the Documentation of Endangered Languages program (DOBES), the Hans Rausing Endangered Languages Project — all post-1992 responses.
Ethnologue's 2024 count: 7,168 living languages. The distribution is wildly uneven.
Papua New Guinea: ~840 languages. The most linguistically diverse nation on earth — roughly 12% of the world's languages spoken by 0.1% of its population. The combination of mountainous terrain, small subsistence communities, and minimal historical state pressure preserved extraordinary diversity.
Indonesia: ~707. Sumatra, Java, Sulawesi, the eastern islands. Bahasa Indonesia, the national lingua franca, is rapidly absorbing speakers.
Nigeria: ~520. Hausa, Yoruba, Igbo dominate, but Niger-Congo and Afro-Asiatic minorities continue.
India: ~447. The 2011 census recorded 121 languages with ~10,000+ speakers; hundreds smaller exist below the count threshold.
The endangered concentrations: Australia (250+ Aboriginal languages, mostly critical), the Americas (especially Amazonia and northern Mexico), Siberia (Russian-language pressure across Yakut, Even, Chukchi, Aleut), the Caucasus, parts of Sub-Saharan Africa.
Of the world's languages, ~40% are spoken by fewer than 1,000 people. ~25% have fewer than 100 speakers. The bottom of the distribution is where most extinction happens.
The top of the distribution is also striking. Just 23 languages account for more than half of all human speakers. Mandarin, Spanish, English, Hindi, Arabic, Bengali, Portuguese — the megafauna of the linguistic ecosystem. Their continued growth is the immediate cause of others' decline.
Marie Smith Jones (1918-2008) was, at her death, the last fluent speaker of Eyak — a Na-Dené language spoken along the lower Copper River in southern Alaska. With her, a language ~3,000 years old, distinct enough from all neighbours that it constituted its own branch of its family, became extinct.
Smith Jones lived in Anchorage. She was an environmental and cultural activist, co-founded the Eyak Preservation Council, attended the United Nations as a delegate. The work of recording her speech — Krauss's project, beginning in the 1960s — produced the dictionary, grammar, and audio archive that constitute the surviving record of the language.
The "last speaker" phenomenon is well-attested. Boa Sr., last speaker of Bo (Andaman Islands), died 2010. Cristina Calderón, last fluent speaker of Yaghan (Tierra del Fuego), died 2022. Hazel Sampson, last native speaker of Klallam (Washington State), died 2014. The pattern: a single elderly individual carrying a complete language; recording urgency; eventual silence.
The role of "last speaker" carries documented psychological weight. In interviews, last speakers often describe the experience as profound loneliness — no one left to dream in their language, joke in their language, hear themselves spoken to. The loss is intimate as well as cultural.
Some "last speakers" are not actually last. Ned Maddrell, last native speaker of Manx (Isle of Man), died 1974 — but Manx is now spoken by ~1,800 people through revival, including children raised in it. The category is contingent.
Languages do not usually die by sudden catastrophe (though some do — disease, genocide, deportation). Most die by gradual functional displacement.
Stage 1. A community is bilingual; the local language and a regional/national one coexist. The local is dominant in the home; the national is required for school, market, government.
Stage 2. Parents begin raising children in the national language for perceived advantage. The local survives among grandparents; passive comprehension persists in the parent generation.
Stage 3. The youngest fluent speakers are middle-aged. The language exists in domains (ceremony, kitchen-table conversation among elders) but no longer in everyday life.
Stage 4. Only elders speak it. Documentation and ceremonial use continue.
Stage 5. Last speakers die. Documentation becomes the only record.
The shift is usually invisible to those inside it. Each generation makes locally-rational choices — to give children opportunity, to fit in, to avoid stigma. The aggregate effect is one of the great cultural losses in human history.
The classic counterfactor is institutional support. Languages with school instruction, government recognition, media presence, and economic utility resist shift. Languages without those resources — most minority languages — do not.
Daniel Nettle and Suzanne Romaine's Vanishing Voices (2000) makes the case that biolinguistic diversity tracks biodiversity geographically: the same regions (tropical, mountainous, isolated) preserve both. The disappearance of small languages tracks the disappearance of small ecosystems and small farming traditions.
Modern Hebrew is the most successful language revitalisation in history — and the only one to have produced a vernacular community of millions from a language with no native speakers.
Hebrew had been a liturgical and literary language for ~1,800 years (since roughly the second century CE) before Eliezer Ben-Yehuda (1858-1922) began advocating its use as a spoken everyday language. Ben-Yehuda's work was both lexical (he coined or adapted thousands of terms for modern objects: mila, dictionary; iton, newspaper; glida, ice cream) and pedagogical (he raised his son Itamar Ben-Avi as the first native speaker of Hebrew in eighteen centuries).
The conditions Ben-Yehuda exploited:
Existing literacy. Religious Jews maintained reading-and-writing competence in Hebrew across the diaspora. The phonology and grammar were not lost — only the speaking community.
Immigration to a new territory. Yishuv (pre-state Jewish settlement in Ottoman/British Palestine) brought together speakers of Yiddish, Russian, Ladino, Arabic, German. Hebrew functioned as a deliberately-chosen common ground.
State support. By the 1948 founding of Israel, Hebrew was already the dominant spoken language of Yishuv. State institutions (the Academy of the Hebrew Language, founded 1953; compulsory Hebrew-medium education) consolidated the shift.
Native Hebrew speakers in 2024: ~9 million. The case is widely cited as proof revitalisation is possible. It is also widely misunderstood: Hebrew's preconditions (continuous literary tradition, ideologically-motivated immigrant population, state founding) are not replicable for most endangered languages.
By 1983, Hawaiian (ʻŌlelo Hawaiʻi) had perhaps 2,000 native speakers, almost all elderly, plus ~200 children on Niʻihau (a privately-owned island where the language survived by isolation). The language had been banned from public schools in 1896, three years after the US-backed overthrow of the Hawaiian Kingdom; three generations of suppressed transmission had nearly finished it.
The 1983 founding of ʻAha Pūnana Leo — "language nest" preschools modelled on Māori kōhanga reo — began the revival. Pūnana Leo schools immerse children entirely in Hawaiian from age 3-5. The first generation graduated into Hawaiian-medium primary, secondary, and (eventually) tertiary education at the University of Hawaiʻi at Hilo.
By 2024, an estimated 18,000+ people speak Hawaiian, including ~2,000 children being raised in it as a first language. The language has reclaimed status as one of two official languages of the State of Hawaiʻi (1978 constitutional amendment). Hawaiian-medium PhDs are now produced.
The work is not finished. Hawaiian remains heavily outnumbered by English on its own islands; intergenerational transmission outside formal schooling is fragile. But the language has crossed the threshold from "moribund" to "living small minority." The case is one of the better recent revitalisation results — partly because of state recognition and university support, partly because the immersion-school model worked.
Larry Kimura, William "Pila" Wilson, and Kauanoe Kamanā are the architects most associated with the program. The work continues.
Welsh (Cymraeg) is the rare case of a language that has stabilised — not grown dramatically, but held a substantial speaker base against decades of English-language pressure.
The 1961 census recorded ~656,000 Welsh speakers (26% of the Welsh population). By 1991, that had fallen to ~509,000 (18.7%). The post-1991 institutional response (Welsh Language Act 1993, S4C Welsh-language television since 1982, devolution of education policy in 1999) substantially arrested the decline. The 2021 census: ~538,000 (17.8%). The number is approximately stable; the percentage is approximately stable.
The institutional architecture:
S4C (Sianel Pedwar Cymru). Welsh-language television channel, broadcasting since 1 November 1982. Daily children's, drama, news, sport. Funded by UK government grant + advertising.
Welsh-medium schools. ~25% of Welsh primary pupils now educated in Welsh-medium schools. The 1944 Education Act allowed but did not require Welsh-medium provision; subsequent decades have built it out, particularly in non-Welsh-speaking areas (the "designated" Welsh-medium school in an English-speaking town).
Cymraeg 2050. Welsh Government strategy launched 2017, target of 1 million Welsh speakers by 2050.
Welsh is the European language whose institutional model is most often studied for transferable lessons. Television, education, official status, and visible signage all sustain a minority language against a regional majority — but they require sustained government commitment, which most endangered languages cannot expect.
Te reo Māori provided the model that Hawaiian later adopted. By 1979, Māori was severely endangered: only ~20% of Māori people spoke it, almost all over 50. The 1982 founding of the first kōhanga reo ("language nest" preschools) began the revival.
The kōhanga reo concept: full immersion preschool, 30-40 hours/week, run by the elders of the community. Children acquire the language as first-language fluent speakers; elders pass on competence the parent generation lacks. The model bypasses the broken parent-to-child transmission link.
By 1990 there were 600+ kōhanga reo. They feed into kura kaupapa Māori (Māori-medium primary schools) and wharekura (Māori-medium secondary schools). The Māori Language Act 1987 made Māori an official language of New Zealand. Māori Television Service has broadcast since 2004.
The 2018 census: ~185,000 New Zealanders (4% of population) report some Māori-speaking ability; ~50,000 report being highly proficient. The numbers are growing in a country where they were declining for the previous century. The 2018 figure was up substantially from 2013.
The model has been exported. Hawaiian (Pūnana Leo, 1983 onward), Hebridean Gaelic (sgoiltean àraich), Mohawk (Kahnawà:ke), Sámi (giellabeassi). The language nest is the most successful single tool of revitalisation work. It does not save every language — many endangered communities lack the elders or the resources to staff one — but where it has been deployed, it has generally worked.
Where revitalisation is impossible, documentation becomes the goal: to leave behind a record from which the language can, in principle, be reconstructed.
The discipline of documentary linguistics, distinct from descriptive linguistics, was formalised in the late 1990s. Nikolaus Himmelmann's 1998 paper "Documentary and descriptive linguistics" defined the field. Documentation prioritises:
Audio and video recordings of natural speech in many genres (narrative, conversation, song, ritual). Not just elicited word-list data.
Time-aligned transcription in the language's own orthography (or IPA where none exists), with translation and morphological gloss.
A grammar and a dictionary derived from the corpus, not just from elicitation.
Ethnographic context — the cultural information without which the language samples are incomprehensible.
The institutional homes:
DOBES (Documentation of Endangered Languages, Volkswagen Foundation, 2000-2018). Funded ~80 documentation projects across the world; results archived at the Max Planck Institute for Psycholinguistics, Nijmegen.
ELDP (Endangered Languages Documentation Programme, SOAS University of London, 2002-present). Hans Rausing endowment. Archive at the Endangered Languages Archive (ELAR).
NSF DEL (Documenting Endangered Languages, US National Science Foundation, 2005-present). Joint with NEH.
The output: tens of thousands of hours of recordings, hundreds of grammars and dictionaries, archived for permanent preservation. It will not save the languages. It will allow descendants to study them.
What documentation looks like in practice. A linguist arrives in a community. The fieldwork is years of relationship-building, transcription, and patient elicitation.
ELAN. Free software, Max Planck Institute. Time-aligns transcription tiers to audio/video. The standard tool of documentary linguistics.
FLEx (FieldWorks Language Explorer, SIL International). Lexical database, dictionary builder, interlinear text annotator.
Praat. Acoustic phonetics — pitch, formants, spectrograms. For phonological analysis.
The orthography problem. Many languages have no writing tradition. The fieldworker negotiates an orthography with the community: which sounds need distinct symbols, what to do about tone, whether to write in IPA or adapt the regional script. The choices have political weight; communities have rejected orthographies designed for them.
The recording. A typical week: 8-12 hours of conversation and narrative recording, twice that of transcription, plus elicitation sessions for grammatical paradigms (verb conjugation, kinship terminology, body-part vocabulary, numerals).
The archive. Recordings are deposited in long-term archival institutions (ELAR, PARADISEC, AILLA, ANLA) with metadata, access conditions, and community consent. The "long-term" means: in fifty years, the recordings should still be findable and playable.
The output of a multi-year documentation project is typically a grammar (book-length), a dictionary (5,000-15,000 entries), a corpus of transcribed texts (30-100 hours), and a community version of all of the above. The work is slow, expensive, and irreplaceable.
A small language is not a small version of a big language. The structures lost when a language dies are often unique to it.
Pirahã (Amazon, ~250 speakers). Daniel Everett's claim that Pirahã lacks recursion (clauses embedded in clauses) was at the centre of a long, intense linguistic debate about Universal Grammar. Whether Everett is right or not, the data come from one community of a few hundred people.
!Xóõ / Taa (Botswana, ~2,500 speakers). The largest phoneme inventory of any known language: ~83 consonants by some counts, including the click consonants (dental, alveolar, lateral, palatal, bilabial) plus ~40 vowels. When Taa is gone, click-consonant phonology will lose its most extreme exemplar.
Guugu Yimithirr (Queensland, ~775 speakers). The language uses absolute (cardinal) directions instead of relative (left/right). Speakers know their orientation continuously and describe space in absolute terms ("there is an ant on your north leg"). The cognitive consequences for speakers — superior dead reckoning, different mental representation of space — were documented by Stephen Levinson's MPI group.
Yupik / Inuit languages. Highly polysynthetic — entire sentences expressed as single morphologically-complex words. Tuluksak Yupik, Greenlandic, Inuktitut: each preserves grammatical structures rare elsewhere.
Kabardian (Northwest Caucasus, ~1.6 million speakers). Three vowels but ~50 consonants. The opposite extreme from polyphonemic systems.
What the global linguistic record loses with each death is not just the words but the structural possibilities. Each minor language is a working sample of what human grammar can do. The field of linguistic typology depends on this diversity remaining alive long enough to be studied.
Pre-1788, Australia had ~250 distinct Aboriginal languages, with another ~600 dialects, spread across a continent with 65,000+ years of human habitation. The languages constitute their own family (or several families); their internal diversity is comparable to that of all of Eurasia.
Today, of those ~250 languages, perhaps 20 are still being learned by children. The 2021 census identified ~167 Aboriginal languages still spoken at home, but the great majority have only elderly speakers. The damage was specific: the Stolen Generations (1905-1970s, ~100,000 children removed from their families and placed in white-Australian institutions where Aboriginal languages were forbidden) destroyed transmission across the country.
Some have recovered. Wiradjuri, once classed as extinct, has been actively revived since the 1990s and is now taught in schools across central New South Wales. Kaurna (around Adelaide) was reawakened from 19th-century missionary records by Rob Amery's revival project (1990s onward). Diyari, Awabakal, Dharug — similar stories.
The institutional support, when it exists, is uneven. AIATSIS (Australian Institute of Aboriginal and Torres Strait Islander Studies) holds a major archive. Indigenous Languages and the Arts Program funds revitalisation. The 2019 International Year of Indigenous Languages and the 2022-2032 Decade of Indigenous Languages have raised attention. National Indigenous Languages Report (2020) documented the field.
The remaining strong languages (Yolŋu varieties in Arnhem Land, Pintupi-Luritja in the Western Desert, Pitjantjatjara/Yankunytjatjara in central Australia) are mostly in remote communities where English contact remained limited. Where contact came, language went.
Before 1492, the Americas held an estimated 1,000-2,000 languages. The post-contact loss was catastrophic: disease, displacement, residential schools, language bans. By the early 21st century perhaps 600 indigenous American languages still had speakers; most are critically endangered.
North America. Of ~300 indigenous languages of the United States and Canada, ~165 still have speakers (Ethnologue 2024). Navajo (Diné Bizaad) is the largest, with ~170,000 speakers — but transmission to children is faltering. Cherokee, Lakota, Ojibwe, Blackfoot, Cree, Inuktitut: all have organised revitalisation but face severe demographic pressure. The US residential boarding schools (1879-1970s) and Canadian residential schools (1830s-1996) systematically broke transmission across the continent.
Mexico and Central America. ~290 indigenous languages, mostly Mayan, Nahuan (Aztec), Otomanguean, Mixe-Zoquean. Nahuatl alone has ~1.7 million speakers but is classed as endangered because urban migration disrupts village transmission.
Amazonia. Astonishing diversity — ~250 languages, often in tiny isolates. The territory's relative inaccessibility delayed language shift but is no longer doing so. The Pirahã (~250 speakers) and Yanomami (~35,000 across multiple varieties) are among the better-known.
Andes. Quechua (~7-10 million speakers across Peru, Bolivia, Ecuador, Colombia, Argentina), Aymara (~2.8 million). Both stabilised by sheer demographic mass and bilingual education programs but losing ground to Spanish in cities.
Brazil and the Guianas. ~180 indigenous languages, mostly endangered. The Tupian, Macro-Jê, and Cariban families are among those losing speakers fastest.
The Americas hold roughly half the languages on UNESCO's "critically endangered" list. The next century will produce most of the remaining loss.
Papua New Guinea. The world's most linguistically diverse nation. Of ~840 languages, perhaps half are healthy (children still learning), the other half are sliding. Tok Pisin, the English-based national creole, is the lingua franca consolidating speakers. Where road and school reach, the local language usually loses.
Indonesia. Bahasa Indonesia, mandated since independence (1945), has displaced hundreds of regional languages from formal domains. The major regional languages (Javanese, Sundanese, Madurese, Minangkabau, Buginese) survive — Javanese has ~80 million speakers — but smaller local languages of Sumatra, Sulawesi, the Lesser Sundas, and Papua are losing ground.
Philippines. ~180 languages. Tagalog (as Filipino) and English dominate; Cebuano, Ilocano, Hiligaynon, Bikol all robust; smaller languages of the Cordillera and Mindanao struggling.
India. ~447 languages. The 2011 census recorded 99 "non-scheduled" languages with 10,000+ speakers; many smaller ones (especially in the Northeast: Tibeto-Burman varieties of Arunachal Pradesh, Nagaland, Manipur) are critically endangered.
China. ~300 languages. Mandarin's institutional dominance has compressed minority languages — Tibetan, Uyghur, Mongolian, Zhuang, the Yi varieties — though demographic mass keeps the largest of these alive. Smaller languages of Yunnan, Tibet, and Xinjiang are in trouble.
Russia and Siberia. The Soviet and post-Soviet language policy, combined with russification and small-community shrinkage, has decimated the indigenous languages of Siberia. Itelmen, Yukaghir, Aleut, Even, Evenki, Nivkh, Ket — all with double-digit speaker numbers in some cases.
Pacific island languages — Tongan, Samoan, Fijian, Tahitian, Marshallese — have varied fortunes, from healthy mid-size populations to critically small ones (Niuean, ~1,500 speakers).
Africa has roughly 2,000 languages — about 30% of the global total — and a relatively low proportion are critically endangered, in international terms. The reasons are structural: many African nations are functionally multilingual, with national languages, regional languages, and local languages all alive in different domains. A typical Cameroonian or Tanzanian speaks three to five languages without any of them dying.
Where endangerment is concentrated:
Click languages of southern Africa. The Khoisan languages — Khoekhoe, Naro, !Xóõ, ǂ'Amkoe, ǁXegwi — are mostly under severe pressure from Bantu and English / Afrikaans. Many have under 1,000 speakers. The click consonants of the Khoisan languages are not found elsewhere.
Pygmy languages. Most Central African forest peoples (Aka, Baka, Mbenga, Bambuti) have shifted to neighbouring Bantu languages over the past century, retaining only specialised vocabulary (forest, music, ritual).
Cushitic and Omotic small languages in Ethiopia and Kenya. Many under 5,000 speakers. Boon, Ongota, Kwegu — critically endangered.
Bantu micro-languages. Across central, southern, and eastern Africa, dozens of small Bantu languages are being absorbed by larger Bantu neighbours.
Africa's relatively stable position is partly artefact: many languages have not been counted carefully. As linguistic surveys catch up (the 2020s have seen substantial new fieldwork in Cameroon, Tanzania, Ethiopia, the DRC) the picture is darkening. The resilience of African multilingualism is real but should not be mistaken for total preservation.
Europe has only ~280 languages, fewer than any other inhabited continent. The endangered ones are minorities within nation-states.
The Celtic family. Welsh stable; Irish under pressure (~1.9 million claim ability, ~70,000 use it daily); Scottish Gaelic in trouble (~57,000 speakers, declining); Manx revived from extinction; Cornish revived from extinction; Breton severely endangered (~210,000 speakers, mostly elderly, French-language hostility historic).
Sámi languages. Nine distinct varieties across northern Norway, Sweden, Finland, and Russia. Northern Sámi (~25,000 speakers) is robust; the smaller varieties — Pite Sámi, Ume Sámi, Akkala Sámi (last speaker died 2003), Ter Sámi, Skolt Sámi — are critically endangered or extinct. Norwegian and Swedish state policies historically suppressed Sámi; current policy supports it.
Romansh. ~60,000 speakers in canton Graubünden, Switzerland. One of four official languages of Switzerland. Holding steady but small.
Basque. ~750,000 speakers in northern Spain and southwestern France. Stable and growing through Basque-medium ikastolas (immersion schools). Recovered substantially from Franco-era suppression (1939-1975, when public Basque was banned).
Frisian, Sorbian, Kashubian. Small Germanic and Slavic minorities surviving with state recognition.
Italian regional languages. Italy's "dialects" — Sicilian, Venetian, Lombard, Neapolitan, Sardinian — are in many cases full languages with millions of speakers. Their unrecognised status as "dialects" has historically suppressed them; the trajectory has been generations of decline.
The European story: state recognition saves languages, lack of recognition kills them. The pattern is more legible in Europe than anywhere else.
"Extinct" is the wrong word for some categories of language loss. A language with no speakers but a written record can sometimes be "reawakened" — brought back into use as a community language.
Cornish. The last fluent native speaker (Dolly Pentreath) died in 1777. Henry Jenner's Handbook of the Cornish Language (1904) and Robert Morton Nance's standardisation (1929) began the revival. By 2010, perhaps 600 people spoke Cornish daily, with some children being raised in it. UNESCO upgraded Cornish from "extinct" to "critically endangered" in 2010 — a rare reverse classification.
Wampanoag (Massachusetts). The last native speakers died in the 19th century. Jessie Little Doe Baird's MIT-supervised Wôpanâak Language Reclamation Project began in 1993, working from 17th-century documents (the King James Bible was translated into Wampanoag by John Eliot in 1663). By 2013 there was at least one child being raised as a first-language Wampanoag speaker — the first in seven generations.
Manx. The last native speaker died in 1974, but before that, recordings had been made by the Irish Folklore Commission (1948), and a small revivalist community carried on. Bunscoill Ghaelgagh, the Manx-medium primary school on the Isle of Man, opened in 2001. Hundreds of children now speak Manx as a first or second language.
Hebrew, again. The largest reawakening. From a liturgical-only language to ~9 million native speakers in one century.
Reawakening is the limit case of revitalisation. It requires extensive written records, community will, and sustained institutional commitment. It also requires a willingness to accept that the reawakened language is a continuation, not a perfect reconstruction — vocabulary will be modernised, contact effects will linger, the language will have changed.
The case is contested even within linguistics. Some arguments:
The scientific argument. Each language is a working sample of what human cognition and grammar can do. The hypothesis space of possible languages narrows with each death. Linguistic typology, historical linguistics, and the cognitive science of language all depend on this diversity remaining studyable. (Krauss; Hale 1992 in Language, "Endangered languages.")
The cultural-heritage argument. Each language carries oral literature, ecological knowledge, kinship systems, and ritual practices that are not translatable. A culture cannot be preserved without its language; museum-style preservation of "the language" without speakers is not preservation. (Nettle and Romaine 2000; Harrison 2007.)
The human-rights argument. Speakers of small languages did not freely choose to abandon them; the choice was structurally imposed by larger states and economies. The right to use one's mother tongue is a human right (UN Declaration on the Rights of Indigenous Peoples, 2007, Articles 13-14). (Skutnabb-Kangas; many others.)
The ecological-knowledge argument. Indigenous languages encode environmental knowledge — plant uses, weather patterns, navigation, agriculture — refined over centuries and not preserved elsewhere. Languages of the Amazon, the high Arctic, and the Pacific encode information climate change makes urgent.
The skeptical argument. Some linguists (notably John McWhorter) argue that linguistic diversity, while regrettable to lose, does not justify coercing communities to maintain languages they are choosing to abandon. The decision is theirs.
The arguments are not mutually exclusive. The strongest case combines all of them: each language is irreplaceable scientifically, culturally, and ethically, and the choice to maintain it should be the speaking community's, supported by the institutions that have historically suppressed it.
The toolkit of working revitalisation programs:
Language nests. Full-immersion preschools run by elders. Kōhanga reo (Māori), Pūnana Leo (Hawaiian), Kahnawà:ke survival school (Mohawk). The most-imitated single intervention.
Master-apprentice programs. One-on-one pairing of an elderly fluent speaker with a younger learner committed to ~600+ hours of structured immersion. Developed by Leanne Hinton and the Advocates for Indigenous California Language Survival (1992 onward); has produced new fluent speakers of Karuk, Yurok, Hupa, others.
Immersion camps. Summer-long full-immersion programs for adults and youth. Scotland's Sabhal Mòr Ostaig (Scottish Gaelic, founded 1973) and equivalents for Cherokee, Ojibwe, others.
Medium-of-instruction schooling. Where critical mass and political will exist: schools that teach all subjects in the language. Welsh-medium, Hawaiian-medium, Māori-medium, Basque ikastolas. Most demanding in resources, most effective in outcome.
Documentation and revitalisation linked. Most modern programs combine archival recording with community pedagogy. The records produced for the linguist also become the textbooks for the community.
Family language plans. Households committing to speak the language at home. The unglamorous core of any revitalisation: without family transmission, the institutional efforts cannot stick.
Standardisation. Many small languages had no agreed orthography. Revitalisation often requires negotiating one — typically through a community working group with linguistic advice. The negotiations can be politically charged.
None of these is a silver bullet. The languages that have recovered (Hebrew, Hawaiian, Māori, Welsh, Basque, Catalan) used several methods at once, sustained over decades, with substantial state or community resources. Most endangered languages cannot mobilise those resources.
The relationship between digital technology and minority languages is complicated.
Against. Digital monolingualism privileges large languages. A user without a computer interface in their language faces practical pressure to switch — keyboards, websites, search, voice assistants, social media all default to majority languages. The 2000s and 2010s expanded this pressure dramatically: smartphone adoption in the global South made daily digital life essentially monolingual for most users.
For. Recording, archiving, and distribution are now cheap. A community can produce its own language-learning app for under $50,000 (the budget of a single 1980s field linguist's monograph). Apps such as Duolingo's Hawaiian, Navajo, and Welsh courses have introduced these languages to millions of casual learners. The language-learning audience is no longer limited to the speaker community.
Open documentation. Initiatives like the Endangered Languages Archive (ELAR), AILLA (Latin American), PARADISEC (Pacific), and the Living Tongues Institute publish open recordings online. Researchers and community members anywhere can access them.
Machine translation and large language models. Google Translate covers ~243 languages (as of 2024); it does not cover the 6,500+ that are smaller. Large language models trained on internet-scale text encode the linguistic majority and replicate it. The tools that could help small languages mostly do not include them.
The "low-resource NLP" research field works specifically on building useful tools (translation, speech recognition, OCR for low-resource scripts) for under-served languages. Progress is real but slow; the demand outstrips the supply of researchers willing to work on Pirahã, Burushaski, or Mansi.
The net effect: technology has made documentation easier and language pedagogy more accessible, but the daily-life dominance of majority languages continues to grow.
The work is not anonymous. Some figures.
Michael Krauss (1934-2019). Director of the Alaska Native Language Center 1972-2000. Author of the 1992 alarm. Spent fifty years documenting Alaskan languages, especially Eyak. Without his archive, much of Alaskan linguistic heritage would not exist in any form.
Ken Hale (1934-2001). MIT linguist, prolific fieldworker, raised his children speaking the languages he documented (Warlpiri, Navajo). His 1992 paper alongside Krauss's was foundational. Worked closely with Aboriginal Australian, Native American, and Central American language communities. The annual Ken Hale Prize (Linguistic Society of America) honours fieldworkers in his name.
Joshua Fishman (1926-2015). Sociolinguist; defined the field of "reversing language shift." His scale (GIDS) is the standard measurement tool of revitalisation. Reversing Language Shift (1991), Can Threatened Languages Be Saved? (2001).
K. David Harrison. Co-founder, Living Tongues Institute. Documented Tofa, Tuvan, and others. When Languages Die (2007). National Geographic films of last-speaker work brought the issue to popular audiences.
Daniel Everett. Pirahã. Disputed but vivid; Don't Sleep, There Are Snakes (2008) is one of the few field-linguistics books to have reached non-specialist readers.
Larry Kimura, William "Pila" Wilson, Kauanoe Kamanā. Hawaiian revitalisation architects, founders of ʻAha Pūnana Leo.
Jessie Little Doe Baird. Wampanoag reawakening. MacArthur Fellow 2010.
Leanne Hinton. Master-apprentice methodology. How to Keep Your Language Alive (2002).
Suzanne Romaine, Daniel Nettle. Vanishing Voices (2000), the most-cited general-audience treatment.
The work is mostly slow, mostly unglamorous, and mostly performed by community members rather than outside academics. Where outside linguists succeed, it is in partnership with the communities they document.
What to read.
One documentary embedded; two more linked. National Geographic's Dying Languages covers the field for a general audience — last speakers, fieldwork urgency, the revitalisation hopes.
National Geographic — Dying Languages
Two more:
• Saving Languages from Extinction — overview of contemporary revitalisation efforts.
• ʻĀhaʻi ʻŌlelo Ola: Hawaiian Revitalization Efforts — inside the Pūnana Leo language-nest movement.
And to read alongside: Krauss 1992 (the foundational alarm, freely available through JSTOR); the UNESCO Atlas of the World's Languages in Danger (online, free); the Endangered Languages Project (endangeredlanguages.com), a community-curated archive launched in 2012 by Google and the Alliance for Linguistic Diversity.
For people considering field-linguistic work: the Endangered Languages Documentation Programme (ELDP at SOAS), the NSF Documenting Endangered Languages program, and the AILLA (Latin American) and PARADISEC (Pacific) archives all welcome volunteer involvement.
Terms used by the field.
The terminology is contested. "Murder" and "suicide" both presume a level of agency that working communities reject; many endangered languages are dying in conditions where neither term fits. The current preference in respectful writing is "shift" or "loss" — neutral about cause.
Pulling it together. Across the documented revitalisations of the past forty years, certain features predict success:
Critical-mass speaker community. A few thousand current speakers is roughly the floor. Below that, the work is heroic but unlikely to produce self-sustaining transmission. Above it, the work is hard but feasible.
Institutional anchoring. Schools, broadcast media, government recognition, official-language status. Hebrew, Welsh, Hawaiian, Māori, Basque all benefited from one or more of these. Languages without any institutional foothold rarely recover.
Long time horizons. Revitalisation is a 30-50 year project minimum. Quarter-decade efforts produce documentation; half-century efforts can produce living communities.
Community ownership. Programs run by speakers and descendants, not by outside experts. Outside linguists can document and advise; only the community can decide to live in the language.
Family transmission. However effective the institutional supports, the language has to be spoken at home or it does not become a first language. The hardest piece, since it requires individual household decisions sustained for decades.
Aesthetic pleasure. Languages that produce literature, music, and humour their speakers want to engage with retain speakers better than languages confined to ceremony or pedagogy. The pop song in Welsh, the novel in Inuktitut, the comedy in Māori — these matter more than they look.
What does not predict success: the language's "linguistic interest" (typological rarity, complexity, beauty) is unrelated to its survival prospects. The least linguistically remarkable languages and the most remarkable are equally vulnerable. The factors that decide are demographic, institutional, and political.
The trajectory.
By 2050. Approximately 1,500-2,000 of the languages alive today will have lost their last fluent native speakers. The losses will concentrate in Australia, Amazonia, Siberia, parts of New Guinea, and indigenous North America.
By 2100. Krauss's prediction — between 50% and 90% loss from the 1992 baseline — will be approximately fulfilled. The lower end (50% surviving) requires sustained revitalisation work at unprecedented scale; the upper end (only 10% surviving) is the trajectory if current rates continue.
The new languages. The 21st century is also producing new languages — pidgins becoming creoles, dialects diverging, mixed languages emerging. Tok Pisin, Bislama, Sango, Jamaican, Singlish, Spanglish. The total inventory is not declining as fast as the loss-rate suggests; new emergence offsets some loss. But the direction of net change is sharply downward.
The documentation legacy. The 1992-2024 documentation push will leave behind, by 2100, the most comprehensive record of human linguistic diversity ever assembled. ELDP, DOBES, AILLA, PARADISEC, ANLA, and dozens of national archives will hold tens of thousands of hours of recordings, hundreds of grammars, thousands of dictionaries. Future scholars will know in detail what was lost.
The ones that come back. Some will. Hebrew did; Hawaiian did; Māori did; Welsh held; Cornish came partly back. The cases prove the work can succeed under sufficient conditions. Most languages will not get those conditions. A few will. The work continues.
Three claims, in conclusion.
Linguistic diversity is irreplaceable. Each language is a working sample of what human cognition and culture can do. The space of possible human languages cannot be reconstructed from the languages that survive; each death narrows what we can know about ourselves as a species.
The current loss is unprecedented. No earlier era of human history has seen language extinction at the rate of the past 200 years. The pre-1500 baseline rate of language loss was probably approximately matched by language emergence; the post-1500 rate has been net catastrophic. The 20th and 21st centuries will likely account for the disappearance of more languages than the previous five thousand years combined.
The losses are not natural. Each language is dying because of specific historical pressures: state policy, residential schooling, economic restructuring, urbanisation, demographic disruption. The implication is that the trend is not inevitable. Different policies — recognising minority languages, supporting bilingual schooling, funding documentation, ceding institutional space — would change the rate. Some have. Most have not.
The question is not whether half the world's languages will be alive in 2100. They will not. The question is which half, and what the survivors will have inherited from those that did not survive. The work of the next century is, in part, the work of giving the survivors as much of the lost record as possible. The work has begun. Whether it scales is up to institutions, communities, and the choices of speakers themselves.
Endangered Languages — Volume III, Deck 7 of The Deck Catalog. Set in Iowan Old Style on a deep moss-and-ember palette. Thirty leaves on what is dying, what has been recorded, and what has come back.
"Surely, just as the extinction of any animal species diminishes our world, so does the extinction of any language" — Krauss, 1992.
↑ Vol. III · Lang. · Deck 7