How Languages Die: The Sociolinguistics of Endangerment and Revival

Roughly 7,000 languages are spoken in the world today. By the most-cited estimate, drawn from Michael Krauss's 1992 paper in Language, between half and ninety percent of them will not survive the 21st century. This is not a forecast in the ordinary sense; it is an observation about who is still teaching their language to children. Languages do not die because their last speakers grow old. Languages die because at some point their speakers stop teaching them to the next generation, and then forty years later the last speakers grow old.

The mechanism is mundane and consistent enough that sociolinguists can name a language's stage of endangerment from a small set of indicators, and from those indicators forecast the trajectory accurately. The pattern is bleak. The exceptions, where a language has reversed course and reentered intergenerational transmission, are rarer and more interesting than the ordinary case.

The dominant-language displacement pattern

The standard story is a version of the following. A community speaks language L, which has historically been the language of family, work, religion, and community life. A more economically dominant language D arrives, usually carried by colonization, conquest, or labor migration. D is the language of state administration, of schools, of formal employment. L speakers learn D as a second language; their children grow up bilingual.

The transition is not when L stops being spoken; it is when L stops being passed to children. Bilingual parents, for entirely understandable reasons, begin to speak D to their children, because D is what will determine the children's economic prospects. The children grow up D-dominant with passive understanding of L. They marry and have their own children, and now there is no L in the home at all. Three generations from full L community to no L speakers under sixty.

This is the pattern that has consumed most of the indigenous languages of North America, Australia, and Siberia in the last century. It is consuming languages of West Africa, the Andes, and Papua New Guinea now. The driving force is rarely overt suppression (though that exists); it is the economic logic of giving children access to the dominant economy.

Krauss's stages

The sociolinguist Michael Krauss formalized the stages of language endangerment in a way that is now standard. Roughly:

• Safe: All generations speak it, and it has institutional support (state recognition, school instruction, media). English, Mandarin, Spanish.
• Stable but threatened: All generations speak it, but its domains are shrinking. Maybe used at home and in church, not in school or work.
• Eroding: Most adults speak it, but children mostly do not. Active in some homes, absent in others.
• Moribund: Only spoken by older adults. No children are learning it as a first language. The language has perhaps thirty to fifty years before its last fluent speaker dies.
• Nearly extinct: Only a handful of elderly speakers remain. Domain reduced to memory and perhaps formal occasions.
• Extinct: No remaining speakers.

The transition from "stable but threatened" to "eroding" is the irreversible-looking moment in most cases. Once intergenerational transmission breaks, mathematical demography takes over. The remaining speaker pool is older than the population average and shrinks faster than it grows. This is why a language with 50,000 speakers can be functionally doomed and a language with 5,000 speakers can be perfectly stable; the relevant number is not how many speak it but how many under thirty speak it.

What is lost

The question "what is lost when a language dies" is not sentimental; it has specific and partial answers. A language carries vocabulary that maps to a community's environment and practices. The Tiwi language of northern Australia has detailed vocabulary for kinship categories that English does not encode. The Quechuan languages of the Andes have precise terms for landscape features and weather patterns specific to the high altiplano. Some of this is recoverable through documentation; much of it is not, because some of the vocabulary describes practices that have themselves disappeared with the language community.

Languages also carry oral literature, song traditions, naming systems, and everyday metaphors that cannot be cleanly translated. They carry the sound and rhythm of the speech community. They carry, in many cases, the mythology and cosmology of a people. Translation rescues some of this; documentation preserves more. But there is a residue of meaning that lives only in fluent use, and that is what dies when the last speaker dies.

There is also a more abstract loss. Each language is a natural experiment in what human grammar can do. Languages with case-marking, with switch-reference, with evidentiality, with click consonants, with whistled registers, with grammatical genders for animacy or shape, with tense systems organized around discourse topology rather than time. Linguistics learns from this diversity what the boundaries of human cognitive capacity are. As languages contract toward the dominant family of European-derived nominative-accusative SVO structures, that diversity shrinks. We find out what humans can do with grammar by surveying the room. Each empty chair is information lost.

The revival cases

The most commonly cited language revival is Hebrew, which is misleading. Hebrew was preserved as a liturgical and scholarly language in Jewish communities throughout the diaspora; it was not a dying everyday language brought back to life so much as a non-everyday language re-domesticated. The revival was the work of a specific person, Eliezer Ben-Yehuda, who arrived in Ottoman Palestine in 1881 determined to raise his son as a monolingual Hebrew speaker. He largely succeeded, and the political project of Zionism gave Hebrew a state's worth of institutional support. By 1948 Hebrew was a national language. By 2000 it was the everyday language of seven million people.

The Welsh case is more typical. Welsh was in steep decline through the 19th and 20th centuries, eroded by English-language schooling and economic migration to England. The reversal began with the Welsh Language Society in the 1960s, which agitated for and won bilingual road signs, Welsh-medium schools, and a Welsh-language television channel (S4C, 1982). The Welsh Language Acts of 1993 and 2011 established Welsh as a co-equal language for public administration. The 2021 census shows Welsh speaker numbers stabilizing for the first time in centuries, with notable growth among under-fifteens. The revival is incomplete and could still fail, but the trajectory has changed.

Maori in New Zealand has a similar story: kohanga reo (language nests) that immerse pre-school children in Maori, kura kaupapa Maori (Maori-medium schools) that continue the immersion, and a national Maori language broadcaster. Maori went from moribund in the 1970s to stable-but-threatened today, an unusual climb up Krauss's ladder.

Hawaiian is the case most directly comparable to many North American Indigenous languages: it had been reduced to a few hundred elderly speakers by the 1980s, was prohibited in schools for most of the 20th century, and has been brought back through Punana Leo immersion preschools (founded 1983) and Hawaiian-medium schools at every grade level. The next generation of fluent native Hawaiian speakers is now in its twenties.

What the revivals share

The successful revivals share a small number of characteristics. They have political support strong enough to win school-medium status, because that is where intergenerational transmission can be re-established at scale. They have parents willing to enroll their children in immersion programs at material cost (job opportunities, peer language match). They have a generation of speakers committed enough to teach despite material disadvantage. And they have a recognized cultural identity that the language is bound up with, which gives the revival meaning beyond linguistic preservation.

None of this is easy to organize, and most threatened languages do not have the demographic mass or the political traction to put the pieces together. The revivals are exceptions, and they tend to be more recent, better-funded, and more politically salient than the average endangered language case.

The default forecast remains the one Krauss gave in 1992: a substantial majority of human languages will not survive the century. The exceptions exist; they are valuable both for the languages themselves and as proof that the trajectory is not fatalistic. Languages die because of human decisions, sometimes individual ones over generations. Languages can also live for the same reason. Whether the rest of the world's linguistic diversity is preserved depends on enough such decisions, in enough such places, in the next forty years.