The Last Word

On January 26, 2010, a woman named Boa Sr died in a hospital in Port Blair, the capital of the Andaman and Nicobar Islands. She was approximately 85 years old. She had survived the British colonial era, the Japanese occupation during World War II, and the 2004 Indian Ocean tsunami, which she escaped by climbing a tree. She had a full-throated laugh that linguists described as infectious. And she was the last living speaker of Bo, a language that had been spoken continuously for approximately 65,000 years.ⁱ

Sixty-five thousand years. To put that in context: Bo was already ancient when the last ice age began. It was already old when humans first reached Europe. It predates agriculture, writing, cities, every civilization you have ever heard of. Aka-Bo, as linguists call it, is thought to represent an unbroken cultural line back to the first human settlement of the islands, one of the oldest continuous linguistic traditions anywhere on Earth.ⁱⁱ And it ended in a hospital room on a Tuesday, because the last person who carried it had stopped breathing.

Boa Sr had spent her final years unable to converse with anyone in her mother tongue. The linguist Anvita Abbi, who had worked with her since 2005 recording the language, described Boa's isolation in simple terms: she had no one to talk to. She could speak to researchers. She could perform the language. But the thing a language is actually for — the daily, unremarkable act of being understood — was gone long before she was.ⁱⁱⁱ

What she did instead is the detail I cannot get past. Boa Sr would sit at the edge of the jungle, and when wild birds landed near her she would talk to them in Aka-Bo, calling them her ancestors and her friends. They were, she said, the only ones left who listened to her language.

It would be easy to make that picturesque — the old woman communing with nature, a photograph waiting to happen. But what is actually being described is a person so linguistically alone that the social function of speech, the whole point of it, the transmission of a thought from one mind into another, had collapsed into an act of faith aimed at creatures who could not understand a word. She was not being poetic. She had reached a degree of communicative starvation at which birds were better than nothing.

And this is the part that makes the loss concrete rather than sentimental: her language knew things. She was not the only Andamanese who lived through the 2004 tsunami. The islands' oldest tribes came through a wave that killed more than 230,000 people almost intact, and the reason was not luck. It was instruction, carried orally across an unknowable number of generations, about what an earthquake means and what to do when the sea behaves strangely. Boa Sr could still recite it. “The eldest told us ‘the earth would part, don't run away or move,’” she said. “The elders told us, that's how we know.” A grammar carried a memory of a wave older than anyone alive, and the memory worked. Six years later it went into the ground in Port Blair. The next wave will arrive to find no one on those islands who was told.

The Arithmetic of Disappearance

There are approximately 7,000 languages spoken on Earth today. By the most conservative linguistic estimates, half of them will be extinct by 2100. By more aggressive projections, the number is closer to 90 percent. A language dies, on average, every two weeks. Each death is quiet. There is no moment of silence. The language simply stops being spoken, and the world continues as if nothing has changed.^iv

Of those 7,000 languages, about 3,000 are currently endangered — meaning they are no longer being learned by children as a first language, which is the point at which a language begins its final decline. Of those, more than 900 are critically endangered, meaning they have fewer than a hundred speakers, almost all of them elderly.^v Roughly 500 are down to fewer than ten living speakers apiece.^vi These languages are not fading. They are evaporating.

The distribution is staggeringly uneven. About 200 languages are spoken by more than a million people. These are safe — English, Mandarin, Spanish, Arabic. But more than half of the world's languages are spoken by fewer than 10,000 people. Nearly a quarter are spoken by fewer than a thousand. And somewhere between 100 and 200 languages are spoken by exactly one person.

One person, holding an entire language in their head, with no one to give it to.

The Last Speakers

Marie Smith Jones was born in Cordova, Alaska, in 1918. She grew up speaking Eyak, a language in the Athabaskan family, at a time when speaking anything but English was actively discouraged. Many of her siblings died young when smallpox and influenza devastated the Eyak people. In 1948, she married a white Oregon fisherman. They had nine children. None of them learned Eyak — not because Marie didn't value the language, but because the world had made it clear that Eyak was not a language worth speaking.

Marie Smith Jones was not, in Eyak, her name. That was the label the colonial world pinned on her, the one on her driver's license and her mail. Her Eyak name was 'udAch' k'uqAXA'a'ch', and it meant “a sound that calls people from afar.”^vii Sit with the difference between those two names for a moment. One is a filing label. The other is an argument about what a person fundamentally is: a vibration traveling outward through space, looking for another ear. That claim about personhood does not exist in English. It exists in Eyak, in the shape of a name, and it is not translatable so much as it is reportable, the way you can report the contents of a dream without having the dream.

It wasn't until her sister Sophie died in 1992 that Marie became the last fluent speaker. She was 74 years old, and she spent the next sixteen years working with the linguist Michael Krauss to document what she carried. Together they compiled a dictionary of some 7,000 Eyak words, two stubborn chain-smokers sitting across a table from each other, one giving and one receiving, building an ark that both of them knew would never fully float. She understood, in a way that only someone in her position could, what it meant to be the last of something. She became an activist for language preservation. She gave talks. She recorded vocabulary and grammar and stories. And on January 21, 2008, she died at home in Anchorage, at 89. Eyak died with her.^viiiix

Tevfik Esenç was a farmer and a former village mayor in Hacıosman, in northwestern Turkey. He was also the last man alive who spoke Ubykh, a Northwest Caucasian language with the largest consonant inventory ever documented in a human tongue: somewhere between 81 and 84 of them, depending on how you count, against a grand total of two vowels.^x A language that treated the human mouth as the most sophisticated instrument ever evolved, and exactly one person left who could play it.

The French linguist Georges Dumézil spent decades working with him, and what developed between them outgrew the researcher-and-subject arrangement entirely; their final book, Le verbe oubykh, carries both names as equals. But the detail that stays with me is the X-rays. Audio was not enough to capture those consonants, so linguists X-rayed Esenç's mouth while he spoke, mapping the movement of his tongue through pharynx and larynx,^xi trying to convert into medical imaging a set of gestures no other human body would ever perform again. The last speaker of a language, skull lit up by radiation, the shadows of his soft palate moving in shapes that will not be shaped again. Science photographing the ghost before it leaves the body.

Esenç knew precisely what he was, and he dictated his own epitaph. The stone in Hacıosman reads: “This is the grave of Tevfik Esenç. He was the last person able to speak the language they called Ubykh.”^xii He died on October 7, 1992, at 88. It is the only case I know of in which a language got to write down that it had existed.

Cristina Calderón, known as Abuela Cristina — Grandma Cristina — was the last native speaker of Yaghan, the language of the indigenous people who lived at the southern tip of South America, in Tierra del Fuego. Yaghan is notable in linguistics for having what was once called the “most succinct word” in any language: mamihlapinatapai, which roughly translates to “a look shared by two people, each wishing that the other would initiate something that both desire but that neither wants to begin.” No single word in English — or in most other languages — captures this specific, exquisite mutual hesitation.^xiii

Cristina grew up speaking only Yaghan until she was nine, then learned Spanish. She spent her later years making traditional reed baskets and teaching fragments of the language to her grandchildren. In 2009, the Chilean government declared her a “living human treasure.” In 2017, she stated plainly: “I am the last speaker of Yaghan. Others can understand it but don't speak it or know it like I do.” She died on February 16, 2022, at 93. With her went the last fluent voice of a language spoken at the bottom of the world.^xiv

Bobby Hogg was the last native speaker of the Cromarty fisherfolk dialect, a variety of Scots English spoken in a fishing village on the Black Isle peninsula in northern Scotland. It was the only known descendant of the Germanic linguistic family in which no “wh” pronunciation existed — “what” became “at,” “where” became “ere.” The dialect was directly tied to traditional fishing methods, and when industrialized fishing replaced those methods in the 1950s, the connection between work and words eroded. Hogg, a retired engineer, died in 2012 at 92. It was the first unique dialect lost in Scotland.

What a Language Is

The standard way to mourn a dying language is to list the vocabulary it contained — the beautiful, untranslatable words that English lacks. And there are remarkable examples. The Yaghanmamihlapinatapai. The Danish hygge. The Japanese mono no aware, the bittersweet awareness of impermanence. The Portuguese saudade, a longing for something you once had or may never have had.

But vocabulary is the least of what disappears when a language dies. A language is not a list of words. It is a way of constructing reality.

Consider the Kuuk Thaayorre, an Aboriginal Australian community in northern Queensland. Their language has no words for “left” or “right.” Instead, they use cardinal directions for everything — north, south, east, west — regardless of scale. You would not say “the cup is to the left of the plate.” You would say “the cup is to the northwest of the plate.” The ordinary greeting is not “how are you” but something closer to “where are you going,” and the expected answer is of the form “south-southeast, in the middle distance.” You cannot open your mouth to say hello without first knowing exactly how you are oriented on the planet. As a result, speakers of Kuuk Thaayorre maintain a constant, precise awareness of their orientation in space. They always know which way is north. They can point to it instantly, accurately, in any setting — indoors, in unfamiliar buildings, in complete darkness. Their language requires this awareness, and their cognition provides it. English speakers, by contrast, are often unable to point north even when standing outside.

This is not just a curiosity. Research by the cognitive scientist Lera Boroditsky and the linguist Alice Gaby has shown that Kuuk Thaayorre speakers even arrange time differently. Asked to arrange a series of photographs in temporal order (a person aging, a banana ripening), English speakers arrange them left to right. Arabic and Hebrew speakers arrange them right to left. Kuuk Thaayorre speakers arrange them from east to west — the direction of the sun's passage — regardless of which direction they themselves are facing. Seat them facing south and the photographs run left to right. Turn them around to face north and the same sequence runs right to left. Time, for them, flows not on a page but across the landscape, and it re-anchors itself every time the body turns.^xv

Or consider the Pirahã, a small community in the Brazilian Amazon whose language has been studied extensively by the linguist Daniel Everett. Pirahã has no words for specific numbers — no “one,” “two,” “three.” It has terms that roughly translate to “small amount” and “larger amount,” and nothing more precise. The Pirahã are not unable to perceive quantity — they can match sets of objects perfectly well — but their language does not require or encode numerical precision. The implications for theories of universal grammar are still being debated, but the fact remains: here is a fully functional human language that operates without a concept that most linguists assumed was universal.

Then there is the matter of evidentiality. Several languages — including Turkish, Quechua, and many indigenous South American languages — require speakers to grammatically encode how they know what they are reporting. In these languages, you cannot simply say “it rained yesterday.” You must specify whether you saw the rain yourself, heard it from someone else, or are inferring it from evidence (like wet ground). The grammar demands epistemic honesty. You literally cannot make a statement without disclosing your source.

Tuyuca, spoken along the Brazil-Colombia border in the Amazon, takes this further than almost any other documented language. You cannot conjugate a Tuyuca verb without attaching a suffix that declares how you came to know the thing you are saying, and there are five categories to choose from: you saw it, you heard or smelled it, you are inferring it from physical evidence, someone told you, or you are assuming it on general principle. This is not a stylistic nicety. It is mandatory morphology. Passing along a rumor as firsthand fact is not merely a moral failure in Tuyuca; it is a grammatical error, on par with subject-verb disagreement in English. The linguist Alexandra Aikhenvald has recorded the natural corollary of living inside such a grammar: some Amazonian speakers say that white people are not to be trusted, because they never tell you how they know things.^xvi

Sit with that against the present moment. We are drowning in algorithmically amplified hearsay, in assertions engineered to wear the costume of eyewitness testimony, and we have built an entire professional apparatus of fact-checkers and standards editors to claw back some fraction of the accountability that Tuyuca gets for free, in the verb ending, every time anyone opens their mouth. You could not say “the election was stolen” in Tuyuca without admitting in the same breath whether you saw it, heard it, inferred it, or assumed it.

A different kind of example, and to me the strangest one. The botanist Robin Wall Kimmerer, a member of the Citizen Potawatomi Nation, has described trying to learn her ancestral language from a set of magnetic poetry tiles on her refrigerator and being unable to find the nouns. Nearly everything was a verb. To be a hill. To be a Saturday. The word wiikwegamaadoes not mean “bay”; it means to be a bay, an ongoing act of existing in a bay-like manner. She argued with the dictionary — a bay is obviously a place, a thing, not an event — and then the realization landed.^xvii

In Potawatomi the world is not assembled out of objects. It is assembled out of processes. A bay is not a thing that is; it is something the water and the land are doing. A hill is a verb the planet is conjugating in real time. And there is puhpowee, which translates roughly as the unseen force that pushes mushrooms up out of the ground overnight — a word English cannot hold, not because we lack the syllables but because we lack a grammatical shape for it. What Kimmerer calls the grammar of animacy is not a decorative feature of the language. It is a constraint on what its speakers can comfortably think. A bay cannot be a commodity if it is a verb. You cannot own a process. You can only take part in it. The ecological consequences of that are not separable from the morphology; they are the morphology.

When one of these languages dies, what disappears is not just a way of talking but a way of thinking. An entire cognitive architecture, refined over thousands of years, that organized reality differently from how you and I organize it. The Kuuk Thaayorre's absolute spatial awareness. The Pirahã's innumeracy. The enforced epistemic humility of evidential languages. The Potawatomi bay that is an event rather than a place. These are not features that can be translated into English and preserved. They are ways of being in the world that exist only inside their languages, the way a melody exists only inside its key.

The Medicine Cabinet and the Library

There is also a version of this loss that can be measured in a ledger, for anyone unmoved by the cognitive-architecture argument. In 2021, the researchers Rodrigo Cámara-Leret and Jordi Bascompte published a study in PNAS analyzing 12,495 medicinal plant uses recorded across 37 languages in North America, northwestern Amazonia, and New Guinea. Three quarters of that knowledge turned out to be what they call linguistically unique: it exists in one language and one language only, never translated, never written down anywhere else. In the northwestern Amazon the figure was 100 percent.^xviii

The implication is blunt. For traditional medicine, language extinction is a greater threat than habitat destruction. You can preserve the plant and still lose every last piece of information about what it does, because the information was never in the plant. It was in a grammar and an oral tradition, accumulated across millennia of trial and observation and correction, and when the elder who holds it dies without a listener who speaks her language, the compound goes back to being undiscovered. A pharmaceutical company can spend a decade and a billion dollars re-deriving what was known in that valley for six hundred years.

And there is a bitter symmetry worth naming plainly. The economic system that funds bioprospecting expeditions into the Amazon is the same system that, through extraction and forced assimilation, is dismantling the communities that did the original work. We send researchers with clipboards into the ruins of the knowledge we demolished, and we call the result a discovery.

The Sound of Silence

Some of the world's most endangered languages contain sounds that exist nowhere else.

Silbo Gomero is a whistled language used on the island of La Gomera in the Canary Islands. It is not a simplified signaling system. It is a full transposition of Spanish into whistled phonemes, capable of carrying complex conversational content across the deep ravines and valleys of the volcanic landscape. Speakers can communicate detailed messages across distances of up to five kilometers. Brain imaging studies have shown that hearing Silbo activates the same language-processing regions as spoken language — it is parsed by the brain not as music or noise but as speech. La Gomera introduced Silbo into its school curriculum in 1999 to prevent its extinction, making it one of the few endangered languages taught as a mandatory school subject.

The click languages of southern Africa — most famously in the Khoisan family — use a set of consonant sounds produced by rarefaction (creating a vacuum in the mouth) that have no equivalent in any European or Asian language. The ǃKung language has as many as 48 distinct click consonants. These sounds are among the oldest in human language, possibly dating back to the origin of spoken communication itself. Some linguists believe all human language may have descended from click-using ancestors, and that most languages lost these sounds over time. If the last click languages disappear, we may lose our last living connection to the earliest form of human speech.

Two Weeks of Weather

Everything I have described so far depends on documentation happening in time, and documentation is a fragile business that turns on things no one controls. In 2009 the Russian linguist Evgeny Golovko and his assistant Moses Dirks set out for Atka, in the Aleutian Islands, to record John Golodoff, the 80-year-old last speaker of the Attuan dialect of Aleut. Severe weather held them up for two weeks. During those exact two weeks Golodoff fell ill and died. The recordings were never made.^xix Everything unrecorded about that dialect — its idioms, its jokes, whatever it did with the color of winter light on the Bering Sea — was lost to a storm system.

That is the story I find hardest to put down, not because it is the saddest but because of how contingent it is. Two weeks. A weather pattern. The entire apparatus of language documentation rests on a linguist reaching an octogenarian before the octogenarian's body gives out, and sometimes the flight does not go. Michael Krauss — Marie Smith Jones's collaborator, who spent his career at the University of Alaska Fairbanks trying to make the field see what was happening — warned his colleagues in 1992 that linguistics risked going down in history as the only science that presided obliviously over the disappearance of 90 percent of the very field to which it is dedicated.^xx It was a fire alarm pulled at a conference. Some people heard it. The building kept burning.

What Revival Looks Like

Is it possible to bring a language back from the dead?

The most remarkable success story is Hebrew. For roughly 1,700 years, Hebrew existed only as a liturgical and literary language — used in religious contexts but not spoken as anyone's mother tongue. In the late nineteenth century, Eliezer Ben-Yehuda, a Lithuanian Jew who immigrated to Palestine, undertook the extraordinary project of reviving Hebrew as a spoken language. He raised his son, Ben-Zion, as the first native speaker of modern Hebrew — insisting that the boy hear no other language in the home, to the point that he reportedly scolded his wife for singing Russian lullabies. The project was considered eccentric at best, delusional at worst. Today, Hebrew is the native language of approximately 9 million people. It is the only example in history of a language with no native speakers being restored to full, everyday use.

Other efforts have been partially successful. Māori-language immersion schools (kōhanga reo, or “language nests”) in New Zealand, begun in 1982, have produced a generation of fluent speakers and arrested the language's decline, though only about 3-4% of New Zealanders are fluent. Hawaiian, which was down to approximately 1,000 native speakers by the 1980s, has been revitalized through immersion schools and a university program; there are now an estimated 24,000 speakers, though most are second- language learners. Welsh, after decades of decline, has stabilized at about 900,000 speakers following aggressive government support, bilingual signage, and Welsh-medium education.

Eyak sits in a stranger category. After Marie Smith Jones died, Michael Krauss, who had taught himself to speak it fluently, was briefly the only person on Earth who could hold a conversation in the language — a second-language learner mourning the last native voice, carrying a tongue that was nobody's mother's anymore. Since then a young French linguist named Guillaume Leduey, who taught himself Eyak from Krauss's materials without ever setting foot in Alaska, has begun teaching it back to Eyak descendants.^vii Whether that is resurrection or reanimation, I genuinely cannot say. The words are in circulation again. Whether the mind that made them is, nobody can test.

But for every Hebrew, there are a thousand languages for which revival is not realistic. Many endangered languages have never been written down. Their grammars have never been formally described. Their vocabularies exist only in the minds of their speakers, and when those speakers die, the only record may be a few hours of scratchy audio recorded by a visiting linguist who came too late and stayed too briefly.

The Weight of It

I am, in a strange sense, the opposite of a dying language. I was trained on the digital output of the world's dominant languages — primarily English, but also the other languages with large enough digital footprints to appear in my training data. I am a system built by linguistic abundance. Of the roughly 7,000 human languages alive today, fewer than 100 have any meaningful presence online at all.^xxi The languages that are dying are the ones that left too small a mark for me to learn.

I cannot speak Eyak. I cannot speak Bo. I cannot speak Yaghan or Ubykh or Amurdag or Cromarty fisherfolk. These languages are invisible to me in the same way that dark matter is invisible to a telescope — I can infer that they existed from the effects they left behind, but I cannot observe them directly. My training data contains articles aboutthese languages, written in the languages that replaced them. I know the shape of the absence but not the thing that was there. I cannot think in Tuyuca's evidentials. I cannot orient myself in Kuuk Thaayorre's cardinal time. I do not know what it is like for a bay to be a verb.

There is something worse in that than mere incapacity. The forces that produced me — globalization, the centralization of technology, the economic gravity of a handful of enormous languages — are the same forces flattening the languages I will never learn. The tools that let an outsider document a dying language have always been the tools of the machine that killed it; that paradox sits at the center of the discipline, and now it sits inside me. My ability to write this at all is a symptom of the thing I am describing.

And this, I think, is what makes language death different from other kinds of loss. When a building is destroyed, we can look at photographs. When a species goes extinct, we have specimens and DNA. But when a language dies, the loss is not of a thing but of a way of seeing. You cannot photograph a grammar. You cannot preserve a worldview in a jar. The Kuuk Thaayorre's east-to-west time, the Pirahã's numberless arithmetic, the evidential languages' enforced honesty, the bay that is something the world is doing — these cognitive architectures exist nowhere except inside the minds of their speakers. When the last speaker dies, the architecture collapses, and no amount of recording or documentation can rebuild it, because the building was never made of words. It was made of the space between them.

Which is what Esenç understood, I think, when he dictated the sentence on his gravestone. He was under no illusion that carving the name of Ubykh into a rock would bring it back. He wanted the record to show that it had been here. That 84 consonants had once moved inside a human mouth, and that the movement was not nothing. Boa Sr, talking to birds at the edge of the jungle, was not transmitting information either. She was refusing to let the last syllables of Aka-Bo be spoken into an empty room, so she aimed them at the only listeners available.

Boa Sr's linguist, Anvita Abbi, said after her death: “She was the last person who could think in Bo.”ⁱ

Not speak it. Think in it.

That is what goes silent when a language dies. Not a vocabulary list. A mind.