Consider this: every language you have ever heard, every tongue spoken by every person alive today, represents a rounding error. Of the roughly 7,500 languages spoken or signed across the planet right now, almost all are the survivors of a vast, quiet extinction event that has been underway for millennia. A study published July 23, 2026 in Science (DOI: 10.1126/science.adx4343) reveals that linguistic diversity peaked between 1,000 and 3,000 years ago at between 20,000 and 75,000 languages. That is up to ten times the number we have today.
The finding overturns two long-held assumptions. The first is that language diversity was highest before agriculture, in the deep prehistoric past. The second is that the current extinction crisis began with European colonialism roughly 500 years ago. Both are wrong. The decline started far earlier, driven by agricultural intensification and the rise of centralized nation-states, especially the Roman and Greek empires. By the time the first colonial ships crossed the Atlantic, humanity had already lost the majority of its linguistic heritage.
But the paper’s most haunting implication is not about what was lost in bulk. It is about what was lost in kind.
Modeling the unspeakable past
Led by Claire Bowern, a linguist at Yale University, and D.E. Blasi and an international team of collaborators, the study began with a clever premise. Since there is no fossil record of spoken language, the researchers used a proxy: more than 170 contemporary hunting-and-gathering societies around the world. In these groups, each community speaks its own language, and group sizes have remained remarkably stable over time, typically ranging from a few hundred to just over 1,000 people. If this pattern held across prehistory, it means the number of languages at any given moment was tied directly to the number of independent human groups.
The team modeled the relationship using known endpoints. Twelve thousand years ago, before agriculture took hold, the global population was roughly 6 million people and the model estimates about 5,000 languages. Today, 8 billion people speak about 7,500 languages. Between these two points, the researchers simulated multiple population scenarios, factoring in disease events, migration patterns, and the time it takes for a language to become distinct after a population splits. That split rate is surprisingly consistent: 500 to 1,000 years.
Under every scenario, the peak came between 1,000 and 3,000 years ago. “That was the golden age of linguistic diversity,” Bowern said. The planet held somewhere between 20,000 and 75,000 distinct languages. Today we have roughly 7,500.
What caused the fall
The study identifies the turning point not with colonialism, but with the intensification of agriculture and the emergence of large, centralized states. As the Roman and Greek empires expanded, smaller language communities were absorbed, displaced, or assimilated. The same dynamic repeated across China, Mesoamerica, and the Andes. A language does not need to be actively suppressed to disappear. It can simply become less useful as trade networks grow, as roads connect villages to cities, as a single administrative language becomes the key to economic survival.
This timeline matters because it challenges the narrative that the current extinction crisis is a modern problem. Roughly half of today’s surviving languages are endangered. About four vanish every year. But the peak was thousands of years ago, which means that almost everything we think we know about the full range of human language is built on a fraction of the original data.
The hidden library
Chiara Barbieri, a geneticist at the University of Cagliari who contributed to the study, captured the vertigo of this realization in a single sentence: “Are there people who developed languages with some traits that we can barely imagine now? It makes your head spin.”
The spin is earned. Every living language belongs to a known family, follows discoverable rules, and operates within an observable range of grammatical and phonological possibilities. But the 60,000 or more languages that no longer exist were not simply more of the same. They may have contained structures, categories, and cognitive architectures that have no parallel in anything we can study today.
Consider what existing languages already teach us about the outer edges of human cognition. Some Australian Aboriginal languages use complex spatial systems that require speakers to keep absolute cardinal directions in mind at all times, reshaping how they think about time, memory, and sequence. The Piraha language of the Amazon has no words for numbers and no relative tenses, challenging universalist theories of grammar. The Tuyuca language of the Amazon uses up to 140 grammatical evidentiality markers, forcing speakers to specify the source of every piece of information they convey. The Hadza language of Tanzania uses click consonants that appear in no other African language family, representing a distinct phonetic lineage that may reach back tens of thousands of years.
Each of these represents a single surviving data point from a vanished forest. Imagine the range when the forest was ten times larger. Languages could have developed grammatical systems based on categories we no longer possess: ways of encoding kinship that blur the line between self and group, tense systems that treat future and past as the same direction, noun classification systems based on sound symbolism or sensory qualities rather than gender or animacy. There may have been languages without nouns, or languages where word order encoded emotional valence rather than grammatical role. There may have been click-based languages far more elaborate than anything in southern Africa, with dozens of click types used to distinguish lexical meaning from grammatical function simultaneously.
What the loss means for cognition
This is not idle speculation. The strong implication of the Bowern study is that the surviving languages represent a radically impoverished sample. If the full dataset of human language once held 60,000 to 70,000 more entries than it does today, and if language shapes thought in the ways that decades of cognitive science have demonstrated, then the cognitive range of humanity has narrowed in parallel with the linguistic one.
The human brain did not change substantially in the last 3,000 years. The same neural architecture that parses Mandarin, Swahili, or Navajo today parsed those lost languages in the past. But the software it ran changed dramatically. The Bowern team’s model suggests that the narrowing happened fastest in the regions where agriculture and empire struck first: the Fertile Crescent, the Mediterranean, East Asia. By the time writing systems emerged roughly 6,000 years ago, they captured only a tiny slice of the linguistic world, and only for languages that already belonged to dominant groups. Everything else vanished without a trace.
The work ahead
The study does not merely revise the timeline of language extinction. It reclassifies the loss from a recent tragedy to a long-term pattern, and it raises the stakes for documentation efforts. If the current 7,500 languages are the last survivors of a far more diverse whole, then every language that disappears today is not one lost out of 7,500. It is one lost out of a heritage that once numbered in the tens of thousands. Each extinction closes the door on a unique solution to the problem of translating thought into sound, a solution that took hundreds or thousands of years to evolve and that can never be reconstructed.
Bowern’s team plans to extend the model to regional scales to understand how diversity collapsed differently in different parts of the world. But the central insight is already clear: the golden age of language was not a distant, unreachable past. It was yesterday, in evolutionary time. And we are living through the hangover.
The question the study leaves us with is not how many languages we have lost. It is what those languages could do that we will never know. As Barbieri put it, the possibilities make your head spin. The tragedy is that we no longer have a way to stop spinning.