Humanity Spoke Tens Of Thousands Of Languages Two Millennia Ago. We’re Down To 7,500

Count the languages alive right now and you get to roughly 7,500, signed and spoken, and about four of them will fall silent this year. That number feels like a floor. Something battered, reduced, the remnant of a richer world that agriculture and then empire and then Europe ground away at for twelve millennia. Turns out the arithmetic runs the other way at the start, and the drop, when it came, was steeper and older than almost anyone had reckoned.

A team working across Spain, the United States and Germany has published in Science a set of modeled trajectories for global language diversity stretching back 12,000 years. Their headline figure: a peak of tens of thousands of languages somewhere between 3,000 and 1,000 years ago, an order of magnitude above today.

The problem they were tackling is not a shortage of interest. It’s that language leaves almost nothing behind. Writing shows up around 6,000 years ago and even then documents a vanishingly thin slice of what people were actually saying; systematic language cataloging is a project of the last century or so. Everything before that is inference. Marcus Hamilton, at the University of Texas at San Antonio, notes that progress has been limited “not by a lack of theoretical interest but by the exceptional difficulty of observing the relevant dynamics empirically”.

So Damian Blasi, Hamilton, Russell Gray and Claire Bowern built a workaround out of demography. If you can’t count languages, count people, and constrain how many people a language can plausibly hold.

The chain runs like this. Start with ethnographic records of 171 hunter-gatherer and fisher societies whose subsistence and mobility hadn’t been wrecked by contact with farmers. Those groups cluster around a median of about 800 individuals, distributed roughly log-normally, and the ceiling is set by foraging itself: energetic demand rises with group size while the distance you can walk to find food does not. Feed that through a Bayesian model, with controls for regional nonindependence and lumpy sampling, and you get a distribution of plausible early-Holocene group sizes.

Fewer Languages Before The Farms

Then divide. The global population 12,000 years ago sat somewhere between 4.4 and seven million. Fit groups of that size into that population, assume in the early Holocene one ethnolinguistic group generally meant one language (a fair assumption then, less so later, when populations swell past the point where everyone knows everyone), and you land on about 4,500 to 6,200 languages. The wider plausible band runs 3,300 to 7,800.

Which is fewer than today. That result knocks out two of the three standing hypotheses in one go: the Megadiverse Paleolithic idea, which had 12,000 or more languages before farming and steady decline ever since, and the Premodern Equilibrium view, in which numbers held roughly flat until Europeans started sailing. Only the Demographic Boost hypothesis survives contact, and only at its upper edge.

Gray, who directs linguistic and cultural evolution at the Max Planck Institute for Evolutionary Anthropology in Leipzig, said the first surprise was that the pre-agricultural world was probably not exceptionally rich in languages, with most likely fewer than exist now, and that diversity then grew alongside the human population for thousands of years.

Grew, and kept growing, through the domestications and the migrations and the first cities. The models let the number of people per language climb over time while independent reconstructions of global population climb faster, and the product of those two curves is the language count. Thousands of trajectories, all of them required only to be monotonic in the ratio, and a functional principal components analysis showed three components mopping up over 96 percent of the variance. The shape they share: a long rise, a peak between 3000 and 1000 years ago at roughly ten times early-Holocene levels, then a fall.

The Fall Started Before The Ships

That fall is the uncomfortable part. It began, on these numbers, during the era of large states and early empires, not with the colonial expansions of the past 500 years. Bowern, at Yale, called it the most unexpected result, that peak diversity may have come “not in the remote Palaeolithic, but during the era of expanding states and early empires”. The rate implied over the last two millennia is worse, faster, than the gloomiest current projections for the year 2100.

Two caveats, and the authors raise both themselves. This is a coarse global reconstruction, not a census; it rests on assumptions about early group sizes and on a monotonic ratio that smooths over every plague, famine and war (the Black Death took perhaps 40% of Europe and the model simply cannot see it). And it is silent on mechanism. It says a vast reduction happened. It does not say which empire, which century, which valley.

What it does say is that the languages we have are not a random sample of the languages there were. Blasi, who led the work from ICREA and Harvard, put it this way: “A linguistic feature may be common not because it is inherently superior, but because it happened to be carried by populations that expanded.” Linguists have long explained the global frequency of some grammatical structure or some vowel by appealing to learnability or communicative efficiency or cognitive fit. That explanation now has a rival that requires no functional story at all, just demography and luck.

And languages travel with other things. Kinship systems, marriage rules, religious practice, ways of organizing a settlement. If the linguistic bottleneck was as tight as these models suggest, a comparable mass extinction of cultural variants ran alongside it, mostly unrecorded, since the records that survive were kept by the winners.

Frequently Asked Questions