News

Linguistic 'golden age' peaked 1,000–3,000 years ago, then empires crushed it

A new study in Science reconstructs 12,000 years of language diversity, finding a peak tens of thousands of languages strong—and a decline driven by states and empires long before colonialism.

August 1, 2026· 2 min read· Source: Yale News
Linguistic 'golden age' peaked 1,000–3,000 years ago, then empires crushed it

A team of linguists, demographers, and evolutionary biologists has reconstructed global language diversity over the past 12,000 years—and the picture is not what most textbooks suggest. The study, published July 23 in Science, estimates that the world's languages peaked between 1,000 and 3,000 years ago, when tens of thousands of distinct languages were spoken. That 'golden age' was followed by a steep decline driven by the rise of states and multinational empires—not, as commonly assumed, by European colonialism starting ~500 years ago.

Using ethnographic data from 171 hunter-gatherer and fisher societies, combined with prehistoric population estimates (4.4–7 million people globally 12,000 years ago), the researchers modeled thousands of possible trajectories linking early-Holocene language counts to the present. Their models consistently show a peak in diversity during that window, followed by a collapse as expanding states spread their languages, cultures, and pathogens to previously independent populations.

'The world immediately before agriculture was probably not exceptionally rich in languages,' said lead author Damián Blasi of Pompeu Fabra University. 'There were most likely fewer languages than there are today. Linguistic diversity then grew alongside the human population for thousands of years.'

The study's key implication: the ~7,600 languages spoken or signed today are a historically biased survivor sample. 'A linguistic feature may be common not because it is inherently efficient or better for communication, but because it happened to be carried by populations that expanded,' said coauthor Russell Gray of the Max Planck Institute. 'Extinction may have shaped linguistic diversity much more profoundly than we previously appreciated.'

For engineers building language models and NLP tools, this is a reminder that the data we train on is not a neutral sample—it's the product of millennia of political and demographic filtering. The languages that survive are the ones that hitched their wagons to empires, not necessarily the ones that are 'best' in any linguistic sense.

A linguistic feature may be common not because it is inherently efficient or better for communication, but because it happened to be carried by populations that expanded.
Manul X Editorial