Estimated peak number of languages: The study estimates that global language diversity peaked at roughly 20,000 to 75,000 languages around 2,000 years ago (roughly 1000 BCE to 1000 CE) . This would have been up to 10 times the approximately 7,500 languages spoken or signed today .

When the decline began: The decline started well before European colonialism. The "golden age" of linguistic diversity ended between roughly 3,000 and 1,000 years ago, driven primarily by the expansion of agriculture and larger, more centralized populations that absorbed or replaced smaller language groups .

Earlier Holocene baseline: At the beginning of the Holocene (~12,000 years ago), the researchers estimate only about 4,500 to 6,200 languages existed — fewer than today — because human populations were much smaller .

Current and projected loss: Today roughly 7,500 languages are spoken or signed . Previous research (not this specific Science paper but complementary work published in Nature Ecology & Evolution) projects that without intervention, language loss could triple within 40 years, with at least one language lost per month and over 1,500 languages potentially lost by the end of the century .

The role of AI in marginalizing smaller languages

The Science study itself focuses on historical reconstruction and does not directly model AI's impact. However, the wider research context cited in the reporting makes clear that AI and large language models (LLMs) are deepening the digital language divide:

• Fewer than 5% of the world's ~7,000 languages are meaningfully represented online .
• Over 6,000 languages remain "digitally disadvantaged" — unable to generate the training data needed for AI representation, creating a vicious cycle of marginalization .
• LLMs have been shown to be systematically "unfair" to less-spoken languages, conferring legitimacy on dominant language varieties while sidelining minority ones .
• The result is a digital language divide where AI tools amplify the dominance of English and a handful of other widely-spoken languages, further accelerating the pressure on smaller language communities .

Bottom line: The Science study reveals that humanity's linguistic diversity was once vastly richer than today — peaking at tens of thousands of languages around 2,000 years ago — and that the decline began with the rise of agriculture, not modern colonialism. Current extinction rates are severe, and AI systems risk accelerating marginalization by excluding most small languages from the digital ecosystem.

Estimated peak number of languages: The study estimates that global language diversity peaked at roughly 20,000 to 75,000 languages around 2,000 years ago (roughly 1000 BCE to 1000 CE) . This would have been up to 10 times the approximately 7,500 languages spoken or signed today .

When the decline began: The decline started well before European colonialism. The "golden age" of linguistic diversity ended between roughly 3,000 and 1,000 years ago, driven primarily by the expansion of agriculture and larger, more centralized populations that absorbed or replaced smaller language groups .

Earlier Holocene baseline: At the beginning of the Holocene (~12,000 years ago), the researchers estimate only about 4,500 to 6,200 languages existed — fewer than today — because human populations were much smaller .

Current and projected loss: Today roughly 7,500 languages are spoken or signed . Previous research (not this specific Science paper but complementary work published in Nature Ecology & Evolution) projects that without intervention, language loss could triple within 40 years, with at least one language lost per month and over 1,500 languages potentially lost by the end of the century .

The role of AI in marginalizing smaller languages

The Science study itself focuses on historical reconstruction and does not directly model AI's impact. However, the wider research context cited in the reporting makes clear that AI and large language models (LLMs) are deepening the digital language divide:

• Fewer than 5% of the world's ~7,000 languages are meaningfully represented online .
• Over 6,000 languages remain "digitally disadvantaged" — unable to generate the training data needed for AI representation, creating a vicious cycle of marginalization .
• LLMs have been shown to be systematically "unfair" to less-spoken languages, conferring legitimacy on dominant language varieties while sidelining minority ones .
• The result is a digital language divide where AI tools amplify the dominance of English and a handful of other widely-spoken languages, further accelerating the pressure on smaller language communities .

Bottom line: The Science study reveals that humanity's linguistic diversity was once vastly richer than today — peaking at tens of thousands of languages around 2,000 years ago — and that the decline began with the rise of agriculture, not modern colonialism. Current extinction rates are severe, and AI systems risk accelerating marginalization by excluding most small languages from the digital ecosystem.