Perception · 2026-09-15 · 1:42

Writing without borders: one million abstracts, one English

Name: Writing without borders: one million abstracts, one English
Uploaded: 2026-09-15
Duration: 1 min 42 s
Description: what converged is the surface: English that reads alike everywhere. Whether the ideas underneath converge too is the question worth watching.

Over one million social-science abstracts (2012 to 2024): writing complexity rose worldwide, non-native English countries converged on the leaders, and by the study's metric China passed the United States.

In favour

Large, open-access evidence that AI writing tools lower the language barrier outside the Anglosphere.
Twelve years of data, before and after ChatGPT, not a single snapshot.

Worth watching

"Quality" means readability formulas (Flesch-Kincaid, Fog, SMOG), not argument or rigour.
One index, three publishers; the LLM signal is a vocabulary proxy, not proof of use.

Our takewhat converged is the surface: English that reads alike everywhere. Whether the ideas underneath converge too is the question worth watching.

Source: Prakash κ.ά., Humanities & Social Sciences Communications, 12:1058 (2025)

← All Focus posts

AI-narrated · AI-generated content

Transcript

Is research writing, all over the world, starting to sound like one English? A very large study of academic abstracts went looking for the answer. It read over one million social-science abstracts, from 2012 to 2024, and tracked how complex the writing was, country by country. The finding has three parts. Writing complexity rose worldwide. Non-native English countries converged on the leaders. And by the study's metric, China passed the United States. Why it matters. This is large, open-access evidence that AI writing tools lower the language barrier outside the English-speaking world. Not an anecdote, a measured pattern. And it covers twelve years of data, before and after the chatbot boom. Not a single snapshot, but a line you can follow over time. Two things to keep straight. Here, quality means readability formulas. They score how easy the text is to read, not whether the argument holds or the work is rigorous. And the data is one index, three publishers. The language-model signal is a vocabulary proxy. A pattern in the words, not proof that anyone actually used a model. Our take. What converged is the surface: English that reads alike everywhere. Whether the ideas underneath converge too is the question worth watching.