In July 2023, PLOS Biology published a survey with a blunt title: “The manifold costs of being a non-native English speaker in science.” Tatsuya Amano and colleagues asked 908 environmental scientists from eight countries, including Bangladesh, Japan, Nigeria and the United Kingdom, how much effort their everyday work took in English [1]. The answers added up to a heavy, measurable tax.

My view is simple. Science should be judged on its ideas and evidence, not on how smoothly the author writes a second language. English as the shared language of science is worth keeping, but its cost falls mostly on the people least able to absorb it, and journals and conferences can share that load. AI writing help can take a bite out of the problem. It will not make it go away.

What the survey measured

The survey ran from June to October 2021 and reached researchers in ecology, evolution, conservation and related fields who had already published at least one first-authored paper in English [1]. Countries were grouped by English proficiency and income, to help separate language from money.

Reading was the first cost. Among researchers who had published only one English paper, those from countries with moderate English proficiency reported spending a median 46.6% more time reading an English paper than native speakers, and those from low-proficiency countries 90.8% more [1]. The authors estimate that a PhD student from a low-proficiency country who reads 200 papers a year could spend up to 19.1 extra working days a year just on reading [1].

Writing showed a gap too, early in a career, though not at later career stages [1]. People do get faster. But the early years are when careers are made or lost.

Rejected for grammar, not for science

The number that bothered me most was about rejection. Among researchers who had published one English paper, about 38% of respondents from moderate-proficiency countries and 36% from low-proficiency countries said they had a paper rejected because of the English writing, against 14.4% of native speakers [1]. In revision, about 42.5% of non-native speakers in both groups said they were often, mostly or always asked to improve their English, against 3.4% of native speakers [1].

Money makes it worse. Respondents from lower-income countries tended not to use paid editing, probably for lack of funding, the authors suggest [1]. A 2020 study of Colombian researchers in biology put a price on it: among the editing and translation services it reviewed, the cost per article was between a quarter and half of a doctoral monthly salary in Colombia [2]. Of the 49 doctoral students and doctorates it surveyed, 43.5% reported a rejection or revision because of English grammar [2].

That gap has nothing to do with the science. It is a toll collected at the door.

The conference you never went to

Conferences are where young scientists meet collaborators and get noticed. In the Amano survey, about 30% of early-career respondents from Japan and Spain together said they often or always decided not to attend an English-language conference because of language, and about half often or always avoided giving a talk [1]. The Colombian survey found the same pattern: 33% chose not to attend international meetings because oral presentations had to be in English [2].

Nobody writes down the collaboration that never happened because someone stayed home.

Keep the scope honest

This evidence has limits. It is one field, environmental science, in eight countries. The answers are self-reported, and the authors list a relatively small sample, likely non-random recruitment, and the difficulty of estimating time as limitations [1]. They also say their numbers “may not be quantitatively applicable to all disciplines” [1]. The Colombian study is smaller still, and its extra writing hours were based on each person’s own sense of time [2].

Still, self-reported does not mean imaginary. Two surveys on different continents pointing the same way is reason to take the pattern seriously, even if the exact percentages would shift in physics or medicine.

The case for one language

There is a real argument on the other side, and both studies make it themselves. Amano and colleagues write that English as a common language “has no doubt contributed to the advance of science” [1]. Ramírez-Castañeda agrees that a common language is important for science communication, and argues for multilingual alternatives that keep a shared channel open [2]. A single language lets a lab in Nepal read a paper from Spain the day it appears. Splitting science across many languages would make work harder to find and check.

I accept that. The goal is to stop treating fluency as a hidden entry exam and leaving the whole cost on individuals. Amano and colleagues say the size of the disadvantage “seems far beyond the level that can be overcome with individuals’ efforts” [1]. They argue that journals, funders and conferences should provide language support and take these disadvantages into account when judging scientists’ work [1].

In practice, reviewers can be told to flag language only when it blocks understanding. Journals can offer editing support instead of a rejection letter. Conferences can accept prerecorded talks or written questions.

Where AI helps, and where it does not

AI writing tools are the obvious new option. Amano and colleagues suggest that tools such as ChatGPT and DeepL could cut the effort and cost of proofreading and translation, and that journals and universities should allow appropriate use of AI for English proofreading [1]. I agree. For a student who cannot afford an editing service, that is a real gain.

I build large language model tools that pull structured information out of clinical records, so what these models do well with text, and where they slip, is part of my work. A tool can polish grammar, but it cannot give someone confidence at a microphone, and it may quietly turn a careful claim into a stronger one.

There is also a new trap. A study in Patterns the same month found that seven AI detectors, on average, labeled more than half of 91 TOEFL essays by non-native English writers as AI-generated, while judging essays by US eighth-graders accurately [3]. When the researchers used ChatGPT to make the essays’ word choices sound more native, the false-positive rate dropped from 61.3% to 11.6% [3]. So a writer who leans on AI may be flagged less, and a writer who does not may be flagged more. I have written about detectors separately. The narrow point here is that AI help and AI suspicion now arrive together, and non-native writers carry both.

If you review papers, separate prose you could not follow from prose that simply does not sound like yours. If you run a meeting, ask what it would take for the quiet early-career researcher to give the talk instead of skipping the trip. These are small changes, and the 2023 survey suggests they would help many people now paying extra just to be heard.

References

[1] T. Amano, V. Ramírez-Castañeda, V. Berdejo-Espinola, I. Borokini, S. Chowdhury, M. Golivets, et al., “The manifold costs of being a non-native English speaker in science,” PLOS Biology, vol. 21, no. 7, Art. no. e3002184, Jul. 2023, doi: 10.1371/journal.pbio.3002184.

[2] V. Ramírez-Castañeda, “Disadvantages in preparing and publishing scientific papers caused by the dominance of the English language in science: The case of Colombian researchers in biological sciences,” PLOS ONE, vol. 15, no. 9, Art. no. e0238372, Sep. 2020, doi: 10.1371/journal.pone.0238372.

[3] W. Liang, M. Yuksekgonul, Y. Mao, E. Wu, and J. Zou, “GPT detectors are biased against non-native English writers,” Patterns, vol. 4, no. 7, Art. no. 100779, Jul. 2023, doi: 10.1016/j.patter.2023.100779.