You might assume that tools like ChatGPT will level the playing field for multilingual scholars. But a new study of reviews from a leading international scientific conference in AI (ICLR)—nearly 80,000 peer reviews in total—suggests otherwise.
The study
- Analysed reviews from 2018–2024, focusing on comments about “clarity” in writing.
- Compared author regions (using institutional data) & reviewer evaluations before/after ChatGPT’s release in late 2022.
- Interviewed 14 multilingual researchers across five continents to understand lived experiences.
The findings
- Papers with more multilingual authors consistently received lower scores, more criticism of clarity & fewer compliments.
- After ChatGPT, there was only a slight reduction in bias—statistically significant only for authors in China.
- Reviewers described using “signs” (e.g. grammar errors, long sentences, certain phrases) to infer authors’ language background, which then influenced perceptions of scientific quality.
- With ChatGPT, these signals changed but didn’t vanish: reviewers instead picked up on “AI style” features (words like delve or overly generic phrasing) & linked them to “non-native” authors.
Why it matters
This echoes wider research on raciolinguistic ideologies (Flores & Rosa, 2015): even when students (or scientists) adapt their language, bias persists because listeners/readers interpret identity through new cues. In other words: fix the grammar & people will still find other “signs” of difference.
Example: A missing plural -s might be taken as evidence of a Chinese author. With ChatGPT, a polished but slightly generic phrase (“this paper delves into…”) now risks the same judgement. The science itself becomes entangled with assumptions about who is writing.
Teacher Takeaways?
- Remind students that “clarity” is often a social judgement, not just a linguistic one.
- Explore whose English counts as “good enough” & why. Discussions about accent, grammar, & register can highlight hidden biases.
- Encourage critical reading: when students critique a text, do they judge the ideas or the language first?
If bias shifts but doesn’t disappear, what does that mean for how we prepare learners to publish, present & participate in global English-medium fields?
Leave a Reply