Those who write about science pay the price of thinking in another language; AI could change everything

For almost forty years, I have been writing about science in a language that is not my own. Every article, every funding application, every reply to a reviewer, every letter to an editor, is first conceived in Italian and then translated, sentence by sentence, into English – a language I have a good command of but which I do not feel as at home in as I do in my own language. This is not a complaint about English as the common language of science: a shared language is, on the whole, a good thing, and I do not believe that European research would be better served by a return to national languages. It is a more specific observation, one that very few non-native English-speaking researchers in medicine have ever expressed with such clarity: writing about science in English does not merely mean it takes longer. It changes what is argued, how forcefully it is argued, and sometimes whether an idea is put on the table at all or not.

This is not a subjective impression. A 2023 survey of 908 environmental scientists across eight countries found that non-native English-speaking researchers face a systematic disadvantage at every stage of scientific work — reading, writing, publishing, presenting at conferences, participation in scientific debate — with the effort required to prepare an article estimated to be several times greater than that of a native speaker, and a much higher likelihood of having a paper rejected on purely linguistic grounds, or of deciding not to speak at a conference due to language-related anxiety.

Another analysis, also from 2023, estimated that non-native-speaking researchers spend around two and a half weeks a year on writing alone — time taken away from actual research. These are not minor obstacles. They constitute a structural burden on the participation of the majority of the world’s scientists — including a significant proportion of Europe’s own scientific workforce — in the language that determines whether their work is read or not.

The debate on language and science has focused almost entirely on one aspect of the issue: access to scientific literature. Machine translation has made real progress in this area, and this should be acknowledged. But this approach treats non-native-speaking researchers as consumers of science who need help to understand it, rather than as authors of science who need to be understood with equal clarity when putting forward an argument. These are two different problems, and the second is the more costly one. A clinical researcher who is able to read a translated article but cannot, in English and in real time, construct an equally incisive counter-response, an equally persuasive project narrative, or an equally confident reply to a sceptical reviewer, is not competing on a level playing field — regardless of how sound the underlying science may be.

Conversational artificial intelligence changes this imbalance in a way that machine translation alone had never done. Used effectively, it allows a researcher to think and argue first in their own language — building the logic, weighing up the evidence, honing the counter-argument — and then produce an English version that carries the same persuasive force, rather than a flat, approximate rendering. I have done this myself: constructing the structure of an argument in Italian, and checking that the English version conveyed the same angle. The difference compared to translating a paragraph that has already been written is not merely cosmetic: it is the difference between exporting a thought and bringing that thought to life, without weakening it, in both languages right from the start.