“You Cannot Sound like Gpt”: How Multilingual Scholars Navigate Cutting Edge Language Ideologies in Scientific Publishing

Indexed computer science research is published almost entirely in English, limiting the participation of readers and writers from most of the world's population. ChatGPT has thus been celebrated as a tool for scholars whose first language is not English to instantly adapt their writing in preparation for English-language publication. Using the case study of ICLR, an influential, international computer science conference, we look to sites of ChatGPT adoption to understand the circumstances and mechanisms of linguistic exclusion in scientific publishing. With computational methods, we analyze the conference's over 80,000 peer reviews over 2018-2024, comparing the prevalence of writing critique by author region before and after the introduction of ChatGPT in late 2022. We supplement our findings with interviews of multilingual authors and reviewers from across five continents to understand their experiences with the (non)use of ChatGPT in scientific writing.

We find the tool to already be entangled in the evaluation of scientific knowledge. Prior to ChatGPT, reviewers described using certain features of writing to infer that an author's first language is not English, citing intelligibility as a reason for critique. After ChatGPT, interviewees describe how these indicator features shift. Certain words or flowery language they associate with the tool- despite syntactically and morphologically matching the norms of the dominant linguistic group- may now come to signify the authors' “non-native," and therefore "untrustworthy" identity. This innovative way of expressing and experiencing language ideologies then becomes a mechanism for continued social and cultural sorting into hierarchies of scientific worth. This use-case of ChatGPT offers a unique lens into the reproduction of scholarly language ideologies which conflate producers of “good English” with producers of “good science.”