"You Cannot Sound Like GPT": Signs of language discrimination and resistance in computer science publishing

TL;DR

This work examines how language background shapes scientific evaluation in a prominent computer science venue. Using nearly 80,000 ICLR peer reviews from 2018–2024 and 14 interviews, it shows a robust bias against manuscripts with substantial non-English-speaking author representation in clarity critiques and ratings, with only modest shifts after the introduction of ChatGPT. The authors frame reviewer judgments as indexical signs that link linguistic features to perceived quality, and they document extensive pre-publication language labor among multilingual researchers. While ChatGPT can mask some language cues, interview data reveal new signs of difference (including AI-generated writing styles) that still index author demographics and influence evaluation. The study highlights the risk that AI tools may reproduce or reconfigure language-based discrimination in scholarly publishing and argues for multilingual publishing and diverse language practices as corrective measures with practical policy implications.

Abstract

LLMs have been celebrated for their potential to help multilingual scientists publish their research. Rather than interpret LLMs as a solution, we hypothesize their adoption can be an indicator of existing linguistic exclusion in scientific writing. Using the case study of ICLR, an influential, international computer science conference, we examine how peer reviewers critique writing clarity. Analyzing almost 80,000 peer reviews, we find significant bias against authors associated with institutions in countries where English is less widely spoken. We see only a muted shift in the expression of this bias after the introduction of ChatGPT in late 2022. To investigate this unexpectedly minor change, we conduct interviews with 14 conference participants from across five continents. Peer reviewers describe associating certain features of writing with people of certain language backgrounds, and such groups in turn with the quality of scientific work. While ChatGPT masks some signs of language background, reviewers explain that they now use ChatGPT "style" and non-linguistic features as indicators of author demographics. Authors, aware of this development, described the ongoing need to remove features which could expose their "non-native" status to reviewers. Our findings offer insight into the role of ChatGPT in the reproduction of scholarly language ideologies which conflate producers of "good English" with producers of "good science."

Figures (5)

• Figure 1: ICLR has seen a rapid increase in submissions and author internationalization. The vertical line marks the release of ChatGPT on November 30, 2022.
• Figure 2: The number of authors from countries which would require a TOEFL score for admission at a U.S. university surpasses authors from TOEFL-exempt countries between 2022 and 2023.
• Figure 3: Sentences from reviews which our method infers are about clarity also make inferences about the identity of authors, echoing our findings in qualitative interviews.
• Figure 4: Shifts in reviewer scores on overall paper ratings, praise of writing clarity, and critiques of writing clarity before and after ChatGPT becomes available in 2022.
• Figure 5: Correlation heatmap of evaluative categories