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

2025-05-12

AI Summary

This study uncovers systemic linguistic bias against non-native English-speaking authors in ICLR peer review, wherein reviewers erroneously conflate linguistic features—such as syntactic complexity and lexical choice—with scientific quality, prompting authors to strategically “erase” non-native linguistic markers. Method: Drawing on nearly 80,000 review comments and in-depth interviews with 14 intercontinental participants, the study employs critical discourse analysis and sociolinguistic frameworks to examine bias mechanisms before and after the adoption of large language models (LLMs) like ChatGPT. Contribution/Results: Contrary to expectations, LLMs do not mitigate bias; instead, reviewers shift attention toward detecting “AI-generated writing styles” and non-linguistic proxies—including citation patterns and authorship ordering—to infer linguistic background. This reveals deeply entrenched academic language ideologies. Critically, this is the first empirical demonstration that linguistic bias exhibits structural geographic patterning and undergoes strategic reconfiguration—not attenuation—following AI integration.

Technology Category

Natural Language Processing: Ethics — Bias, Fairness, Transparency & PrivacyMachine Learning: Ethics, Bias, and FairnessComputer Vision: Language and Vision

Application Category

Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web searchUser Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendationSocial Networks and Social Media: Fairness and bias in social network and social media analysis

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."

Problem

• Detecting linguistic bias in peer reviews against non-native English authors
• Examining ChatGPT's role in masking versus reinforcing language discrimination
• Exploring how reviewers use writing style to infer author demographics

Innovation

• Analyzed 80,000 peer reviews for linguistic bias
• Interviewed 14 global conference participants
• Examined ChatGPT's role in language discrimination