The profound impact of Large Language Models (LLMs) extends across various aspects of human life, including its implications for scientific publications. This study seeks to investigate its influence on scientific publications authored by individuals who are not native-English speakers. Using the release of ChatGPT 3.5 in November 2022 as a natural experiment, we present an analysis of paper metadata sourced from the Social Science Research Network (SSRN) spanning seven distinct subject areas. Our findings indicate a significant increase in the use of AI in paper writing following the introduction of ChatGPT 3.5. Then, we employ difference-in-differences (DID) to discover that, EFL(English-as-a-Foreign-Language)-authored papers experienced a more noticeable improvement in language proficiency after the advent of ChatGPT 3.5, compared to non-EFL authored papers. This convergence in language proficiency suggests that LLMs play a beneficial role in reducing the language barrier in academia. This effect is present across various disciplines and is particularly pronounced in the Social Science discipline. This paper contributes to our understanding of AI's augmentation of human capabilities, and provides a way to mitigate the academic language barrier.
