Linguistic and dialectal bias in large language models: a review of evidence, mechanisms, and implications (2018–2025)

Eric Zhao

doi:10.54254/2753-7080/2026.31302

Abstract

Large Language Models (LLMs) increasingly mediate communication, evaluation, and decision-making across social and institutional domains. While prior work has documented various forms of bias in language technologies, linguistic variation has received comparatively less integrated treatment despite its deep sociolinguistic significance. This review synthesizes recent research on how LLMs reflect and reinforce linguistic bias toward nonstandard dialects, with a focus on African American English and other marginalized varieties. Drawing on studies spanning natural language understanding, reasoning, speech recognition, and conversational systems, we show that LLMs exhibit consistent performance disparities when processing dialectal inputs, even when semantic content is held constant. Beyond accuracy degradation, emerging evidence demonstrates that dialect functions as a social signal that triggers differential judgment, stereotyping, and covertly racist decision-making in high-stakes contexts. We organize the literature around four themes: dialect-based performance gaps, dialect-triggered social evaluation, measurement frameworks for linguistic bias, and proposed mitigation strategies. While recent benchmarks and adaptation methods improve robustness, we argue that many approaches treat dialect bias as a technical deficiency rather than a sociotechnical problem rooted in language ideology and power. By synthesizing research across technical, social, and evaluative dimensions, this review identifies methodological gaps and proposes directions for developing linguistically and culturally robust LLMs.

Keywords: linguistic bias , dialectal variation , large language models , raciolinguistic bias , AI fairness

Article Information

Received: 13 January 2026

Published: 13 January 2026

DOI: https://doi.org/10.54254/2753-7080/2026.31302

Article Type: Research Article

Cite this article

Zhao,E. (2026). Linguistic and dialectal bias in large language models: a review of evidence, mechanisms, and implications (2018–2025). Advances in Humanities Research,13(1),73-80.

Data availability

The datasets used and/or analyzed during the current study will be available from the authors upon reasonable request.

Disclaimer/Publisher's Note

The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of EWA Publishing and/or the editor(s). EWA Publishing and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

About this Volume

Volume Title: AHR Vol.13

Part of Series: Advances in Humanities Research

ISSN: 2753-7080 (Print) / 2753-7099 (Online)

Copyright & License

[1]Gallegos, I. O. (2024). Bias and fairness in large language models: A survey.Computational Linguistics, 50(3), 1097–1153. https: //doi.org/10.1162/tacl_a_00760

[2]Ziems, C., Chen, J., Harris, C., Anderson, J., & Yang, D. (2022). VALUE: Understanding dialect disparity in natural language understanding. In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (ACL 2022) (pp. 3701–3720). https: //doi.org/10.18653/v1/2022.acl-long.258

[3]Lin, F., et al. (2024). Assessing dialect fairness and robustness of large language models in reasoning tasks. arXiv. https: //arxiv.org/abs/2410.11005

[4]Markl, N. (2022). Language variation and algorithmic bias: Understanding algorithmic bias in British English automatic speech recognition. In Proceedings of the 2022 ACM Conference on Fairness, Accountability, and Transparency (FAccT 2022).

[5]Kantharuban, A., Vulić, I., & Korhonen, A. (2023). Quantifying the dialect gap and its correlates across language technologies. In Findings of the Association for Computational Linguistics: EMNLP 2023.

[6]Hofmann, V., et al. (2024). AI generates covertly racist decisions about people based on their dialect. Nature. https: //doi.org/10.1038/s41586-024-07856-5

[7]Fleisig, D., et al. (2024). Linguistic bias in ChatGPT. arXiv. https: //arxiv.org/abs/2406.08818

[8]Labov, W. (1972). Language in the inner city: Studies in the Black English vernacular. University of Pennsylvania Press.

[9]Rickford, J. R. (1999). African American Vernacular English: Features, evolution, educational implications. Blackwell.

[10]Lippi-Green, R. (2012). English with an accent: Language, ideology, and discrimination in the United States (2nd ed.). Routledge.

[11]Flores, N., & Rosa, J. (2015). Undoing appropriateness: Raciolinguistic ideologies and language diversity in education.Harvard Educational Review, 85(2), 149–171. https: //doi.org/10.17763/0017-8055.85.2.149

[12]Ziems, C., et al. (2023). Multi-VALUE: A framework for cross-dialectal English NLP. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (ACL 2023) (pp. 741–758). https: //doi.org/10.18653/v1/2023.acl-long.44

[13]Ziems, C., Held, W., & Yang, D. (2023). TADA: Task-agnostic dialect adapters for English. arXiv. https: //arxiv.org/abs/2305.16651

[14]Hovy, D., Yeh, Y.-T., et al. (2025). Culturally aware and adapted NLP: A taxonomy and a roadmap. Transactions of the Association for Computational Linguistics.