Abstract
This paper examines how automated speech systems such as voice assistants, AI transcription tools, and smart devices systematically favor standardized accents—particularly General American and Received Pronunciation—while marginalizing speakers with regional, global, or non-native English accents. By analyzing the technical architectures behind speech recognition, the paper uncovers how data-driven models reinforce linguistic hierarchies and diminish the intelligibility of diverse speech patterns. The study also addresses how such bias shapes users’ linguistic behaviors, promotes assimilationist language ideologies, and widens the digital divide for multilingual communities. Ultimately, the paper calls for inclusive AI design that values linguistic diversity as an asset rather than an obstacle.
References
1. Koenecke, A., Nam, A., Lake, E., Nudell, J., Quartey, M., Mengesha, Z., & Goel, S. (2020). Racial disparities in automated speech recognition. PNAS, 117(14), 7684–7689.
2. Tatman, R. (2017). Gender and dialect bias in YouTube’s automatic captions. Ethics in NLP Workshop, ACL.
3. Blodgett, S. L., Green, L., & O’Connor, B. (2016). Demographic dialectal variation in social media. arXiv:1608.08868.
4. Herring, S. C. (2019). Linguistic Diversity in the Digital Age: Challenges and Opportunities. Language Resources and Evaluation, 53(2), 239–256.
5. Eubanks, V. (2018). Automating Inequality: How High-Tech Tools Profile, Police, and Punish the Poor. St. Martin’s Press.
6. Mozilla Foundation. (2023). Common Voice Project. Retrieved from https://commonvoice.mozilla.org
7. UNESCO. (2022). World Languages Report: Digital Inclusion of Linguistic Minorities. Paris: UNESCO.
8. Pennycook, A. (2010). Language as a Local Practice. Routledge.
9. Zuboff, S. (2019). The Age of Surveillance Capitalism. PublicAffairs.
10. Bender, E. M., & Friedman, B. (2018). Data statements for NLP: Towards mitigating system bias and enabling better science. Transactions of the ACL, 6, 587–604.