U.S. Southerners’ Attitudes Towards Algorithmic Analysis of Voice Data for High-Stakes Employment and Education Evaluations

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

The rise of algorithmic decision-making in workplaces and schools has drawn attention to potential benefits and harms for stakeholders. Additionally, uneven performance of language technologies for non-standardized or underrepresented dialects has resulted in calls to engage speakers of those dialects. Our survey study explores the attitudes of 111 American English speakers from the southern U.S. towards voice-specific applications of algorithmic analysis in four high-stakes decision-making use cases: evaluation of candidates in hiring and college admissions interviews, employee performance evaluations based on presentations, and grading of students’ pronunciation on English final exams. We elicited descriptions of potential benefits and harms with prompted consideration for dialect and speech and categorized responses inductively as well as deductively, based on existing concepts of sociotechnical harms of algorithmic systems. We contribute focused insights into an understudied but critical area: the performance of language technologies for low-resource dialects. While participants acknowledged the relevance of metrics such as pronunciation and communication skills to employers and schools, many conveyed concerns about evaluations that could negatively evaluate their voices and result in harms such as economic loss. Many participants expressed a preference for human involvement in evaluations. Building on our insights, we make recommendations for researchers and policymakers to mitigate sociotechnical harms of algorithmic evaluations that may affect various linguistic and speech communities.