Biography
Biography pending.
Conference contributions 1
LINGUISTIC BIAS IN AI SPEECH AND TEXT PROCESSING:
A literature review on who AI works best for
As AI language tools become central to everyday communication, a critical question emerges: do these systems work equally well for all users? This paper presents a systematic literature review arguing that current AI speech recognition and text processing technologies are not linguistically neutral. They systematically favor certain speaker profiles while disadvantaging others. Drawing on sociolinguistics, psycholinguistics, and human-AI interaction research, this study examines the key dimensions along which AI performance varies: nativeness, age, gender, dialect, language resource availability, and socioeconomic background. Studies consistently show that AI tools trained predominantly on data from young, educated, native speakers of high-resource languages underperform for users outside this profile. Non-native speakers, elderly users, dialect speakers, and speakers of low-resource languages such as Arabic and Amazigh encounter higher error rates, miscommunication, and unequal access to AI-mediated communication. This review synthesizes existing findings across these dimensions, identifies gaps in current research, and calls for a more inclusive approach to AI language tool development. The findings have direct implications for developers, educators, and policy makers advocating for linguistic justice in AI design.