Linguistic technologies and the “bad speakers”

This article explores the relationship between language technologies (such as automatic translation, speech recognition, and language corpora) and the social distinction between “normative speakers” and so-called “bad speakers”. It argues that when these technologies are built primarily on data from “standard” or “correct” language users, they reproduce and reinforce linguistic bias. As a result, non-standard accents, speech disorders, and marginalized varieties are excluded, leading to a form of epistemic injustice in which certain speakers are rendered invisible or less legitimate. The article defends the need to design more inclusive language technologies by incorporating real linguistic diversity, including non-normative forms of speech, in order to avoid discrimination and promote fairness.