Governance Node Seminar - Nina Markl, University of Edinburgh

In this talk I will present some of my PhD research on algorithmic bias in automatic speech recognition (ASR). I show how performance disparities for different varieties of English relate to language variation and pre-existing language ideologies and linguistic discrimination. I also discuss how potential harms of ASR can be mitigated during dataset and system design by considering the wider (sociolinguistic) context the system is embedded in.