Language technologies increasingly mediate everyday communication, but they often fail to adequately accommodate the diversity of real-world language use. This course, positioned at the intersection of sociolinguistics and speech technology, aims to uncover some of these technology biases. It focuses on how systems such as Automatic Speech Recognition (ASR), voice assistants, and large language models deal with linguistic variation in humans. Students will learn about such concepts as linguistic identity, variation, and accommodation in the context of conversational AI (e.g., Siri, Alexa, ChatGPT), as well as explore experimental methods (e.g., Wizard-of-Oz). Key theoretical pillars of the course are linguistic accommodation, salience, indexicality, and enregisterment. The course will include critical readings, hands-on analysis, and project work, in order to understand how social factors like gender, class, ethnicity, and region are reflected in the performance of language technology. Topics include fairness and bias in ASR, dialect representation in training corpora, dialect erasure, and communicative adaptation in human-machine interaction.
