Talks/ Teaching/ Collaborations

Thanks to inclusion of large language datasets and advances in deep learning and natural language processing, voice-enabled technologies have recently dramatically improved (Jurafsky and Martin 2025). Consequently, voice AI is transforming our communicative routines in various domains of social life. At the same time, studies demonstrate that automatic speech recognition (ASR), which sits at the core of these technologies, still exhibits ethnoracial (e.g. Koenecke et al. 2020), gender or regional dialect biases. In this talk, I focus on Automatic Speech Recognition (ASR), to unpack why voice AI still listens with an accent, what social consequences this entails, for whom and what can be done to make it more inclusive.

To do so, I first review how ASR works and what factors have been reported to affect its performance. I assess how well ASR can handle the many ways people speak, both across different groups and within individuals. By focussing on selected case studies, I examine how what we assume about a language and its speakers affects the design of ASR systems. I argue that for ASR to become more ethical, we must tackle its built-in accent biases, acknowledging that imposing ‘an accent upon another is violent’ (Lawrence 2021), helps encode aural discrimination and may disproportionately affect those at the margins. Finally, I discuss selected practices that oppose ASR discrimination and carve a way towards voice AI where humanity and multiplicity of each voice is recognised. 

References:

Jurafsky, D. and J. Martin (2025) Speech and Language Processing: An Introduction to Natural Language Processing, Computational Linguistics, and Speech Recognition with Language Models, 3rd edition. Online manuscript released January 12, 2025. https://web.stanford.edu/~jurafsky/slp3.

Koenecke A, Nam A, Lake E, et al. (2020) Racial disparities in automated speech recognition. PNAS 117(14): 7684–7689. 

Lawrence, H. (2021) “Siri disciplines.” In Your Computer Is On Fire, Thomas S. Mullaney, Benjamin Peters, Mar Hicks, and Kavita Philip (Eds.). MIT Press, 179– 198. 

In this talk, Dr. Kozminska examines the current state of AI-powered speech recognition and language processing technologies. She asks:

 1. How do these technologies actually work?

2. Who benefits most from these tools, and who might be left behind?

3. Why does some people's speech get recognized more accurately than others? 

This talk explores how what we assume about a language and its speakers affects the AI tools we build. Dr. Kozminska will assess how well AIs can handle the many ways people speak, both across different groups and within individuals. Ultimately, she claims, for AI to become more ethical and inclusive, we must tackle the built-in language biases in current AI. 

Drawing on her recently published book, Soundings and the Politics of Sociolinguistic Listening for Transnational Space Dr Koźmińska explores how the focus on the practices of sounding and listening helps better understand the role of language in contemporary vernacular-cosmopolitan transformations.

The talk will examine how the act of bringing different logics and rhythms of seemingly similarly positioned transnational actors with links to Poland, but residing in the UK in relation with one another, informs how semiotics of the voice may work today. It will also examine how sensorimotor capacities are entangled in bodily experiences and how they are embedded in their sociocultural and technological contexts. The talk adds to the debates on how the use of communicative resources operating in multiple and co-present frames influences the changing ideals of linguistic legitimacy.

Dr Koźmińska begins by foregrounding the need to connect research on scale in migration contexts with studies of embodied soundwork and of stance in semiotics. She argues that such a focus helps unpack how language both perpetuates and, through emerging alternatives, challenges systems of exclusion and inclusion. 

Her talk will then discuss the situated production of perspectives on norms and practices, highlighting contradictory tendencies and processes through which the change in location changes both the transnational actor’s perspective and units of comparison and discuss what dimensions and contexts are relevant to whom, and how these fit into the ways in which larger stories are enacted and explained. This in turns helps reveal how we occupy and push the limits of normative structures in the context of changing research possibilities, positionalities, economies, ecologies of media, and aesthetics.

Dr Koźmińska proposes to see sociolinguistic listening as curatorship, potentially enabling healing of multiple publics brought in dialogue in the times of increased politicization of movement. 

Birkbeck, University of London:

Sociolinguistics (BA, MA): 2021-ongoing; Analysing Language Structure and Use (BA): 2019-ongoing; Contemporary and Global English (BA): 2025-ongoing; Language, AI and social justice (MA): new module to begin in 2025/26; Dissertation (MA, team-taught): 2021-ongoing; Research Methods and Design (MA, team-taught): 2021-ongoing; Linguistics Final Year Project (BA): 2021-ongoing; Independent Literature Review (MA, team-taught): 2021-2025; Empirical Research Skills Training Workshop (MA, team-taught): 2021-ongoing; Investigating Language (BA, team-taught): 2021-ongoing; Reading Transnational Cultures (BA, team-taught): 2022-ongoing; Understanding the City (MA Urban Studies, team-taught): 2022/23; Representations of Gender, Sexuality and Love (BA, team-taught): 2022/23; MA sessions in intersectional sociolinguistics for Introduction to Applied Linguistics: 2021-ongoing.

University of Oxford:

Prelims: Sociolinguistics (lectures) – 2014-16, 2019-21.

Tutorials for freshmen: Sociolinguistics – 2014-16, 2019-21.

FHS Paper XII: Sociolinguistics (lectures) for Undergraduates and Graduates – 2014-2016, 2019-21.

Advanced tutorials for finalists: Sociolinguistics – 2014-2016, 2019-21.

Option paper: Sociolinguistics (Seminar sessions for MPhil and MSt students) – 2019-21.

Yearly tutorials for freshmen: Phonetics and Phonology – 2019-21.

Other teaching:

University of Brighton, January-July 2017:

Researching English language and use: seminar sessions for advanced undergraduate students of linguistics and English.

Language and media: seminar sessions for advanced undergraduate students of linguistics and English.

Text design: lectures for undergraduate and graduate students of English language and linguistics.

London Interdisciplinary Social Science Doctoral Training Partnership, QMUL, KCL, October 2017 – May 2018: Introduction to Qualitative Research: seminar sessions for PhD students from across social sciences.

Aarhus School of Architecture, Denmark, February 2019: Introduction to Anthropology: workshops with students of architecture.

Universität Greifswald, Germany, August 2018: Language and migration: seminar sessions for undergraduate and graduate students of linguistics, history and cultural studies.