- Convenors:
-
Zahra Mughis
(Lahore School of Economics)
Ahmad Nawaz (Lahore School of Economics)
Send message to Convenors
- Format:
- Paper panel
- Stream:
- Digital futures: AI, data & platform governance
Short Abstract
Language serves as a socio-digital and developmental infrastructure that reflects and reinforces existing inequalities. When combined with AI-enabled systems, it creates new forms of social and epistemic injustices that shape human agency and everyday lived realities.
Description
Language has long served as a crucial axis of inclusion and its intersection with technology is far from neutral. This panel explores how language-based socio-digital inequalities shape access to services, markets, and the state, when tools, algorithms, and interfaces misread, misrecognize, or erode the languages of those most in need of its potential benefits.
From access to participation, language support determines who can benefit from the economic, socio-cultural, civic, and personal affordance of digital technologies. However, chatbots, translation tools, recommendation systems, and automated governance heavily rely on language as a computational capital that is unevenly distributed among individuals, communities, and countries. Limited linguistic diversity and knowledge coverage and embedded biases in training data feed language-based marginalization into AI systems and reinforce them in use by limiting agency in who can engage with these systems, interpret and contest their processes and outputs.
Foregrounding language as a socio-digital and developmental infrastructure, this panel invites interdisciplinary contributions unpacking how linguistic digital divides intersect with economic means, skills, and other structural variables to deepen opportunity gaps and create new forms of social and epistemic injustice. Conceptual and empirical work exploring the intersections of language, artificial intelligence, and socio-digital inequalities, and their implications for inclusive, equitable, and plural digital futures, particularly in diverse and low-resource settings, from both Global North and South. Works-in-progress are also welcome.
Accepted papers
Paper short abstract
This paper draws on a 14-month linguistic ethnography of a migrant-advice group in a low-income UK town. Taking a metapragmatic approach, it explores how interlocutors conceptualise digital translation technologies as participants in advice encounters, in theory and in practice.
Paper long abstract
This paper draws on a 14-month linguistic ethnography of a migrant-advice group, in a low-income coastal in the UK. Amidst financial and bureaucratic pressures, organisations that provide advice on accessing economic and civil services in the UK are subject to logics of efficiency (Koch and James, 2022). This is particularly true of advice organisations that aid refugees, asylum seekers and new migrants, in towns already lacking in state infrastructure, where such logics manifest linguistically in the form of standardised (“scripted”) conversations and mediation through digital translation technologies. Attempts to make advice interactions “efficient”, however, are regularly opposed by the linguistic and emotional complexities of clients’ cases.
Amidst contradictions between standardised expectations and complex linguistic realities of advice encounters, this paper asks: how do interlocutors conceptualise digital translation technologies as participants in advice encounters, in theory and in practice? Drawing on sociolinguist Jan Blommaert’s (2010) concept of “truncated speech” – the patchworking of multiple speakers and multiple “bits” of linguistic competence to work through an encounter – it explores the ways in interlocutors switch between working with and working against translation technology to generate the advice clients need.
By offering a metapragmatic (Silverstein,1979; Gal and Irvine, 2019) approach to digital language technologies – that moves away questions of what language technologies are doing, on a pragmatic or technological level, and toward what participants perceive them as doing/being able to do – that we can begin to theorise how they might fit into otherwise human encounters about livelihood security.
Paper short abstract
Kenya’s multilingual education faces a widening gap between policy and practice as AI tools rarely support Indigenous languages. Limited digital access reinforces English dominance. Inclusive, community-driven AI and stronger Indigenous data governance are needed to support multilingual learning.
Paper long abstract
Kenya’s multilingual education system is increasingly shaped by AI-mediated communication, where translation technologies offer the potential for broader language access. However, the gap between language-in-education policies and actual classroom practice has widened due to inadequate digital infrastructure and the limited inclusion of African Indigenous languages in mainstream AI systems. These shortcomings contribute to renewed linguistic marginalisation despite national commitments to multilingualism. This study investigates how Competency-Based Curriculum (CBC) language policies function in digitally influenced learning environments, identifies infrastructural constraints affecting multilingual instruction, and explores how AI translation tools might be redesigned to better align with local language priorities. A mixed-methods design was used, combining surveys and semi-structured interviews with educators (n=120), students (n=100), policymakers (n=30), and community leaders (n=50) in Kisii and Homabay counties. The data examined experiences with policy implementation, access to digital and AI resources, and user perceptions of translation technologies. Quantitative analysis assessed relationships between language use and learning outcomes, while qualitative perspectives revealed systemic challenges and unequal technological access. Guided by Vygotsky’s sociocultural theory, the study highlights the role of language in shaping learning processes. Critical language policy analysis draws attention to power structures influencing language technologies, and decolonial computing brings Indigenous data sovereignty to the forefront. Findings indicate persistent gaps between multilingual policy goals and classroom realities, alongside minimal digital support for Indigenous languages. Current AI tools largely reinforce English dominance, yet community-led language documentation offers promising alternatives. The study recommends strengthening Indigenous data governance and investing in culturally grounded AI systems.
Paper short abstract
This paper argues that AI transcription reproduces socio-digital inequality by treating language as computational capital. Dialect erasure and misrecognition render marginalized speakers algorithmically illegible, calling for participatory and linguistically just digital infrastructures.
Paper long abstract
Language is increasingly positioned as a core socio-digital infrastructure through which access to rights, services, and political recognition is mediated. Yet in contemporary digital systems, language is operationalized as computational capital, measurable, standardizable, and unevenly distributed. This paper theorizes AI-mediated transcription as a critical site where language-based socio-digital inequalities are produced and normalized, particularly in low-resource and high-stakes contexts.
Drawing on conceptual insights from linguistic anthropology, critical data studies, and migration scholarship, the paper reframes transcription not as a neutral technical process but as a socio-technical practice embedded in power relations. Automated transcription systems, trained on limited and hierarchical language datasets, privilege standardized linguistic forms while misrecognizing dialectal variation, affect, and strategic ambiguity. These omissions disproportionately affect marginalized speakers, transforming linguistic difference into algorithmic illegibility and constraining who can meaningfully engage with, interpret, or contest digital systems.
Using forced migration as a critical analytic lens, the paper shows how AI transcription operates as a gatekeeping mechanism within humanitarian and legal infrastructures, shaping credibility, visibility, and institutional legibility. Language misrecognition thus becomes a mechanism of epistemic injustice, whereby certain forms of speech are rendered inaudible within automated governance regimes.
The paper advances a theoretical intervention by proposing a shift from extractive, automation-centered language technologies toward participatory and plural models of linguistic mediation. By foregrounding language as a relational and political infrastructure rather than neutral data, it contributes to debates on inclusive digital futures and highlights the necessity of embedding linguistic justice and human agency into AI-driven systems.