The Language Tax: Comparative Transparency, Language Shift and Linguistic Justice in AI-Mediated Learning

Generative AI systems perform less well in many languages than in English, and learners who notice this often respond by switching to English. Earlier work proposed comparative transparency, making such quality gaps visible to learners, as a remedy for unequal access to AI-mediated explanation, but left open whether transparency might accelerate the very language shift that narrows learners' interpretive resources. This conceptual paper takes up that unresolved problem. I introduce the concept of a language tax: the epistemic cost that learners whose language is not English pay whichever way they choose, either accepting weaker explanations in their own language or obtaining better explanations at the price of working outside the language and frames through which they understand. I argue that the injustice lies in the existence of the tax rather than in the learner's choice, and that switching to English, although rational for each learner, generates a collective dynamic of language shift that no individual can reverse. I then show that leading theories of linguistic justice, including Van Parijs's account of parity of esteem and Patten's account of equal recognition, address the status and rights of languages but not the epistemic position of learners who depend on AI. In response I propose a standard of linguistic epistemic parity, which pairs transparency with obligations to reduce quality gaps and with translanguaging pedagogies that allow learners to use English-language outputs without abandoning their own interpretive frames. The case of Arabic, where the diglossic gap between Modern Standard Arabic and spoken dialects complicates the very idea of a learner's own language, is used to test the account.

Authors

Institutions

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-21
DOI
https://doi.org/10.5281/zenodo.24423532
Primary Topic
Multilingual Education and Policy
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

The Language Tax: Comparative Transparency, Language Shift and Linguistic Justice in AI-Mediated Learning

Mona Basal
Zenodo (CERN European Organization for Nuclear Research)
Multilingual Education and Policy
preprint

The Language Tax: Comparative Transparency, Language Shift and Linguistic Justice in AI-Mediated Learning

Mona Basal
preprint en

Abstract

Generative AI systems perform less well in many languages than in English, and learners who notice this often respond by switching to English. Earlier work proposed comparative transparency, making such quality gaps visible to learners, as a remedy for unequal access to AI-mediated explanation, but left open whether transparency might accelerate the very language shift that narrows learners' interpretive resources. This conceptual paper takes up that unresolved problem. I introduce the concept of a language tax: the epistemic cost that learners whose language is not English pay whichever way they choose, either accepting weaker explanations in their own language or obtaining better explanations at the price of working outside the language and frames through which they understand. I argue that the injustice lies in the existence of the tax rather than in the learner's choice, and that switching to English, although rational for each learner, generates a collective dynamic of language shift that no individual can reverse. I then show that leading theories of linguistic justice, including Van Parijs's account of parity of esteem and Patten's account of equal recognition, address the status and rights of languages but not the epistemic position of learners who depend on AI. In response I propose a standard of linguistic epistemic parity, which pairs transparency with obligations to reduce quality gaps and with translanguaging pedagogies that allow learners to use English-language outputs without abandoning their own interpretive frames. The case of Arabic, where the diglossic gap between Modern Standard Arabic and spoken dialects complicates the very idea of a learner's own language, is used to test the account.

Zenodo (CERN European Organization for Nuclear Research)
University of the People (US)
Multilingual Education and Policy
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon