The Monolingual Machine: Internal Translation, Unequal Access, and the Architecture of Sovereign AI

Abstract: The claim that large language models are multilingual describes their user interface. Their internal architecture remains monolingual. Commercial systems generate text across dozens of languages, encouraging institutions to treat them as neutral global infrastructure. Beneath that polyglot surface, modern architectures operate as an English-centric monopoly. Training corpora drawn from the commercial web concentrate resources in English, rendering the planetary majority of languages computational noise. Inside the network, intermediate processing layers translate non-English queries into English representations before projecting answers back into the user's language. This architectural monopoly produces compounding inequalities. Non-Western scripts are shattered into fragmented processing units, imposing economic and memory penalties on non-English users. Safety guardrails trained in English collapse when prompts are translated into lower-resource tongues. Meanwhile, translated benchmarks certify capability by testing foreign fluency in Western cultural norms. Sovereign AI initiatives in India, Aotearoa New Zealand, and Africa reject this extractive pipeline. They demonstrate that linguistic self-determination requires native pretraining and enforceable regulatory parity. Keywords: Linguistic Imperialism, Multilingual Large Language Models, Latent Language Hypothesis, Internal Translation, Tokenization Disparity, Cross-Lingual Safety, Benchmark Parity, Indigenous Data Sovereignty, CARE Principles, Sovereign AI Infrastructure, Bhashini, BharatGen, Te Hiku Media, Masakhane, Algorithmic Monoculture, Computational Equity, Regulatory Certification, Model Auditing, Data Extraction, Language Reclamation

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-17
DOI
https://doi.org/10.5281/zenodo.24303495
Primary Topic
Ethics and Social Impacts of AI
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

The Monolingual Machine: Internal Translation, Unequal Access, and the Architecture of Sovereign AI

Felix Velasco, Josie Jefferson
3 links
Zenodo (CERN European Organization for Nuclear Research)
Ethics and Social Impacts of AI
preprint

The Monolingual Machine: Internal Translation, Unequal Access, and the Architecture of Sovereign AI

Felix Velasco, Josie Jefferson
preprint en

Abstract

Abstract: The claim that large language models are multilingual describes their user interface. Their internal architecture remains monolingual. Commercial systems generate text across dozens of languages, encouraging institutions to treat them as neutral global infrastructure. Beneath that polyglot surface, modern architectures operate as an English-centric monopoly. Training corpora drawn from the commercial web concentrate resources in English, rendering the planetary majority of languages computational noise. Inside the network, intermediate processing layers translate non-English queries into English representations before projecting answers back into the user's language. This architectural monopoly produces compounding inequalities. Non-Western scripts are shattered into fragmented processing units, imposing economic and memory penalties on non-English users. Safety guardrails trained in English collapse when prompts are translated into lower-resource tongues. Meanwhile, translated benchmarks certify capability by testing foreign fluency in Western cultural norms. Sovereign AI initiatives in India, Aotearoa New Zealand, and Africa reject this extractive pipeline. They demonstrate that linguistic self-determination requires native pretraining and enforceable regulatory parity. Keywords: Linguistic Imperialism, Multilingual Large Language Models, Latent Language Hypothesis, Internal Translation, Tokenization Disparity, Cross-Lingual Safety, Benchmark Parity, Indigenous Data Sovereignty, CARE Principles, Sovereign AI Infrastructure, Bhashini, BharatGen, Te Hiku Media, Masakhane, Algorithmic Monoculture, Computational Equity, Regulatory Certification, Model Auditing, Data Extraction, Language Reclamation

Zenodo (CERN European Organization for Nuclear Research)
Reduced inequalities
Ethics and Social Impacts of AI
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