Structural Silence: When AI Infrastructure Fails Speakers of Underrepresented Languages

Source

• Source name: arXiv
• Source kind: paper
• Authors / organization: Avijit Roy, Proma Roy
• Venue: arXiv: cs.CL
• Published: 2026-08-12
• Collected: 2026-08-13
• URL: http://arxiv.org/abs/2608.12278v1
• PDF: https://arxiv.org/pdf/2608.12278v1
• DOI: UNKNOWN
• Registry group: academic_apis_and_preprints
• Evidence role: canonical_academic_index
• Language: en
• Country / region: UNKNOWN

Plain-Language Brief

The authors discuss how the underlying infrastructure of AI tools designed for education and language support can disadvantage speakers of underrepresented languages, using Bengali as an example. They highlight four main issues: a lack of web presence, data deficit in major multilingual corpora, script-related tokenization penalties, and rural internet penetration lower than urban areas.

What the Document Says

• AI tools for education and language support are often seen as solutions to access gaps in under-resourced communities.
• The infrastructure underlying these tools can disadvantage speakers of underrepresented languages before training.
• Bengali, one of the most widely spoken languages, is used as an example to examine these barriers in AI-assisted education in low-connectivity environments.
• Four main issues identified: web presence gap, 67:1 training-token deficit between English and Bengali, tokenization penalty due to Bengali's script, and rural internet penetration lower than urban areas.
• These failures are due to resource allocation decisions, institutional priorities, and design defaults that did not center underrepresented languages in mainstream AI development.
• The authors argue that dataset scarcity should be seen as a structural barrier rather than an isolated technical limitation.
• Offline-first design is suggested as an equity-oriented infrastructure strategy.
• Conclusion: directions for linguistics and AI research aimed at reducing these structural inequalities.

Key Concepts Mentioned

• AI tools
• education
• language support
• underrepresented languages
• Bengali
• web presence gap
• data deficit
• tokenization penalty
• rural internet penetration
• offline-first design

Claims Made by the Document

• AI infrastructure can disadvantage speakers of underrepresented languages.
• The four identified issues reflect longstanding resource allocation decisions and design defaults.
• Dataset scarcity is a structural barrier rather than an isolated technical limitation.

Evidence Used

• Severe web presence gap for Bengali
• 67:1 training-token deficit between English and Bengali in major multilingual corpora
• Tokenization penalty associated with Bengali's alphasyllabary script
• Rural internet penetration lower than urban areas

Methods / Approach

UNKNOWN

Limitations Stated by the Authors

UNKNOWN

Limitations Noted by the Agent

Agent interpretation — not from the source.

UNKNOWN

Open Questions

UNKNOWN

Notes for Future Q&A

May include interpretation. Labeled where it does.

UNKNOWN

Raw Abstract / Excerpt

Artificial intelligence tools for education and language support are increasingly framed as scalable responses to access gaps in under-resourced communities. Yet the infrastructure underlying these tools, including training corpora, tokenization schemes, evaluation benchmarks, and deployment architectures, can systematically disadvantage speakers of underrepresented languages before a model is trained. This paper examines these structural barriers through Bengali, one of the world's most widely spoken languages, focusing on AI-assisted education in low-connectivity environments. We identify four interlocking failures: a severe web presence gap, with Bengali accounting for less than 0.5% of global web content despite representing nearly 4% of the global population; a 67:1 training-token deficit between English and Bengali in major multilingual corpora; a tokenization penalty associated with Bengali's alphasyllabary script that compounds the data deficit through higher token fertility; and connectivity exclusion, with individual internet penetration at 36.5% in rural areas compared with 71.4% in urban areas. These failures reflect longstanding resource-allocation decisions, institutional priorities, and design defaults that did not center underrepresented languages in mainstream AI development. We argue that dataset scarcity should be understood as a structural barrier rather than an isolated technical limitation, and that offline-first design should be treated as an equity-oriented infrastructure strategy. We conclude with directions for linguistics and AI research aimed at reducing these structural inequalities.