Signal Structural Silence: how AI infrastructure fails underrepresented languages

Summary

This paper examines structural barriers facing speakers of underrepresented languages in AI systems, using Bengali as a case study for AI-assisted education in low-connectivity environments. It identifies 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. It documents a 67:1 training-token deficit between English and Bengali in major multilingual corpora, compounded by a tokenization penalty from Bengali's alphasyllabary script. It also notes connectivity exclusion, with individual internet penetration at 36.5% in rural areas versus 71.4% in urban areas. The paper argues dataset scarcity should be understood as a structural barrier rather than an isolated technical limitation, and proposes offline-first design as an equity-oriented infrastructure strategy.