Linguistic Diversity and Emergence: Where Does LLM Intelligence Come From?

This position paper proposes that emergence in large language models is not primarily a product of scale, but a product of linguistic diversity. When a model learns a sufficiently diverse set of languages simultaneously, the intersections and tensions between those languages cross a threshold, and new representations appear that exist in no single source language alone. This, the paper argues, is the mechanism of emergence. The argument is developed across nine sections: a critique of the scale hypothesis (Schaeffer 2023; Hoffmann 2022), the core hypothesis grounded in the weak Sapir-Whorf framework, human evidence from the bilingual brain literature with explicit engagement with the bilingual advantage replication crisis (Lowe 2021; Paap 2022; Nichols 2020), recent multilingual scaling research that revises the curse of multilinguality (ATLAS — Longpre, Kudugunta, Muennighoff et al. 2025; Chuang et al. 2025), application to LLMs as the first entities to hold thousands of linguistic frameworks simultaneously, and implications for AI safety research. The paper is offered as a position paper inviting empirical scrutiny rather than as a verified result. The methodology to causally establish the relationship between linguistic intersection density and emergent capabilities does not yet exist. Yet industry behavior (Meta's NLLB and MMS), recent ATLAS results on positive cross-lingual transfer, and human multilingualism research provide directional support.