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

Fuente: Zenodo
Gespeichert in:
Bibliographische Detailangaben
1. Verfasser: Kim, Hongsung
Format: Recurso digital
Sprache:Englisch
Veröffentlicht: Zenodo 2026
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866901792974962688
author Kim, Hongsung
author_facet Kim, Hongsung
contents <p>This position paper proposes that emergence in large language models is <br>not primarily a product of scale, but a product of linguistic diversity. <br>When a model learns a sufficiently diverse set of languages <br>simultaneously, the intersections and tensions between those languages <br>cross a threshold, and new representations appear that exist in no single <br>source language alone. This, the paper argues, is the mechanism of <br>emergence.</p> <p>The argument is developed across nine sections: a critique of the scale <br>hypothesis (Schaeffer 2023; Hoffmann 2022), the core hypothesis grounded <br>in the weak Sapir-Whorf framework, human evidence from the bilingual <br>brain literature with explicit engagement with the bilingual advantage <br>replication crisis (Lowe 2021; Paap 2022; Nichols 2020), recent <br>multilingual scaling research that revises the curse of multilinguality <br>(ATLAS — Longpre, Kudugunta, Muennighoff et al. 2025; Chuang et al. <br>2025), application to LLMs as the first entities to hold thousands of <br>linguistic frameworks simultaneously, and implications for AI safety <br>research.</p> <p>The paper is offered as a position paper inviting empirical scrutiny <br>rather than as a verified result. The methodology to causally establish <br>the relationship between linguistic intersection density and emergent <br>capabilities does not yet exist. Yet industry behavior (Meta's NLLB and <br>MMS), recent ATLAS results on positive cross-lingual transfer, and <br>human multilingualism research provide directional support.</p> <p>Keywords: large language models, emergent abilities, multilingualism, <br>linguistic diversity, scaling laws, Sapir-Whorf hypothesis, AI cognition, <br>AI safety</p>
format Recurso digital
id zenodo_https___doi_org_10_5281_zenodo_20073919
institution Zenodo
language eng
publishDate 2026
publisher Zenodo
record_format zenodo
spellingShingle Linguistic Diversity and Emergence: Where Does LLM Intelligence Come From?
Kim, Hongsung
large language models
emergent abilities
multilingualism
linguistic diversity
scaling laws
Sapir-Whorf hypothesis
AI cognition
<p>This position paper proposes that emergence in large language models is <br>not primarily a product of scale, but a product of linguistic diversity. <br>When a model learns a sufficiently diverse set of languages <br>simultaneously, the intersections and tensions between those languages <br>cross a threshold, and new representations appear that exist in no single <br>source language alone. This, the paper argues, is the mechanism of <br>emergence.</p> <p>The argument is developed across nine sections: a critique of the scale <br>hypothesis (Schaeffer 2023; Hoffmann 2022), the core hypothesis grounded <br>in the weak Sapir-Whorf framework, human evidence from the bilingual <br>brain literature with explicit engagement with the bilingual advantage <br>replication crisis (Lowe 2021; Paap 2022; Nichols 2020), recent <br>multilingual scaling research that revises the curse of multilinguality <br>(ATLAS — Longpre, Kudugunta, Muennighoff et al. 2025; Chuang et al. <br>2025), application to LLMs as the first entities to hold thousands of <br>linguistic frameworks simultaneously, and implications for AI safety <br>research.</p> <p>The paper is offered as a position paper inviting empirical scrutiny <br>rather than as a verified result. The methodology to causally establish <br>the relationship between linguistic intersection density and emergent <br>capabilities does not yet exist. Yet industry behavior (Meta's NLLB and <br>MMS), recent ATLAS results on positive cross-lingual transfer, and <br>human multilingualism research provide directional support.</p> <p>Keywords: large language models, emergent abilities, multilingualism, <br>linguistic diversity, scaling laws, Sapir-Whorf hypothesis, AI cognition, <br>AI safety</p>
title Linguistic Diversity and Emergence: Where Does LLM Intelligence Come From?
topic large language models
emergent abilities
multilingualism
linguistic diversity
scaling laws
Sapir-Whorf hypothesis
AI cognition
url https://doi.org/10.5281/zenodo.20073919