One Language, Two Scripts: Probing Script-Invariance in LLM Concept Representations

Sripad Karne · 2026

Read One Language, Two Scripts: Probing Script-Invariance in LLM Concept Representations by Sripad Karne (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-08869.txt.

Description

Sripad Karne asks whether sparse autoencoder features track meaning or mere spelling. Serbian, written in Latin and Cyrillic with near-perfect letter mapping but zero shared tokens, lets meaning stay fixed while script changes. Same-meaning sentences in different scripts fire highly overlapping features, far above random baselines, and script swaps disturb features less than paraphrases do, with the effect growing with model size.

Catalog record

CREATORSripad Karne
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE26012
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.08869

Browse the full library · About the Tower of Babel