When Symbol Names Should Not Matter: A Logistic Theory of Fresh-Symbol Classification

Wenjie Guan, Jelena Bradic · 2026

Read When Symbol Names Should Not Matter: A Logistic Theory of Fresh-Symbol Classification by Wenjie Guan, Jelena Bradic (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-2605-07120.txt.

Introduction

Wenjie Guan and Jelena Bradic study whether classifiers can ignore symbol names and learn the underlying template instead. Test items may use entirely fresh symbols that training never saw, yet share the same hidden pattern. Unlike prediction tasks, the classifier need not output the new symbols, only apply a renaming-proof decision rule.

Analyzing regularized kernel logistic classification, the authors split the learned rule into an ideal template classifier plus a disturbance from accidental token overlaps. A graph records those collisions, and margin guarantees follow. Vocabulary size sets how often collisions happen, but their geometry decides whether accuracy survives, a distinction that also judges prompting tricks.

Catalog record

CREATORWenjie Guan, Jelena Bradic
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE228 KB
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.07120

Browse the full library · About the Tower of Babel