Cultural Binding Heads in Language Models

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Floro, Avrile, Benedetto, Luca
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916055987781632
author Floro, Avrile
Benedetto, Luca
author_facet Floro, Avrile
Benedetto, Luca
contents LLMs often default to equal treatment across cultural groups, even though context warrants differentiation: this is a lack of difference awareness. Using mechanistic interpretability and a factorial design on the N4 cultural appropriation benchmark from Wang et al. (2025), we identify 2-3 mid-layer attention heads per model that contribute causally to cultural binding across eight models (four architectures, base and instruct). Cultural binding is the process of associating cultural items with the appropriate identity. Knockout of the identity-to-item edges on these heads lowers the binding strength by 9-23%. The identified heads transfer from instruct to base models, suggesting that cultural binding is created at pre-training. An $α$-scaling shows a graded dose-response and moderate amplification steering at generation ($α= 2-3$) increases cultural differentiation accuracy by 1-3 pp while leaving neutral reasoning mostly intact. A knowledge probing task shows that models know 3-5 times more than they act upon it, indicating that the bottleneck lies in routing and not knowledge.
format Preprint
id arxiv_https___arxiv_org_abs_2605_28543
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Cultural Binding Heads in Language Models
Floro, Avrile
Benedetto, Luca
Artificial Intelligence
Computation and Language
Machine Learning
LLMs often default to equal treatment across cultural groups, even though context warrants differentiation: this is a lack of difference awareness. Using mechanistic interpretability and a factorial design on the N4 cultural appropriation benchmark from Wang et al. (2025), we identify 2-3 mid-layer attention heads per model that contribute causally to cultural binding across eight models (four architectures, base and instruct). Cultural binding is the process of associating cultural items with the appropriate identity. Knockout of the identity-to-item edges on these heads lowers the binding strength by 9-23%. The identified heads transfer from instruct to base models, suggesting that cultural binding is created at pre-training. An $α$-scaling shows a graded dose-response and moderate amplification steering at generation ($α= 2-3$) increases cultural differentiation accuracy by 1-3 pp while leaving neutral reasoning mostly intact. A knowledge probing task shows that models know 3-5 times more than they act upon it, indicating that the bottleneck lies in routing and not knowledge.
title Cultural Binding Heads in Language Models
topic Artificial Intelligence
Computation and Language
Machine Learning
url https://arxiv.org/abs/2605.28543