CuMA: Aligning LLMs with Sparse Cultural Values via Demographic-Aware Mixture of Adapters

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Sun, Ao, Wang, Xiaoyu, Tan, Zhe, Li, Yu, Zhu, Jiachen, Su, Shu, Jia, Yuheng
Formato: Preprint
Publicado: 2026
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866914240974028800
author Sun, Ao
Wang, Xiaoyu
Tan, Zhe
Li, Yu
Zhu, Jiachen
Su, Shu
Jia, Yuheng
author_facet Sun, Ao
Wang, Xiaoyu
Tan, Zhe
Li, Yu
Zhu, Jiachen
Su, Shu
Jia, Yuheng
contents As Large Language Models (LLMs) serve a global audience, alignment must transition from enforcing universal consensus to respecting cultural pluralism. We demonstrate that dense models, when forced to fit conflicting value distributions, suffer from \textbf{Mean Collapse}, converging to a generic average that fails to represent diverse groups. We attribute this to \textbf{Cultural Sparsity}, where gradient interference prevents dense parameters from spanning distinct cultural modes. To resolve this, we propose \textbf{\textsc{CuMA}} (\textbf{Cu}ltural \textbf{M}ixture of \textbf{A}dapters), a framework that frames alignment as a \textbf{conditional capacity separation} problem. By incorporating demographic-aware routing, \textsc{CuMA} internalizes a \textit{Latent Cultural Topology} to explicitly disentangle conflicting gradients into specialized expert subspaces. Extensive evaluations on WorldValuesBench, Community Alignment, and PRISM demonstrate that \textsc{CuMA} achieves state-of-the-art performance, significantly outperforming both dense baselines and semantic-only MoEs. Crucially, our analysis confirms that \textsc{CuMA} effectively mitigates mean collapse, preserving cultural diversity. Our code is available at https://github.com/Throll/CuMA.
format Preprint
id arxiv_https___arxiv_org_abs_2601_04885
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle CuMA: Aligning LLMs with Sparse Cultural Values via Demographic-Aware Mixture of Adapters
Sun, Ao
Wang, Xiaoyu
Tan, Zhe
Li, Yu
Zhu, Jiachen
Su, Shu
Jia, Yuheng
Computation and Language
Artificial Intelligence
Machine Learning
As Large Language Models (LLMs) serve a global audience, alignment must transition from enforcing universal consensus to respecting cultural pluralism. We demonstrate that dense models, when forced to fit conflicting value distributions, suffer from \textbf{Mean Collapse}, converging to a generic average that fails to represent diverse groups. We attribute this to \textbf{Cultural Sparsity}, where gradient interference prevents dense parameters from spanning distinct cultural modes. To resolve this, we propose \textbf{\textsc{CuMA}} (\textbf{Cu}ltural \textbf{M}ixture of \textbf{A}dapters), a framework that frames alignment as a \textbf{conditional capacity separation} problem. By incorporating demographic-aware routing, \textsc{CuMA} internalizes a \textit{Latent Cultural Topology} to explicitly disentangle conflicting gradients into specialized expert subspaces. Extensive evaluations on WorldValuesBench, Community Alignment, and PRISM demonstrate that \textsc{CuMA} achieves state-of-the-art performance, significantly outperforming both dense baselines and semantic-only MoEs. Crucially, our analysis confirms that \textsc{CuMA} effectively mitigates mean collapse, preserving cultural diversity. Our code is available at https://github.com/Throll/CuMA.
title CuMA: Aligning LLMs with Sparse Cultural Values via Demographic-Aware Mixture of Adapters
topic Computation and Language
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2601.04885