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

TL;DR

This work tackles the problem of aligning LLMs to culturally diverse user groups by identifying Cultural Sparsity and Mean Collapse as core obstacles to universal alignment. It introduces CuMA, a Cultural Mixture of Adapters that uses demographic-aware routing to learn a Latent Cultural Topology, effectively disentangling conflicting gradients into specialized expert subspaces via a Mixture of LoRA Adapters. Across WorldValuesBench, Community Alignment, and PRISM, CuMA achieves state-of-the-art results, reduces mean collapse (lower entropy, higher Distinct-2), and demonstrates robust zero-shot generalization to unseen demographics. The approach offers a scalable path toward pluralistic, culturally resonant LLMs, while highlighting limitations such as reliance on demographic data and the need for richer expert capacity to cover global diversity.

Abstract

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.

Figures (4)

• Figure 1: Mechanism of Mean Collapse and the CuMA Solution. (A) Human values exhibit Cultural Sparsity, forming distinct modes (e.g., Traditional vs. Secular). (B) Standard dense models suffer from Gradient Interference when optimizing for conflicting modes simultaneously. This forces the model into Mean Collapse (the "Diluted Middle"), producing generic responses that fail to resonate with any group. (C) CuMA addresses this via Demographic-Aware Routing, explicitly disentangling gradients into specialized experts. (D) Consequently, the model generates distinct, culturally resonant outcomes for diverse users, effectively restoring value diversity.
• Figure 2: Architecture of CuMA. The framework disentangles cultural values by conditioning the routing mechanism on both semantic hidden states and demographic embeddings, effectively isolating gradients into specialized experts.
• Figure 3: Quantitative Verification of Mean Collapse.(Left) Dense baselines (e.g., LoRA, DoRA) exhibit high prediction entropy ($H \approx$ 1.38), indicating probability mass dispersion typical of mean collapse. CuMA significantly reduces uncertainty ($H \approx$ 1.17). (Right) In open-ended generation, CuMA achieves the highest Distinct-2 score, confirming that it avoids repetitive, generic templates by accessing specialized cultural vocabularies.
• Figure 4: Emergence of Latent Cultural Topology. t-SNE projection of expert activation patterns across 65 nations. Without explicit supervision, the router spontaneously organizes demographic profiles into coherent clusters that align with sociological frameworks (e.g., the African-Islamic and Confucian spheres). This geometric structure facilitates zero-shot generalization by routing unseen demographic profiles to experts trained on culturally proximate groups. Details on the visualization protocol are provided in Appendix \ref{['sec:appendix_analysis']}.