PLURAL: A Global Dataset for Value Alignment

2026-07-08

AI Summary

This study addresses the pervasive embedding of Western values in large language models, which hinders their alignment with diverse global cultural contexts. Leveraging the Integrated Values Survey (IVS) spanning 92 countries, the authors introduce a novel approach to transform large-scale cross-national value data into learnable preference signals, constructing a culturally aligned dataset comprising approximately 500,000 synthetic preference triplets across 20 representative nations. A two-stage generation pipeline—integrating automated evaluation with blind human assessments from multiple countries—ensures fine-grained and non-Western-centric cultural representation. Experimental results demonstrate that fine-tuning models on this dataset reduces cultural value misalignment error by up to 27.7% in target countries, with human evaluations confirming that generated content significantly better reflects local values.

Abstract

Large language models (LLMs) are used worldwide, yet disproportionately reflect Western values, limiting their ability to represent diverse value systems. We introduce PLURAL, a large-scale, value-focused preference dataset grounded in the Integrated Values Survey (IVS), a nationally representative survey spanning 92 countries. Using a two-stage generation pipeline, we transform survey responses into synthetic preference triplets that preserve normative value signals while producing realistic scenarios. We release an initial version of PLURAL containing ~500,000 preference triplets representing people in 20 diverse countries. We evaluate PLURAL in three ways: (i) dataset-level validation showing that it preserves both cross-country value differences and within-country diversity from the original survey; (ii) automated evaluation showing that training on PLURAL improves alignment with target countries' cultural profiles, reducing mean absolute error by up to 27.7% relative to strong baselines; and (iii) blind human evaluation with 176 evaluators in India, Brazil, and Japan, who judge PLURAL-aligned responses as more representative of their national values. Together, these results show that PLURAL contains learnable signal for value steering, offering a scalable resource for pluralistic alignment. Dataset: https://huggingface.co/datasets/agdhruv/plural-alignment

Problem

• value alignment
• large language models
• cultural diversity
• global representation
• pluralistic values

Innovation

• value alignment
• preference dataset
• cultural diversity
• synthetic data generation
• large language models