Large language models (LLMs) are widely used globally but often reflect Western values, limiting their ability to represent diverse value systems. To address this, 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:
- Dataset-level validation showing it preserves both cross-country value differences and within-country diversity from the original survey;
- Automated evaluation indicating that training on PLURAL improves alignment with target countries' cultural profiles, reducing mean absolute error by up to 27.7% relative to strong baselines;
- 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 demonstrate that PLURAL contains learnable signals for value steering, offering a scalable resource for pluralistic alignment.
Dataset link: PLURAL Dataset
Blogger's Review: The introduction of the PLURAL dataset provides a significant foundation for value alignment in large language models, especially in multicultural contexts. This dataset not only enriches the corpus but also offers more representative value signals for future model training, and I look forward to its performance in practical applications.