Title: Culturally Grounded Personas in Large Language Models: Characterization and Alignment with Socio-Psychological Value Frameworks

Abstract:

While Large Language Models (LLMs) are increasingly utilized to simulate human behavior, it remains unclear how accurately their synthetic personas mirror the world and moral value systems shaped by diverse cultural contexts. This study explores the concordance between these culturally embedded synthetic agents and established theoretical frameworks, namely the World Values Survey (WVS), the Inglehart-Welzel Cultural Map, and Moral Foundations Theory. We design and generate LLM personas defined by interpretable variables derived from the WVS. These personas are evaluated using three distinct analytical perspectives: first, by mapping them onto the Inglehart-Welzel dimension to highlight stable interpretive differences linked to cultural conditioning; second, by assessing demographic-level consistency with the World Values Survey, observing that their response distributions generally align with human group trends; and third, by deriving moral profiles via a Moral Foundations questionnaire. Through a culture-to-morality mapping, we analyze how moral responses shift across various cultural configurations. This methodology for generating and analyzing culturally grounded personas facilitates the assessment of cross-cultural structures and moral variability.

Source: arXiv Generated at: 2026-06-04 00:00:00 UTC