ACE-Align: Attribute Causal Effect Alignment for Cultural Values under Varying Persona Granularities

Jiatang Luo, Bingbing Xu, Rongxin Chen +5

Ensuring that large language models (LLMs) reflect diverse cultural values is important for globally deployed NLP systems. However, existing approaches often treat cultural groups…