Dual Perspectives in Emotion Attribution: A Generator-Interpreter Framework for Cross-Cultural Analysis of Emotion in LLMs

A new research paper titled 'Dual Perspectives in Emotion Attribution: A Generator-Interpreter Framework for Cross-Cultural Analysis of Emotion in LLMs' reveals significant cultural biases in how large language models understand human emotions. Authored by Aizirek Turdubaeva and Uichin Lee, the study introduces a novel framework that evaluates emotion attribution from both expression (generator) and interpretation perspectives, addressing a critical gap in prior research that focused mainly on interpretation while assuming universal emotional expression.

The researchers systematically evaluated six major LLMs using data from 15 countries, finding that performance variations depend heavily on both emotion type and cultural context. Their analysis shows generator-interpreter alignment effects are present, but more importantly, the generator's country of origin has a stronger impact on performance than the interpreter's background. This challenges the assumption of universal emotional expression and reveals how current AI models may misinterpret emotions across cultural boundaries.

The study's findings have significant implications for developers building AI systems for global markets. The authors call for culturally sensitive emotion modeling in LLM-based systems to improve robustness and fairness in emotion understanding across diverse cultural contexts. As AI systems increasingly handle tasks requiring emotional intelligence—from customer service to mental health support—this research highlights the need for more nuanced approaches to emotion recognition that account for cultural differences in expression and interpretation.