New study reveals that LLMs struggle with cross-cultural emotion understanding, with a generator's cultural origin mattering more than the interpreter's perspective.

Researchers tested six LLMs across emotion attribution tasks using data from 15 countries, finding significant performance variations based on emotion type and cultural context

The Generator-Interpreter framework addresses a key gap in prior research by considering both how emotions are expressed culturally and how they are interpreted, rather than assuming universal emotion patterns

The study found that the country of origin of the emotion generator has a stronger impact on LLM performance than the interpreter's background, highlighting the importance of cultural expression in AI systems

Current LLMs often fail to account for cultural norms that shape how different nations express and perceive emotions, limiting their effectiveness in cross-cultural applications