Probing Cultural Signals in Large Language Models through Author Profiling
This work addresses cultural bias in LLMs for applications with societal impact, though it is incremental as it focuses on probing existing models rather than developing new mitigation methods.
The study probed cultural biases in large language models by evaluating their zero-shot ability to infer singers' gender and ethnicity from song lyrics, finding that models like Ministral-8B showed strong ethnicity bias while Gemma-12B was more balanced, with performance assessed on over 10,000 lyrics.
Large language models (LLMs) are increasingly deployed in applications with societal impact, raising concerns about the cultural biases they encode. We probe these representations by evaluating whether LLMs can perform author profiling from song lyrics in a zero-shot setting, inferring singers' gender and ethnicity without task-specific fine-tuning. Across several open-source models evaluated on more than 10,000 lyrics, we find that LLMs achieve non-trivial profiling performance but demonstrate systematic cultural alignment: most models default toward North American ethnicity, while DeepSeek-1.5B aligns more strongly with Asian ethnicity. This finding emerges from both the models' prediction distributions and an analysis of their generated rationales. To quantify these disparities, we introduce two fairness metrics, Modality Accuracy Divergence (MAD) and Recall Divergence (RD), and show that Ministral-8B displays the strongest ethnicity bias among the evaluated models, whereas Gemma-12B shows the most balanced behavior. Our code is available on GitHub (https://github.com/ValentinLafargue/CulturalProbingLLM).