The Culture Funnel: You Can't Align What isn't in the Data
For researchers and practitioners working on cultural alignment of LLMs, this paper highlights a critical data bottleneck in training pipelines that must be addressed to achieve balanced cultural representation.
The paper identifies a 'cultural data funnel' in LLM pipelines where explicit cultural signals decline sharply during post-training, while geographically concentrated, task-specialized data dominates. Using a multidimensional tagging framework, they show that their tags improve downstream cultural benchmark performance.
Current cultural alignment approaches focus on inference-time interventions, assuming models already contain sufficient cultural knowledge. We argue modern LLM pipelines suffer from a cultural data funnel. Using a multidimensional tagging framework across pretraining, fine-tuning, alignment, and reasoning datasets, we show explicit cultural signals decline sharply during post-training, while geographically concentrated, task-specialized data dominates. Multilinguality enhances geographic diversity of cultural knowledge but does not ensure balanced representation. Our tags improve downstream cultural benchmark performance, demonstrating that advances require shifting focus in training data pipelines. To facilitate future research, we release our culturally tagged dataset with 5.6M samples at https://huggingface.co/datasets/CohereLabs/CultureMarkers.