Cowen · Nature 2021 · Observational cross-sectional machine-learning study · n=6 million videos (participant count ?)

Sixteen facial expressions occur in similar contexts worldwide.

Level 3 - non-randomized controlled study

Large-scale observational cross-sectional study using automated machine-learning analysis of naturalistic video data.

PubMed 33328631 · doi:10.1038/s41586-020-3037-7 · record verified 2026-08-26

What was done

Deep neural networks were applied to 6 million naturalistic videos sourced from 144 countries across 12 world regions. The researchers conducted two experiments to measure whether 16 predefined categories of facial expressions systematically co-occurred with thousands of real-world social contexts (such as weddings or sporting competitions).

What was found

Each of the 16 facial expression categories showed distinct associations with specific social contexts. These context-expression associations were 70% preserved across 12 world regions. Differences in the frequency of specific facial expressions across regions varied as a function of which contexts were most salient.

Why it matters

Demonstrates that dynamic facial expressions have systematic, predictable associations with specific social situations globally, providing large-scale real-world data supporting cross-cultural commonalities in human expressive behavior.

Limits

The abstract does not report the number of unique human participants across the 6 million videos. Analysis relies entirely on automated deep neural network inference of expressions and contexts rather than direct measurement of internal emotional states. Internet-sourced video data may carry substantial demographic, cultural, and upload-selection biases.