Want to read more?
We value our content, so access to our full site is only available on subscription.
Your subscription entitles you to 7-day-a-week access to our website, plus a full digital copy of that week’s paper to read on your pc/mac or mobile device.
And there’s more – your subscription includes access to digital archive copies from 2006 onwards.
Our facial expressions reflect our emotions, be it a smile of happiness or a frown of disapproval.
Such emotional signals are easily interpreted, a fact that is perhaps more obvious now that they are hidden by masks to prevent Covid-19 transmission.
Nevertheless, they are thought to shape our relationships right from birth, and are important to our health and well-being throughout life.
Around 150 years ago Charles Darwin controversially proposed that these facial expressions form a universal language of social life.
This spawned lively debate over whether the reflex use of facial muscles to signal feelings is universal among humans or whether their use and interpretation show cultural and regional variation.
To date, small studies using participants to label photos of posed expressions have not yielded concrete conclusions, but now AI scientists have entered the fray.
In a massive study they minimised participant bias by applying machine learning to analyse facial expressions in online videos showing real-world situations across many different cultures (1).
Specifically, they investigated 16 patterns of dynamic facial expression, including amusement, awe, contentment and triumph, in thousands of natural contexts by analysing six million videos from 144 countries in 12 world regions.
The results showed that in certain natural settings such as weddings, fireworks displays and sports competitions, particular patterns of facial expressions, like awe, contentment and triumph, were preserved across all 12 world regions examined.
Overall, there was 70 per cent overlap between facial expression and specific social context, lending strong support to Darwin’s suggestion of their universality.
But because these findings are based on online video content, they could be biased by the cultural globalisation of Western media.
However, regional analysis did not support this criticism since greater similarity was seen between neighbouring, rather than remote, regions.
For example, the Middle East and India showed 85 per cent preservation of facial expression/context patterns, but only 64 per cent and 54 per cent overlap respectively with North America.
These data indicate that facial expressions denoting our emotions probably developed early in human evolution, and have been preserved as a means of communicating certain emotional challenges because they enhanced our survival.
(1) Cowen et al Nature 589, 251-257. 2020.