Have you ever encountered someone who just looks like their name? Do you know someone who you couldn't imagine being named anything other than "John" or "Sally"? Most people have experienced this kind of intuition about names before, but is it a real phenomenon or just a figment of our imaginations?
Well, it turns out there may be some truth to it after all, if the success of new software developed by researchers at Cornell University is any indication, reports New Scientist. The software is capable of guessing a person's name just by analyzing the person's face, and could lead to improved ways of combing online pictures for more accurate tagging.
Though many people claim to be able to guess a person's name by their face, it's difficult to make sense of how this could be possible. Surely names don't have any sort of ability to shape the faces of the people they're assigned to. Right?
"First names are not given to babies at random," explained Andrew Gallagher, one of the researchers on the project. Many names are gender-specific, for instance. Others are passed down along family lines, or indicative of one's ethnicity or culture.
Some names carry obvious connotations. A person named "Carlos" isn't likely to be a female born to Russian parents, for instance. Furthermore, common names proliferated down family lines are more likely to be tied to certain phenotypes, since the naming is also associated with genetic heritage. But how specific a description can be implied by a name, especially in modern times when naming has become less traditional?
Gallagher and colleagues first trained a computer to scour tagged photos on Flickr to look for commonalities among people with the same names. For instance, they found that Alejandras tend to have darker hair and skin than Heathers, while Ethans tend to be younger than Davids. Using information like this, the computer was then asked to make educated guesses about peoples names just by analyzing their features.
To make matters more precise (and more manageable), the team focused only on 100 of the top names in the U.S. – 48 male, 48 female and four gender-neutral – which, astoundingly, actually encompasses the names of about 20 percent of the U.S. population.
So given these parameters, how did the computer do? Well, it was able to predict names with about a 4 percent accuracy. That might not sound all that great, but it's actually four times better than chance, and a third better than human intuition. So it's significant. As the database expands, the results could be expected to improve.
Eventually Gallagher would like to include all U.S. names in the computer's database, not just the names that are most common. It's possible that more distinctive names will be easier to attach to a face, though the sheer volume of possible matches will also present a challenge.
The most likely application for this kind of software will be for online photo-sharing tools, such as on Flickr and Facebook. Facebook has already experimented with facial recognition software to aid its users in tagging photos, for instance.