Recently, some engineers at the visual tech firm Curalate set out to make a tool that could automatically caption Instagrams with emoji. It was a chance to mess around with computer vision - complex models that consume such vast troves of data that they can recognize objects and suss out their meanings.
The tool, Emojini, did that pretty well: Given an Instagram of a bouquet, for instance, it would suggest the flower emoji. Given an Instagram of a horse race, it served up the horse and jockey.
But if you upload a picture of a can of Red Bull, for instance, Emojini associates it with the "open hands" emoji, not a glass or an ox. And if you upload a picture of a flower tattoo, it suggests a syringe - not a blossom, tulip, sunflower, rose or hibiscus.
In the course of developing the Emojini, Curalate's engineers had accidentally cracked a little-discussed code: They had uncovered the secondary, non-semantic meanings for emoji - the ways people use certain symbols and icons that have nothing to do with their designer's intent or the physical object that they correspond to in the real world.
Aside from the Red Bull/open hands and tattoo/syringe pairings, they found that the goat is typically used to caption photos of athletes ("greatest of all time" = G.O.A.T.) and the raised fist often accompanies Instagrams of people dirt-biking. People use the "busts in silouette" - they look like the Facebook icon - to mark selfies with strong backlighting. "Water closet" has become code for "woman crush." The dash symbol = vaping.