Microsoft trained an algorithm to judge the 'New Yorker' caption contest
LatestNew Yorker cartoons are the Vegemite of humor: they’re an acquired taste that is somewhat questionable for any person to want to acquire.
Like the cartoons or not, Microsoft has written computer software to determine which captions are the funniest.
Bloomberg reports two Microsoft researchers worked with longtime New Yorker comics editor Bob Mankoff to design a computer program that tries to determine which submissions to the magazine’s cartoon caption contest are the funniest. Mankoff told Bloomberg he gets 5,000+ submissions a week and can’t keep an assistant for more than a year because of the strain of sorting the droll from the uncouth.
Microsoft’s algorithm uses human-tagged descriptions of the cartoons to determine what they’re about, then compares each each of the captions Swiss tournament-style to rank them in terms of humor.
The researchers’ paper says the algorithm can eliminate about 2,200 of the submissions each week as not being up to the contest’s standards. But its shortlist for best caption only matched that of the human editors 55 percent of the time.
The paper provides an example of a caption contest shortlist assembled by a human editor for a cartoon about a car with hairy legs and a monstrous mouth. The winning caption was:
“Relax! It just smells the other car on you.”
Whereas the algorithm selected:
“Just listen to that baby purr.”
It’s a true computer science breakthrough: We’ve finally trained a computer to be as un-funny as a human.