From the article:<p>> “The stereotypical image we might have of a meme is the image with captions at the top and bottom,” says Wark. “But memes have gotten a lot weirder over the last few years. Many don’t really have a punchline like a joke does [...] “A program that can classify these [memes] with a 92% accuracy rate could be extremely useful for meme consumers with visual impairment,” she says.<p>A large part of the memes I see day to day have no text whatsoever, and their humor comes from the context of the conversation they are being applied to. Many are simple edits whose meaning I think would be near impossible to convey through language.<p>And why memes? They might provide a good set of data for building algorithms that can truly grasp the context of content while parsing it's meaning, but most memes require you to be part of an in-group, have knowledge of their history and their values, and even then, a single meme could be a humorous homage for one group of people, and a mocking joke for another.