Steve Carell's final episode of The Office aired
last night, and without him, we're left with a hole in our lives where those "that's what she said" jokes (hereby referred to as TWSS jokes) should be. Luckily, a couple of computer scientists have created some software that recognizes the opportunity to make a TWSS joke in the midst of natural language. It sounds like a huge effort, but definitely has the potential to fill that hole. (That one was a gimme.)
The software actually sounds pretty tricky to write—it's a natural language analyzer, sort of like a more specific and not so terrifying version of IBM's Watson. Created by Chloé Kiddon and Yuriy Brun of the University of Washington (make sure to read their report if you want to see an NBC sitcom properly cited in an academic work), the system is named DEviaNT, which stands for Double Entendre via Noun Structure—a bit of a stretch, but we've seen worse.
To figure out if a seemingly innocuous sentence could have a second, more sexual meaning, Kiddon and Brun first analyzed lots of text from both erotic and non-erotic sources, then set the software to evaluate which nouns, adjectives, and verbs have a high "sexiness" possibility—words like "big," "meat," and "satisfy" would all qualify. They also fed it jokes from TWSS websites (yes, there are compendiums of TWSS jokes, of course there are compendiums of TWSS jokes).
At the moment, DEviaNT has about a 70% accuracy rate, which leads to some jokes that don't really make sense (that's what she said), but the researchers are confident that with more data, they can boost its abilities to near-Michael-Scottian levels.
Popular Science is your wormhole to the future. Reporting on what's new and what's next in science and technology, we deliver the future now.