For decades computer scientists have struggled to program machines that can understand natural language. IBM, for example, had to teach Watson how humans speak in order to beat Ken Jennings at Jeopardy. Now two researchers from the University of Washington have invented a software system aimed to get computers to understand one particular intrinsic aspect of human communication: “That’s what she said.” In order to get a bunch of ones and zeroes to recognize when it’s funny to tack “that’s what she said” on the end of a sentence, the pair began by analyzing 1.5 million erotic sentences and 57,000 sentences from standard literature to look for nouns, adjectives, and verbs with a “sexiness function.” For example, rod and wet.
After spending what we imagine were a few free afternoons coming up with the name Double Entendre via Noun Transfer, or DEviaNT for short, the system is now capable of the same level of sophistication as Michael Scott. Forgetting, for a moment, that they should have based it on "sounds like my wedding night," we think this is awesome and obviously a great use of everyone’s time. Who wants to be ruled by robot overlords that can’t understand our dumb jokes?
That’s what she said: Software that tells dirty jokes [New Scientist]