Scientists Train Computer to Understand "That's What She Said" Jokes
As science marches on, your computer might soon find itself giggling every time you say the words "hard drive" and "motherboard" in the same sentence.
Chloé Kiddon and Yuriy Brun, a pair of computer scientists at the University of Washington, have finally brought humor into the 21st century by creating a program that can analyze blocks of text for the most inviting places to insert yourself ... in the form of a timely "that's what she said!"
The problem the pair faced was not just teaching the computer to know when it was possible to add a "that's what she said" to the end of a sentence, but also to know when it would actually be funny. To do this, they fed the computer 1.5 million unintentionally naughty sentences, along with examples of plain and unfunny text. Then, to increase the program's understanding of what words make for good euphemisms, the researches applied a "sexiness" function to the raunchiest-sounding nouns and adjectives, with words like "rod," "meat," "hot," and "wet," being given high sexiness functions.
The researchers further trained the program, which they named Double Entendre via Noun Transfer (or DEviaNT), on stories collected from TWSSstories.com [http://twssstories.com/]. After testing, they found that DEviaNT had a 70 percent accuracy rate when analyzing text for dirty bits, which the pair believe they can improve to 99.5 percent with the addition of more data. They also believe that DEviaNT could be modified to recognize "other types of double entendres and other forms of humor."
Well, thanks a lot, science. The ability to laugh was going to be one of the only ways to tell man from machine, and now you've gone and given them a sense of humor. An infantile one, yes, but you've gone and made our inevitable war with the machines that much more difficult. You've made it longer, harder, and more dangerous.
And that, my friends, is what she said.
Source: New Scientist [http://www.newscientist.com/blogs/onepercent/2011/04/software-works-out-whether-tha.html?]
(Image) [http://wearscience.com/design/robot/]
Permalink
Chloé Kiddon and Yuriy Brun, a pair of computer scientists at the University of Washington, have finally brought humor into the 21st century by creating a program that can analyze blocks of text for the most inviting places to insert yourself ... in the form of a timely "that's what she said!"
The problem the pair faced was not just teaching the computer to know when it was possible to add a "that's what she said" to the end of a sentence, but also to know when it would actually be funny. To do this, they fed the computer 1.5 million unintentionally naughty sentences, along with examples of plain and unfunny text. Then, to increase the program's understanding of what words make for good euphemisms, the researches applied a "sexiness" function to the raunchiest-sounding nouns and adjectives, with words like "rod," "meat," "hot," and "wet," being given high sexiness functions.
The researchers further trained the program, which they named Double Entendre via Noun Transfer (or DEviaNT), on stories collected from TWSSstories.com [http://twssstories.com/]. After testing, they found that DEviaNT had a 70 percent accuracy rate when analyzing text for dirty bits, which the pair believe they can improve to 99.5 percent with the addition of more data. They also believe that DEviaNT could be modified to recognize "other types of double entendres and other forms of humor."
Well, thanks a lot, science. The ability to laugh was going to be one of the only ways to tell man from machine, and now you've gone and given them a sense of humor. An infantile one, yes, but you've gone and made our inevitable war with the machines that much more difficult. You've made it longer, harder, and more dangerous.
And that, my friends, is what she said.
Source: New Scientist [http://www.newscientist.com/blogs/onepercent/2011/04/software-works-out-whether-tha.html?]
(Image) [http://wearscience.com/design/robot/]
Permalink
