2011年6月27日 星期一

UW programs computer for 'that's what she said' jokes

Story Published: May 2, 2011 at 8:39 PM PDT


Story Updated: May 2, 2011 at 8:39 PM PDT

SEATTLE -- Who says computer science can't be fun?

Two University of Washington researchers decided to put their coding skills to the test and develop a computer program that can identify whether a seemingly benign sentence might turn risque if you add on "that's what she said."


That's right -- software that tells "that's what she said" jokes.


Recently popularized by Michael Scott, Steve Carell's character on "The Office," TWSS is a delicate and nuanced form of humor. As researchers Chloe Kiddon and Yuriy Brun wrote in their research paper, developing a computer program with a TWSS sense of humor "is complex and would require both deep semantic and cultural understanding to recognize the vast array of double entendres."


But they wanted to see if they could do it. (That's what she said.)


"We worked on this project because we want to help computers understand human speech better," Brun said in an e-mail to seattlepi.com. "Computers today can understand some things humans say, but they have a hard time recognizing humor and sarcasm."


So, they got to work on a system called Double Entendre via Noun Transfer, or DEviaNT, which would take written content from popular joke websites -- twssstories.com, textsfromlastnight.com, fmylife.com and wikiquote.org -- and compute whether it would be funny to add on "that's what she said.


From their research paper:


We observe two facts about the TWSS problem. First, sentences with nouns that are euphemisms for sexually explicit nouns are more likely to be TWSSs. For example, containing the noun "banana" makes a sentence more likely to be a TWSS than containing the noun "door". Second, TWSSs share common structure with sentences in the erotic domain. For example, a sentence of the form "(subject) stuck (object) in" or "(subject) could eat (object) all day" is more likely to be a TWSS than not.


They translated those and other assumptions into computer algorithms -- including mathematical representations such as "noun sexiness" (NS) and "verb sexiness" (VS) -- then ran the possible jokes through DEviaNT. Their program, it turned out, told a true TWSS joke 71.4 percent of the time.


"With TWSS identification, it is important to never incorrectly say a sentence is a TWSS," Brun wrote. "However, it is OK to sometimes incorrectly say a sentence is not a TWSS. (Think about being at a party and the embarrassment of yelling 'that's what she said' after someone says something that's not a TWSS, as opposed to not saying anything after a sentence that could have been followed by 'that's what she said.')


"Accordingly, our technique focuses on minimizing false positives: saying a sentence is a TWSS when it is not."


Out of nearly 21,000 possible jokes from the websites, 262 of which could be turned into TWSS jokes, DEviaNT returned 28 positives -- six of which were actually false positives. According to the research paper, these true TWSS jokes included the sentences, "Yes give me all the cream and he's gone," "Yeah but his hole really smells sometimes" and "Don't you think these buns are a little too big for this meat?"


The researchers plan to continue work on DEviaNT, which employs "metaphorical mapping" to decide whether something will be deemed humorous. The technique "may be generalized to identify other types of double entendres and other forms of humor."


But will their DEviaNT fit? (That's what she said.)


Kiddon plans to present their research at the annual Association for Computational Linguistics conference on Human Language Technologies in June in Portland.


"We had a lot of fun working on this project," Brun said. "At the same time, this is serious research. Humor identification is a hard problem in the field of natural language understanding. The (computer science) community enjoys fun research but also recognizes the importance of this problem and of our findings."


Seattlepi.com is a media partner of KOMO News.


http://seeebook.com/

沒有留言:

張貼留言