if (That’s What She Said) then go to (Now Computers Want Your Funny Bone)

In February of this year, audiences watched as IBM’s supercomputer known as “Watson” matched wits against two human competitors on the game show Jeopardy. Watson bested former champions Ken Jennings and Brad Rutter by a significant margin, showing the huge strides made in speech recognition and comprehension technology. Watson won a tidy sum for charity, but the game is simply a demonstration of a wide variety of uses for computers like Watson.

According to IBM, the same technology that powers Watson may one day assist doctors in diagnosing diseases. The amount of bio-medical literature that exists in the field doubles every seven years, making it nearly impossible for doctors to memorize and apply quickly to their patients. Smarter computers could store data and retrieve it much faster, like Watson did on Jeopardy. The system probably wouldn’t be able to replace a human doctor, but it could be a valuable tool in saving lives. And the uses for this question-answer technology apply to engineering, science, business, and a variety of other fields.

The natural language algorithms and speech recognition technology used by Watson allowed it to dominate a game that features frequent wordplay. From there it’s not a stretch to imagine applying it to something that appeals to human emotion. Could a machine recognize what makes people laugh and why? Couldd computers tell jokes?

According to Chloé Kiddon and Yuriy Brun, two computer scientists at the University of Washington, machines may be able to dish out gags fairly soon. They’ve developed a system that recognizes a specific type of joke, namely the “That’s What She Said” joke. This type of language interpretation presents an interesting set of problems for the programmers, as the joke is based on double-entendres, words and phrases that could have multiple meanings. The program is called DEviaNT, which stands for Double Entendre via Noun Transfer, and works by recognizing nouns that are euphemisms for more sexual terms and compares the structure to phrases in the erotic domain.

For example, the word “banana” has a higher probability of having a sexual connotation than “apple”. The sentence structure “[subject] put the [object] inside” is more likely to have a literal and metaphorical meaning. They also used recognition of adjectives and verbs that have sexual meanings like “hot” or “doing”.

In order to program DEviaNT, Kiddon and Brun fed the computer tons of data from both erotic and non-erotic sources. They compiled information from websites filled with “That’s What She Said” (TWSS) jokes to assist the system in what they call “TWSS recognition”. They even used episodes of The Office, a show known for using TWSS jokes frequently.

The result was that DEviaNT was able to analyze text and produce TWSS jokes with about 70% accuracy. The programmers say that even that number is deceptively low because much of the data they gathered didn’t consist of TWSS jokes. With more consistent data, the system could achieve accuracy of up to 99.5%. They believe that their technology could be expanded to recognize other jokes, including puns, metaphors, and other types of humor.

It’s a far cry from Watson helping doctors to diagnose diseases, but maybe for DEviaNT, laughter is the best medicine.

More To Read:

Leave a Reply

Your email address will not be published. Required fields are marked *