Computers Programmed to Get the Joke

Aug. 28, 2007 — Computers are socially awkward. They have a hard time with informal communication and the subtleties of natural language — especially when it comes to jokes.

But now a group of researchers have equipped a computer with a sensor of humor. The technology could lead to programs that can solve problems that are informally stated, as well as to robots that are able to interact with humans more naturally.

"We rely on computers more and more, yet they don't seem to handle the way we communicate," said Julia Taylor, a Ph.D. candidate for computer science and engineering at the University of Cincinnati, Ohio. "I think it would be great for computers to understand natural language the way we use it," she said.

Taylor developed the program with associate professor Lawrence Mazlack, coordinator of the university's Applied Artificial Intelligence Laboratory.

Whether computer or human, getting a joke depends on the knowledge one has of the events mentioned in the joke.

Since programming a computer to have total world knowledge is a little overwhelming, Taylor and Mazlack restricted the domain of humor to children's jokes that have similar sounding words, the way puns and knock-knock jokes often do.

The program consists of two parts: a knowledge base derived from a children's dictionary and a collection of children's texts, and an algorithm that takes into account how the word sounds, how it's spelled, and what it means.

The knowledge base, called an ontology, represents an innovative, and more complex approach, said Christian Hemplemann, chief scientific officer at Hakia, an Internet search engine company.

Hakia conducts searches based on meaning, instead of popularity of key words or phrases. More popular approaches rely on statistics, which analyzes millions of words in a text and looks at what words occur frequently around other words.

For example, "bank" can mean a financial institution or land beside a river. Words such as "teller," "check" and "account" indicate financial institution, while "fish," "rapids" and "water" indicate land beside a river.

With an ontology, the researchers must build a database that includes all of the things and events in a given world — in this case, the world of children's jokes — and how they relate to each other. The relationships are categorized in a hierarchical structural from general to a more precise meaning.

"Humor is a very specific form of meaning that's related to a specific emotional response in humans that occurs in specific social situations," said Hemplemann. "If you understand how to do it artificially, that may give you an idea of how it works for the real thing."

To test the computer's understanding, Taylor enters text into the system and then lets the program tell her if it thinks the text is a joke or not. See what you think:

Mother to boy: "Johnny, you've been working in the garden a lot this summer."

Boy: "I know. My teacher told me to weed a lot."

Because "weed" sounds similar to "read," the program picks up on the wordplay and flags the text as a joke.

Taylor and Mazlack are currently working to build the knowledge base for the computer so that it can eventually recognize more sophisticated jokes. And one day, it may even be able to generate jokes of its own.

Getting a computer to recognize whether or not a joke is funny is a whole different matter.

Hakia Search Engine

University of Cincinnati's Applied Artificial Intelligence Laboratory