Air Date: Week of July 6, 2007

A new computer program separates fact from opinion in texts. Ian Gray reports.

Transcript

[DRAGNET THEME]

GRAY: He wanted the facts. They were hard to come by...he got his from shifty-eyed shoeshine boys in dark alleys and dingy bars. His only weakness besides two-timing blondes and a pack of smokes - he wasn’t a computer. Now a virtual detective was taking over his beat on the means streets of America. The new detective in town: a computer program!

[SCIENCE NOTE THEME]

Scientists at Cornell University are designing a computer program that will distinguish opinion from fact. The research hinges on how computers convert human language into computer language. Our words turned into 0s and 1s. The Cornell team will try to teach their computers the difference between subjects, objects and other parts of written language. Their goal is to develop algorithms that recognize key phrases of opinion like “according to” or “it is believed that...”

This type of programming, called information extraction can be used to locate specific types of information from sources like emails, blogs and other online forums. Information extraction is a growing cornerstone of the security and surveillance industry – and the Cornell study is funded by the Department of Homeland Security. But Cornell scientists say that their research is only focusing on online newspaper text. For instance, their algorithms could be used to find out what newspapers around the world are saying about the American occupation in Iraq. They can also find out how much of what online sources are saying is based on factual statements versus statements of opinion. By cross-referencing thousands of texts at once, the computers could even determine whether some media are presenting opinions as though they were facts.

That’s this week’s note on emerging science. I’m Ian Gray.