Watson Software Makes Online Research Elementary

You’re sitting at your desk, trying to write a report for the boss and sifting through zillions of documents your search engine turned up that may or may not be useful.

Wouldn’t it be neat if the computer could automatically figure out which of those documents are relevant by simply paying attention to whatever you’re writing?

That’s just what computer scientists at Northwestern University in Evanston, Ill., say they have come up with.

They call their software Watson, and like Sherlock Holmes’ sidekick, it will do the tedious part of research for you while leaving you free to compose those pearls of wisdom.

Watson serves as a link between the browser and the word processing program, monitoring the latter to see what it needs to collect from the former, according to Kristian Hammond, director of the university’s Intelligent Information Laboratory.

“Essentially, it just sits there in the background, and it does an analysis of the document you are working on,” Hammond said. “It figures out the content of the document and then goes and does a set of searches on the Internet to find other documents that would be useful to you. Then it tosses those up in its little window. At any point, you can click on those documents and see what it has suggested.

“So while you are writing, it’s off doing research for you.”

Since Watson knows the context of the query, if “you are writing a paper on construction equipment and you ask Watson to find documents related to ‘Caterpillar,’ it will never return pages related to fuzzy insects,” Hammond said.

It amounts to a real-time research program that refines and modifies the search terms according to its analysis of whatever the user is writing, Hammond said.

“What’s really cool is once it’s figured out what you’re writing, or browsing, it will allow you to type in a query,” he said. “If you are looking at a document having to do with electric cars, you can type in a query like ‘pictures’ and, because Watson knows what you are working on, it builds the context around the query and it goes out and looks for pictures of electric cars.

“It looks for whatever it believes your current context is,” he said.

Watson goes considerably beyond the new GuruNet program (https://www.gurunet.com), which allows the user to look up the meaning of a word or phrase while using a browser, word-processing or e-mail program. With Watson, not just a dictionary or an encyclopedia or thesaurus is available--but the entire Internet.

During tests at the lab comparing Watson to computer experts who know the right buzzwords to use for a highly advanced search, Watson was “about twice as good,” Hammond said.

The software, developed by graduate student Jay Budzik, is available free on the Internet at https://dent.infolab.nwu.edu /infolab/downloads/software /infolab16.htm, but the current version works only with a browser, Internet Explorer. In a couple of weeks, Hammond expects to have the software available to link that browser with Microsoft Word.

“We have our internal prototype, and we’re using it here in the lab, but there are a couple of things we want to clean up before letting it out on the world,” he said.

A Netscape version will also be available soon, and eventually the system will be expanded to other word processing programs.

The software requires only a tiny bit of memory, Hammond said, because it is just a conduit between the browser and the word processor.

The university is giving the program away, he said, primarily to attract attention to the “InfoLab.”

“We want to free people from the tedium of search,” Hammond said. “We want to build systems that understand you, and understand what you are doing, and can help you proactively without your having to ask for the information.”

His researchers are already working on a more advanced version of Watson, called “Eye-to-Eye,” which could take the Information Age to a higher-and more intimate-level.

That program will work in much the same way as Watson, except it will also make a note of who else is looking at the same documents.

“It will say, ‘Here are people you might want to talk to’ ” because they are researching the same subject, Hammond said.

That might be a little too intrusive for some users, but Hammond notes that if people don’t want to talk, they don’t have to.

Lee Dye can be reached by e-mail at leedye@ptialaska.net.