Speaking the Natural Lingo

Making Internet search engines respond to the way people actually talk is key to improving the usefulness of results.

Making Internet search engines respond to the way people actually talk is key to improving the usefulness of results.

Many searchers can get by on keywords, said David Harris, vice president of marketing at EasyAsk, but business users want to submit long queries and get back relevant results. When voice processing comes to the Internet, search engines that use more natural language will become a necessity, not an option, he added.

Executives at EasyAsk and LingoMotors, two companies that specialize in natural-language search, said the goal is to understand not only the language syntax, but also its context. If the context is understood, the search engine can pursue the meaning of the question, not just hits on keywords.

"Asking what stocks tanked on the Nasdaq could produce thousands of answers. Whats needed is an understanding of tanked, " said John Hanselman, president of LingoMotors. Hanselmans company has produced algorithms that try to divine the meaning of words in context. Behind the processing are dictionaries built from an individual business terminology and product contexts.

At an outdoor sporting goods store, its important to realize that a query on "hunting boots" means "insulated, waterproof and good traction," even though the stores site doesnt have anything in the "hunting" category, Hanselman said. By understanding the context of the query, the search engine can bring back appropriate responses, he noted.

How to improve the relevancy of results was one of the key issues debated at Search Engine Strategies 2001, sponsored by Internet.com, last week in Boston.

When asked whether Inktomis search engine could respond to a natural-language question, Tim Mayer product manager at Inktomi, said Inktomi users prefer to search on one or two key words, not full sentences.

If you were to ask AltaVista for "Books by Warhol," you will get back more than 19,000 results, Hanselman said. But if you parse the query through LingoMotors and then send it to AltaVista, you get back about 24 responses, he said.

Thats one reason why AltaVista partners with the Cambridge, Mass., firm to offer LingoMotors to its enterprise customers. Hanselman said he is talking to AltaVista about adopting LingoMotors TurboSearch natural-language processor for the Web site. "Both sides are highly interested," he said.