Speaking In Tongues

Language can be an imprecise form of communication. Many words have more than one meaning, and our reliance on slang, dialect, foreign phrases and jargon can create even more confusion. A dish can be a vessel for serving food, the food itself, a nice-looking individual or a verb that invites us to gossip. No wonder computer search engines sometimes fail us — they don't understand us.

Now comes Albert (www.albert-inc.com), a Swiss software company that's developed an everyday-language search engine that understands how we speak — in 20 different languages. "It is up to the machine to understand the man. The only thing that counts is human language," explains founder and chairman Jean-Michel Livowsky, a cognitive scientist.

Albert focusses on the rapidly growing knowledge-management market with what it calls "corporate searchware," products that can either work as a stand-alone search engines or interface with existing ones. The Albert Interfacer is designed to search the databases of intranets and extranets; the Albert Searcher, the result of a partnership with the Norwegian information retrieval company FAST, can zip through more than a billion Internet documents.

Common to both products is technology devised by Livowsky, which uses algorithms — complex mathematical formulas — to decode human language. Albert software also learns users' behavior patterns so that it can continually sharpen its results, listed by relevance. "The more you use it, the better it works," notes CEO Beth Krasna, a former venture capitalist who joined the start-up in January. For example, it will quickly surmise that employees in the marketing department want different documents than the technicians in the lab, even if their searches are worded the same way. Albert software takes into consideration the context in which a key word is used. For instance, it would know that "automobile" or "auto" is usually used in the context of trade groups or shows; "vehicle" often refers to regulations; and "car" is typically associated with buying, selling or repairing. And it's also forgiving of human error. It can decipher flagrant misspellings and egregious typos.

Since its creation in 1999, Albert has amassed $23.5 million in three investment rounds, and has struck partnerships not only with FAST, but with IBM Europe. Krasna says she has about a year's worth of cash left, and will likely seek additional funding at the end of the year — though she envisions first profits sometime in 2002. Early clients include French Internet service provider free.fr, France Telecom and French overseas broadcaster RFO. Beyond France, it has offices in the U.S. and the U.K., and expects to enter Germany soon. And next year, after it adds Chinese, Japanese and Korean to its lexicon, it want to expand into Asia.

Revenues mainly come from licensing fees and a 15% annual maintenance fee. The former can range from $50,000 to $250,000 depending on the company's query load. Moreover, it charges a minimum of $2,500 for each server it must crawl through. Talk may be cheap, but the cost of comprehension is a bit more dear.