Comment Articles

Pinpointing locations

Michael Kanellos CNET News

Published: 29 Mar 2004 16:35 BST

What's worse is that, in many documents, locations aren't described with much specificity. Instead, a text might say "land mines 22 miles north of Um" or "Indian Plate." The MetaCarta software essentially has to translate these statements into navigational coordinates to achieve the right results.

"A lot of these places aren't even on the map," Frank said. The difficulty also explains why the company's software currently sells in the six-figure range.

The MetaCarta and Language Weaver efforts essentially address the central paradox of search: the more you know, the less you know. The amount of information out there and the ways people want to use it are so wide-ranging that there are plenty of technology opportunities.

"It is much larger than I ever thought," Google chief executive Eric Schmidt said. "There is no single platform strategy that will win out at the expense of others."

For its part, Google plans to integrate its Orkut social network to its main search services so that experts can provide answers to questions that can't readily be answered by standard searches. Intel and others, meanwhile, are fostering research projects that will allow people to conduct searches using images or audio clips instead of keywords.

Like established technologies, both Language Weaver and MetaCarta rely on probability to generate results. In a Spanish-to-English translation, Language Weaver first converts a Spanish phrase ("Que hambre tengo!") into a likely equivalent in broken English ("Have I that hunger!") by comparing how the sequences of words have been used in Spanish and English documents in a database. It then does another probability analysis to convert that sentence into standard usage ("I am so hungry!").

Language Weaver's database of documents for performing European or Chinese translations based on probability is fairly extensive. For Somali, the sources are more limited -- the company had to use the Bible.