Google Labs is doing some interesting things.

http://labs.google.com

I especially like the "sets" function. It gives the illusion of generating -- or rather, it does generate -- a real semantic network, producing items that are near each other in a real-world human meaning system. It's an illusion because the machine, as I understand it, has no model of "meaning", and no semantics built into it at all. Such models have always been insanely hard to create (or very specialized, and very hand-crafted). Much work was done, with no great success.

This "Sets" tool just skips the idea of meaning altogether. If items are found together in the documents and lists that humans make (the ones they put on the web, anyway), then if we sift the documents the right way, our method can suggest things that might group together meaningfully, _for humans_. The computer doesn't have any claims to meaning; it just has this method. And it's nothing but a method, there's no oversight of the results -- they are often bad. Try typing in the words "crane" and "shovel". What works well is to type in a few names; do the names of a couple of jazz greats.

The Google Glossary is maybe less interesting theoretically, but it's actually damn useful.