Your example of word proximity analysis does not necessarily solve the problem, does it? You still want to be able to type "Jaguar" and have Google know you mean cat, not car. There's no guarantee that even with proximity analysis the results would be any different than what you're getting now. Or am I missing something here?

Hand-optimization, IMO, is a "bad thing" inasmuch as it essentially breaks the Google model. Some queries are "optimized," some are not; the algorithm used in one search is not the same as that used in others. Moreover, the human touch implies judgement by humans who _still_ might not see the world the way that people do who are actually doing the searches. And that's assuming that Google could even keep up with their Top 100 searches, which surely change by the second.

I stand by my statement that Google already has powerful syntactical tools that can help you find dang near anything, and that a two-word search will winnow your list by a huge percentage. Based on the Google searches that bring people to my blog (/BlogGoogleSearches.aspx), it seems that people frequently type MORE, not less, than they need. Perhaps that's a self-selected audience, but still.

I guess I'm in the minority here, but I don't actually see much of a problem. Given the a) incredible syntactical tools that Google allows you to construct your search with and b) the Google API that anyone is free to leverage and, say, add their own front end, it just isn't that hard to zero in on what you want. When I look something up and it isn't first or second in the list, I reflexively look at the number of hits, and it it's in the gajillions, I refine my search. Add one more word to your search (any word remotely connected with your topic) and you're golden in 99.999% of the cases. (Statistics (c) 2005 Mike Pope, any resemblance to real people or numbers strictly coincidental.)

Don't forget that all Google is really doing is showing you the results of a popularity contest for your term. Just because YOU didn't mean "Jaguar-the-mispronounced-car-name" doesn't mean others -- the majority of others, poor blokes -- were not searching for overpriced vehicles.

I'm also interested in how you think Google could actually implement categories. Google has no metadata about content except the stuff that goes into the PageRank system, and they're already using that. If I understand correctly, any after-the-fact categorization in any search engine is done with human intervention. (?)

Incidentally, there's a certain irony here in that Google's initial success was precisely that their ranking algorithm was spookily prescient about what you meant, as distinct from engines that weighted pages based on, say, word count. We sure have become spoiled ... :-)

PS Google Suggest would in this case not help -- it wouldn't be until the second word that the search would be sufficiently refined ... other than that you'd know it before hitting I Feel Lucky, I guess.

Your example of word proximity analysis does not necessarily solve the problem, does it?

I think it does, since that's how http://www.clusty.com appears to work..

Well, I typed "Jaguar" into Clusty and the first hit was for the car. They pull up Wikipedia's (not their) page as a disambiguator -- is that the result of the proximity analysis?