> That those errors cannot be automatically detected and eliminated is actually relevant to the overhyped semantic web, no?

No. That would require your computer to understand sentences meant for humans; the whole point of the Semantic Web is that it's machine-readable.

If you could make it understand sentences meant for humans, it would be machine-readable!

This was the topic of my degree thesis, but I gave up on it (far too complex to parse everything, and finding useful applications is difficult).

In fact, I think the Semantic Web has application problems. They say things like auctions, but I don't see what's wrong with SQL for those.

And if you used Google to search the web for millions of currently-listed auctions on webpages in XML format, you'd have a difficult time trying to find what you want.

> They say things like auctions, but I don't see what's wrong with SQL for those.

For local data sets, SQL's great. For remote, centralized data sets you can use something similar.

But once you want to combine data sets spread out over several sites (say, combining IMDB with local movie listings) you need to express that data in a format with globally unique identifiers and distributed extensibility. That's RDF. You can still have a nice query language, see SPARQL (which is much nicer than SQL, IMO).

> And if you used Google to search the web for millions of currently-listed auctions on webpages in XML format, you'd have a difficult time trying to find what you want.

Sure, if you're just doing full-text search like Google does. Of course, the point of having machine-readable data is that you're not limited to full-text search.