At StartupSchool '08 Peter Norvig answered a question about Google's take on the Semantic Web. The question was accusatory in the sense of "Hey, why aren't you playing nice?" Norvig disparaged the need for structured data on the web, presumably because his natural language lab is good enough that Google doesn't need explicit annotations to grab semantics.

I think this technology is the same as Direct Answers from three years back. It could do capitols and geography facts back then. It seems there is a new module for genealogical relationships.

If by new, you mean three years old, then sure.

http://www.pcworld.com/article/120362/google_intros_qanda_se...

I'm not entirely convinced that's the same thing. The examples don't match up. Maybe, though!

Well, three years have passed. I wouldn't be surprised if it's just the natural progression of the prior tech.

Here's a Yahoo BOSS mashup that does this without any semantic technology: http://bossy.appspot.com/qa?query=who+started+ycombinator

still needs some work...

http://bossy.appspot.com/qa?query=what+is+the+capital+of+flo...

This kind of structured data is the next wave, you shouldn't associate "semantic" with W3C's semantic web stack. That's just one possible vision. Microformats is another. The reality is probably somewhere in between.

Google explicitly believes in pulling semantics without explicit structure. It's what people do, and Google's getting really good at it. Peter Norvig claims to have the best natural language processing lab in the world and I'm inclined to believe him.