Okay when I wrote the post the other day questioning whether the current Semantic Web approach was the way to go I didn't know about this. But thanks to El Reg I've now got a nice reference point to what I mean. The rather smart chaps at Google claim to have the best human language translators around, even for languages that they don't understand. Now in comparison with matching an address line to another address line the task of doing human to human seems to me (with my one double unit at university on semantics, linguistics and computer representation of knowledge) to be a much bigger challenge.
So are formal ontologies the way to go, or smart contextual searches? Me I'm backing the searches.