I completely agree. Having worked with some AI systems in the past (mostly ANNs although I'd love to even begin to understand SVMs), I think that there's more "practicality" in just building robots/software that (based on various pattern analyses/etc/etc) figures out where a malignant tumor is, for example, rather than a robot that can understand coherence relations or what-have-you.
So I don't fault the CS community from stepping away from linguistic puzzles and getting on with more useful stuff. After all, that's what engineering is. It looks like Levesque disagrees, but I'm not sure he's right.
SVMs are lickety-split simple. You're drawing the best line/plane/hyperplane between some points. You can turn this into a nice convex optimization program, given some conditions. If this thing isn't fully separable, you can fudge it a little with some penalty terms, or you can cheaply project these points into some space where such a separation does exist. The hard part will always be your feature extraction, labeled data collection, and the parameter tuning for everything I just waved my hand at.
So I don't fault the CS community from stepping away from linguistic puzzles and getting on with more useful stuff. After all, that's what engineering is. It looks like Levesque disagrees, but I'm not sure he's right.