I really like GeekDad. It’s a great blog that I’m sure I’ll use when I have geeklings of my own running around the house with lightsabers. But the story Why Watson’s Jeopardy! Win Is Mostly Meaningless is rife with bad thinking. I would assume a lesser site was trolling. That’s how bad this stuff is:
Programming a computer to parse Jeopardy! “answers” is of course the only real advance here, in terms of natural language processing, which is very cool. However, as it’s confined to the form of the show’s clues, which aren’t all or even most written in normal conversational English, all the Watson team has really demonstrated is that a genetic algorithm can be effectively applied to a highly restricted form of language processing, which isn’t really that impressive. I mean, there are many thousands of previous Jeopardy! clues and correct responses that can be used to hone the algorithm, since the way the show’s clues are crafted hasn’t changed much over the years.
What he means:
- This isn’t really parsing language; it’s parsing language that Jeopardy uses.
- Language that Jeopardy uses is not real language.
- I mean, there are thousands of examples of Jeopardy language Watson could have practiced on.
This is colossally stupid. Jeopardy language is a particular subset of language. So is the language I use when I order a pizza, or call tech support. Or use on a quiz show. Interactions in a singular context are, of necessity, reliant on a particular subset of language.
As to the second point, I thought Jeopardy’s clues were difficult to understand specifically because the language used is often couched in verse or obscure allusions. I don’t know how you can make the claim that Jeopardy’s clues are somehow easier to grasp than ordinary English with a straight face.
Except by resorting to the third point: there are so many examples of Jeopardy language that Watson could learn from, that an algorithm was as much an inevitability as an innovation. If that statement is true about this fabled, transparent “Jeopardy language,” than the statement is a thousand times more applicable to the English language. For every book of Jeopardy-language clues Watson could have practiced on, there is a library of English language to practice on.
Reasonable minds can disagree, but to call this meaningless is at best intellectually dishonest, and at worst, simple trolling for hits. Either way, I expect better from Wired.