It's funny. Lately I've been working with NLP systems and in the last few years there are a few really good parts-of-speech taggers that are about 99% accurate. All the ones I know of are based on hidden markov models, which definitely would disappoint Chomsky.
Part of the trouble w/ Chomsky is that real language doesn't draw a clear line between syntax and semantics. Even though an HMM doesn't correctly model the nested structures that are common in natural language, it makes up for it by encoding semantic information.
Another trouble is that human beings are innately probabilistic when it comes to language. A sentence written/spoken by humans does not have to be gramatically correct, to convey it's meaning, and does not always follow the strict rules that Chomsky talks about.
It's not the language that defines how we communicate, it's how we communicate defines the language.
But I also disagree with peter when he says the why is not important, it is this why or the understanding of the matter that separates us from the machines like watson, since our sole purpose in life is not to win at a game, but play/enjoy the game and most importantly "reuse the understanding" gained in some other facet of life, a feat that I beleive no machine is capable of.
>It's funny. Lately I've been working with NLP systems and in the last few years there are a few really good parts-of-speech taggers that are about 99% accurate. All the ones I know of are based on hidden markov models, which definitely would disappoint Chomsky.
No, it wouldn't disappoint him at all. In fact, one of his earliest works in linguistics discussed how transition probabilities could be used for chunking and categorization. (See http://www.tilburguniversity.edu/research/institutes-and-res... ) It's not as if Chomsky ever presented part of speech tagging as a poverty of the stimulus argument.
Part of the trouble w/ Chomsky is that real language doesn't draw a clear line between syntax and semantics. Even though an HMM doesn't correctly model the nested structures that are common in natural language, it makes up for it by encoding semantic information.