Almost every adult in the world can speak and comprehend speech, but most of us have no understanding of or appreciation for quite how amazing and complex this capability is. I'm not a linguist, but I've studied a lot of linguistics as part of my AI research and I'm constantly impressed by how intractable the problem is. Some people have made completely incredible and impossible claims regarding computers and language, but for the most part we really have no idea how the human brain does what it does.
Much of the difficulty in language stems from the need to disambiguate words and phrases that can have multiple meanings. Human language is incredibly redundant; there are a hundred ways to say the same thing, and any given word or sentence can have a hundred different meanings. Our brains learn (over the course of years, but still at a startling rate) how to perform these tasks, but we're still not able to teach (program) our computers to do the same things. (And I doubt we ever will be.)
Neal Whitman, who apparently is a linguist, presents an excellent example of just how confusing language can be. AI is a long way from being able to learn/generate/comprehend this type of result (other than with a rule-based approach, which is impractical), and we need more than an accumulation of baby steps to get there -- we need some dramatic breakthroughs that will radically change the way we think about natural language processing, because I think we're on an entirely wrong track.
Comments should be working again.
Hey,
Great entry! I'm a student at the University of Georgia which has a great, new, A.I. department. They're really making great strides over there. I was just currious, what made you decide to post on this topic? Something in the news?
-Will
www.wkirby.com
Will: I'm an AI grad student at UCLA, and I thought some people might be interested in how AI relates to what Neal Whitman wrote about linguistics.
michael, could you post more often about a.i.? this entry was very interesting.
s: I can post more about AI, sure. Give me some questions and I'll be happy to answer them to the best of my ability.
An excellent illustration of the complexity of the language problem.
I've worked on NLP for some years now, and it all comes down to context, I think. I can't see the point of dreaming up new ways of tagging parts of speech, building complex grammars etc, when the problem is nothing to do with sentence structure and lots to do with the meaning of key terms in the conversation context and the relationship between those terms.
Unfortunately this is well beyond the scope of AI at the current time (well, apart from Jim Wightman, of course).
That's not to say, though, that NLP can't be commercially useful. I believe it can, and is, even if we're only on the first step of the ladder. It's just not anywhere as good as we'd like it to be - but then nothing really is!