The NLP Wall
June 12, 2012 7:34 AM
What are some good references (papers or books) that address the difficulty of computers to understand natural language?
When Chomsky and others published their theories a half a century ago, it looked like we were well on our way to computers that could understand natural language. This turned out to be WAY too optimistic. Presumably the core difficulty is that so much context (in the form of human experience) is needed to fully grok even a simple sentence.
I'm looking for reasonably scholarly accounts of this huge wall that natural language researchers hit in the 1970s.
When Chomsky and others published their theories a half a century ago, it looked like we were well on our way to computers that could understand natural language. This turned out to be WAY too optimistic. Presumably the core difficulty is that so much context (in the form of human experience) is needed to fully grok even a simple sentence.
I'm looking for reasonably scholarly accounts of this huge wall that natural language researchers hit in the 1970s.
It is a little old now but "Hal's Legacy" got a number of academics to write about which aspects of 2001 a Space Oddysey had taken off technologically, and which ones had not. It includes some discussion on the difficulties of natural language recognition.
posted by rongorongo at 8:34 AM on June 12, 2012
posted by rongorongo at 8:34 AM on June 12, 2012
Jurafsky and Martin's _Speech and Language Processing_ may be a good start. I don't have my copy handy, but it has plenty of references to papers about the history of NLP techniques. (It's an excellent textbook, too.)
posted by silentbicycle at 8:46 AM on June 12, 2012
posted by silentbicycle at 8:46 AM on June 12, 2012
Although this is more popular than scholarly, you might enjoy David Auerbach's "The Stupidity of Computers" from n+1. He explores the history of trying to teach computers how to understand natural language, as well as various pitfalls encountered along the way. You can read part of it online, although (sadly) much of it is behind a paywall.
Personally, I find buying issues of n+1 worthwhile but YMMV.
posted by brackish.line at 10:49 AM on June 12, 2012
Personally, I find buying issues of n+1 worthwhile but YMMV.
posted by brackish.line at 10:49 AM on June 12, 2012
« Older Should we break up? The middle aged version... | Looking for a family doctor/pediatrician for... Newer »
This thread is closed to new comments.
If I tell my phone "text my wife and tell her i missed the six o'clock train, I'll be on the 6:44" it can handle the task reasonably well
If you're talking about being able to program using standard english, well, there's a large variety of high level / scripting type languages that are pretty close, but the "problem" is always that computers do what you tell them to do, not what you want them to do.
If I wanted to write a script that combed through my contacts to find all of my "friends" that I haven't talked to in a "while", I'd still have to define a group for "friends" and a time period for "while".
There was a great article I read in a programming class about this a little ways back basically saying that languages like Ruby and Python are awfully close to "natural language", but still require you to be specific and think out the actual steps for what you're trying to do.
posted by Oktober at 8:10 AM on June 12, 2012