Saturday, May 20, 2006

Natural Language Understanding is Hard

Sorry for the gap in posts. I will try to get more consistent.

If you ever wonder why it is so difficult for computers to understand natural language, check out this article by George Miller, creator of WordNet. In one example from his article, he points out that the following couplet from a Robert Frost poem . . .
"But I have promises to keep, and miles to go before I sleep."
. . . has over 3 trillion possible meanings.

3 Comments:

At 8:41 PM, Anonymous Anonymous said...

> . . . has over 3 trillion possible meanings.

Isn't that a bit melodramatic? Those numbers include 3 meanings of the word "I" and 5 meanings of the word "And", which seems a bit artificially high.

 
At 11:11 PM, Blogger John De Oliveira said...

Yes, I'd agree it's a little overstated. But parsers that attempt to convert English to something like predicate calculus choke way before even a few thousand interpretations. So, it it's a trillion or a billion or a million -- it's out of reach without taking just the right approach, including defining context as precisely as possible up front and adjusting it as more is known.

 
At 3:24 AM, Anonymous Anonymous said...

Why does George Miller assume language is made up of words with independent meanings, which must then be combined to create meaning in a text?

 

Post a Comment

<< Home