Ah, it’s time to roll up our sleeves and dive in to another issue of “Intractable Problems”. Let’s brush the dust off our thinking cap and see if we can make some sense out of the language we all speak. In this issue I want to look at some of the ambiguities and complexities of our natural language.
We care about language because our computer applications must interface with people. We are frequently asked to find documents on certain topics from a database, extract information from messages or articles on certain topics, translate documents from one language to another, or summarize a thousand page proposal into two or three pages.
Today, most of our application systems solve these problems by using simple word matching techniques. They do not understand the language. While such techniques may produce useful applications, it is unlikely that they can be extended to handle complex query tasks, such as the query “Find me all the articles on software projects involving more than 10 million dollars that were attempted but failed during 1990 and 1995.” To handle such queries, the system needs to read the articles and extract enough information from them to understand which articles meet the criteria. A crucial component of understanding involves computing a representation of the meaning of sentences and texts.
If I said to you “We painted all the walls with cracks.” would you think that cracks were painted onto the walls, or that cracks were somehow used as an instrument to paint the walls?
Here is another one for you, called a garden path sentence. “The raft floated down the river sank.” When you read the word sank, you probably realized that the sentence interpretation that you had constructed so far in your mind was not correct. Yet the sentence does have a reasonable interpretation corresponding to “The raft that was floated down the river sank.” The initial sentence is grammatically correct, it simply uses a reduced relative clause for the word floated. Go back and read the sentence again and you will understand.
For another one like the above, try “The horse raced past the barn fell.”
How about the sentence “I thought it would rain yesterday.” Most of us will interpret this as yesterday was when it was thought to rain, rather than the time of thinking.
These are all examples of ambiguous sentences. Any natural language is full of ambiguity. I suppose, in some sense, it’s our ability to understand language ambiguity that makes us human. At least that’s what I try and believe whenever I listen to politicians, economists, and other fortune tellers.
The problem with language ambiguity is that it stops us from making the appropriate inferences needed to understand. Ambiguity resolution in natural language is an intractable problem.
It’s quite paradoxical to use language to understand how little we know about understanding language.
This leads me into our next issue’s topic on interesting paradoxes. If you thought that you were tongue-tied with language, then I promise you that your brain will be tied in knots with these paradoxes.
1. Allen, J. Natural Language Understanding, 2nd edition, Benjamin Cummings Publishing Co., 1995.