UT Linguist Helps Computers Understand Natural Language [Watch]


They’re not taking over yet, but computers are rapidly becoming more intelligent.

Katrin Erk, a computational linguist in UT’s College of Liberal Arts, has developed a new method that will allow computers to better understand natural language and, more specifically, word meanings.

“The central point in my research is that … word senses are not so clearly distinct,” Erk says. “Sometimes they are far apart, but sometimes you have senses that really blend into one another.”

Even to humans who have a lifetime of knowledge regarding word meanings, figuring out each different meaning of a word can be difficult. Take the word ‘run,’ for instance. Though it’s a word you may think has only a few meanings, the Oxford English Dictionary lists more than 80 distinct definitions—and that’s just for the verb. All that variation leads to one very important question.

“How can a program figure that out?” Erk asks. “How can the computer understand when you mean run a race versus run the company?”

And that’s what Erk hopes to enable computers to do. Instead of teaching computers rigid, definite word meanings, she has created a way for them to visualize words in a high-dimensional space. That way, computers can use implicit connections between words to map relationships.

“If you can present these individual instances … as points in space,” Erk says, “then you can really just deal with all these different degrees of similarity and difference between the words.”

But don’t think this mapping technology is just for linguists. Similar technologies already help millions of people everyday via Web searches and the like. Now it’s set to be even more widespread, allowing for things like the automated summarization of articles and automated extraction of useful information in those articles.

“There’s actually a lot of ways in which a program that automatically understands text can help you,” Erk says. “In the long run, what I hope to help the computer do is to draw better conclusions from text.”

Video courtesy The University of Texas at Austin.

 

Tags: , , , ,

 
 

No comments

Be the first one to leave a comment.

Post a Comment


 

 
 
Menu