| Who Needs Berlitz? |
 |
Part 4: The Quest for a Universal Translator
|
|
|
|
| Join the Discussion |
"Come si traduce 'ain't' in italiano? È inglese o americano?"
AERDNA74
|
|
|
|
|
|
|
|
|
|
|
| |
|
Mary Harper, who develops computer-language models at Purdue University, points out that even human beings can hear the exact same sounds in several different ways depending on their expectations.
Researchers are fond of the phrase "how to wreck a nice beach" or is that "how to recognize speech"?
Given the ambiguities of human language, she suspects it may never be possible to craft flawless computerized representations of what we mean. But artificial-intelligence expert Jack Mostow of Carnegie Mellon contends that even severely limited language-recognition programs can be surprisingly smart.
Synthetic interviews, which present relevant selections from a previously filmed interview in response to spoken questions, consistently leave interlocutors thinking they have talked to a real person. Judge for yourself: Grand Illusion Studios has demonstrated synthetic interviews with Albert Einstein.
So Mostow created the Reading Tutor, a computer program that coaches children in reading by using a variety of conversational tricks. The software follows the child's spoken words and makes encouraging noises to give the impression it is listening carefully.
It also takes into account how fluently or haltingly a student speaks in addition to whether individual words are correct. In a small pilot study, students advanced an average of two years in reading skills over eight months.
Next Page > Talking and Listening Computers > Page 1, 2, 3, 4, 5
Newsletter
