A team of computer scientists from Microsoft Research and colleagues from Cornell University and the University of California, Irvine are trying to determine if it is possible for computers to know what humans are going to say before they finish their sentences.
The researchers devised a series of machine-learning algorithms for analyzing and understanding sentences, which use Good Turing frequency estimation. They trained the algorithms on the archive of all material published by The Los Angeles Times from 1985 to 2002, which consists of approximately 1.1 billion words, and presented them with a series of comprehension tests.
The algorithms successfully filled in the blanks 53 percent of the time when working with SAT questions and 52 percent of the time when presented with passages from Sherlock Holmes novels. "Encouragingly, one third of the errors involve single-word questions, which test the dictionary definition of a word," the team says. Those mistakes could be reduced with a few tweaks, the researchers believe. Still, about 40 percent of the errors were associated with some level of general knowledge.
From V3.co.uk
View Full Article
Abstracts Copyright © 2012 Information Inc., Bethesda, Maryland, USA
No entries found