<$BlogRSDUrl$> Marcus P. Zillman, M.S., A.M.H.A. Author/Speaker/Consultant
Marcus P. Zillman, M.S., A.M.H.A. Author/Speaker/Consultant
Internet Happenings, Events and Sources


Friday, September 17, 2004  

Noisy Channels Models Provide Short Answers to FAQs
http://www.economist.com/printedition/displayStory.cfm?Story_ID=3127462

Eric Brill is guiding his Microsoft research team to build a system capable of providing 50-word answers to questions such as "What are the rules for qualifying for the Academy Awards?" by using an approach called a "noisy channel model." Noisy channel models, which are already used in spell-checking and speech-recognition systems, work by modeling the transformation between what a user means (such as the word he or she intended to type) and what he does (the garbled word as it was actually typed). By analyzing many pairs of correct and misspelled words using statistical techniques, it's possible to predict how such transformations work in general cases, and to design a system to work the process backwards, so that, given a misspelled word, the system can guess what that word is most likely to be a misspelling of. Brill's question-answering system works in the same general way. Many question-and-answer pairs exist on the Web in the form of FAQs ("frequently asked questions"), and Dr. Brill trained his system using a million such pairs to create a model that, given a question, can work out various structures that the answer could take, and then to use those structures to generate search queries. The matching documents found on the Web are then scanned for possible answers to the question. This has been added to the articles section of Deep Web Research Subject Tracer™ Information Blog.

posted by Marcus Zillman | 4:15 AM
archives
subject tracers™