jps@cat.cmu.edu (James Salsman) (06/27/89)
I am searching for neural-net algorithms for spelling correction. Given a spelling, the net should indicate whether it is a vaild word, and if not, suggest plausible alternate spellings. I would like to know how big a net would be required to perform this kind of task. Also, speed and space comparisons to traditional programs that suggest alternate spellings (like "ispell") would be interesting. There exist many human neural-networks that are very good at this task. I am not one of them. This is such an obvious Neural-Net/Associative Memory task that there must be something in the connectionist literature, somewhere. :James -- :James P. Salsman (jps@CAT.CMU.EDU)