mwilkins@jarthur.Claremont.EDU (Mark Wilkins) (03/17/90)
In article <5315@mnetor.UUCP> frank@mnetor.UUCP (Frank Kolnick) writes: >Ah, that makes me feel better! I don't have Grammatik, but I have both >Sensible Grammar and MacProof. The above description could be applied to them >as well. I tried both on various manuals and a book I was writing, and >eventually gave up. To sort out the poor advice from the good, you have to >know more than the program does in the first place. I now doubt whether >any program (now or in the near future) can analyze style and/or grammar. Well, the approach I would take if I were to attempt a new and innovative grammar-checking program would be to create a simulated (or hardware-implemented) neural network. An example of a similar project was done by a man named Sejnowski, who created a system called NETalk, which was able to read aloud, and in fact learned almost all of the common pronunciation rules of English in about sixteen hours. At any rate, a grammar-checking system could work in much the same way: the system would be "trained" by the programmers and the rules solidified, then the distributed program would use the neural net to apply the rules. One question would be whether such a thing could be efficiently implemented in software. Assuming that the minimum acceptable rate of checking is about the printing speed of a laser printer on 12 point text, or 3000 words per minute. Could it be done on a small computer like a Mac? Followups to comp.ai.neural-nets -- Mark Wilkins mwilkins@jarthur.claremont.edu