I semantics we have no measure of the ability to qualify partial matches. It was one of those things, words do not have prices otherwise banker bots would be reading. So the match process when translating must be well defined in a general convolution of two semantic nets.
Quantify the match function with an allotment of commutative power.
All grammars, we assume, have some sort of stem and pre/suffex system. Any given word can be categorized by how well and often it uses the designated stem/modifier system. The computer can read widely and build likelihood trees for stem variations, find a measures of variability on a per stem basis. The measure tells us how deep into a semantic net is needed before a negative match is declared. We have actually priced the link traversal which is what we want. Then our reading comprehension problem oils down to finding the minimal spanning tree that represents the translation.
No comments:
Post a Comment