There are some things which computers can't do too well because they are missing a human understanding of context and how the world works. Since computers tend to rely on syntax to "understand" language, but don't have any concept of what would be the most "logical" interpretation of a given sentence, they tend to fail at this point. I guess my point is that you need some knowledge about the world to be able to intelligently disambiguate synactically ambiguous sentences. What most linguistic researchers have done up to now is use a hand-tagged training corpus to disambiguate and give filter out unlikely (atypical) interpretations of a syntactic structure.
The idea of dealing with typical cases is basically what you're suggesting here, I think.