Usefulness of categorizing words based on occurances in different parts of the MS?

[Anton]

So this is an interesting idea, but I think you would struggle to usefully assign semantic values to the words. There's no way to know what any given word might mean and little internal validation for the guesses you'll be forced to make.

With the discussed approach, one does not need to assign semantic values, it would be enough to classify each vord into one of several broad categories based on pre-defined quantitative criteria.