Words are represented by postags in rules. Every word one postag.
But there are many multi-word words, as wel as single ‘words’ representing multiple postags (abbreviations).
So I think LT could benefit from an extra layer of tagging. Maybe even multiple layers (tree like)
Think of noun groups, adjective groups, adverb groups.
Of course code would be needed to dissect sentences in groups like these. Having a list of valid groups, this should be feasible.
I performed a small experiment. In this, one metatag sentence validates 130000 valid postagged sentences.