In part, it's because it's a proof-of-work - there's no way to automatically do it. I've been musing about using word embeddings + dynamic programming to do just that, however: you take a list of all possible synonyms with their embeddings, and calculate every rewrite that results in lining up exactly, and then choose the ones with the minimum cumulative embedding distance between the original words & their replacement.

Wouldn't that just be done by a text editor?

Eg in emacs C-u 70 M-x fill-region. (There's probably a better way)