I guess you could skirt around this by using something to tag the various parts of speech in your original text (using something like Python's NLTK) and replace them with randomly picked synonyms from a thesaurus?

Pretty sure it would obscure the original writer although possibly at the cost of obscuring the original meaning.

If you use wordpress.com from Tor and use Google Translate from Tor, what do they learn more about you than just using wordpress.com?

(I have no clue)

I think what we’re concluding here is that using Google to obscure the linguistic style is flawed, because a state actor could obtain the original linguistic style from Google records, or from their own records of snooped traffic.

In other words: the blog should find a way to obscure linguistic style offline.

They can see the original text, which can the be analyzed.