How is grammar processed with a transformer?

I am a computerlinguist working on grammar. I do not understand completely how a transformer processes grammar. Is it true, that grammar is solely processed due to positional encoding? Because, without positional encoding, you can change the order of the words without any effect.

The positional vector is small compared with the word embedding vector. On which reacts the transformer more: on inserting a new word, perhaps an additional adjective, or on a reordering of the words?

And can the words be reordered randomly without changing the output much?