Considering that the vast majority of 'research reporting' involves rephrasing press releases, this is a difference without a distinction. While the article is

Considering that the vast majority of 'research reporting' involves rephrasing press releases, this is a difference without a distinction. While the article is about generating abstracts rather than press releases, I don't see why the same approach wouldn't work for both problems. I'm sure a neural net could be trained to write in press release style instead of abstract style.