The Hainan example in the article is especially impressive, the generated summary uses completely different expressions compared to the source text, yet it is s

The Hainan example in the article is especially impressive, the generated summary uses completely different expressions compared to the source text, yet it is spot-on. Of course those are probably cherry-picked results, but still. As a side note, it would be interesting to see how the algorithm performs with longer sources.