Interesting topic, unfortunately not a lot of substance in the article. Reminds me of the news piece a while ago : Google's Artificial Intelligence Translates Poetry
http://www.npr.org/2011/01/16/132959095/googles-artificial-i...
http://www.npr.org/2011/01/16/132959095/googles-artificial-i...
I could go into much more detail, but I think my biggest point was the necessity of moving towards a bottom-up approach. Text generation by itself is too advanced - it's like trying to teach a child physics before elementary science. There's root work to be programmed in first.
But thanks for your feedback!