Comment Mixing a couple of capabilities here (Score 1) 134
There's really two different capabilities being discussed here. One (the Northwestern example) is the actual generation of prose from an underlying data asset. There are certain well structured domains of information (baseball games, earnings announcements, etc) where this will most likely work quite well.
The second capability is automating the analysis of new content. NewsScope falls into that category. It takes raw news (written by humans) and extracts key terms, entities and events to make that content more easily consumable by machines. If you're interested you can use the same Thomson Reuters tools that are under NewsScope on your own content. My site uses them to analyze news from feeds, throw most of it away and put the rest in the right places. Thomson makes this capability available to anybody for free at a project called OpenCalais (see http://viewer.opencalais.com/ to play with it). Another group has built it into a complete publishing platform called OpenPublish.