Can you determine which articles in a corpus of news, or other documents, agree/disagree with one another? Can you cluster and sub-cluster along statements of fact being made? Can you then trace a statement of fact back to it's "original source" based on when that entry was acquired?
There's a lot of opportunity in that space...