I really don't see this replacing "insider trading" anytime soon... especially since Reuters is disseminating the same information (presumably at the same time) to every algorithm hooked to their feed. I've toyed with latent semantic analysis for pruning unstructured news articles and creating a structure of the semantics. It works much better than phrase searching and there are ways of using it to determine a bias in news with probabilistic models of archived stories. http://lsa.colorado.edu/papers/dp1.LSAintro.pdf But computers are dumb... Speech recognition software, for example, has no understanding of what's being said (think about John Searle's Chinese room). Natural language comprehension is still impossible with traditional AI.