A friend who works for a Washington-state software company of some reknown, and who has worked on natural language processing for them for many years, said (with permission to share it): ------------------------------ I actually worked on exactly this problem (the sentiment analysis part, not the scarry spooky stuff) for a couple of years, until very recently. It works fairly well - if you're willing to miss out on a few low-confidence cases, you can get pretty good precision in evaluating the sentiment of a sentence / article. (For obvious reasons, it's important to businesses to be able to track what people like & don't like about things). I don't really know that we're back to 1984 though. I mean, they're talking about monitoring publically available sources, i.e. things that people published. If they start running this on my e-mail, of course (and given events in the recent past this doesn't seem completely unlikely), then I'm not particularly psyched about it, but it's the fact that they're doing ANYTHING on my mail that I'm unpsyched about, not the sentiment analysis in particular. ------------- See also: http://research.microsoft.com/~anthaue/pulse_paper.pdf -eg