Spams using a new trick

I use Fastmail for my email, and I'm very happy with them. They recently updated their anti-spam filters with a new Bayesian one. It was catching virtually all of the roughly 125 or so spam I was getting each day.

I've noticed over the past few days that most of the spam that gets through is using the same trick. The message includes a paragraph of randomly selected real English words. Those "non-spam" words must be given enough credit in the message to overcome the normal spaminess. My lastest spam used this list:

igor prague vestry centroid asylum butt mathematic plentiful inhabitation puppeteer sybarite pittsburgh carney delete playwright noun teamwork vestal vote diatomic crouch drank crotch widen damsel ephesus construct bail amplify dupe sung intermittent sparrow what'd yard metzler gross collector ditzel cascara merrimack pour cheek quirinal carolinian mcelroy spatula splay actress crease bequeath turbulent crotch antithetic delicious bayed

You could detect that the words were not in a sentence by checking frequencies of pairs of words. However, it seems pretty clear that the next step will be to randomly select sentences or paragraphs from Public Domain books. How the heck are we going to detect that? They can make this section arbitrarily large, to improve their ratings. I bet that spam of the future will be a unique combination of penis enlargement and classic literature.