deadsea - 4:11 pm on Oct 13, 2011 (gmt 0)
The bigram post by option1138 is quite interesting. He claims that certain pairs of words indicate spam on web pages.
Based on the number of legit emails I get that my email client marks as spam using its Bayesian filter, a big percentage of the web would get marked down as spam if a similar approach were taken by Google.
Lends a lot more credence to the "poison words" theory that I saw a thread here about last week.