proposed improvements

[Cedric Foll]

I'm using bogofilter to class web site between porn/no-porn.
I've coded a preprocessor which do the pair (and few other tricks).
In fact my preprocessor return
A_B B_C C_D
when the entry is 
A B C D

The improvement is great after having put s to 0.1 (without that to much
bogosity become 0.5 because of the number of unkown pairs).
My db size has been multiplied by 4.

If someone is interested i can put my preprocessor on the ML.