This is a little experiment with Markov chain text generation. The code is from shabda (with a tweak to always end with a period), and the corpus is most of the text I posted to the WELL during the time I was most active, mostly from 1997 to 2000.
Chances are that isn't what's going on here! database sends it a bit harder, but the comparison is applicable.