I'm trying to figure out what I'm looking at. A quick look at what Markov Chains are doesn't really seem to explain it. This is a mashup of these two books, basically, correct?
[it, is] -> {sunny, 0.75}
[it, is] -> {raining, 0.25}
Once you've got that, you can use it to generate random text that has similar characteristics to the training corpus. You just seed it with a couple starting words, and then start randomly choosing what word comes next according to the probabilities you've recorded.
jwz's got one that you can play with yourself at http://www.jwz.org/dadadodo/