Date: 10/28/08 (Algorithms)    Keywords: no keywords

A well-known algorithm for generating humorous texts using Markov chains is exist.
For example:
http://www.eblong.com/zarf/markov/
This algorithm is very simple and it is working only at word level.
What if I would like to have such algorithm, which works not only at word level, but also at sentence level.
It can be said that this algorithm remix words randomly. So then, I would like to remix sentences first, then words inside of each sentences.
At top level, statistical table for sentences will be build. At lower level: table for words.
But how to connect these two levels?
Probably, something already exist in theory, not just for humorous texts.
Thanks in advance for any comment about where to read more.

Source: http://community.livejournal.com/algorithms/100435.html