Markov chaining is a way of putting together words that matches the probability they would go together in real life. For example, it is a good bet that the words 'I love' will be followed by 'you' (especially when said by me in your presence), even though they could, in theory, be followed by 'elephant poo'(?!). This site uses Markov chaining to generate all-new Garfield strips. Cool idea.
My nonsense-generating program, JanusNode (available from JanusNode.com) also uses Markov chaining.
