A Markov Chain seeded from the 2008 submission to the Blender of Love.

Make new love poems by picking the next words seeded from the two most recent words. (or let the computer pick “random”-ly). The more frquently the next word completes the two current words the larger it shows as a selection.

This is a very primitive version of how LLMs such as ChatGPT (which helped write this version) work. They have an incredibly richer sense of what words/syllables/concepts are related, but mostly they're just thinking of what sounds most likely to say next given the current context.