****
Haikus are easy
But sometimes don’t make sense
Refrigerator
****

DISCLAIMER: Programmatic syllable counting is a difficult problem. Occasionally, this work will generate sentences with one too many or one too few syllables. We have tried to make the syllable counting algorithm as robust as possible, but, due to the quirks of the English language, there will always be more cases that miscount. (And if you get 6 instead of 5, it’s value for money).

Originally made for January 28th, with the prompt “Generative Poetry”, this is a complete overhaul of the original throwaway Genuary project (https://www.fxhash.xyz/generative/24355) for #genuary32nd. This ground up rebuild features new backgrounds, a new method of Haiku generation, and a brand-new syllable counting function.

The Haikus are generated using a Markov Chain, which creates sentences based on a large database of haikus. The Markov Chain Method guarantees that the generated strings are locally sensible (as in, adjacent words will make sense together), however, the whole string won’t necessarily make sense. We feel that this whimsical sentence structure somewhat accents the style of the haiku, by providing (accidentally) artistic phrases. Afterwards, the strings are screened with a homemade syllable counter, which attempts to guarantee a 5-7-5 structure of the Haiku. The generated haiku is then displayed in a field of letters, with various color palettes and animations possible, which are displayed as fxHash Features. Try and collect the Legendary Matrix + Rain combo!

Credits to xiboon, and his excellent Markov Chain library, kurwov (https://github.com/xiboon/kurwov). This framework forms the backbone of the generation. The haikus database was taken from https://github.com/herval/haikuzao/blob/master/inputs/haiku.txt and https://github.com/sballas8/PoetRNN/blob/master/data/haikus.csv , and then cleaned by us.