NaNoGenMo
Feed Me A Stray Cat
Named after the scene from American Psycho... because this is probably going to be random, weird, and a little violent.
My program will generate a novel based on source material fed into it, using next-word prediction and style analysis with some adjustable settings for randomness and arc structure. The goal is to get it to produce something that is (at a minimum) readable; reasonably enjoyable; and to see how often it actually surprises me.
Had a thought today -- I'll need to develop some requirements so that I have specific metrics to hit. That way I'll know when I'm done and (hopefully) won't suffer from feature creep. Some thoughts to guide that:
- Word prediction generates a novel that looks like normal sentences/paragraphs/chapters, a la lorem ipsum, but with real words. First major milestone before any of the pseudo-intelligence can be implemented.
- Getting copycat sentences below a certain threshold. Check generated text against source material to reduce the length of unoriginal chains (straight copies).
- Avoiding repetition of uncommon words. Based on source material, give each word a uniqueness count (number of words since last occurrence) and then arbitrarily set an acceptable percentage of that value before it can appear again.
- Reducing the number of nonsensical sentences. Train FMASC to avoid certain combinations of words that have been flagged during consumption of source material. This might be a stretch goal, and also depends on how good the word prediction algorithm actually turns out to be.
Philosophical question: Is creativity really just a matter of finding that sweet spot between convention and chaos?
I'm fascinated by your idea of quantifying the success of your novel-generation process! I thought about calculating the "novelity" (?) of generated text before, but have never collected ideas on how to do that, and yours seem pretty solid. Would be interested to see your creativity-gauge code when you have it!
Today, it became smarter than me.
I couldn't figure out where the third occurrence of "the great" was showing up without adding some debugging messages, and was convinced I had made a mistake in counting/entering somewhere. Upon closer inspection, lines 194 and 195 of the txt file show a newline delimited instance that CTRL-F wasn't catching. So it works!