tl;dw—Vonnegut plots the well-being of characters over time, and claims that there are a few basic sine-wave-esque shapes that emerge over and over.

The Shapes of Stories with ChatGPT - Superb Owl

submited by
Style Pass
2023-03-18 16:30:04

tl;dw—Vonnegut plots the well-being of characters over time, and claims that there are a few basic sine-wave-esque shapes that emerge over and over. You can also check out the extended version.

I’ve always wanted to test this claim empirically, but struggled to do so. Old-school NLP tools (like sentiment analysis and named entity recognition) don’t seem to be up to the task, for reasons I’ll explain below. Fortunately, LLMs like ChatGPT are far more sophisticated!

I used the ChatGPT API to get ratings for the well-being of characters in a few dozen works of classic fiction. You can play with interactive graphs of the data, or keep reading to learn how they were made. You can also jump down to the Results section of this post to see the highlights.

In my first attempt, I ran sentiment analysis on each passage of the text. A very naive sentiment analysis algorithm will simply look for words like “good”, “bad”, “happy”, “terrible”, etc. and provide an overall score. More sophisticated methods can take things like negation or sarcasm into account.

Leave a Comment