
139 - Coherent Long Story Generation, with Kevin Yang
NLP Highlights
00:00
How to Evaluate a Story on a Computer
We asked humans on mechanical Turk did this we just like gave them two different stories that are based on the same premise and then asked them to look to like read over both stories quickly and then evaluate. There's a few there's a few metrics we asked the end dates like which story is more interesting or coherent from the perspective of the high level plot. The results were definitely a bit noisy because it's on mechanical Turk and we didn't have any great ways to like enforce them but even so you can still see quite significant differences between using our system and other systems.
Transcript
Play full episode