I wonder if you could apply this to training data. Like here is an example of a common mistake and why that mistake could be made, or here is a statement made in jest and why it could be found funny.