Predictive Punning

I tell many, many, many bad puns, as anyone who has hung out with me knows. What many may not know is how much preparation and mental remapping has gone into this process.

The two key factors in the success of a pun are Timing and Obscurity*.

By Timing, I mean that the pun has to be said close enough to the sentence it is riffing on so that the short term memory of the listeners is willing to go back and look and compare, to find the humour/reference. If you wait too long, you risk your listener timing out** and ignoring you, as they have already forgotten the specifics of the original sentence. Too soon, and the listener has not finished understanding the meaning of the original sentence, and the pun sentence will pass them by.

By Obscurity, I mean that puns which are too obscure will cause the listener to think about the pun for a brief while, then time out and move on. Puns which are too obvious will cause a groan as the pun wave collapses, and the listener will move on. Only a pun somewhere between these, where the listener is subtly forced to engage their brain will get the reaction you desire***.

Complicating matters is that Obscurity is defined differently for each listener****, as each listener will have different amounts of knowledge in each area. So, you’re constantly juggling what you know of the knowledge levels of each of your listeners, and trying to find puns that will fit inside enough of the Timing and Obscurity windows of your audience.

What can help is Sentence Prediction. Just like Amazon can tell that you will need toothpaste before you do, you can predict what words someone will say in a sentence before they know themselves. Once a person has started a sentence and is about halfway through, it is remarkably simple to predict how they will finish the sentence*****. More importantly, it is easy/possible to predict the exact words****** they will use, as you will need the exact words they will use in order to generate your pun.

So, you’re listening to someone speak. Partway through their sentence, you fill in their sentence with what they’re going to say. You then spend the next couple/few seconds planning your pun, you wait until they’re done speaking, and then you strike! Mental chaos ensues! Coyote is happy.

Want to hear more about this? Let me know in the comments below!

*I use ‘Obscurity’ instead of ‘Difficulty’ here because a specific pun will have different ‘Obscurity’ levels for each listener, depending on the specific shape of their knowledge/experiences.

**I’m using ‘Timing Out’ in the sense of the computer term ‘Timeout‘, where after a certain defined period of time, the computer will simply go and do something else. If you want a great example, watch two cats interact. You will see one or both of them timing out on a regular basis.

***If they start hitting you, you’re probably doing this right. Or wrong. It’s all the same. Dada is the anti dada.

****Timing is probably different for each listener as well, but I haven’t studied that in as much depth.

*****I’m sure someone has studied this, but I can’t find a link.

******Incidentally, I quite enjoy the feeling of ‘cache miss‘/’branch misprediction‘ that I get when someone uses a word I don’t expect. It jumbles my neural net and makes me think.

7 thoughts on “Predictive Punning

  1. I didn’t realize you thought so much about it Bryan, I thought it was more innate. It probably is and you’ve built up this predictive punning model over the years to explain what you inherently do and improve it. So the question is, has anyone in natural language processing done research on this? Or in cognitive science. Why do (some) people enjoy puns and others don’t. And could you build an automated system that tries to generate puns in response to its model of the speaker? This would be much more convincing in a chat bot than the usual tricks of being vague or rude. Even if people find puns annoying, it indicates that the punner is really listening to and understanding what you are saying because they have been able to connect your statements to another topic and provide an unexpected connection. Hmmm. What’s a good pun for Hidden Markov Model?

    1. Yes, you’re totally right. I test out as very intuitive, but I have this very intense left brain training to try to help me make sense of things and explain what I”m doing. πŸ˜€ I don’t know of any research on these things, but then again, I’m not in any of those fields. I feel like some of it could be really useful, if only so we could automated pun generation and spread the joy more widely.

      We should talk about this and make it happen. πŸ˜€

      P.S. ‘Concealed Anthony Catwalk.’

      1. Friend of Markov: So if you’re only allowed to remember one thing at a time, how can you learn from your past?
        Markov: Stoking beard, considering the question, then says. “Hmmmm.”

  2. That bit where you described how you anticipate the end of people’s sentences reminds me of that line from Frozen…

    Hans: We’re in such perfect sync, we finish each others…

    Ana: Sandwiches!

    Hans: …That’s just what I was going to say!

Leave a Reply

Your email address will not be published. Required fields are marked *