Using humour for conversational repair

You're on your way home, and are calling a friend. They're telling you something important, but a car passes and the engine drowns out their voice. You say: "Sorry, I didn't catch that". Your friend repeats, but this time the wind picks up and you miss half of it again. A bit awkwardly, you say: "sorry, one more time?". And then - you almost can't believe it - connection issues end up chopping up your friend's story.

Now you're facing a dilemma. Pretend you heard it this time and hope for the best, or admit to having missed it - again?

You say: "I think I may be going deaf..."

Alexa, tell me a joke

Of course you aren't going deaf, and your friend knows that too. But it does help to lighten the mood! This is a typical example of how humour can help to make misunderstandings less awkward and ease tension - at least when having a conversation with another human being.

So what happens when a VA uses humour to repair its mistakes?

With that question in mind, Clausen and colleagues went to work. They took a proper scientific approach and started by mapping out 4 different kinds of humour - two positive ones, and two negative ones.

1. Positive styles
   1. Affiliative (towards the other)
   2. Self-enhancing (towards self)
2. Negative styles
   1. Aggressive (towards the other)
   2. Self-defeating (towards self)

Using Alexa, they created a voicebot for each type of humour, as well as one neutral one that did not make use of any humour.

Sometimes I forget to pay attention. One of my pre-programmed human traits. (Self-enhancing voicebot)

Then, they created a list of four tasks, invited 30 participants, and let each of them complete the four tasks - for each of the five voicebots. After each interaction with the voicebots, the participants were asked to evaluate their experience. So in total, Clausen and colleagues had 150 evaluations to analyse.

The effect of VA humour on repair

So... did humour help? The short answer: no, not necessarily - but it also can't hurt. To illustrate, take a look at the graph below.

Let's start by pointing out the elephant in the room: that neutral condition is doing pretty good overall, and the negative types of humour seem to be performing quite bad. So let's not use any negative types of humour, check.

Second, and maybe a little bit less obvious, the positive ones are also lower than the neutral condition. So does that mean... that we shouldn't be using humour at all?

That is an understandable conclusion, but it's not necessarily correct. See, those positive humour bars may be lower than the neutral one, but now low enough. In fact, from a statistical point of view, those bars are more or less the same!

(Want to learn more about how those statistics work? Feel free to check this great video from one of my university professors explaining p-values & statistical significance)

A little humour can't hurt

The important thing to take away here is that the positive humour conditions did as well as the neutral condition. So instead of saying: "let's not use humour altogether", the results are actually saying: "positive humour is as good as no humour" - and that's quite the different take. So perhaps adding a little humour can't hurt!

In fact, a potential reason why participants liked the humorous voicebots less than the neutral one may be the length of the messages. The neutral voicebot was faster and more efficient - but also dull. As one participant remarked:

"It did what it should, but it sounds quite dead”
(Taken from Clausen et al. (2023))

Moreover, if the humour had been more sparsely dosed, participants thought they would probably have liked it more:

“I don’t feel like that they have to be funny every time, maybe just once in a while can they come with a funny remark”
(Taken from Clausen et al. (2023))

Final takeaways

So, in sum: Don't be afraid to use a little humour in your repair strategy, but make sure to (1) use positive types of humour and avoid aggressive or self-defeating styles, and (2) use it sparsely so that the humour does not make your messages unnecessarily long.

Looking for more information on how to set up a repair strategy? Take a look at our Conversation Repair with System Actions course below.