Listen on: Apple Podcasts | Libsyn | Overcast | Pocket Cast | Soundcloud | Spotify Or search for “Knowable” in your favorite podcast app.

Read more: Bob Holmes writes about the long-overlooked importance of interjections in our everyday speech (and how AI gets it wrong).

Transcript

Adam Levy: Hello, this is Adam Levy, co-host of the Knowable podcast, with an audio story about interjections. And I’m speaking strangely. You see, right now I’m reading from a script and I’ve got my best podcast voice on, so I’m avoiding all those silly little noises we make when we normally talk, like uh, um, right? — except it turns out avoiding these fragments, these interjections, isn’t just unnatural. It actively impacts our ability to communicate, because these long-overlooked grunts and utterances might just be integral to speech.

But enough from me and my unnatural introduction. I spoke with two researchers about interjections, and I kicked things off the way I typically do for podcast interviews:

Adam Levy: To start with, could you just give your name and your affiliation?

Mark Dingemanse: My name’s Mark Dingemanse, and I am an associate professor at the Center for Language Studies at Radboud University in Nijmegen, the Netherlands.

Adam Levy: Now, when you were speaking there, I try to do what I always try to do when I’m speaking to someone in a podcast, which is stay silent, which is what I learned to do. Can you explain how this fits or doesn’t fit with more genuine conversation, more everyday conversation?

Mark Dingemanse: Yeah, sure. So part of being a good listener is also showing that you are listening by, you might be nodding, you might be going, “Mm-hmm.” We do that incredibly often, and they help the speaker to know that the other’s on the same page is with them, whether we need to rephrase or whether we should shut up.

Adam Levy: Well, you said there that they happen incredibly often. How often are we talking?

Mark Dingemanse: Yeah, so it depends on what words you’re looking at, but one thing that I’ve looked at is just all the things we use as standalone items. So think of it as one-word utterances. So that could be an mm-hmm, it could also be oh, it could also be huh, and there’s a bunch more of these. And I looked at this in a range of languages around the world, and what I found was that about one in seven turns of our conversations is or has one of those items.

Adam Levy: Do we have anything which shows how communication is affected when we do versus when we don’t have these kinds of interjections? And I’m asking partly selfishly, as someone who wants to know whether I’m interrupting my own podcast interviews by avoiding them.

Mark Dingemanse: Yeah, there’s some lovely research on this. So for instance, by Janet Bavelas and colleagues in the early 2000s, where they had people tell stories in one of two possible settings.

In one setting you’d be telling your story, you’d have a listener opposite who’d be doing their thing as a proper listener, nodding, going, “Mm-hmm,” going “OK,” and so on. In the other setting, you’d also have a listener opposite you, who’d also be looking at you very attentively. But unbeknownst to you, their instruction was to count the number of words starting with t.

And what they found in that experiment was, in the second case where you have this listener trying to count the number of words starting with t  in your story, the delivery of the story just breaks down. People don’t formulate complete sentences anymore: They restart, there are stops and starts, hiccups, hitches, because the speaker doesn’t get this immediate feedback all the time, which we normally do in everyday conversation when we tell a story.

Adam Levy: So do you think then that there would be some impact if I changed my podcast interview technique to, yeah, to try and give a bit more vocal acknowledgement of people rather than doing what I’ve learned to do, which is keep the audio clean and stay quiet?

Mark Dingemanse: My feeling is, as a scientist of talk, that it’s probably a good idea in a podcast-like format, in a dialogue, to stay fairly close to the kinds of things we do in everyday conversation too.

Adam Levy: Now we’re talking in and, I suppose, about English. Do we notice any differences in what those interjections actually are? Is mm-hmm a particularly English interjection or are there patterns in those as well?

Mark Dingemanse: Yeah, so one interesting thing is that some of these seem to be quite similar across languages, and it turns out that in almost every spoken language for which we have any data, you find that one of the things you can do at this place in a story to show your listenership is something with one of those nasal sounds — so mm-hmm, mm-hmm, mm and so on. So that appears to be a fairly strong pattern across languages.

Adam Levy: Why might that be, that that’s a strong pattern? What is it about those noises?

Mark Dingemanse: So there’s something very neat about this, which is, so when we take turns in conversation, normally when I stop speaking, it’s your turn, you take up your turn. We are very fast at this, it’s pretty amazing. But a story is something different.

It’s not this A-B, A-B, it’s not this back and forth between you and me. It’s to tell a story, I need to produce a bunch of turns in succession. How do I do this? Well, I need your cooperation for this. How do you show that you’re cooperating? Well, what would be neater than a word where you literally have your lips sealed and where you’re minimally obstructive, where you just show, “I am here, I am present, I am listening, but I’m not taking a turn. I’m letting you go on.” That is essentially what mm-hmm is doing. So I find it a very neatly designed word. It’s like streamlined for that particular function.

Adam Levy: That was Mark Dingemanse, who we’ll hear from again in a moment. But I also spoke with Martina Wiltschko of the Pompeu Fabra University in Barcelona. You see, as a podcast host, it’s not just my silence when I’m listening that’s unnatural.

Typically when we speak, we also use interjections, stuff like um to indicate when we are pausing but not finished or you know to check that the listener is still, well, listening.

Another thing that I’ve learned as a podcast host and I’ve been doing for hundreds if not thousands of interviews is trying to ask my questions as cleanly, again in quotes, as possible without any unnecessary words added. Are these kinds of markers also important in active speech and in questioning?

Martina Wiltschko: Yes, totally. We start by saying, OK, we have an ideal conversation, which is where you are asking me a question and I assume you don’t know the answer, and you assume that I do know the answer. But in the real world, things are not always like that. Sometimes you might have a hunch what my answer would be, or sometimes you might not be sure that I have an answer.

And these departures from the typical conversation, they’re marked. It also holds for when you tell me something — say, you’re telling me about the weather, and you might think that I don’t know what the weather is like, then you can say, “It’s sunny out.” But if you see that, OK, I have a window and maybe I know what the weather is like, then you say, “It’s sunny out today, isn’t it?”

In Canada, they would say, “It’s nice out, eh?” That final eh, that is a way of making sure that, OK, I’m not telling you, like I would say, I’m not shoving the information down your throat, I’m indicating that there’s a chance that you already know this and then if so, do you agree with me?

Adam Levy: I’m going to try my hardest at this point to try and speak in a way which is much more natural to me. It’s very difficult for me because I’m just in podcaster mode right now. But you’ve drawn there some comparisons between different languages, but these kinds of interjections more broadly, they’re pretty universal across languages, right?

Martina Wiltschko: Yes, yes, yes. So I have looked at, well, I don’t know, at least 15, 20 languages — and they’re from all over the world — and I haven’t found a single one that doesn’t make use of such markers.

And those functions of interaction, of conversational interaction, they seem to be universal, that what we are trying to do is we’re trying to build common ground, meaning I know something and I want you to know it and I want you to tell me what you know, and we are building something that we both know then, and regulating turn-taking. That’s conversation, and that’s conversation all around the world as far as we know.

Adam Levy: And that’s actually something I noticed we weren’t necessarily doing very well, in the beginning of this interview, is turn-taking, I would start asking my questions and because I hadn’t been making any of these noises or anything like that, you weren’t sure that I was about to start asking a question and so you were continuing speaking.

Martina Wiltschko: Right, right, exactly. It’s not natural and especially Zoom is not exactly the natural conversation, so things are a little bit off.

Adam Levy: I’d like to talk about maybe even more unnatural conversations, by which I mean completely unnatural conversations, because we now have podcasts designed and recorded purely by artificial intelligences, which are trying to do their best at having the kind of conversation we are having right now. And to my untrained ear, the first thing I notice is that they can sound quite incredible.

Recording of AI conversation: You know, they actually help to make the conversation smoother (aah), especially in storytelling (right). So imagine you’re listening to someone telling this crazy story…

Adam Levy: Can you explain what aspects of these kinds of conversations are still quite difficult for these artificial podcasts?

Martina Wiltschko: At first glance, yes, they sound like a natural conversation, but if you start listening, what I’ve noticed what is unnatural is that sometimes what should be a marker is really just they’re producing a grunt. And sometimes they sounded to me like just going, “Uh!” — which is not great. My first impression when I listened to this, was like looking at those AI-generated pictures where at first sight they looked good, but on closer inspection there is one finger too many, or something is off.

Adam Levy: But it makes sense, because it’s a part of speech that we don’t really talk about so much, and by we, I mean we in general. I’m sure you talk about it plenty.

Martina Wiltschko: That’s the cool thing about these markers, because they’re so under the radar and we have this unconscious knowledge about them. That’s what I find so fascinating about language — that we know so much about our language that we are not even aware that we know. And there’s so much information packed into these little grunts, they’re not grunts, but it’s unbelievable.

Adam Levy: Martina Wiltschko there. Now, back to my conversation with Mark.

Why, if these interjections are so common, so important, why are they so overlooked?

Mark Dingemanse: I mean, there’s a couple different reasons, but one of them is to do with the history of my discipline, of linguistics, the study of language. And that discipline has its roots in studying language as it is written down.

And so when linguists started thinking about linguistic structure, that was where their focus was, because at the time, it was impossible to even make recordings — we’re talking 100, 150 years ago at least. And so that’s one simple reason that they overlooked this bit of the linguistic machinery. And it’s taken quite a long time before people started thinking, “Wait, maybe there is something in how we use language in everyday conversations that might also be worth studying.”

Adam Levy: And to you, just how worth studying is it? And to you, just how big a part of language is it?

Mark Dingemanse: Yeah, so one argument I’ve made — deliberately provocative, I should say — but one argument I make is that interjections, these little words that streamline our everyday language use, that they’re at the heart of language. We would not have complex language were it not for these humble words. This is entirely different from how people used to look at these things, if they looked at them at all. And so it’s fascinating to me that it’s so easy to overlook, whereas we would struggle to have any coherent conversation if we did not have these little words.

Adam Levy: And like, I guess, a lot of the most fascinating science questions, the question is there staring you right in front of your nose, and you just have to ask it, and ask, “What are these things that I’m doing and hearing constantly?”

Mark Dingemanse: Indeed, yeah.

Adam Levy: Well, this has all made me somewhat self-conscious, I suppose, about how I conduct podcast interviews, and now I’m really trying to “mm-hmm” as we go along.

Mark Dingemanse: Yeah, it’s terrible, that’s one of the afflictions of this line of work is you start, at least for a moment, being self-conscious. It’ll wear off, I’m sure.

Adam Levy: OK, good.

This has been a special one-off episode of the Knowable podcast with Mark Dingemanse and Martina Wiltschko.