When you pay close attention to a verbal exchange, you’ll realize that the participants often sprinkle their dialogue with numerous small, seemingly meaningless sounds or words, such as mm-hmm, um, huh?. These expressions have traditionally been dismissed by linguists as insignificant noise—mere language debris that surfaces when speakers lack eloquence.

However, these tiny verbal cues might play a more significant role than previously thought. Some modern linguists argue that these elements are not merely linguistic leftovers but are essential in directing the flow of conversation and facilitating mutual understanding. This places them at the very core of linguistic interaction, and they represent a challenging aspect of language for artificial intelligence to grasp.

“This is a phenomenon that’s been right under our noses, one we hardly paid attention to,” notes Mark Dingemanse, a linguist at Radboud University in the Netherlands. “It turns out to be a game changer for our understanding of complex language.”

Support Science Journalism

If you find this article engaging, please consider supporting our distinguished journalism by subscribing. Your subscription helps secure the future of influential stories about the breakthroughs and concepts that shape our modern world.

Historically, linguistics has concentrated on the study of written language, mainly because it was what was available for analysis. But with the advent of audio recordings, researchers have been able to scrutinize spoken language in the same detailed way they examine text.

“To streamline your conversations, these tools are indispensable.” —Mark Dingemanse, linguist

Upon examining spoken language, researchers find that interjections—brief words or phrases that stand alone—are remarkably common in daily speech. “One out of every seven phrases is one of these,” Dingemanse remarks, who discusses the role of interjections in the 2024 Annual Review of Linguistics. “You encounter these small elements about every 12 seconds in conversation. Evidently, they’re necessary.”

Many of these interjections help manage the pace and flow of dialogue. “Think of them as a toolkit for managing social interactions,” explains Dingemanse. For instance, when a speaker says um or uh, it indicates a forthcoming pause, signaling they haven’t finished their thought. Conversely, a listener’s quick huh? or what? might indicate a breakdown in communication that needs addressing.

This necessity is universal: In a study covering 31 languages globally, Dingemanse and his team discovered that all languages use a short, neutral syllable like huh? for clarification, likely because it’s simple and quick to utter. “In moments of confusion, you need the simplest question, and that’s huh?,” he says. “It appears that all cultures develop this tool for the same reasons.”

Other interjections act as what some linguists term “continuers,” such as mm-hmm, which are signals from the listener indicating engagement and encouraging the speaker to continue. The formulation of mm-hmm, produced with a closed mouth, clearly shows that the listener does not intend to speak.

In sign languages, continuers are often managed differently, notes Carl Börstell, a linguist at the University of Bergen in Norway. In Swedish Sign Language, for instance, listeners frequently sign yes to indicate continued interest, often positioning their hands lower to minimize disruption.

Moreover, different interjections can send nuanced signals. For example, Allison Nguyen, a psycholinguist at Illinois State University, describes a scenario where one person is explaining how to assemble IKEA furniture. Here, mm-hmm might suggest that the explainer should continue with the current step, whereas yeah or OK might signal readiness to move to the next step.

More Than Meets the Ear

Interjections are not just about politeness; they are pivotal to the structure of conversation, asserts Dingemanse. He refers to a classic study where 34 students listened to a story. Some listeners actively showed they were paying attention, while others, tasked with counting words starting with t, appeared distracted. The lack of typical listener responses resulted in less coherent stories, demonstrating the significant impact of these small words.

Nguyen concurs that these expressions are far from trivial. “They significantly enhance mutual understanding and the flow of conversation,” she states. She is currently investigating whether emojis in text conversations serve analogous roles.

The function of interjections extends beyond mere conversation regulation; they also assist in establishing the conversational context or “grounding.” Each time people engage in a conversation, they must align on their respective knowledge bases and expectations. A lot of this alignment is facilitated by interjections, according to linguists.

“We are creating shared understanding and taking turns.” —Martina Wiltschko, theoretical linguist

“If I’m sharing a story and you react with a ‘Wow!’ I might be encouraged to elaborate,” says Nguyen. “But if you respond with ‘Uh-huh,’ I might take it that you’re not interested in more details.”

Establishing mutual knowledge is a critical aspect of grounding, explains Martina Wiltschko, a theoretical linguist at the Catalan Institution for Research and Advanced Studies in Barcelona, Spain. Some languages, like Mandarin, explicitly differentiate responses based on the listener’s prior knowledge. In English, this differentiation often relies on interjections.

Wiltschko highlights the Canadian eh? as a prime example: “If I inform you about your new dog, it’s generally not news to you, so it feels odd for me to mention it,” she explains. “But adding ‘eh?’ removes the awkwardness by indicating that I’m not assuming it’s news to you.”

Additionally, certain interjections might suggest that the speaker knows they are not fully addressing the listener’s inquiry. “If you ask about the weather in Barcelona and I respond with ‘Well, I haven’t been outside yet,’ the well acknowledges that I’m not answering directly,” says Wiltschko.

After studying over 20 languages, Wiltschko and her students have found that all languages use these small words for such negotiations. “I’ve yet to find a language that doesn’t address these three aspects: what I know, what I think you know, and taking turns,” she notes. These elements are essential for regulating conversations, she adds: “We are establishing common ground and alternating turns.”

These details are not merely academic; they are crucial for achieving fluency in a second language, Wiltschko points out, though language instruction often overlooks them. “In language teaching, using ums and uhs can cost you points for lack of fluency,” she says. “However, native speakers use them because they are functional. They should be included in teaching.”

Artificial intelligence also struggles with using interjections effectively, which can be the best way to tell apart a machine from a human. Moreover, interjections can reveal much about interpersonal dynamics, potentially offering valuable insights in contexts like couple’s therapy. For example, the interjection oh often signals confrontation, as in the difference between ‘Do you want to go out for dinner?’ and ‘Oh, so now you want to go out for dinner?’

Ultimately, these small verbal cues are at the heart of language and its purpose. “Language exists so we can interact,” says Börstell. “That’s the main reason for its success.”

Dingemanse takes it a step further. Interjections not only aid our conversations but also define how language facilitates discussion about itself. “With huh? you’re not just saying ‘I didn’t understand,'” he explains. “It’s more like ‘I know you’re trying to communicate something, but I didn’t catch it.’ This level of reflexivity allows for more complex speech and thought. Without these simple words, complex language might not be possible at all.”

Can AI Master the Use of Interjections?

In an effort to make artificial intelligence sound more human-like, developers are incorporating interjections into AI responses. For instance, Google’s NotebookLM can simulate a podcast discussion between two AI-generated hosts, summarizing scientific papers.

At first listen, the program performs well; the hosts engage in laughter, jokes, and insert Mm-hmm and Wow! at seemingly correct intervals. However, to a trained linguist, there are subtle discrepancies. (Listen to an example.)

“They almost work, but not quite,” observes theoretical linguist Martina Wiltschko from the Catalan Institution for Research and Advanced Studies. “It’s reminiscent of AI art where the anomalies—like too many fingers—aren’t immediately obvious, but upon closer inspection, something seems off.”

A noticeable flaw is that when one AI host responds with mm-hmm or laughs, the other host pauses, creating an unnatural break in the conversation. “It feels almost like the uncanny valley,” says Wiltschko. “It’s close, but not quite there.”

A major challenge is the AI’s fluctuating grasp of who knows what during the conversation. “It’s not just about what’s being said, but who is speaking, in what context, and who holds what knowledge,” Wiltschko explains. “I’d be surprised if AI could ever fully manage that—something humans do with ease.”

This article originally appeared in Knowable Magazine, an independent journalistic endeavor from Annual Reviews. Sign up for the newsletter.