The scientists cracking the code of interspecies communication

In the Karoo, South Africa’s vast semidesert, an African striped mouse basks in the morning warmth outside the bush it calls home.

Nearby, audio equipment casts a long shadow on the rust-colored earth and emits a string of high-frequency squeaks inaudible to human ears, interrupting the rodent’s tranquil morning routine. The mouse recognizes the call as coming from a mouse in a neighboring nest bush, just as the scientists broadcasting it had intended.

The striped mouse rises up on its hind legs, an expression of measured vigilance. When the researchers play a call from a mouse in the same nest, however, the striped mouse carries on sunbathing, unperturbed.

“When it’s a vocalization from a neighboring individual, they pay a lot more attention. They really look at the speaker. They are disturbed,” said Nicolas Mathevon, a professor at the University of Saint-Etienne in France, who led the research on African striped mice.

“If it’s from a complete stranger, then we see an even stronger reaction, like the mouse fleeing into the bush because they are really surprised.”

The research is the first to decode the hidden sounds of mice in the wild, and it’s one of several studies in the past few years that are revealing just how sophisticated vocal communication between animals can be — even when the sounds are imperceptible to human ears.

“Not so long ago, people thought that animals were not communicating at all, or very simple things,” said Mathevon, who is the author of “The Voices of Nature: How and Why Animals Communicate.” He has studied animal communication in birds, dolphins, monkeys, hyenas and crocodiles, and even attempted to parse the cries of human babies for meaning. Hippos are next on his list.

Equipped with sophisticated recording equipment, machine learning algorithms, and a deep well of resolve and patience, bioacousticians are finding communication patterns between animals once thought unique to humans. The discoveries challenge the ideas about what makes human language special.

Ultimately, researchers say they will not only listen in on animal speech but develop the ability to talk back, like the fictional character Dr. Dolittle. Experts disagree on whether two-way communication will ultimately benefit animals.

The hidden chatter of African striped mice

The hidden chatter of African striped mice

0:51

In 2023, Mathevon and his colleagues recorded 122,619 squeaks from dozens of African striped mice over 12 days and nights using 23 microphones spaced out over four nest bushes. The vocal repertoire consisted of at least seven different squeak types. The mice used some of the squeaks within their nests and others on the fringes of their territory.

The researchers used the information to train an artificial neural network — the same system that underpins large language models such as ChatGPT. The network allowed them to uncover that each nest of mice had a specific vocal signature. Further study has uncovered that individual mice had unique signatures.

“Machine learning is absolutely essential because you have too many calls, too many vocalizations, you cannot handle them,” Mathevon said.
He explained that the sounds he and his team decoded represent “static” information about the identity of the mice that doesn’t change over time. The next goal would be to try and decipher “dynamic information” he believes is coded in the calls, such as information about stress levels, which varies.

The work on the African striped mice is one of four research projects listed as finalists for this year’s Dolittle Prize. It awards $100,000 to honor significant advances in deciphering animal communication. The prize, sponsored by British billionaire businessman Jeremy Coller, also promises an award of either a $10 million investment or $500,000 in cash if a team can demonstrate that a species communicates independently with the researchers without recognizing that it is communicating with humans.

“The vision is supposed to be fluent two-way communication, where humans can interact with wild animals in the way that they interact with each other, creating a kind of genuine, meaningful contact,” said prize judge Jonathan Birch, a professor of philosophy and director of The Jeremy Coller Centre for Animal Sentience at the London School of Economics. “We recognize we are a way from that goal.”

This year’s winner will be announced on June 25. The inaugural winner in 2025 was a team that found a language-like communication system in the whistles of wild dolphins in Sarasota, Florida. While much research on animal communication focuses on large mammals thought to have relatively rich communication systems, such as primates and cetaceans (whales and dolphins), Birch said the judges try to cast a wide zoological net.

In addition to the striped mice researchers, this year’s finalists include scientists decoding the communication of two species of great ape in Ivory Coast and the Democratic Republic of Congo, and boisterous red-beaked songbirds known as zebra finches, native to Australia but kept in captivity in California. A 2025 finalist deciphered gestures made by cuttlefish in a French laboratory.

“There’s probably all kinds of complexity out there that we’re only just beginning to even notice,” Birch said.

The biggest challenge researchers face, he added, is getting enough data. “What we’ve seen in the human case is that AI systems, on the face of it, once looked much too simple to do anything interesting, but once there’s enough training data, you see these extraordinary emerging capabilities.”

Wild animals, of course, have not been busy generating training data for our human-made, trillions-of-words-large language models. Getting enough information in the wild for even a simple model can take decades.

Catherine Crockford, head of the Ape Social Mind Lab at the Institute of Cognitive Sciences in Lyon, France, started her professional life as a speech and language therapist working in London hospitals. But in the late 1990s, she got interested in the evolution of language and began studying the wild chimpanzees that roam the Taï National Park in Ivory Coast. “The idea that their communication was interesting was not fashionable,” she said. “We’ve come a long way since then.”

Crockford and her collaborator Roman Wittig, a senior scientist and research group leader at the Max Planck Institute for Evolutionary Anthropology in Leipzig, Germany, have spent years gathering around 20,000 of hours of recordings of chimpanzee calls. The duo’s latest work suggests that the apes can create new meaning by stringing sounds together in pairs — something akin to syntax, which was once thought to be a unique hallmark of human language.

“We have 150 chimpanzees that we know personally. Many of them you know from birth to death by now, because we are in the third generation of individuals that we are observing,” Wittig said.

Crockford said the team’s latest work found that, while the chimps only have 12 call types in their repertoire, they have the flexibility to combine these calls in all sorts of ways to modify meaning or generate new meanings, “which really gives them much more capacity to say things.”

The researchers found 16 two-call combinations. Some of them added to or clarified one of the call’s meaning — for example: “I’m feeding and resting,” the team worked out. Meanwhile, four calls were combined to create an entirely new meaning. “Hoo,” a call made when resting, combined with a “panted-grunt” used as a greeting, for example, means “Let’s build a nest.”

Mélissa Berthet, a postdoctoral researcher at the University of Milan, spent nine months observing bonobos in the remote rainforests of the Democratic Republic of Congo. She and her colleagues found that the ape — a close relative of the chimpanzee but with a dramatically different, female-dominated social structure — uses similar paired calls.

Over the course of nine months, Berthet got up at 4 a.m. to follow the apes through dense forest, where they travel several miles each day before nesting. She made detailed voice notes of what had happened just before the bonobos made their calls: Was the bonobo feeding? Was he traveling? Was he alone or with others? For each call she recorded, she ticked off a list of more than 300 parameters.

It wasn’t until she was back at her then-base at the University of Zurich in Switzerland that a clearer picture began to take shape. Berthet used a mathematical technique to analyze hundreds of hours of recordings. She captured 700 calls from different bonobos and created a visual map of the single and combination calls that could be used to investigate their meaning.

For example, she found when a “peep” call, used to suggest a course of action, is combined with a “whistle,” used singly to help keep the group together when traveling, the combination had a totally different meaning.
Instead, it denotes a tense social situation.

“They used it during very sensitive social context, for example, when someone is threatening someone else,” she said. “I think it’s a way to say, ‘I would like that we make peace.’”

Crockford said the team had begun to use artificial intelligence tools and found they “massively speed up the process of managing large data sets.”

“One thing that makes the chimpanzee vocal repertoire difficult is that the vocalizations grade into one another. You can have grunts that grade into ‘hoos’ or grade into barks, and barks that grade into screams and so on,” she said.

“So we gave AI this problem: classifying these grunts according to the context in which they’re emitted. And it did a really decent job,” Crockford said.

Together, the findings in chimpanzees and bonobos suggest a rudimentary form of syntax, the rules that govern word order in human language and give it flexibility and creativity. For example, in English, “ape goes” and “go ape” use the same words and have totally different meanings.

Many scientists believe that the vocal systems of great apes were too limited to be considered precursors of human language, but the work of Crockford, Berthet and their colleagues suggests otherwise. Getting a better understanding of these dynamic vocalizations also lays the groundwork necessary to achieve the goal of communicating with animals. Without understanding the meaning and function of animal vocalizations, two-way communication would be impossible.

Berthet, however, is torn on whether the benefits of speaking with animals will outweigh potential harms. While being able to communicate with domesticated and zoo animals, whose lives are deeply intertwined with humans, could improve their care, she was less sure about wild animals.

“I work in this field to understand animals and obviously it would be a dream to talk with them, but I think we have to be very careful,” she said. “I would be afraid that if we take this path we end up with tourism that is for talking with chimps, bonobos and then gorillas. And then you have very messed-up animals that don’t understand why they are being told to go here or there or why someone wants to play with them.”

With these concerns in mind, Berthet and her colleagues are cautious about conducting playback experiments in the wild, similar to those Mathevon conducted with the African striped mice, that would allow them to confirm their findings. The researchers fear that the experiments are too invasive and could alter the social dynamics of the bonobo group.

“We don’t want individuals to hear their own voice, right? These are species we work with. They’re very intelligent, they’re very socially aware, and I want to avoid, at any price, to mess with their heads,” said Martin Surbeck, an associate professor in the department of human evolutionary biology at Harvard University, who in 2016 helped establish the bonobo research site at Kokolopori, where Berthet conducted her field work. “It would be quite unsettling and might have consequences we really cannot anticipate.”

Groups such as Project CETI or the Cetacean Translation Initiative, which works with sperm whales, are already looking at the legal and ethical implications that might arise if humans become better able to understand animal communication.

Julie Elie, an associate project scientist in the department of neuroscience at the University of California, Berkeley, now works with zebra finches in captivity after studying them in the wild. She ventured that it would be easier to crack the code of interspecies communication with birds than with primates or whales.

“They are really vocal animals and they are easy for us to observe. We can constrain the space in which they live without, you know, oppressing them and still have natural behaviors,” she said. “It’s very difficult to keep a whale in a swimming pool.”

While whale song has fascinated scientists for decades and AI has been used to discover something akin to a phonetic alphabet in the clicks used by sperm whales, what they are saying largely remains a mystery. That’s because it’s much harder to track and observe whales in their natural environment and therefore link vocalizations to specific actions.

In her work with zebra finches, Elie has categorized 11 of the birds’ calls, linking them to distinct meanings such as hunger, danger, bonding and social conflict.

Her catalog agreed with the calls documented in the wild by the late ornithologist Richard Zann. However, she wanted to go a step further and know whether the birds agreed with her categorization.

To do this, she devised an experiment using zebra finches trained to peck a button. Each time the button was pressed, a different call would sound. The specific call-type that Elie wanted to study on a given day was rewarded with a seed.

“Each day the bird has to find out what call-type … is associated with the reward,” Elie explained. Usually, the bird started off by listening to several calls in their entirety until they got to the rewarded vocalization. After this happens a few times, the zebra finch “quickly learns to interrupt the playback of non-rewarded sounds to trigger the next one” — not unlike how a human might swipe through uninteresting videos, she added.

“What we showed is that they are able to do this classification task across every call-type that we, humans, identified in their repertoire,” she said. “For the first time, we were able to ask animals if we have correctly identified the ‘words’ or building blocks of their ‘language.’”

The vocal systems of zebra finches

The vocal systems of zebra finches

0:49

She also discovered that the birds weren’t perfect. Sometimes they interrupted vocalizations that would have given them a reward, or they waited until the end of the playback of a non-rewarded vocalization — but the errors followed a distinct pattern that suggested the birds were mixing up similar meanings rather than similar sounds.

Elie gave an example using human language: If the bird were to classify the words hurt, heart, love and glove, it would be more likely to make mistakes between heart and love (with similar meanings and different sounds) than love and glove (which have similar acoustics but different meanings).

“By demonstrating that zebra finches have a mental representation of the meaning of their calls, we are starting to break the wall between our own species and the rest of the animal kingdom.”

For the next step in her research, she hopes to design a robot that would move, sound and look like a zebra finch. However, even if successful, she predicts there will be limits to that communication. “It’s important to remind ourselves that each species is living in its own world. What it makes sense to ask a human might not make sense to ask a bird,” she said.

It’s clear that artificial intelligence will yield a greater understanding of animal communication and allow scientists to better mimic animal speech. But it’s far less certain whether these tools will enable two-way communication with animals in the way humans, brought up on “The Jungle Book,” might desire, said Yossi Yovel, a professor at Tel Aviv University’s department of zoology, who studies bat communication and chairs the Dolittle Prize’s judging panel.

“I think interspecies communication, it will exist. It’ll just be much more boring than some people maybe imagine,” he said. “I would love to speak to my cat. Unfortunately, it might be a limited conversation.”

Berthet agreed. “They’re not us. They have different needs and different interests. And so it’s very unlikely even if one day we managed to have this two-way communication that we can talk about war or ecology.”

It’s also possible that humans might not like what they hear, the researchers added, particularly when it comes to domesticated animals, pets or those kept in captivity. “If we want to talk with animals, are we willing to listen to what they actually say?” Surbeck asked.

Interacting with animals comes with a large degree of uncertainty about what is being communicated and its effects on the animals in question, which could result in emotional or physical harm, said Mark Ryan, a senior digital ethics researcher at Wageningen University and Research in the Netherlands.

“I feel that a lot of animal communication research is done for the right reasons: to get a better understanding of the non-human world, to develop scientific knowledge about animals and our place in the world, and as a rationale for conservation,” Ryan, who has coauthored a paper on the ethical risks of trying to “speak whale,” said via email.

“However, I am much more sceptical about the ‘two-way’ communication being proposed in competitions such as this.”

Melanie Challenger, an ethicist and vice president at the Royal Society for the Protection of Animals or RSPCA, said the study of animal communication was “fascinating,” but she didn’t believe it would necessarily result in a fairer future for animals.

“There’ll be all kinds of surprises that it throws up,” said Challenger, who is the author of the upcoming book “Alive: The Hidden Intelligence of the Living World.” “But I think we’re still in danger of falling into the trap of expecting them to be like us, and if they’re not like us enough, then we don’t think we need to consider them.”

Mathevon argued, however, that even if we’re never able to talk to animals in a human way, understanding the complexities of animal communication, and attempting to mimic or tap into it, would benefit animals.

For example, in an effort to prevent collisions, trains in Japan emit warning signals based on the snortlike sound deer use to alert others when they’re in danger, he noted.

Information encoded in vocal systems could reveal useful details about stress levels that could help improve the welfare of farm and lab animals without invasive procedures such as blood tests.

One day, he expected, it would be possible to have a “Dolittle machine” that can communicate with animals. “For the species that are very well studied, I think there is no technical, specific limitations.”

CNN’s Max Burnell produced the video in this report.

Sign up for CNN’s Wonder Theory science newsletter. Explore the universe with news on fascinating discoveries, scientific advancements and more.