Will there ever be a Google Translate for pets?
The tech world is on the case – but there’s no guarantee that our animals will have anything interesting to say
Jan 3rd 2025
By Imogen West-Knights
Something was wrong with Bunny. Her owner, Alexis Devine, an artist and jewellery designer in her 40s, had just brought her back from an autumn walk near their home in Tacoma, Washington. Normally Bunny, a bubbly young sheepadoodle, would be exhausted by the exercise, but today she wouldn’t settle down. For most dog-owners, this would be the beginning of a frustrating guesswork process. Was Bunny hungry? Did she need to go to the toilet? Was she bored?
Devine, however, had a secret weapon. Bunny trotted over to a patchwork of brightly coloured hexagonal mats on the floor of the living room, and placed her paw on one of a series of buttons. “Mad,” it said in Devine’s disembodied voice. The dog looked expectantly at her owner.
“Why mad?” Devine asked.
Bunny pressed another button, prompting the voice to say “ouch”.
“Where ouch?” Devine replied.
The dog pressed the buttons for “stranger” and “paw”. Devine checked Bunny’s paws. Embedded in some matted fur on the bottom of her front left paw was the “stranger”: an inch-long foxtail spike.
A video of this interaction went viral on TikTok in October 2020. Of course it did: who wouldn’t want to see a video of a talking dog? As long as humans have had imaginations, they have dreamed of communicating with other animals, of chatting with a cheetah and cursing in fluent kangaroo (as a song from a musical adaptation of “Doctor Dolittle” put it). Any apparent progress towards that goal sets pet-owners’ pulses racing.
There’s a long, grubby history of animals being trained to “speak” in ways that humans have claimed to understand. This time felt different, though. Bunny is one of a growing cohort of dogs worldwide who are now using these buttons, known as Augmentative and Alternative Communication (AAC) boards, to express themselves. It’s an open question whether they are intentionally communicating, memorising combinations of actions or doing something else entirely – but an answer may be forthcoming. Two thousand of these dogs are currently being studied by researchers at the University of California, San Diego, as part of an unusually massive citizen-science study (in which members of the public submit their own data).
Over the years many have tried to crack the challenge of what it takes to communicate effectively with animals. Will it soon be possible to call our pets not just our companions, but our confidants?
If you think of the books and films you enjoyed as a child, it’s highly likely that the protagonist was a talking animal: Babar, Aslan, Peter Rabbit, Baloo, any of the 101 Dalmatians. Or maybe they were a human who could talk to animals – Dora the Explorer, for instance, can yap for hours with her best friend, a monkey named Boots. There are countless fables, fairy tales and stories from around the world in which humans possess the ability to speak to animals, and vice versa. Egyptian and Greek gods were always turning themselves into animals to interact with people; Japanese foxes and tanuki (a type of canid also known as a raccoon dog) would magic themselves into humans to play tricks on them.
The seeds for the UC San Diego study were sown in 2013, when Leo Trottier, an entrepreneur and cognitive scientist, adopted two cats, called Jonas and Salk (the first and last names of the inventor of the polio vaccine). Like many animal lovers, Trottier dreamed of tapping into his pets’ intelligence. Later that same year, while pursuing his PhD at UC San Diego, he founded a company called CleverPet, which makes stimulating toys for animals. Its flagship product, a machine that releases kibble to reward a dog for pressing light-up buttons in a certain order, proclaims itself to be “the world’s first game console for dogs”.
In autumn 2019 Trottier, who shares the healthy glow of many California-based techies, noticed a new social-media trend: people trying to start a conversation with their pets. Some intrepid owners had developed simple, labelled buttons that their pets pressed when they wanted something, like water. He got in touch with a few people to find out more about how their animals were interacting with these buttons – including Devine, Bunny’s owner.
Trottier was impressed with the owners’ inventiveness, but also sensed a business opportunity. “They were really calling out for some purpose-built devices,” he told me. By June 2020 Trottier had launched FluentPet, a company that makes customisable AAC boards, specifically designed for dogs. (The company now also markets its buttons to cat-owners.) Devine helped Trottier develop a button system that could be used by dogs of all shapes and sizes, and Bunny switched to FluentPet’s products soon after. (Devine is now an influencer for the company, and receives a percentage of sales if people buy the buttons through her referral link on social media.) For $65, FluentPet sells a starter kit of buttons, to which owners can affix labels featuring common commands and expressions, such as “outside”, “play”, “cuddle”, “water” and “potty”; the owner then customises the buttons by adding a recording of their own voice. As the dog becomes more proficient at using the kit, the owner can buy extra buttons and programme them to express more complex concepts, such as a wish for a favourite stuffed fox or a walk in the park. (Bunny, who is now five years old, is working with a hundred buttons, including time expressions like “soon” and question words like “why?”)
As FluentPet got off the ground, Trottier kept a close eye on his device’s early adopters. Bunny seemed to be doing things that scientists had generally thought dogs couldn’t do, like combining “tokens” (the term linguists use for basic units of language) to make new expressions. Trottier recalled how Bunny would often pair “sound” with “walk”, leaving Devine mystified. Over time they noticed that Bunny used this combination of words only at the start of a longer “conversation”. They surmised that Bunny was signalling that she wanted to use the board: to “walk” on the buttons and make “sound”. Once they gave Bunny a button for “talk” – the word she needed – she stopped using the other two buttons. Trottier was surprised by how sophisticated the request seemed: “It’s just so metaphorical, right?”
He felt that Bunny’s actions could be of scientific interest. Was she able to come up with new expressions independently of her owner? Was she able to understand what words like talk, sound and walk actually referred to, and if so, to what degree? “It felt like it would be almost malpractice not to start collecting data as quickly as possible,” he said.
Trottier contacted Federico Rossano, a comparative psychologist (someone who researches the behaviour and thought processes of different species) at UC San Diego. A cheerful Italian in his 40s, Rossano has studied how non-human primates communicate with one another: how orangutans put in a lunch order, and how baby bonobos ask their mothers to carry them. Despite his long-standing interest in animal communication, however, Rossano was initially hesitant to pursue a study of FluentPet’s button users. “I knew all the drama and ethical concerns associated with animal-communication studies, and my immediate reaction was, absolutely not, I don’t want to get myself involved,” he told me.
Scientists first started to get interested in animal communication in the second half of the 19th century. In 1872 Charles Darwin examined the evolutionary connection between animal and human physical expressions of emotion, such as smiling. Twenty-five years later Ivan Pavlov demonstrated that dogs could be trained to associate an unrelated cue (a ringing bell) with a reward (food). From the 1960s there was a surge of interest in animal speech, fuelled in part by the “cognitive revolution” in psychology, when scientists first began to assert that animal intelligence differed from human intelligence only in sophistication, not kind.
Many of the resulting studies have since been discredited. Some were based on the behaviour of a single creature. Often these animals – usually primates, who were prized for their cognitive abilities and close genetic relationships to humans – were isolated from the rest of their species in a lab and trained for hours a day to use sign language, recognise symbols or even to imitate human speech. Some of the chimps became famous after having been trained to “speak”, typically at the expense of their mental and physical wellbeing. (Washoe and Nim Chimpsky, who were taught American Sign Language in the 1970s, became hunched over and withdrawn, which were interpreted as signs of dejection and loneliness.)
As public opinion turned against these sorts of studies, scientists changed tack. Today, instead of trying to train one or two animals in a language or communication pattern used by humans, many studies focus on listening to animals use their own “language” in as noninvasive a way as possible. (For instance, venturing into the forest to hear the call of a certain bird rather than trapping a bunch of them and taking them back to the lab.) The goal is no longer to teach animals how to speak to humans, but to study how they communicate in their own environments.
Trottier assured Rossano he would take care to avoid the greatest traps in animal-communication studies. The pets would live at home, meaning that their routines could continue with minimal disruption, and that they would have their owners keeping them company. He was also confident he could rustle up a large sample size, increasing the accuracy and breadth of the study. Even so, Rossano remained wary: “I thought, well, I don’t know if there’s anything to see, probably not.”
Today, around 2,000 pets from 47 countries are participating in the UC San Diego study. Owners are asked to film the dogs using the buttons and fill in surveys on their pets’ behaviour. Periodically, research assistants visit the dogs’ homes to verify the owners’ claims. Rossano and his team hope to establish whether dogs can indeed learn to associate the words on the buttons with their meanings and make simple requests with them. If they manage to do that, Rossano wants to explore more complex questions – including whether FluentPet’s boards could be used to help pets have two-way conversations with their owners.
The study, which began in 2020, will run for several more years. In the meantime, Rossano’s team have published two papers describing their work and process. Rossano made it clear to me that they are not claiming that dogs have language – a systematic form of communication that deploys a particular grammar and vocabulary – or the ability to speak. “I really don’t want you to think I’m some crazy person that goes out there and says yeah, dogs talk, because dogs don’t talk,” he explained. “But they communicate” – or transfer information – “just like any animal communicates.”
He has been surprised by what the animals in his study seem to communicate about: some dogs refer to things and people that are not present, make requests for food or assistance on behalf of other animals in their household, or seemingly “remember” past events. Cache, a golden retriever involved in the study, lives with his owner, Christina Lee, on the outskirts of San Francisco. Lee told me Cache often uses the buttons to talk about past events. “There will be something that upsets him the day, and we will come home and four hours later he will reference that event.” Once, said Lee, Cache’s fur became matted by burrs as he walked through a field; it took her three hours to get him untangled. Later that day Cache pressed the button he had with the name of the field, as well as the one for “worried”.
Although Rossano is careful not to get carried away, he can’t help but get excited by such stories. “I’m a scientist. I am still sceptical,” he said. “I don’t think that they are communicating like adult human beings. But I’ve seen certain things to make me convinced there are things that they do that are way more compelling than we give them credit for.”
Dogs are thought to have the linguistic capacities of a human toddler. But what about my cat, Sushi Two, a two-year-old mackerel tabby? As a paid-up member of the crazy-cat-lady club, I would obviously like to be able to hold long and meaningful conversations with Sushi Two about her hopes and dreams. I’d also like to find out why an unexpected movement or noise – a slight cough, for example – will send her running from the room, and to reassure her that everything’s going to be OK.
I may be waiting a long time, however: cats, it seems, are a tougher nut to crack than dogs. Jussi Karlgren, a Swedish computational linguist with an interest in animal communication, told me that cats don’t have as many facial muscles with which to make expressions as dogs do. “Cats are stiff” in their faces, he said, which is why they express themselves through meowing.
Interestingly, while kittens will meow at their mothers, adult cats reserve their meows for humans. (When adult cats do communicate with each other, out of aggression or fear, it’s usually through hissing or yowling.) “Either they meow to us because they think we’re their mums, or alternatively think we’re kittens,” explained Karlgren. Other researchers – most prominent among them is another Swedish linguist named Susanne Schötz – believe that cats use different kinds of meows, depending on the circumstances, in order to elicit specific responses from humans.
Javier Sanchez, an engineer from Washington state who worked on Amazon’s Alexa device, was listening to the radio a few years ago when his ears pricked up. Schötz was giving an interview in which she claimed that humans could interpret their cats’ meows. It gave Sanchez, who has five cats – Bear, Mongo, Timmy, Candy and Concha – an idea. He remembers thinking, “Well, I work at Alexa, I know how audio-recognition technologies work. We can do this for cats.”
Sanchez got in touch with Stavros Ntalampiras, a data scientist who had written a paper demonstrating that algorithms could be used to distinguish between different types of cat vocalisations. Cats are believed to have up to 21 distinct meows, spanning feelings such as fear, contentment and friendliness. As Sanchez said, “think of sounds that you and I would make if we were startled, scared, we hit our toe on something around the house; we would all make sort of universal sounds like ‘ah’, ‘ouch’…cats have the same thing.” Cats can combine these common vocalisations to make a sort of personal vocabulary that only their owner can understand, said Sanchez. A certain sequence of meows, for instance, might mean “let me outside”.
Sanchez got to work developing a device that could translate cats’ meows. His first idea was for a gadget that a cat would wear on its collar and that would play the translation out loud. (“Specifically, I wanted Samuel L. Jackson’s voice,” he told me.) This proved too technically difficult to implement, though, so instead Sanchez designed an app for smartphones. Using a Google program for classifying sounds called YAMNet, the app distinguishes meows from other noises. It then “translates” the meow according to a cat/human dictionary Ntalampiras and Sanchez have developed, based on datasets that Ntalampiras had collected. “I’m in love!” a cat might be trying to tell you; or if it’s displeased with you, “We’re fighting.”
Eager to hear what Sushi Two had to tell me, I downloaded the app, which is called MeowTalk. I quickly became frustrated with it. The phone had to be in the same room as my cat to hear her meowing clearly – but, as any cat-owner knows, cats can’t be persuaded to do things they don’t want to do. The meows that it did manage to translate were Sushi Two’s demands to be let into a room. Since I could hear her outside the door, making noises and clawing at the carpet, this was hardly a revelation.
Naomi, a woman in her 30s who lives in London with her seven-year-old British Shorthair named Waffle, also downloaded MeowTalk. Waffle’s favourite phrase, apparently, is “Don’t ignore me,” which Naomi finds ironic: “I’m literally always paying him attention and begging him to hang out with me.”
The truth, Sanchez told me, is that cats don’t have all that much to say. “You’re not going to get complex sentences, and you’re not going to get a conversation. That’s just not there.” Even Schötz has acknowledged that about 90% of cats’ meows are simply cries for attention.
I couldn’t help but be disappointed with this answer; I wanted there to be more to discover about my cat’s inner life than her desire to have her basic needs fulfilled. So did Naomi. Even though she was not particularly surprised by MeowTalk’s insights, she continues to use it. “I think it’s just another way to obsess over these animals that give us so much joy,” she said.
Much of the latest research into animal communication is not on pets, but on cetaceans, or aquatic mammals, such as whales and dolphins. These creatures have highly developed brains that are good at problem-solving, and they possess sophisticated social skills. Although cetaceans are significantly more intelligent and better at communicating than cats and dogs, the technology that scientists are developing to understand them could also one day be applied to pets. Could a talking whale get me one step closer to understanding Sushi Two? I travelled to Santa Cruz, California, to find out.
Ari Friedlaender studies whale behaviour at the University of California, Santa Cruz’s coastal-science campus, a row of low buildings overlooking the sea at the edge of the Monterey Bay National Marine Sanctuary. The bay teems with cetaceans: grey whales migrate here in the early spring, followed by humpbacks, then fin and blue whales in the summer. Friedlaender also monitors whales in the Antarctic, and has made almost 50 trips there in 25 years. Online photographs show him leaning off boats and poking whales with long poles. They bring to mind images of spear-wielding whalers, but Friedlaender is engaged in a different kind of hunt.
He uses the poles to attach suction-cupped tags with trackers and recording devices to the whales. These tags collect all sorts of information, measuring the whales’ movements up to 400 times a second, as well as how deep they are swimming. The tags also contain a hydrophone for listening to whale song, and video cameras to give us a window onto their world. “It’s a continuous, very detailed look at a day in the life of a whale,” said Friedlaender.
He and his team plan to hand over the data about the whales’ behaviour to their research partners at the Earth Species Project, a California-based organisation that brings together scientists working in artificial intelligence (AI), conservation and biology. The goal of the Earth Species Project is to create software akin to Google Translate for animals, using large language models (LLMs), a type of AI that can identify and generate text. By feeding communication data from a host of animals, including whales, into an LLM, the group hopes to track similar components between different animal languages – and lay the groundwork for future bilateral communication between all sorts of species, be they humans and whales, humans and dogs, or even whales and dogs.
There are others in this game. Members of the Cetacean Translation Initiative, an interdisciplinary group whose research takes place on the Caribbean island of Dominica, have claimed that we will be able to converse with whales as early as 2026. But not everyone believes this kind of technology is necessary. At a zoo near Stockholm I spent some time with a group of researchers who are trying to interpret the whistles that dolphins make to one another. As the researchers fed a pod of dolphins a lunch of fresh fish, I asked one of them, Josefin Larsson, whether we will ever have enough shared context to talk to dolphins.
“You want to speak their language, you mean?” she asked. “Because we communicate with them every day.” The researchers are adept at reading dolphin body language, and can tell when the creatures are happy or restless. The cetaceans have similar powers of discernment: Larsson told me that at times when she has been stressed, the dolphins become more agitated themselves. We learn to understand these cues whenever we know another animal well, whether they’re human or not – and being able to figure out a creature’s needs in this way is what allows us to understand how they experience the world.
In late 2023 I met Ovi, a seven-year-old schnauzer, and his owner, Mika Agnihotri, at their home in north London. Agnihotri and Ovi had been participating in the UC San Diego study for three years. At the time, Ovi was using just shy of 50 FluentPet buttons, which included words such as “hi”, “scratches” and “puzzle” (to refer to a toy he likes). He presses “love you” to open a request, explained Agnihotri: “So it’ll be like, ‘love you’, then ‘scratches’,” which apparently means he wants to be stroked.
That’s not all Ovi can do with the buttons. On drizzly days, said Agnihotri, the dog will repeatedly paw at the “rain” button – apparently wanting to make small talk about the weather. Agnihotri is under no illusion Ovi is capable of engaging in a “complete conversation like you and me”, and still relies on body language to work out what her pet wants. But she believes the device has improved Ovi’s quality of life: “They live for so little time, and if I could just do anything that would make his life a bit better, I .”
Ovi seemed excited by having a stranger in the house and showed interest in the contents of my backpack (a sandwich) – but unfortunately he refused to use his buttons in front of me. Maybe he’s more interested in having cosy chats with his owner than spouting off to any random passerby. This matches with the observations of the FluentPet users I spoke to – that the device had increased the bond specifically between them and their dog. Christina Lee admits Cache may not fully understand the words on the buttons (he now uses 130 of them), but that doesn’t mean he’s not using them to communicate on some level. She told me Cache uses the button “friend” to refer to her, probably because he knows she’ll stroke him in response. “People are like, Cache doesn’t understand the concept of a friend. I’m like, yeah, neither does a child,” Lee explained. “They just show up and see the same person twice at the playground, and you call it a friend. But people want the dog to be able to have a dictionary definition of what friendship means.”
Ultimately, the results of Rossano’s study won’t matter much to many of FluentPet’s users. Agnihotri is simply grateful that she and Ovi have their own private code to work with. “At the end of the day…if we can just communicate a bit better with each other in this additional way, and not everybody understands, I’m fine with that,” she said.
And maybe it is this intimacy – and inscrutability – that keeps us enthralled by our pets. If I knew what caused Sushi Two to suddenly dart from a room, or what she meant when she meowed at nothing in a corner, perhaps the mundanity of her thoughts would disappoint me. As it is, I enjoy the puzzle.
https://www.economist.com/1843/2025/01/03/will-there-ever-be-a-google-translate-for-pets?