Lost in translation: Can AI help us understand what our pets are trying to say?

The urge to converse with animals is age-old, long predating the time when smartphones became our best friends

MeowTalk is the product of a growing interest in enlisting additional intelligences – machine-learning algorithms – to decode animal communication. Illustration: Melanie Lambrick/The New York Times

My cat is a bona fide chatterbox. Momo will meow when she is hungry and when she is full, when she wants to be picked up and when she wants to be put down, when I leave the room or when I enter it, or sometimes for what appears to be no real reason at all.

But because she is a cat, she is also uncooperative. So the moment I downloaded MeowTalk Cat Translator, a mobile app that promised to convert Momo’s meows into plain English, she clammed right up. For two days I tried, and failed, to solicit a sound.

On day three, out of desperation, I decided to pick her up while she was wolfing down her dinner, an interruption guaranteed to elicit a howl of protest. Right on cue, Momo wailed. The app processed the sound, then played an advertisement for Sara Lee, then rendered a translation: “I’m happy!”

I was dubious. But MeowTalk provided a more plausible translation about a week later, when I returned from a four-day trip. Upon seeing me, Momo meowed and then purred. “Nice to see you,” the app translated. Then: “Let me rest.” (The ads disappeared after I upgraded to a premium account.)

READ MORE

The urge to converse with animals is age-old, long predating the time when smartphones became our best friends. Scientists have taught sign language to great apes, chatted with grey parrots and even tried to teach English to bottlenose dolphins. Pets – with which we share our homes but not a common language – are particularly tempting targets. My TikTok feed brims with videos of Bunny, a sheepadoodle who has learned to press sound buttons that play pre-recorded phrases such as “outside”, “scritches” and “love you.”

MeowTalk is the product of a growing interest in enlisting additional intelligences – machine-learning algorithms – to decode animal communication. The idea is not as far-fetched as it may seem. For example, machine-learning systems, which are able to extract patterns from large data sets, can distinguish between the squeaks that rodents make when they are happy and those that they emit when they are in distress.

Applying the same advances to our creature companions has obvious appeal.

“We’re trying to understand what cats are saying and give them a voice,” says Javier Sanchez, a founder of MeowTalk. “We want to use this to help people build better and stronger relationships with their cats.”

To me, an animal lover in a three-species household – Momo the cranky cat begrudgingly shares space with Watson the overeager dog – the idea of a pet translation app was tantalising. But even MeowTalk’s creators acknowledge that there are still a few kinks to work out.

In a 2019 study, Stavros Ntalampiras, a computer scientist at the University of Milan, demonstrated that algorithms could automatically distinguish between the meows that cats made in three situations. Illustration: Melanie Lambrick/The New York Times

Making Meowsic

A meow contains multitudes. In the best of feline times – say, when a cat is being fed – meows tend to be short and high-pitched and have rising intonations, according to one recent study, which has not yet been published in a scientific journal. But in the worst of times (trapped in a cat carrier), cats generally make their distress known with long, low-pitched meows that have falling intonations.

“They tend to use different types of melody in their meows when they try to signal different things,” says Susanne Schötz, a phonetician at Lund University in Sweden who led the study as part of a research project called Meowsic.

And in a 2019 study, Stavros Ntalampiras, a computer scientist at the University of Milan, demonstrated that algorithms could automatically distinguish between the meows that cats made in three situations: when being brushed, when waiting for food or when left alone in a strange environment.

MeowTalk, whose founders enlisted Ntalampiras after the study appeared, expands on this research, using algorithms to identify cat vocalisations made in a variety of contexts.

The app detects and analyses cat utterances in real time, assigning each one a broadly defined “intent”, such as happy, resting, hunting or “mating call”. It then displays a conversational, plain English “translation” of whatever intent it detects, such as Momo’s beleaguered “let me rest”. (Oddly, none of these translations appear to include “I will chew off your leg if you do not feed me this instant.”)

A pet translation by the MeowTalk mobile app. Photograph: MeowTalk via The New York Times
A history of pet translations by the MeowTalk mobile app. Photograph: MeowTalk via The New York Times

MeowTalk uses the sounds it collects to refine its algorithms and improve its performance, the founders say, and pet owners can provide in-the-moment feedback if the app gets it wrong.

In 2021, MeowTalk researchers reported that the software could distinguish among nine intents with 90 per cent accuracy overall. But the app was better at identifying some than others, not infrequently confusing “happy” and “pain,” according to the results.

Assessing the accuracy of a cat translation app is tricky, saya Sergei Dreizin, a MeowTalk founder. “It’s assuming that you actually know what your cat wants,” he says.

In 2021, MeowTalk researchers reported that the software could distinguish among nine intents with 90 per cent accuracy overall. Illustration: Melanie Lambrick/The New York Times

I found that the app was, as advertised, especially good at detecting purring. (Then again, so am I.) But it’s much harder to determine what the calls in each category mean – if they carry a consistent meaning at all – without actually having a way of, you know, communicating with cats. (Cat-ch-22?)

After all, the precise purpose of purring, which cats do in a wide variety of situations, remains elusive. MeowTalk, however, interprets purrs as “resting”.

“But to be candid,” Sanchez says, “it can mean…” He rephrases: “We don’t know what it means.”

At times, I found MeowTalk’s grab bag of conversational translations unsettling. In one moment, Momo sounded like a college acquaintance responding to a tossed-off text message: “Just chilling!”. In another, she became a Victorian hero: “My love, I’m here!” (This spurred my fiance to begin addressing the cat as “my love,” which was also unsettling.) One afternoon, I hoisted Momo off the floor and, when she mewed, glanced at my phone: “Hey baby, let’s go somewhere private!”

“A lot of translations are kind of creatively presented to the user,” Ntalampiras says. “It’s not pure science at this stage.”

Schötz says that over the years she has seen several cat translation products, but that she is yet to find one that truly impresses her. “I’m looking forward to seeing something that really works, because that would be just brilliant,” she says.

In the meantime, Sanchez says he had also heard from users who had found an unexpected use for the app, which stores recordings of the meows it captures: listening to these recordings after their animal had died. It’s a “very magical experience”, he says.

Zoolingua, a start-up based in Arizona, is hoping to create an artificial-intelligence-powered dog translator. Illustration: Melanie Lambrick/The New York Times

Deciphering dogs

Dogs could soon have their own day. Zoolingua, a start-up based in Arizona, is hoping to create an artificial-intelligence-powered dog translator that will analyse canine vocalisation and body language.

Dog owners have been overwhelmingly enthusiastic about the concept, says Con Slobodchikoff, founder and CEO of Zoolingua, who spent much of his academic career studying prairie dog communication. “Good communication between you and your dog means having a great relationship with your dog,” he says. “And a lot of people want a great relationship with their dog.”

(But, he added, not everyone: “One small minority says, ‘I don’t think that I really want to know what my dog is trying to communicate to me because maybe my dog doesn’t like me’.”)

Still, even sophisticated algorithms may miss critical real-world context and cues, says Alexandra Horowitz, an expert on dog cognition at Barnard College in the US. For instance, much of canine behaviour is driven by scent. “How is that going to be translated, when we don’t know the extent of it ourselves?” Horowitz says in an email.

Even sophisticated algorithms may miss critical real-world context and cues, says Alexandra Horowitz, an expert on dog cognition at Barnard College. Illustration: Melanie Lambrick/The New York Times

The desire to understand what animals are “saying”, however, does not seem likely to abate. The world can be a lonely place, especially so in the past few years. Finding new ways to connect with other creatures, other species, can be a much needed balm.

Personally, I would pay at least two figures for an app that could help me know whether my dog truly needs to go outside or just wants to see if the neighbour has put bread out for the birds. (Maybe what I really need is a canine lie-detection app.) For now, I will simply have to use my own judgment and powers of observation.

After all, our pets are already communicating with us all the time, Horowitz says. “It’s far more interesting to me to learn my own dog’s communications,” she says, “especially the idiosyncrasies that are formed between particular people and particular animals, than pretend that an app can — presto! — translate it all.” – This article originally appeared in The New York Times.