In early September 2020, lightning strikes turned the forests of the West Coast of the United States into matchbooks. The fires recoloured the skies and their smoke veiled the sun. When that bright disk in the sky reappeared, at least one dog seemed to notice.
Bunny, the social media celebrity sheepadoodle, looked outside the window, then sauntered over to . Each button resembled a tap light: a press of the paw played a recorded word. Bunny pressed 鈥淚S鈥� and looked at her owner. She then yawned and pressed 鈥淲ENT.鈥� Bunny鈥檚 owner tried to make sense of the sentence: 鈥淲hat is went? Who is went?鈥� The dog returned to the window, and in the Instagram video her owner can be heard figuring it out. 鈥淎re you talking about the smoke?鈥�
Bunny returned to her collection of buttons and pressed the one that said 鈥淥UCH.鈥�
鈥淎re you talking about the smoke ouch? Oooh! Oh my God! The smoke is went! It was ouch! Look, it鈥檚 clearing up!鈥�
Bunny is the poster girl for talking dogs. Her account has 1.6 million followers; her , 8.5 million. Dog parents the world over are riding the coattails of this sensation and US$ 16 per button. The goal is to train their dog to press the correct button to express themselves. Sure, dogs wag their tail, moan, and scratch the door in an attempt to let us know what they want; but these new buttons are meant to share a language between the two species and avoid ambiguity. Communication, which many animals can do, is about sending out a signal; but language is about putting thoughts in order, communicating about the past and future, and passing on abstract notions. Can these buttons unlock a new level in human-canine relations?
The way they work is simple. The human records themselves saying a single word into the button鈥檚 microphone鈥攑ractical words like 鈥淥UTSIDE鈥� and 鈥淧LAY鈥�, but also more abstract ones like 鈥淪TRANGER鈥� and 鈥淪AD鈥濃�攁nd train their dog to tap it when that is what they want. When pressed, the button plays back the owner鈥檚 recorded word on its built-in speaker, giving the dog the voice of the owner. Clever dogs are free to combine multiple button taps to form what their owners hope are meaningful sentences. 鈥淪AD, PLAY, OUTSIDE鈥� becomes 鈥淚 am sad and I want to go play outside.鈥�
One research team in San Diego, tapping into an international collective of dog owners, is testing to see if man鈥檚 best friends are indeed talking here or if all we鈥檙e seeing is the TikTok generation version of a famous horse from a century ago. I am reminded of the evergreen words of The Who:
鈥淚鈥檒l get on my knees and pray
鈥淲e 诲辞苍鈥檛 get fooled again.鈥�
Hans, Koko, and Alex: what have we learned?
In 1904 in Berlin, a became a star attraction for doing something that we humans typically abhor: math. At the tail end of four years of training by his owner, who was a math teacher, Hans could tap his hoof the correct number of times to count the people who had come to watch him perform these feats of intelligence. He could read a clock, add up numbers, even answer multiple choice questions. Play a melody and ask him which of these four composers wrote it, and Hans would knock it out of the park.
It鈥檚 no wonder they called him Clever Hans and that the German board of education put together a commission to investigate if the horse鈥檚 prowess was an elaborate hoax. It was not, but something was amiss and it was a skeptical professor by the name of Oskar Pfungst who figured it out. Hans was clever but only insomuch as he could read a person鈥檚 body language. If Hans had to count up to five and no further, he could see his handler betraying that he had reached the right number鈥攑erhaps by seeing his eye relax, his lip go up a bit, his hand tense up. Despite the public鈥檚 hope, Clever Hans could neither count nor understand the complex questions that were put to him, and his story ended tragically when he was drafted into World War I as a war horse and was either killed in battle or, I鈥檓 sorry to report, became a 鈥淗ansburger鈥� to feed hungry soldiers.
The 20th century saw at getting animals to use words to better communicate with us, efforts that, like with Hans, looked impressive at first but that eventually revealed that we had, once again, been fooling ourselves. Koko the Gorilla was taught sign language; her signed words, however, never became language. Alex the Parrot grew his vocabulary but his accomplishments were criticized for simply being trained choices reinforced by rewards.
Researchers outfitted chimps with computers and dolphins were trained to understand the English language. Fiction showed us worlds in which this indeed worked, like in Michael Crichton鈥檚 Congo or on the television show seaQuest DSV. In reality, there were cracks that led to the field decaying in the 80s and 90s and garnering a bad reputation for sloppy research.
When you spend years working with an animal on linguistic communication, you become personally invested in the outcome. You are likely to overinterpret what is being said. When your dog presses 鈥淚S, WENT鈥� as smoke clears, your brain will look for it to make sense and ignore the possibility that this is just nonsense. Handlers are also likely to subconsciously cue their animal, much like with Hans鈥� owner. An animal can thus respond to a change in your body without understanding the words you are speaking.
And when we look specifically at dogs using buttons to speak on social media, we must be aware of edits. We are never seeing the whole picture: meaningless tapping before or after gets edited out to present the 鈥渉it鈥� in a glorious artificiality. Imagine you spent an hour trying to draw a perfect circle and only posted your best attempt on Instagram. You would look like a genius.
This is why studying dog talking buttons requires rigour. One , based at the University of California, San Diego, certainly mentions rigour in its media interviews, and the team鈥檚 papers, I think, bear out its members鈥� desire to avoid getting fooled. These researchers are methodically studying every aspect of this phenomenon, like they are disassembling an engine and examining each one of its parts to make sure it is functioning as it should. The problem with this sort of rigorous, incremental science is that it鈥檚 not sexy. Yet, very basic questions need to be answered first before we can show that dogs can use language to express abstract concepts, future needs, even the content of their dreams鈥攚hich is something claimed by .
The team鈥檚 was humble. It showed that many dogs do indeed recognize the meaning behind pressing the 鈥淧LAY鈥� and 鈥淥UTSIDE鈥� buttons... but strangely enough not the 鈥淔OOD鈥� button. Maybe the dogs were already full, the researchers wondered, or did not expect to receive food outside of their usual feeding times. Additionally, this study provided evidence that asking dog owners to perform this experiment in their home and report the results to the team led to similar findings as sending research staff to the dog owner鈥檚 home and performing the test, which convinced these scientists that a 鈥渃itizen science鈥� component鈥攚here participants received instructions and collected the data themselves鈥攚as viable moving forward, thus allowing more dogs to be tested at little cost.
Their had a similarly basic goal and seemed to show that dogs are not merely repeating the button presses of their owners or spamming their soundboard, hoping for a treat. But when the dogs tapped two buttons in quick succession, did they really understand what they were doing? We still do not know.
This preliminary work is encouraging. Yet, news coverage of these talking buttons is scarily comparable to how most journalists cover unsubstantiated claims of a fringe practice healing cancer. Here, we start with a dog owner and their wonderfully talkative pet before giving voice to the scientists, who are careful but hopeful and explain the work they are doing. Two thirds of the way down, we get a cursory line or paragraph about skeptics before being told how to try these devices for ourselves. Despite the presence of skeptics, we must, after all, end on a hopeful note.
The problem here is that all of this hope gets quickly deflated when we realize that the speakers on the most popular brand of talking buttons may not be up to par.
Unable to hear themselves think
It was a team based in Hungary that popped the balloon for me. Their , published last year, explains that the scientists were worried about sound degradation. Could a dog even make out the name of their toy when it came out of a tinny speaker? So, they compared seventeen dogs鈥� responses to hearing the word live from a human being; hearing it played back through a dedicated speaker; and, crucially, hearing it coming out of the most popular dog talk button, the FluentPet Classic Speak Up Button, as delivering 鈥渢he loudest and clearest sound.鈥� (It is rare for academic scientists to mention the brand name that they test, for legal reasons. Kudos to this team for being transparent. You can watch an example of the experiment performed in Japan .)
There was a striking drop in understanding when the button were used, which suggested to the authors that 鈥渄ogs do not recognize the words when they hear them from this device.鈥� They measured that, while dogs can hear frequencies up to 45,000 Hertz, the sound coming out of a button鈥檚 speaker was much more limited and lost most of its power above 5,000 Hertz. In other words, the sound is muffled, compressed, and lacking in higher pitches, which can make understanding words coming out of it difficult. Even though some gifted dogs can recognize commands they have been trained on when they are spoken by someone with a different accent from their owner, and even though the Hungarian researchers also tested seven such gifted dogs, recognition of the words dropped by 70% when using a button compared to a live command. With a live command, the gifted dogs would almost always perform the required action; with the button, it was more like one out of three or one out of four times, depending on the dog.
If these results are real and are reproduced, it casts major doubt on the idea that dogs decide to 鈥渟peak鈥� words that they know by pressing their corresponding button and that they can even string them together into meaningful clusters.
I鈥檓 also concerned by FluentPet鈥檚 influence in the research coming out of San Diego. While the two papers they have released have relied on a strong methodology, they include as authors a number of FluentPet employees and consultants. The CEO of FluentPet, Leo Trottier, even co-wrote a with these researchers on the history of animal language studies. When asked about this , the head of the San Diego lab, Professor Federico Rossano, says that Trottier, who had been a student at UCSD, reached out to him to pitch the idea of doing a study on the buttons Trottier was selling. Rossano was a bit of a skeptic: he was well aware of the failed attempts and knew that we had learned very little, scientifically, from trying to get apes to use words. And some of his colleagues told him dipping a toe back into this contaminated pool would be a bad idea.
Then COVID-19 happened. Asking dog owners to perform experiments in their homes using buttons they already owned suddenly became attractive. While FluentPet is , Rossano鈥檚 laboratory is said to crunch the numbers themselves and publish the results, and that Rossano鈥檚 team is not paid by FluentPet. In our era of shrinking academic budgets, these sorts of public-private partnerships are more and more common, and while we should not dismiss the research that comes out of them, we do need to know who, in the long process of collecting data, analyzing it, and getting it published, has influence.
FluentPet is not the only company reaching out to Rossano鈥檚 team. Purina, the dog chow brand owned by the megacorporation Nestl茅, invested in Professor Rossano鈥檚 lab. You can imagine that if someone proves that dogs can articulate their needs clearly with the help of a technology, there is a large market awaiting any corporation ready to swoop in.
How not to get fooled again
When reported on these buttons, they showed a dog that had been taught how easy it was to push the lid off of a Tupperware container and access the goods inside. When the lid was snapped shut, however, the dog couldn鈥檛 do it. That鈥檚 when the buttons came in handy. They pressed 鈥淗ELP鈥� and 鈥淟OOK鈥� to draw the attention of their owner. Pretty convincing stuff, especially since dogs do understand many words. But we have to be careful.
So, when you see a celebrity dog on TikTok using buttons to express a complex thought, ask yourself the following questions.
Has the owner trained their dog to do this whole routine for social media? After all, popular accounts get ad revenue and sponsorship deals. There鈥檚 money to be made here.
What are you not seeing that took place before and after the video itself? Are there cuts in the video where nonsensical button presses could have been removed? If a button is broken鈥攍ike for the word 鈥淏EACH鈥濃�攁nd the dog miraculously presses a new combination that carries a similar message鈥攍ike 鈥淗ELP, WATER, OUTSIDE鈥濃�攃ould it be that they simply pressed the buttons near the broken one?
Is the owner subconsciously cueing the dog to press the correct button? Can you even see the owner on camera, in a high enough resolution, and confirm that they are not doing this?
Do you have any idea how much footage of dogs pressing the buttons at random 诲辞苍鈥檛 get posted to social media because they 诲辞苍鈥檛 fit the story being told?
Could it be that the dog doesn鈥檛 understand the word being played on the speaker鈥攁nd we have data to indicate that this is likely鈥攂ut simply associates the position of the button on the floor with being let out to pee?
Don鈥檛 rely on journalists covering this phenomenon, since most 诲辞苍鈥檛 see this is as a science story but merely one of human鈥攐r canine鈥攊nterest. 鈥淚sn鈥檛 this fascinating?鈥� they ask. 鈥淪ure, some people are skeptical, but wouldn鈥檛 it be great?鈥�
The risk with these buttons, besides spending a fair amount of money in a terrible economy, is to infantilize our pets and forego the rewarding process of studying how they naturally communicate鈥攚ith their eyes, their tail, their movements, their moans and growls. By trying to teach them our language, we may forget that they already 鈥渢alk鈥� to us.
In the meantime, the research is thin on the ground, very preliminary, and not completely detached from commercial interests. Let鈥檚 be skeptical so that, just like with Clever Hans, we 诲辞苍鈥檛 get fooled again.
Take-home message:
- Dog talking buttons are little devices that a dog can tap and that play back a single word, like 鈥渙utside鈥� and 鈥渉elp鈥�, and which dog owners believe allow their dog to speak.
- Research into whether or not dogs using them understand language is too new and preliminary to conclude anything, and the company that makes some of these buttons has been involved in the research.
- An independent study showed evidence that the speaker inside these buttons is bad enough that the dog usually does not recognize the word that comes out of it.
