by Leo Trottier
One of the most frequent questions we get is “do dog buttons really work”? As a cognitive scientist with degrees from the University of Toronto and UC San Diego, when I saw the dog Stella pressing buttons back in 2019, I was skeptical, too. But then I saw one particular video that really stood out: it showed Stella looking and smelling where the removed “Beach” button had been, and then pressing “help” “water” “outside”.
Taken at face value, this dog’s response to a missing button left me floored. Stella had dozens of buttons at the time, which meant that either she got exceptionally lucky with the three buttons she pressed or she was using these buttons in a way that no one previously thought dogs could. If Stella was really doing what it looked like she was doing, this would be the first documented example in a dog of language ability far beyond what was previously thought possible. Consider that for Stella to have done what it appears she did, she would have needed to have 1) recognized that the button was missing, 2) recognized that she would need to come up with a different approach to communicate what she was trying to express, and 3) correctly identify multiple other buttons that, taken together, could be interpreted as having a meaning equivalent to the one of the button that wasn’t working.
Each on its own, none of the videos of Stella posted online can constitute definitive scientific evidence. But if we believe a) that her human Christina Hunger wasn’t “cherry picking” from hundreds of daily button presses, and b) that Stella had previously demonstrated some consistent associations with the buttons’ meanings (e.g. pressing a button for “water” and then running over to her water bowl), then at the very least what we’ve seen so far should make us very curious about what could be going on.
Since the early days in 2019 and 2020, at FluentPet we’ve seen thousands of videos that suggest something extremely interesting is happening here. While any single video on could be a fluke, every additional video and story we hear lowers the chances that the buttons are being pressed meaninglessly. While I and other scientists would no doubt prefer solid experimental evidence, it’s worth keeping in mind that the kinds of language-like behaviors we’ve seen from Stella, Bunny and other “talking” dogs can be very challenging to study scientifically. Scientific experiments benefit most from data gathered repeatedly in controlled conditions, but the meaning of a button press is always sensitive to its context. For example, asking “What’s your name?” once is likely to elicit a different reaction than repeating “What’s your name? What’s your name?”, which might lead to a sense of surprise or annoyance in a listener. Compare this to testing, for example, whether giving praise or giving a food reward is more effective in dog training: you can repeatedly try out each with one or more dogs and just measure how quickly the dog learns.
None of this means that scientifically studying language ability in non-human animals is impossible. It’s just that doing so is harder and slower going. In the case of studies of language ability in dogs, the most compelling have been those examining language comprehension. For example, the dogs Rico and Chaser were shown to have a vocabulary of hundreds (Rico) or over 1000 (Chaser) words, but the reason this aspect of their language ability could be tested was because they were effectively responding to repeated commands to go fetch—their language ability was integrated into a game they were willing to play over and over again. But this isn’t how communication works: in real-world communication individuals remember what was previously said so that they don’t repeat themselves.
With our deployment earlier in 2023 of FluentPet Connect, we now have the ability to run much more systematic examinations of generative language ability, studies which we hope to see bear fruit in 2024. In the meantime, I believe we can learn a lot from a selection of videos and anecdotes for people to begin to decide for themselves whether buttons are real or not.
Until recently, we’ve been reluctant to highlight these examples because we’ve wanted to prevent people from explicitly training their dogs or cats to do these “interesting” things, since if they were explicitly trained then whatever behaviors they demonstrated could be chalked up to being a “pet trick”. However, we have collected enough data at this point that I believe we can start discussing the behaviors we believe are the most intriguing and that do the best job of demonstrating that dogs aren’t just button mashing but instead are likely demonstrating sparks of animal language ability.
Before getting into these, though, there are a few points worth noting: if you see a video from us or a FluentPet Guide, you can be assured that they haven’t been deceptively edited or cherry picked from thousands of meaningless button press videos. We are committed to authentically presenting button press behavior, and so while we and Guides choose to show button press behaviors that are more interesting, these typically represent one of at most around 20 interactions per day, most of which tend to “make sense” in context.
To keep things concise, I’ll primarily focus on dogs in this discussion. However, there are also a number of intriguing examples of cat behavior that I’ll touch on later.
Button Presses Worth Noticing
The first thing to notice when watching genuine button pressing videos is how the dogs are using the buttons. It is pretty easy to tell when a dog is being cued from beyond the camera’s view screen—when the dog presses the button, it looks like they’re responding to a command: they're not really bothering to look down at which button they intend to press, and instead they're just continuously staring pointedly at their human. By contrast, if you watch a video of Bunny and other dogs pressing buttons, you can see them often appear to be searching for a button, then look away as they think about how to respond, as if they’re trying to focus and avoid being distracted. The case of multi-button presses is even more interesting here—dogs will often continue to look away from their human until after they’ve “finished” pressing the buttons they intend to press, or even do a “thinking circle” in which they ignore their human as they work out which button they’re going to press next.
Indeed, we’ve seen many examples of dogs using buttons when no one is around. If dogs were pressing just to get a reward, why would they press when there’s no one there to hear them? And if they were just doing it because they expected to get a reward despite no one being nearby, we’d at least expect to see them look around expectantly, or go searching for their human “treat” dispenser. But they don’t do this in these videos—instead, they will press buttons associated with wanting their human to come back, or in response to something that they’ve heard in their environment.
We’ve even seen videos and heard stories of learners “studying” their boards and exhibiting “babbling” like behavior after buttons have been changed or added, which they will do while ignoring their humans. In these cases, learners use the buttons in an entirely different way: they don’t look up after 1-2 presses, they press a large number of buttons and appear to just be listening. This is exactly the kind of behavior you’d expect to see from a dog who is concerned about what each of the buttons means, and hard to explain as just pressing buttons to get a treat.
Another great example is learners who press buttons that make reference to information that humans either don’t or can’t know. In this video we can see one cat look in the direction of another that’s approaching, pressing the button that refers to the approaching cat, and the human correcting the cat’s assessment of the situation (because the human can’t see the approaching cat). In another striking example related by Alexis Devine (@whataboutbunny), while she was away and Bunny was being watched by a friend of hers, Bunny mysteriously pressed “concerned” and “bird” repeatedly. The friend then noticed that a bird had fallen down the chimney and was stuck in their wood stove. Bunny had over a hundred buttons to choose from at the time, and only a few would have been so appropriate in this situation.
So do buttons really work? Personally, these and the hundreds of other examples we’ve seen have compelled me to believe that dogs, and potentially cats as well, are using sound buttons in ways that go beyond mere random pressing and simple requests for things like treats or water. Do we know what’s going on in their minds when they’re pressing buttons? Certainly not. But is something extremely interesting going on? I can’t help but conclude: yes.