Soundboards and Talking Buttons in Dogs: What the Evidence Shows

1. Why Talking Buttons Raise Scientific Questions

1.1 What a Soundboard Is

A soundboard for dogs is a set of large, flat buttons that play a short recorded word when pressed. The owner records words such as "outside", "play", "food" or "walk", places the buttons on a mat on the floor and teaches the dog to press them. Over time, many owners add more buttons: names of people, places, feelings, time words such as "later" or "now", and abstract concepts such as "help" or "love you". Some boards in home use can grow to dozens of buttons arranged in grids.

In the research literature these devices are called augmentative interspecies communication (AIC) devices, a term borrowed from symbol boards for people who cannot speak. The term already assumes that what happens at the board is communication. Whether that holds, and in what sense, is what the research tries to find out.

1.2 Three Questions, Not One

Public discussion usually asks whether dogs can talk with buttons. That question bundles at least three separate questions, each needing different evidence.

The first question is whether dogs learn button-outcome associations. Does a dog learn that pressing a particular button is followed by a particular event, such as the door opening or a toy appearing? This is a question about operant learning, and the answer is very likely yes, because dogs learn such relations in many other contexts (how reinforcement schedules maintain behavior).

The second question is whether dogs respond to the recorded words themselves. When someone else presses the "outside" button, or says the word aloud, does the dog act as if the word refers to going outside, even when no other cues are present? This is a question about word comprehension, and it has to be tested under controlled conditions.

The third question is whether dogs combine buttons meaningfully and communicate intentionally. When a dog presses "outside" and then "play", is it expressing a combined request directed at a listener? This is the strongest claim, and it is the one most videos imply.

These levels build on each other but do not follow from each other. A dog can learn that a button opens the door without understanding the word on it, and it can produce non-random sequences without those sequences having the structure of language.

1.3 Social Media Clips and Data

The soundboard trend grew mostly on social media, where clips show dogs apparently commenting on their day, asking about absent family members or expressing feelings. Such clips are observations and can generate hypotheses. But a clip is a selected moment, chosen because it looks impressive and usually filmed by someone who knows what the "right" answer would be. The presses that made no sense, and the context that made the "answer" predictable, are rarely shown.

The researchers who have studied soundboard dogs most systematically made this point themselves. They note that early interspecies communication studies with great apes were "heavily criticized for their training methods, testing procedures, and claims", and they set up a citizen science project to address these weaknesses through larger samples, documented training and controlled testing (Bastos & Rossano, 2023). The question for this article is therefore not whether the clips are convincing, but what the data collected under controlled or at least systematic conditions show.

2. A Short History: Symbols, Keyboards and Prompting

2.1 Apes, Signs and Lexigrams

Soundboards have a history that predates social media. Researchers tried for decades to teach great apes sign language or keyboards with arbitrary symbols, called lexigrams, and the methodological lessons are directly relevant to dogs.

In one well-known study, two bonobos acquired lexigrams through observation of everyday communication rather than through explicit training. The researchers reported that the bonobos understood spoken English words and could identify the matching lexigram on hearing the word, and the older animal produced combinations of several symbols (Savage-Rumbaugh et al., 1986). This line of work tested comprehension separately from production, which is the same distinction this article draws for dogs.

The other side of this history is equally instructive. Terrace and colleagues analyzed more than 19,000 multisign utterances of a young chimpanzee called Nim. At first, some two-sign combinations seemed to show regular patterns, for example particular signs tending to occur in particular positions. Analysis of videotapes then showed that most of Nim's utterances were prompted by the teacher's preceding utterance, and the authors described similar patterns in other signing apes (Terrace et al., 1979). The apparent structure was to a large extent a reflection of the human side of the interaction.

The lesson for soundboards is that the question is not whether combinations occur, but what produces them.

2.2 Sofia at the Keyboard

The first systematic study of a dog using arbitrary symbols came from Brazil. Rossi and Ades (2008) trained a female mixed-breed dog to press keys on a keyboard, each key associated with a particular object or activity. According to the authors, the dog used the keys in line with her motivational state, used the keyboard mainly when a researcher was present, and looked at people more often after pressing a key. The authors described this as the first systematic investigation of arbitrary sign learning in a dog.

The study showed that a dog can learn to use arbitrary keys to make requests, and it looked for signs of communication beyond the press itself: an audience effect and gaze toward the human. As a study of a single dog, it shows what is possible, not how common it is.

2.3 From Single Cases to Citizen Science

For a long time, research on unusual cognitive skills in dogs consisted of single-animal studies. The word learning literature changed this through international recruitment: one project identified 41 dogs that performed above chance on a toy-name test over five years and showed that a citizen science model can turn anecdotes into evidence (Dror et al., 2023).

The soundboard research follows the same logic. Thousands of owners have trained their dogs with buttons, and many keep records. A research group at the University of California San Diego set up a project to collect such data from dogs and cats, with the aim of moving from associative button learning to questions about welfare and symbolic capacity (Bastos & Rossano, 2023). The two empirical studies published from this project in 2024 are the core of the current evidence and are discussed in detail below.

3. Question 1: Do Dogs Learn Button-Outcome Associations?

3.1 Operant Learning Explains a Lot

The basic training of a soundboard follows a familiar pattern. The owner presses the "outside" button before opening the door, many times. Then the dog's own press, even an accidental one, is followed by the door opening. Pressing the button becomes a behavior that produces a valued outcome. In learning terms, this is operant conditioning: a behavior is strengthened because of its consequences.

Nothing about this process requires the dog to understand the recorded word. A dog could learn the same relation with a silent button or a bell on the door, and nobody would describe ringing a bell as language. The press is a request in a functional sense: it is maintained because the outcome follows.

This does not make the first level trivial. It means that the dog can learn a new, arbitrary action to influence what people do, which is a form of control over its environment. Control and predictability are relevant to stress and well-being (why control and predictability matter for stress), and this is one of the strongest arguments for the practical value of buttons, independent of any language claims.

3.2 What This Level Does and Does Not Imply

At this level, the most likely explanation for a dog pressing "outside" and then going to the door is that the press has been followed by going outside many times. The dog may press when it wants to go out, but it may also press when it is bored, when it has learned that pressing produces attention, or simply because pressing has been reinforced in many situations. In everyday use, owners often respond to any press with attention, which can turn pressing itself into an attention-getting behavior (how attention can maintain behavior).

A dog that presses many different buttons may therefore be pressing because pressing as such pays off, not because each button carries a distinct meaning.

3.3 Evidence Status

No published controlled study isolates this first level in soundboard dogs, because the result is not in serious doubt: the single-dog keyboard study showed that a dog could learn to use keys to make requests (Rossi & Ades, 2008), and learning theory makes the same prediction. The decisive point is what follows. If dogs learn button-outcome relations, most everyday soundboard behavior can be explained without any understanding of words, and claims about the second and third levels need evidence beyond what operant learning already explains.

4. Question 2: Do Dogs Respond to the Recorded Words Themselves?

4.1 The 2024 Comprehension Study: Design

The most direct test of the second level is a study published in PLOS ONE in 2024 (Bastos, Evenson, et al., 2024). The researchers asked whether soundboard-trained dogs respond appropriately to food-related, play-related and outside-related words, and whether this depends on who presents the word and how it is presented.

In the first part, researchers visited 30 dogs in their homes. In the second, 29 dogs were tested by their owners, who followed the same procedure as citizen scientists and submitted videos. Words came from three categories: outside, play and food. A made-up nonce word with no learned meaning served as a control.

Two factors were varied. The first was the identity of the person presenting the word: the dog's owner or an unfamiliar person. The second was the mode: the word was either produced by pressing a button or spoken. In the researcher-run study, the owner stayed in the room for the dog's comfort but "wore a sleep mask and listened to music on noise-cancelling headphones so they could not see any of the trial procedures". The buttons were covered with colored stickers so that their identity was concealed. Videos were recorded and coded by trained coders who did not know which word had been presented.

The coded behaviors were directed at the door, toys or the food area; head tilts were also recorded.

4.2 Results

For two of the three word categories, dogs responded in a contextually appropriate way. According to the paper, dogs showed approximately seven times more play-directed behavior in the play condition and approximately seven times more outside-directed behavior in the outside condition (Bastos, Evenson, et al., 2024). These responses did not depend on whether the owner or an unfamiliar person produced the word, or on whether it came from a button or was spoken.

For food words, the result was negative: the authors found "no conclusive evidence" that dogs showed more food-directed behavior in the food condition than in the other two. They discuss possible reasons such as the timing of the tests and satiation, but the result as it stands is that food words did not produce a clear effect. The head tilt analysis was inconclusive because there were too few observations.

The authors described the accuracy of the responses as "comparable to dogs' accuracy in responding to human pointing", lower than that of specialized detection dogs, and attributed this partly to more variable reinforcement in everyday homes. Pointing comprehension is itself a skill on which dogs perform above chance but far from perfectly (how dogs read human gestures).

4.3 What the Results Mean

This study supports a specific conclusion: owner-trained soundboard dogs can respond to at least some recorded words in a way that fits their learned outcome, even when the word is presented by an unfamiliar person and the owner cannot see what is happening. That rules out the simplest Clever Hans explanation for these responses, namely that the dog reads the owner's body language rather than the word. The authors' own summary is that dogs "respond appropriately to these words even when they are presented in the absence of any other cues, such as the owner's body language" (Bastos, Evenson, et al., 2024).

The study does not show that dogs understand the words the way people do. The measured orienting and approach behaviors are consistent with the word activating an expectation of the outcome, but they do not reveal what representation lies behind the response. The results also apply only to two categories of highly familiar, frequently reinforced words, not to abstract buttons such as "love you" or "later".

4.4 What Word Learning Research Adds

The soundboard finding fits into a larger literature on how dogs learn words. That literature shows both what dogs can do and how unevenly the ability is distributed (what the research on word learning in dogs shows).

A Border Collie named Rico retrieved objects by name from a vocabulary of roughly two hundred items and inferred the referent of a new word by exclusion (Kaminski et al., 2004). Another Border Collie, Chaser, learned 1,022 object names over three years of daily training of four to five hours and, in fourteen double-blind trials, combined object names with action commands correctly (Pilley & Reid, 2011). These are carefully controlled single-animal studies.

The ability appears to be rare. In a study that trained dogs intensively on toy names over three months, all six dogs previously identified as gifted word learners learned new names, while only one of the typical dogs did, and the authors described the variation as qualitative rather than quantitative (Fugazza et al., 2021). Gifted dogs learned six new names in one week in one condition and twelve in another and typically retained them over months (Dror et al., 2021). In the international sample of 41 such dogs, 74% had acquired their vocabulary spontaneously in play rather than through deliberate training (Dror et al., 2023).

For soundboards, this means two things. The words soundboard dogs respond to are outcome words trained by many repetitions in fixed contexts, a different task from learning dozens of object names. And because dogs differ considerably in word learning, it would be surprising if all soundboard dogs performed alike.

4.5 Neural Evidence in Ordinary Dogs

Neuroimaging and electrophysiology add another angle. In an fMRI study, dogs showed a left-hemisphere bias for meaningful words, a right auditory region involved in processing intonation, and reward-related responses when praise words were spoken with praising intonation (Andics et al., 2016). This shows that dogs' brains process both what is said and how it is said, which matters for buttons: a recorded word on a button has a fixed intonation, while spoken words vary.

A more recent EEG study tested whether dogs form expectations about objects when they hear object words. Owners named a known object and then showed either the matching object or a different one. In the 18 dogs included in the analysis (of 27 tested), the mismatching object produced a neural response resembling the human N400 effect, which in people is associated with semantic processing. The effect was stronger for better-known words and emerged irrespective of how many object words a dog was reported to know (Boros et al., 2024). This suggests that hearing a familiar word can activate some representation of its referent even in dogs that would perform at chance in a behavioral object-choice test.

The neural evidence also shows limits. In an fMRI study with 12 dogs trained on two object names, activation was greater for unfamiliar pseudowords than for trained words, interpreted as novelty detection (Prichard et al., 2018). These studies did not test soundboards.

A 2025 study tested one technical issue that matters directly for soundboards: playback quality. Dogs responded most reliably to trained verbal action cues when a person spoke them directly, less reliably through a high-quality loudspeaker, and least reliably through commercially available AIC buttons. Acoustic analyses showed that the buttons removed substantially more speech information. Device quality also affected recognition of familiar words in Gifted Word Learner dogs. This does not overturn earlier soundboard findings, but it adds an important limit: a dog may know a learned word yet fail when the recording is acoustically degraded (Higaki et al., 2025).

4.6 The Limits of Recorded Words

Two further findings are directly relevant to buttons. In an ERP study with 17 dogs, brain responses distinguished familiar instruction words from very different nonsense words, but not from nonsense words differing by a small phonetic detail (Magyari et al., 2020). If this applies to soundboards, similar-sounding buttons could be confused, and a "correct" press may depend more on position or routine than on sound. How dogs hear speech sounds in the first place is a separate question (how dogs hear).

The second finding concerns how dogs learn words. In three studies modeled on experiments with human infants, four word-experienced dogs did not learn new words when the only information available was the human's referential behavior, displaced in time from the object (Tempelmann et al., 2014). Dogs appear to rely on close association between word and event. That fits the way soundboards are trained: the button is pressed immediately before the outcome. It also suggests that buttons for concepts that cannot be paired with a concrete, immediate outcome, such as "yesterday" or "love you", are unlikely to acquire a stable meaning for the dog in the sense the owner intends.

Dogs also generalize learned responses to similar stimuli in ways that are hard to predict (how dogs generalize learned responses). For the second level, the evidence supports a modest but real conclusion: some familiar, outcome-linked recorded words elicit appropriate responses on their own. It does not support the idea that every button carries a meaning for the dog.

5. Question 3: Do Dogs Combine Buttons Meaningfully?

5.1 The 2024 Combination Study

The third level was addressed in a study published in Scientific Reports in 2024 (Bastos, Houghton, et al., 2024). Its design was observational, not experimental. Owners reported their dogs' button presses, and their own, in a smartphone app built for logging presses. The researchers analyzed data from 152 dogs that had at least 200 reported interactions, covering 194,901 dog presses over 21 months. The buttons in different homes carried different labels, so the researchers grouped them into 68 broad concept categories, for example combining "kibble", "dinner" and "food" into one food concept.

The study asked whether presses were non-accidental, non-random, and more than repetitions of the owners' presses. Two-button combinations were taken from sequences of three or more presses, ignoring order, and each dog's network of combinations was compared with 1,000 simulated random networks.

5.2 What Non-Random Means

The central results, in the words of the abstract, were that "at the population level, soundboard use by dogs cannot be explained by random pressing, and that certain two-button concept combinations appear more often than expected by chance at the population level". Dogs' presses were also "not perfectly predicted by their owners'", which the authors took as an indication that the dogs were not merely repeating human presses. They concluded that this suggests "that dog soundboard use is deliberate" (Bastos, Houghton, et al., 2024).

The most frequently pressed concepts related to routine activities and needs, such as going outside, food and play.

Non-random is a meaningful result: soundboard dogs do not simply hit buttons at random, and the logged data are not pure noise. But it is a low bar, because any learned behavior is non-random. The study shows that button use is structured, not what the structure means.

5.3 What It Does Not Show: Order, Syntax and Reference

Several features of the design limit the conclusions. The order of buttons was ignored, so the study cannot speak to anything like syntax, which depends on order. The data were reported by owners, who may log interesting presses more often than routine ones; the authors acknowledge this, while arguing that population-level patterns make it an unlikely explanation for all results. The dogs were experienced users with at least 200 logged interactions, which means the results may not apply to beginners. And there was considerable individual variation, with some dogs' networks close to random.

Most importantly, the study did not test reference. It did not check whether a dog's press of "outside" was followed by the dog actually wanting to go outside, or whether the combination "food" plus "play" corresponded to any specific request. The authors list this question as open. Without such a test, the meaning of combinations is inferred from the labels the owners chose, not from the dog's behavior.

The label on a button is a human word. When a dog presses "mad" and then "walk", the label invites the reading that the dog is annoyed about a walk. But the dog may have pressed two neighboring buttons, or two buttons that have both been followed by attention. The label tells us what the owner hoped the button would mean, not what it means to the dog (why behavior is not a direct readout of emotion).

5.4 What Intentional Communication Would Require

Research on dog-human communication has developed behavioral criteria for intentional communication that do not depend on words. In an experimental study of "showing" behavior, dogs watched food or a toy being hidden out of their reach. When the owner and the hidden object were both present, dogs showed gaze alternation, defined as "changing the direction of the gaze from the location of the food (toy) to looking at the owner (or vice versa) within 2 s" (Miklósi et al., 2000). The authors interpreted this as a sign of functionally referential communication.

Another relevant criterion is sensitivity to the audience. In a forbidden-food paradigm with 12 dogs, dogs behaved differently when a human was not watching than when she was (Call et al., 2003). Dogs track human attention, which is a prerequisite both for intentional communication and for reading human cues.

The keyboard study with Sofia already looked for such criteria (Rossi & Ades, 2008). A soundboard study could test them systematically: Do dogs press more when someone is attending? Do they look at the person after pressing? Do they repeat or change the press when the response does not match? So far, no published soundboard study has tested this.

5.5 The Lesson from Nim

In Nim's case, combinations that first looked structured turned out to be largely prompted by the teacher (Terrace et al., 1979). Soundboard households have a similar structure: owners model buttons and often reply to the dog's presses with their own. The 2024 combination study tested whether dogs' presses were fully predicted by owners' presses and found that they were not (Bastos, Houghton, et al., 2024). That is a partial answer to the Nim problem. It does not exclude more subtle forms of prompting, such as owners creating situations in which certain combinations are likely or reinforcing some combinations more than others.

The current evidence for the third level can therefore be summarized as follows: dogs' combinations are not random at the population level, and they are not simple copies of the owner's presses. Whether they express combined meanings intended for a listener has not been tested.

6. The Clever Hans Problem

6.1 Hans and the Questioner's Head

The classic case of a misread animal is Clever Hans, a horse that appeared to solve arithmetic problems by tapping his hoof. Pfungst (1911) showed that the horse was reading the questioner. When the questioner knew the answer, 90 to 100% of responses in the various series were correct. When the questioner did not know the answer, at most 10% were. The cue was a slight upward jerk of the questioner's head when the correct count was reached, and the questioners did not know they were giving it. The detailed history and its implications for dog research are covered elsewhere (what the Clever Hans effect means for dogs).

For soundboards, the problem works in two directions: the dog may respond to the owner's unintentional cues, and the owner may interpret presses in the light of expectations. Neither requires any dishonesty.

6.2 Owner Cues in Dog Studies: A Mixed Picture

Controlled research on owner cueing in dogs shows that the effect is real but not automatic. In a pointing study with 69 dogs, owners' beliefs about the correct choice did not change how well dogs followed an experimenter's pointing, and when only the owner knew where the food was, without pointing, dogs performed at chance. When owners deliberately directed their dogs, 10 of 16 dogs individually followed the owner, but only 2 of 15 did so when the experimenter pointed the opposite way (Schmidjell et al., 2012). A study from another lab came to a similar conclusion: owners with an incentive to influence their dogs' choices did not distort the results of a brief pointing task (Hegedüs et al., 2013).

This does not mean that owner cues never matter. In a study with 18 detection dog teams, no search condition contained the target scent, so every alert was incorrect. There were 225 incorrect responses, and more alerts were called at locations that handlers had been falsely told were marked. The authors concluded that "human more than dog influences affected alert locations" (Lit et al., 2011). The mechanism could not be separated: dogs may have responded to handler cues, or handlers may have interpreted the dogs' behavior in line with their beliefs.

6.3 The Interpreter Problem

The second direction of the Clever Hans problem is often more important for soundboards. In a study of the "guilty look" with 14 dogs, the behaviors owners interpret as guilt depended on whether the owner scolded the dog, not on whether the dog had actually disobeyed (Horowitz, 2009). Owners' beliefs shaped both their own behavior and their interpretation of the dog's behavior (what the research on the guilty look shows).

Soundboard interpretation has the same structure. When the dog presses "outside" while the owner is putting on shoes, the press looks meaningful; a press at midnight may be filed as noise. Selective interpretation of this kind can create the impression of a coherent conversation from behavior largely shaped by routine and reinforcement.

6.4 Selection and Reporting

A further layer comes from what gets recorded and shared. Owners may log remarkable presses more reliably, a limitation the authors of the combination study discuss (Bastos, Houghton, et al., 2024). On social media the selection is much stronger: the clips that circulate are those that look most like human conversation.

None of this means that soundboard dogs learn nothing. It means that home recordings are not evidence beyond what controlled studies have shown, and that the more a clip resembles a human conversation, the more carefully it deserves examination.

7. What Controlled Testing Looks Like

7.1 Remove the Knowing Human

The principle goes back to Pfungst: the person who could give cues must not know the answer, or the animal must not be able to perceive that person. The 2024 comprehension study applied it with hidden button identities, a blindfolded owner and blind video coding (Bastos, Evenson, et al., 2024).

7.2 Controls Against Simpler Explanations

Blinding addresses cueing, but other simple explanations remain. A dog may follow a button's position rather than its sound, which can be tested by swapping recordings between buttons. A dog may respond to any button sound with excitement, which requires control words like the nonce word in the comprehension study. And a dog may respond to the time of day, which requires presenting words when the outcome is not expected.

Careful measurement is part of this. A study must define in advance what counts as an appropriate response, how it is coded and by whom (how dog behavior is made measurable). "He looked at the door" is only evidence if looking at the door was defined before the trial and coded by someone who did not know what the dog had heard.

7.3 What Blind Testing Can Reveal

Performance under informal conditions can be very different from performance under blind conditions. In a trial of cancer detection dogs, ten dogs were trained and three reached the second stage. In the double-blind test with unfamiliar samples, the sensitivities of two dogs were 0.13 and 0.25, no better than chance, and the authors suggested the dogs may have memorized individual samples during training (Elliker et al., 2014). This is a different task, but the lesson transfers. Performance that looks convincing in training may depend on features that are not the ones the trainer believes are being learned.

7.4 Replication and Sample Size

Comparative cognition as a field has a replication problem. Published effect sizes tend to be inflated, samples are often small, and many claims are, in the words of one methods review, "practically unfalsifiable" without replication infrastructure (Farrar et al., 2020). The soundboard studies are larger than most earlier work on interspecies communication, with about 30 dogs per comprehension study and 152 dogs in the combination analysis. They are still the work of one research group, and their findings have not yet been independently replicated in published form.

8. Social Media Claims Versus Data

8.1 What Is Supported and What Is Not

Supported, at least in part: dogs learn to press buttons to obtain outcomes; some soundboard dogs respond to familiar outcome words such as "outside" and "play" without owner cues; and experienced dogs' button use is non-random and not a copy of their owners' presses. Not supported by published data: grammar or syntax, abstract buttons for feelings or time used as their labels suggest, reports about past events or absent people, and the idea that every dog can learn to respond to recorded words, which the word learning literature makes unlikely (Fugazza et al., 2021). These claims are not disproven. They are untested.

8.2 A Simple Rule for Reading Clips

A useful rule is to ask, for any clip, which of the three levels it is meant to show and whether a simpler level explains it. A dog pressing "outside" and going to the door is consistent with the first level. A dog going to the door when a stranger presses "outside" while the owner cannot see is evidence for the second. A dog pressing "outside" plus "play" and then playing outside is a combination, but the third level requires showing that the combination was directed at a listener and carried a combined meaning, which a single clip cannot do.

9. What Follows for Owners Who Use Buttons

9.1 Buttons as a Tool for Structured Choice

The strongest practical case for soundboards does not depend on language. A button that reliably produces an outcome gives the dog a clear, learnable way to influence its environment. This is a form of agency, and the research on controllability suggests that being able to influence outcomes matters for stress. The same principle underlies cooperative care, where dogs learn signals that let them start or pause a procedure (how cooperative care gives dogs a say).

In practice: In our training experience, buttons work best when they are few, clearly tied to concrete outcomes and answered consistently. A small board with "outside", "play" and "water" that is honored reliably gives the dog more real choice than a large board with dozens of abstract labels. This is an observation from practice, not a finding from controlled studies.

9.2 Keep the Outcome Honest

Operant learning only works if the press predicts the outcome. If "play" is answered with play only sometimes, the button loses its predictive value. If the outcome cannot be delivered, the dog may experience frustration, especially if pressing has been strongly reinforced in the past (what is known about frustration in dogs).

In practice: We recommend deciding in advance which buttons can always be honored and which only at certain times, and making the latter visible, for example by covering or removing them when they are not available. A button that is "on" but often does not work tends to increase pressing rather than reduce it. This comes from practical experience, not from a study.

9.3 Watch for Pressing as Attention-Seeking

Because owners often react to every press with interest, pressing can become a reliable way to get attention, independent of the button's label. That is not a problem as such, but it changes what the board is. If a dog presses many buttons in quick succession and settles once the owner responds, attention may be the main outcome.

In practice: A simple one-week log of which buttons the dog presses, when and with what result often shows that a few buttons carry most of the board. This is a suggestion from practice, not a research method.

9.4 Test Before You Believe

Owners who want to know whether their dog responds to the words themselves can borrow the logic of the controlled studies. Someone other than the usual trainer presses the button while the owner is out of sight; the button positions are changed; a control button with a nonsense word is included; and someone who does not know which button was pressed judges the response.

In practice: In our experience, some dogs respond clearly to one or two words in such tests, while others respond to position or routine. Neither result is a failure; it shows what the board actually is for that dog. This is an observation from practice, not a controlled study.

9.5 Keep the Rest of Communication in View

Buttons are one channel among many. Body posture, facial expression and vocalizations often say more about a dog's state than a button press does (what dogs' vocal signals convey). A board that draws attention away from those signals would be a step backward. A board used as one structured addition to careful observation can be a useful tool.

10. Summary at a Glance

The research on soundboards in dogs supports a graded set of conclusions. At the first level, dogs learn button-outcome associations, as operant learning predicts and a single-dog keyboard study showed (Rossi & Ades, 2008). At the second level, a controlled study with 30 and 29 dogs found that dogs responded appropriately to play-related and outside-related recorded words, regardless of whether the owner or an unfamiliar person presented them and regardless of button or voice, but not clearly to food-related words (Bastos, Evenson, et al., 2024). At the third level, an observational analysis of 152 dogs found that button presses were non-random at the population level and not simple copies of owners' presses (Bastos, Houghton, et al., 2024). Whether combinations carry meaning or are intentionally directed at a listener has not been tested.

Word learning is real but rare (Fugazza et al., 2021), ordinary dogs show neural signs of referential expectations for familiar words but process limited phonetic detail (Boros et al., 2024; Magyari et al., 2020), and the Clever Hans effect and selective interpretation are the central risks for home recordings (Pfungst, 1911; Horowitz, 2009).

11. Research Gaps and Critical Appraisal

The evidence base is small: two empirical peer-reviewed studies on modern soundboards in dogs, both from one research group, neither independently replicated in published form. The comprehension study tested 30 and 29 dogs, the combination study analyzed 152. That is more than most earlier interspecies communication research, but not enough to map individual differences.

The populations are self-selected: owners who chose to train with buttons and to take part in research, and in the combination analysis only dogs with at least 200 logged interactions. The findings cannot be transferred to dogs in general or to dogs new to buttons.

The measures are limited. The comprehension study coded approach and orientation from video, reasonable indicators of an expectation but not of the underlying representation. Only three word categories were tested, food without a conclusive effect, and no study has tested abstract buttons.

The combination study is observational and depends on owner logging. Order was ignored, reference was not tested, and selective logging cannot be ruled out. The conclusion that soundboard use is deliberate is an inference from non-random and non-imitative patterns, not a direct test of intention.

Key behavioral criteria for intentional communication have not been tested in soundboard dogs. These include audience effects (pressing more when someone is attending), gaze alternation between the button and the person, and persistence or repair when the response does not match the press. Such criteria were used in earlier work on dog-human communication (Miklósi et al., 2000) and in the single-dog keyboard study (Rossi & Ades, 2008), and they could be applied to soundboards.

The background literature has its own limits. The best-known word learning studies are single-animal studies (Kaminski et al., 2004; Pilley & Reid, 2011), and the neural studies tested 12 to 27 dogs with spoken words, not buttons. The owner cueing studies give a mixed picture: little owner influence in brief pointing tasks (Schmidjell et al., 2012; Hegedüs et al., 2013), a clear effect of handler beliefs in detection work (Lit et al., 2011). How strongly owner cues affect everyday soundboard use has not been measured.

Welfare effects have not been studied. The project paper names welfare as a research goal (Bastos & Rossano, 2023), but there are no published data on whether soundboards reduce frustration or increase a sense of control. The role of buttons as a tool for choice remains an extrapolation from the research on controllability.

Finally, comparative cognition as a field tends to produce inflated effect sizes in small samples (Farrar et al., 2020). Multi-site, preregistered replications with blinded testing, button-position controls and predefined behavioral criteria would be the most informative next step.

12. Conclusion

Soundboard dogs are not talking in the sense the videos suggest, but they are not pressing at random either. Dogs learn what buttons do, some respond to some recorded words without help from the owner's body language, and experienced dogs use their boards in a structured way that is not simply a copy of their owners' presses.

What has not been shown is that dogs combine buttons into meaningful messages, use abstract buttons as their labels suggest, or communicate intentionally through the board in a way that goes beyond requesting outcomes. These questions are open, and they can be tested with methods that already exist. Until they are, the most useful way to think about a soundboard is as a structured tool for choice and prediction: a way for a dog to influence its everyday life, which may be valuable in its own right.

For owners, this means keeping the board small and honest and testing before drawing conclusions. For anyone watching clips, it means asking which level a clip claims to show, and whether a simpler explanation is enough.

Key Insights (Takeaways)

• "Can dogs talk with buttons?" bundles three separate questions: button-outcome learning, response to the recorded word itself, and meaningful, intentional combination.

• Button-outcome learning is well supported by learning principles and an early single-dog keyboard study (Rossi & Ades, 2008).

• In a controlled study, soundboard dogs responded appropriately to play-related and outside-related words regardless of who presented them and whether by button or voice, but not clearly to food-related words (Bastos, Evenson, et al., 2024).

• In 152 experienced dogs, button presses were non-random at the population level and not simple copies of owners' presses; order, syntax and reference were not tested (Bastos, Houghton, et al., 2024).

• Word learning in dogs is real but unevenly distributed, and most dogs do not learn object names even with intensive training (Fugazza et al., 2021).

• The Clever Hans effect works in two directions in soundboard homes: dogs may read owners, and owners may read meaning into presses (Pfungst, 1911; Horowitz, 2009).

• Controlled testing requires blinded presenters and coders, control words and position controls; claims beyond these tests remain open.

• The strongest practical value of buttons lies in structured choice and predictability, not in proof of language.

References

Andics, A., Gábor, A., Gácsi, M., Faragó, T., Szabó, D., & Miklósi, Á. (2016). Neural mechanisms for lexical processing in dogs. Science, 353(6303), 1030–1032. https://doi.org/10.1126/science.aaf3777

Bastos, A. P. M., Evenson, A., Wood, P. M., Houghton, Z. N., Naranjo, L., Smith, G. E., Cairo-Evans, A., Korpos, L., Terwilliger, J., Raghunath, S., Paul, C., Hou, H., & Rossano, F. (2024). How do soundboard-trained dogs respond to human button presses? An investigation into word comprehension. PLOS ONE, 19(8), e0307189. https://doi.org/10.1371/journal.pone.0307189

Bastos, A. P. M., Houghton, Z. N., Naranjo, L., & Rossano, F. (2024). Soundboard-trained dogs produce non-accidental, non-random and non-imitative two-button combinations. Scientific Reports, 14, 28771. https://doi.org/10.1038/s41598-024-79517-6

Bastos, A. P. M., & Rossano, F. (2023). Soundboard-using pets? Introducing a new global citizen science approach to interspecies communication. Interaction Studies, 24(2), 311–334. https://doi.org/10.1075/is.22050.pin

Boros, M., Magyari, L., Morvai, B., Hernández-Pérez, R., Dror, S., & Andics, A. (2024). Neural evidence for referential understanding of object words in dogs. Current Biology, 34(8), 1750–1754.e4. https://doi.org/10.1016/j.cub.2024.02.029

Call, J., Bräuer, J., Kaminski, J., & Tomasello, M. (2003). Domestic dogs (Canis familiaris) are sensitive to the attentional state of humans. Journal of Comparative Psychology, 117(3), 257–263. https://doi.org/10.1037/0735-7036.117.3.257

Dror, S., Miklósi, Á., Sommese, A., & Fugazza, C. (2023). A citizen science model turns anecdotes into evidence by revealing similar characteristics among Gifted Word Learner dogs. Scientific Reports, 13, 21747. https://doi.org/10.1038/s41598-023-47864-5

Dror, S., Miklósi, Á., Sommese, A., Temesi, A., & Fugazza, C. (2021). Acquisition and long-term memory of object names in a sample of Gifted Word Learner dogs. Royal Society Open Science, 8(10), 210976. https://doi.org/10.1098/rsos.210976

Elliker, K. R., Sommerville, B. A., Broom, D. M., Neal, D. E., Armstrong, S., & Williams, H. C. (2014). Key considerations for the experimental training and evaluation of cancer odour detection dogs: Lessons learnt from a double-blind, controlled trial of prostate cancer detection. BMC Urology, 14, 22. https://doi.org/10.1186/1471-2490-14-22

Farrar, B. G., Boeckle, M., & Clayton, N. S. (2020). Replications in comparative cognition: What should we expect and how can we improve? Animal Behavior and Cognition, 7(1), 1–22. https://doi.org/10.26451/abc.07.01.02.2020

Fugazza, C., Dror, S., Sommese, A., Temesi, A., & Miklósi, Á. (2021). Word learning dogs (Canis familiaris) provide an animal model for studying exceptional performance. Scientific Reports, 11, 14070. https://doi.org/10.1038/s41598-021-93581-2

Hegedüs, D., Bálint, A., Miklósi, Á., & Pongrácz, P. (2013). Owners fail to influence the choices of dogs in a two-choice, visual pointing task. Behaviour, 150(3–4), 427–443. https://doi.org/10.1163/1568539X-00003060

Higaki, F., Faragó, T., Pogány, Á., Miklósi, Á., & Fugazza, C. (2025). Sound quality impacts dogs’ ability to recognize and respond to playback words. Scientific Reports, 15, 14175. https://doi.org/10.1038/s41598-025-96824-8

Horowitz, A. (2009). Disambiguating the "guilty look": Salient prompts to a familiar dog behaviour. Behavioural Processes, 81(3), 447–452. https://doi.org/10.1016/j.beproc.2009.03.014

Kaminski, J., Call, J., & Fischer, J. (2004). Word learning in a domestic dog: Evidence for "fast mapping". Science, 304, 1682–1683. https://doi.org/10.1126/science.1097859

Lit, L., Schweitzer, J. B., & Oberbauer, A. M. (2011). Handler beliefs affect scent detection dog outcomes. Animal Cognition, 14(3), 387–394. https://doi.org/10.1007/s10071-010-0373-2

Magyari, L., Huszár, Z., Turzó, A., & Andics, A. (2020). Event-related potentials reveal limited readiness to access phonetic details during word processing in dogs. Royal Society Open Science, 7(12), 200851. https://doi.org/10.1098/rsos.200851

Miklósi, Á., Polgárdi, R., Topál, J., & Csányi, V. (2000). Intentional behaviour in dog-human communication: An experimental analysis of "showing" behaviour in the dog. Animal Cognition, 3(3), 159–166. https://doi.org/10.1007/s100710000072

Pfungst, O. (1911). Clever Hans (the horse of Mr. von Osten): A contribution to experimental animal and human psychology (C. L. Rahn, Trans.). Henry Holt.

Pilley, J. W., & Reid, A. K. (2011). Border collie comprehends object names as verbal referents. Behavioural Processes, 86, 184–195. https://doi.org/10.1016/j.beproc.2010.11.007

Prichard, A., Cook, P. F., Spivak, M., Chhibber, R., & Berns, G. S. (2018). Awake fMRI reveals brain regions for novel word detection in dogs. Frontiers in Neuroscience, 12, 737. https://doi.org/10.3389/fnins.2018.00737

Rossi, A. P., & Ades, C. (2008). A dog at the keyboard: Using arbitrary signs to communicate requests. Animal Cognition, 11(2), 329–338. https://doi.org/10.1007/s10071-007-0122-3

Savage-Rumbaugh, S., McDonald, K., Sevcik, R. A., Hopkins, W. D., & Rubert, E. (1986). Spontaneous symbol acquisition and communicative use by pygmy chimpanzees (Pan paniscus). Journal of Experimental Psychology: General, 115(3), 211–235. https://doi.org/10.1037/0096-3445.115.3.211

Schmidjell, T., Range, F., Huber, L., & Virányi, Z. (2012). Do owners have a Clever Hans effect on dogs? Results of a pointing study. Frontiers in Psychology, 3, 558. https://doi.org/10.3389/fpsyg.2012.00558

Tempelmann, S., Kaminski, J., & Tomasello, M. (2014). Do domestic dogs learn words based on humans' referential behaviour? PLoS ONE, 9(3), e91014. https://doi.org/10.1371/journal.pone.0091014

Terrace, H. S., Petitto, L. A., Sanders, R. J., & Bever, T. G. (1979). Can an ape create a sentence? Science, 206(4421), 891–902. https://doi.org/10.1126/science.504995