Some Dogs Eavesdrop on Their Owners to Learn New Words
An Anonymous Coward writes:
https://arstechnica.com/science/2026/01/these-dogs-eavesdrop-on-their-owners-to-learn-new-words/
[ Links in article ]
"Under the right conditions, some dogs present behaviors strikingly similar to those of young children."
When it comes to cognitive ability, not all dogs are created equal. Most dogs can learn simple action cues like "sit" or "down." But so-called "gifted word learner" (GWL) dogs exhibit a remarkable ability to learn the names of objects-for example, learning the names of specific toys so well that they can retrieve them from a large pile of toys on command. And according to a new study published in the journal Science, they can even learn labels for new toys just by overhearing their owners talking about those toys. Per the authors, this suggests that GWL dogs have sociocognitive skills that are functionally comparable to those of 18-month-old human toddlers.
Co-author Claudia Fugazza of Eotvos Lorand University in Budapest, Hungary, has been studying canine behavior and cognition for several years as part of the Genius Dog Challenge. For instance, the group's 2022 study discovered that dogs store key sensory features about their toys-notably what they look like and how they smell-and recall those features when searching for the named toy. Prior studies had suggested that dogs typically rely on vision, or a combination of sight and smell, to locate target objects. GWL dogs can also identify objects based on verbal labels.
In that 2022 study, all the dogs-regardless of whether they were GWL dogs or typical dogs-successfully picked out the target toys in both light and dark conditions, though it took them longer to locate the toys in the dark. Most relied on visual cues, even though dogs possess an excellent sense of smell. However, the dogs sniffed more frequently and longer when searching for the toy in the dark.
The GWL dogs were also able to select the named toys when commanded by their owners, with similar reliance on visual cues-what the toy looks like-augmented by their sense of smell (what the toy smells like), particularly in dark conditions. This confirmed that when dogs play with a toy, they record its features using multiple senses, creating a "multistory mental image." They prefer to rely primarily on visual cues, but dogs can incorporate other sensory cues, most notably smell, when the conditions call for it.
Last fall, Fugazza's group built on those findings and concluded that certain dogs can not only memorize the names of objects like their favorite toys, but they can also extend those labels to entirely new objects with a similar function, regardless of whether or not they are similar in appearance. It's a cognitively advanced ability known as "label extension," and for animals to acquire it usually involves years of intensive training in captivity. But the dogs in this new study developed the ability to classify their toys by function with no formal training, merely by playing naturally with their owners. It's akin to a person calling a hammer and a rock by the same name, or a child understanding that "cup" can describe a mug, a glass, or a tumbler because they serve the same function.
This time, Fugazza et al. wanted to investigate how well GWL dogs listen to and learn from human verbal interactions. So they adapted an experimental protocol used to study this ability in children for their canine subjects. All dogs tend to get very excited about new toys, so the subjects (10 GWL dogs) were allowed to play with two new toys first. Then the owners would spend one minute holding the toy and identifying it with a simple label ("This is Toy A" or "This is Toy B") while directly addressing their dogs. The dogs then played with the toys (both with their owners and on their own) for several minutes with no further mention of the labels.
This training protocol was repeated twice a day for four nonconsecutive days, followed by a label comprehension test. The two new toys were placed on the floor with nine familiar toys in a room out of the owners' line of sight. Owners would first ask their dogs to retrieve a familiar toy and then ask them to retrieve a new toy using its label ("Can you bring ?"). The label comprehension test was repeated 12 times for both the familiar and new toys.
Next, the entire experiment was repeated with one key variation: This time, during the training protocol, rather than addressing the dogs directly when naming new toys, the dogs merely watched while their owners talked to another person while naming the toys, never directly addressing the dogs at all.
The result: 80 percent of the dogs correctly chose the toys in the direct address condition, and 100 percent did so in the overhearing condition. Taken together, the results demonstrate that GWL dogs can learn new object labels just by overhearing interactions, regardless of whether the dogs are active participants in the interactions or passive listeners-much like what has been observed in young children around a year-and-a-half old.
To learn whether temporal continuity (a nonsocial factor) or the lack thereof affects label learning in GWL dogs, the authors also devised a third experimental variation. The owner would show the dog a new toy, place it in a bucket, let the dog take the toy out of the bucket, and then place the toy back in. Then the owner would lift the bucket to prevent the dog from seeing what was inside and repeatedly use the toy name in a sentence while looking back and forth from the dog to the bucket. This was followed by the usual testing phase. The authors concluded that the dogs didn't need temporal continuity to form object-label mappings. And when the same dogs were re-tested two weeks later, those mappings had not decayed; the dogs remembered.
But GWL dogs are extremely rare, and the findings don't extend to typical dogs, as the group discovered when they ran both versions of the experiment using 10 non-GWL border collies. There was no evidence of actual learning in these typical dogs; the authors suggest their behavior reflects a doggy preference for novelty when it comes to toy selection, not the ability to learn object-label mappings.
"Our findings show that the socio-cognitive processes enabling word learning from overheard speech are not uniquely human," said co-author Shany Dror of ELTE and VetMedUni universities. "Under the right conditions, some dogs present behaviors strikingly similar to those of young children. These dogs provide an exceptional model for exploring some of the cognitive abilities that enabled humans to develop language. But we do not suggest that all dogs learn in this way-far from it."
Science, 2025. DOI: 10.1126/science.adq5474 (About DOIs).
Read more of this story at SoylentNews.