[Cet article a initialement été publié le 11 janvier
2026]
Some dogs understand dozens, even hundreds of words, well beyond simple classic commands. This capacity, long considered exceptional, has just been analyzed in a study published on January 8, 2026 in Science by a team of researchers from Eötvös Lorönd University (Hungary) and the University of veterinary medicine of Vienna. Their work demonstrates that certain dogs called GWLs (for “Gifted Word Learners”) can learn new words without direct interaction, simply by listening to exchanges between humans.
This ability, previously observed in children from 18 months of age, is based on complex social skills such as eye tracking or detection of communicative cues. These results raise a central question: to what extent can non-human species access learning mechanisms that were thought to be specific to humans, particularly those involved in the emergence of language?
Observational learning comparable to that of young children
The new study therefore demonstrates that some dogs learn words by observing human interactions, without direct intervention. These so-called GWL dogs can associate a word with an object only by hearing this word in a conversation between humans. The phenomenon was documented in ten dogs selected for their ability to recognize the names of several toys.
The researchers, led by Shany Dror (Eötvös Loránd University and University of Veterinary Medicine Vienna), designed experiments based on the protocols used with 18-month-old children. This choice made it possible to compare the learning mechanisms of two very different species, but sharing a social proximity with humans.
The dogs observed their owner chatting with another person while handling a new toy. At no time were they requested. They remained behind a barrier or in their basket, without physical contact or direct gaze.
Despite this social distance, 7 of the 10 dogs were able to recognize and correctly report the two toys heard during subsequent tests. The success rate was around 80%, close to the results obtained during traditional learning.
This behavior mimics that observed in young children capable of learning by watching two adults interact. According to Dror, in a press release, this indicates that these dogs “are able to interpret complex social interactions and extract relevant information».
A controlled and rigorous experimental method inspired by cognitive sciences
The researchers designed a methodology with 3 experiments to distinguish real learning and simple preference for novelty. In the first, called the “addressed” condition, owners presented a new toy to their dog by repeating its name for one minute. Then, a three-minute play phase followed, without mention of name, then 20 minutes of free exploration. This sequence was repeated twice a day over four non-consecutive days, for each new toy. In total, each dog received 8 minutes of exposure to the word.
© Shany Dror et al., 2025
Overview of tested methods for teaching dogs the names of toy labels.
The second experiment, called “passive listening”, reproduced the same structure, but without direct interaction. Two humans handled the toys, named them, and chatted with each other, under the eyes of the spectator dog. The researchers took care to eliminate any non-verbal communication with the animal (looks, directed intonation, gestures).
The recognition tests took place in another room. The dog had to choose the correct toy from two new and nine old toys. The rate of correct responses reached 90% in the addressed condition and 80% in the passive listening condition, with no statistically significant difference.
A third, more complex experiment tested learning without simultaneity between the object and the word. The toy was shown then hidden. The word was then pronounced when the object was no longer visible. Despite this discontinuity, 5 out of 8 dogs made the association correctly. The controlled experiment, validated by the ethics committee, excludes learning bias and shows that GWL dogs form lasting mental associations, even without traditional reinforcement.
Rare cognitive performances linked to exceptional individual skills
The study highlights the existence of rare cognitive abilities, far from being generalizable to all dogs. To verify this, the researchers repeated the “passive listening” experiment on 10 ordinary Border Collies, with no history of learning object names. The results were clear: none of these dogs succeeded in learning the names of new toys.
When these dogs made a choice, they selected one of two new toys. This indicates a preference for novelty, not an understanding of the word heard. Statistically, their performances did not exceed chance.
On the other hand, GWL dogs, identified via the project Genius
Dog Challengehave an initial vocabulary of up to 100 toy names, acquired naturally in their daily life. Some can memorize up to 12 new words per week, and retain them for more than two years, according to Dror.
These abilities appear to result from a mixture of individual factors, probably including motivation, life context, social reinforcement and innate cognitive skills. Dror evokes a telling comparison. “It’s like comparing a bicycle to a car. Both are moving forward, but not with the same engine.”
Affected dogs do not all share the same breed, although Border Collies are overrepresented. The panel studied also included a Labrador, a German Shepherd and a Blue Heeler. This heterogeneity suggests that the phenomenon is based less on race than on very specific individual traits. Traits still poorly understood at this stage.
A new avenue for understanding the evolution of human language
The observation that some dogs can learn words without direct interaction raises a key question. What social skills precede the appearance of language? According to the team, the results suggest that certain bases of language learning rely on cognitive mechanisms shared between species, prior to the emergence of articulate language.
In the study, GWL dogs were able to recognize words even when the object was no longer visible. This involves abilities like associative memory, understanding human intentions, and the ability to follow a complex social context.
These mechanisms are also observed in very young children, from 18 months. At this age, they can learn words by passive listening. The comparison does not imply biological equivalence, but similar functioning in the use of social cues.
The researchers note that the evolution of dogs, living alongside humans for millennia, may have favored the development of such traits. As Dror explains:
“Dogs that were able to understand our social cues had an advantage in surviving alongside us.”
The scope of this study goes beyond canine ethology. She suggests that understanding language might rely on general social skills. And not on human linguistic capacity alone. These results pave the way for new research in cognitive comparison, to determine whether other species possess these skills. They also question the way in which humans used these foundations to develop complex language.
Source : Shany Dror et al., “Dogs with a large vocabulary of object
labels learn new labels by overhearing like 1.5-year-old
infantsâ€. Science 391,160-163 (2026).
