What Changes When Dogs Repeat a Learning Task?
Table of contents
When dogs repeat a learning task, improvement may show up in one part of the task and not another. A 2026 study followed 105 dogs through the same structured hand-touch task twice, about 30 to 35 days apart. The clearest group-level change appeared in reversal learning, where the rewarded hand switched. Discrimination learning, extinction-related persistence, and the study's operational frustration-like component showed limited overall change.
The question behind the study
The researchers asked whether dogs from different breed clades changed their performance after repeating the same learning task. They also asked whether the gaps seen between clades during the first test stayed the same, and whether some clades changed more than others.
This is a single-study brief, not a systematic review. The evidence applies to one structured hand-touch task, the dogs who completed it, and the two testing occasions in this paper. It does not provide a general intelligence score or a household training assessment.
What the dogs did
The researchers initially recruited 151 dogs from five breed clades. After exclusions, 111 dogs completed the first testing occasion and 105 completed both occasions. The final group included 21 UK Rural dogs, 19 Asian Spitz dogs, 22 Retrievers, 20 New World dogs represented by German Shepherds, and 23 European Mastiffs. The dogs ranged from 0.5 to 11.58 years old.
Owners performed the sessions at home while a researcher guided them virtually. The task used the owner's hands and food rewards:
- During acquisition, dogs learned to touch one presented hand with the nose for a food reward.
- During discrimination learning, both hands appeared and only one hand was reinforced.
- During reversal learning, the reinforced hand switched, so the dog had to update a previously learned choice.
- During extinction, food reinforcement stopped and the researchers measured how the dog continued to respond.
| Species | Domestic dogs (Canis familiaris) |
|---|---|
| Final sample | 105 dogs completing both testing occasions |
| Breed grouping | Five breed clades, with the New World group represented by German Shepherd dogs |
| Timing | Test 2 took place approximately 30 to 35 days after Test 1 |
| Task phases | Acquisition, discrimination learning, reversal learning, and extinction |
| Analysis | Comparable task-specific component scores and repeated-measures mixed models |
What changed on repeat exposure?
The main result was selective. Average reversal-learning difficulty decreased from Test 1 to Test 2. In the paper's component score, a higher value means more difficulty, so a lower score represents easier performance on that part of the task. The reported contrast was 0.721 plus or minus 0.176 for Test 1 minus Test 2, with an adjusted p-value below 0.001.
Discrimination learning showed no statistically significant time change. The descriptive values also show why a single accuracy number needs its task and denominator beside it: dogs averaged 75.9 percent correct choices in Test 1 and 69.3 percent in Test 2 during discrimination, while the average sessions needed to reach the criterion were 2.22 and 2.61. These descriptive changes did not establish a significant improvement.
| Component | Reported pattern | Careful reading |
|---|---|---|
| Discrimination learning | No significant average time change | Repeat exposure did not automatically improve the first choice rule. |
| Reversal learning | Average difficulty improved from Test 1 to Test 2 | Updating a previously reinforced choice was the clearest group-level change. |
| Perseverance | No comparable overall change in the extinction component | The result is tied to this extinction procedure, not persistence in every situation. |
| Operational emotionality | No comparable overall change; one breed-clade pattern differed | The label summarizes coded behaviors and is not a direct measure of private emotion. |
Evidence-led insights
- Practice can be component-specific. The study found its clearest improvement when the reward rule changed, not during the initial hand discrimination phase. A reasonable inference is that repeated experience may help a dog update a familiar rule without improving every part of a compound task. The paper did not test a home training program, so this is not a promise that repetition will produce the same result in every dog.
- One test can make breed differences look more fixed than they are. Most initial between-clade differences were no longer evident at the second test, although some differences persisted. The authors found limited evidence that clades differed in the magnitude of their change. This supports a cautious reading of one-time breed comparisons: prior exposure and task familiarity can shape the result.
- The task defines the meaning of improvement. Reversal learning here meant switching away from a previously rewarded hand. That is narrower than learning in general. The familiar human-hand stimulus, the order of the phases, and the repeated reward changes may have affected the outcome. The authors specifically caution that the result may not transfer to other reversal tasks.
- Labels need their measures beside them. The paper's component called emotionality combined coded frustration-like behaviors, such as vocalization, jumping, pawing, or nipping, with a statistical reduction of several measures. It is useful for comparing dogs within this experiment. It is not a diagnosis or a direct readout of what an individual dog feels.
What the study cannot tell us
The dogs were tested only twice. The researchers could not control whether some owners practiced or reinforced the hand-touch behavior between sessions. The multi-phase design also means that fatigue, boredom, carryover from earlier phases, or a mixed reinforcement history may have influenced later performance.
The sample was organized into breed clades, and some individual breeds were represented more heavily than others. The results therefore cannot rank breeds or explain how every dog in a clade will learn. The study also does not show that a dog with a different result has a problem, lacks intelligence, or needs a particular intervention.
The external data belong to the study authors. They are not data collected by Viva Essence Pet, and the paper does not compare products, trainers, food brands, or household routines.
A careful takeaway for dog families
For everyday learning, a reasonable inference is to track one defined cue over several similar sessions rather than judge a dog's general ability from one attempt. Keep the goal and reward rule clear, note what changed between sessions, and separate the dog's response style from whether the task was completed. That approach matches the paper's central lesson: experience can change some parts of performance while leaving others relatively stable.
If a behavior concern affects daily life, this study cannot replace an individualized assessment. Its value is narrower and practical: it shows why the same dog may look different when a task is repeated, and why a single score should always be read with the task, timing, sample, and limits beside it.
Sources and transparency
Azadian A, Protopopova A. Changes in cognitive performance following repeated exposure to a hand-touch learning task across breed clades of domestic dogs (Canis familiaris). Animal Cognition. Published January 16, 2026; 29(1):20. DOI: 10.1007/s10071-026-02044-6.
The open-access full-text record provides the methods, results, and limitations. The publisher record confirms the journal article and publication details. The source was checked again on September 29, 2026. Reported findings and inferences are kept separate in this brief.