From training trainers… back to training dogs (to count!)
I’ve always been fascinated by the cognitive abilities of animals. I used to teach a graduate course on animal intelligence. The popularity of the course was indicative of how interested people are in this topic.
Throughout my career I’ve been fortunate to work on a number of cognitive studies. I’ve co-designed studies on echolocation in dolphins, on match-to-sample with a sea lion, and on imitation in dogs. However, in most cases, I was not the researcher but the trainer, helping cognitive scientists train various species to participate in projects. I learned first-hand that the scientific peer-review process is a very rigorous, but necessary, step to validate that research is conducted properly.
Over the past 15 years my exposure to scientific research led to my interest in conceptual training with dogs. At the start, my focus wasn’t on research, but on having fun with the animals and helping others train concepts such as modifiers, adduction, matching-to-sample, and mimicry. It was only after many years of teaching people how to train these concepts that I became acutely aware of the lack of published information on some aspects of the cognitive abilities of dogs.
In 2013, I began what I thought would be a brief project: train a few dogs to understand the concept of counting. Previously, I had trained several dogs to count to three, a task that is not particularly difficult. Science has shown that dogs have the ability to recognize quantities of three easily. I had no idea where this journey would take me over the next several years! It certainly would not have been possible without my enthusiastic partner, Coral, a three-year old, highly reactive, rescued Airedale mix.
At ClickerExpo 2014 I presented the early results of teaching Coral to count. It was a simple project, but many trainers were fascinated because they had not seen it trained previously.
Coral learned to look at objects placed in a tray and then touch a target, indicating how many objects were in the tray. A blue circle indicated one object, a green rectangle indicated two, and a black triangle indicated three. Essentially, the number of objects in the tray became Coral’s cue, signaling which target shape she should touch. Coral achieved an accuracy rate of 98% over many trials.
After my presentation, colleagues Dr. Susan Schneider and Dr. Susan Friedman, separately, gave me great feedback and suggestions for improving the project design and turning the project into publishable research. They both believed in the project and gently encouraged me to keep going, despite my insistence that I was not interested in taking the project further.
Phase 2—shared anew
Six months later, I started over with the project, with a number of new protocols in place. This time, Coral learned to indicate the number of items in the tray by touching a whiteboard with magnetic dots on its surface. After Coral looked at the objects in the tray, she had to touch the whiteboard with the corresponding number of dots. To succeed at this, I presumed that Coral would have to “count” twice—first the items in the tray, and then the dots on the board.
I set up what scientists refer to as “double-blind trials” that prevented me from communicating the right answers to Coral unknowingly, or biasing her responses in any way.
The sequence of events and the protocols used in a typical trial:
- I give Coral the cue “How Many?”
- Coral pokes her head into a curtained-off area, where a second person has placed a number of objects in the tray.
- I cannot see the tray of objects or the person placing the objects, and never know the correct answer to the question.
Person #2 (who is placing objects in the tray) is not able to see the whiteboards, and has no ability to prompt or assist Coral inadvertently as she picks the correct answer. The number of objects and the specific objects placed in the tray is determined randomly by a computer program. A third person, unseen by Coral, me, or Person #2, has a one-way view of the boards and marks with a clicker if Coral makes a correct choice. I reinforce Coral if I hear the click. I do nothing if there is no click. Either way, I set her up for another trial. Each session consists of 10 trials. The sequence of the boards is random and changes frequently in each session. The position of the dots on each board is changed regularly to guard against possible pattern recognition.
We progressed to the number eight. Coral’s success remained above 95% for numbers one through five, and dropped to 90% with numbers six through eight. These response rates were far above chance, and I felt certain that I was in a solid position to suggest that Coral had learned to count. I presented this work at ClickerExpo 2015 and received very enthusiastic responses.
More colleagues offer feedback
ClickerExpo faculty member Dr. Jesús Rosales-Ruiz invited me to present my canine counting research at an annual meeting at the University of North Texas. I presented the project to an invited group, and looked forward to feedback from the students and the leading behavior analysts and educators who were in attendance.
While the audience members applauded the work that had been accomplished, they had many suggestions and questions. They acknowledged that Coral was doing something quite extraordinary, but they doubted it could actually be called counting. Although I had a good double-blind set-up for trials, people in the audience helped me see where biases or other factors might be influencing Coral’s responses. They encouraged me to publish—with a few little tweaks…
Once again, I said that I had no interest in taking this project further. I had been thwarted and derailed several times in past years attempting to publish my work with imitation in dogs; it was not a process I was eager to go through again.
To continue or not?
Despite my original reluctance to continue with this project (my outright refusal to even consider it at one point), I could not help but be intrigued by the excellent suggestions I had been given in Texas. Perhaps the most helpful advice was to research similar work that had been done with children. This idea helped me recognize that defining counting is neither easy nor clear. Cognitive psychologists still debate the term counting. No wonder I was having trouble convincing people that Coral was counting!
There were other challenges to resolve. Coral had been adopted out to a loving home. I had not seen her for a long time, and she was no longer part of a structured training program. Plus, I did not have access to the training space where our previous training took place, or to the staff who could assist me.
Despite these obstacles, a voice in my head said there would never be a better time. I had trained many dogs to count, but none had Coral’s enthusiasm and focus. How could I not take advantage of her eagerness, and discover how much more she could learn?
Phase 3—to be shared
How many rings?
How many Kongs?
How many balls?
In the summer of 2015 I embarked on a new, and hopefully improved, phase of the counting project. There were many obstacles to overcome and many new suggestions to implement. The most significant changes included:
- Using “matching-to-sample” as a mechanism to ask Coral multiple questions about a single set of objects placed in a tray
- Varying the size and shapes of the “dots” on the whiteboard to further interrupt possible pattern recognition
- Introducing Coral to multiple “new” or never-before-seen numbers in a single session with no pre-training
We tested Coral in trials that included as many as 22 objects. Preliminary data indicates that she achieved an accuracy rating of 79% with numbers up to 14. As we progressed to larger numbers, her accuracy began to drop quickly. Although she still performed with accuracy above chance, quantities higher than 14 began to frustrate Coral.
Interpreting the results
At ClickerExpo, I will be presenting a Session called Are You As Smart As A Dog? In the presentation I will share videos of the counting training and a representative sampling of many of the trials. I will also share what I discovered about published studies of counting with children, and try to correlate or compare what I have learned about Coral’s abilities. At this point it is difficult to draw any definite conclusions about counting in dogs, but Coral showed remarkable skill at quantities up to 14.
Is Coral actually counting? Is there any significance to the fact that she could recognize quantities of 14 easily but not higher than that? Does the fact that she was a highly reactive rescue dog have a correlation to these abilities? Did positive reinforcement training have an important influence on Coral’s ability to learn as much as she did? Where should others take this information, and what additional research might this project prompt? I will attempt to answer these questions and more.
series of Phase 3 trials
I believe that animals are capable of far more than we give them credit for. A good relationship with an enthusiastic animal, combined with a thoughtful training plan, can produce astonishing results. I am excited to share Coral’s amazing journey and hope to work with others to take this project further.