Stationary behavior: What are we really looking for?
Recently, I watched a man working on duration of a behavior—his dog's front feet, stationary, on a target. Watching his training session, I did not see anything out of the ordinary. But there was a problem. The trainer said he had achieved 5-7 seconds of duration, yet when the class instructor asked for a demonstration, he could only demonstrate the barest fraction of a second of standing still on the target.
Hey, I understand that this happens. Dogs are not able to do as well when they are called upon to demonstrate as they do when they are practicing.
Later, I watched back the video of this team working and I noticed something critical. The trainer was treating the dog intermittently while the dog's feet were on the block. To the trainer, it seemed as if the dog stayed on the block for several seconds, but the dog had not learned to stay as long as 5-7 seconds. Because the dog was being treated every one or two seconds for a total of 5-7 seconds, the trainer thought he had achieved a significant length of time. However, when he was asked to demonstrate 5-7 seconds of duration, he could not.
When you teach duration of a stationary behavior, you have to ask: What exactly is the concept that is being taught? Is it:
- As long as you remain there, I will bring you treats?
- Remain there until you are cued to do something else?
- Offering to stay there will earn a click?
- Fast training loops that are repeated with no delays?
- Or, some combination?
ClickerExpo clarity for me!
I began to reflect on my own stationary behavior/duration journey as a clicker trainer after observing the training team with problems. I, too, fed the dog treats while the dog was in the desired position. Inadvertently, I even taught my first small dog to break a stay by doing that. I was coming in and giving him the treat about an inch too high. He had to flex his rear-end muscles and rise up to eat the treat. Given that he was a hairy Havanese (I wasn't able to see his rear end levitate that small amount), it took me three months to figure out why he kept breaking his stay whenever I walked back to treat him!
After some reflection, I was able to come up with the history, and the contributors to that history, of how I moved on from doing intermittent treating for duration behaviors. I progressed to more reliable training methods with the help of Karen Pryor and other established clicker trainers and faculty members, some at ClickerExpo during my first two or three years of attendance at the conference.
I was stunned the first time I heard Karen Pryor say that giving treats to a dog during duration of a stationary behavior is slovenly training, and that the treat does not carry the key information! At the time, that was how I was training duration. Like everything else Karen has said that has shaken my thought process, I parked the information in the back of my mind for mulling over. And, as always happens when Karen tosses me her wisdom, the sense percolated to the surface—I got it! The power of information is the click. All we have to do is click for an established criterion.
In an Expo panel, Jesús Rosales-Ruiz discussed how Skinner always said behavior has a beginning, a middle, and an end. The behavior is not complete until the dog is in a position to repeat the behavior. This dovetails nicely with Alexandra Kurland's description of "loopy" training. Training loops follow the pattern: cue-behavior-click-treat-cue-behavior-click-treat-cue and so on—all occurring in quick, clean loops.
Another "ah ha!" moment for me came during Karen's first presentation of her adventures with the neuroscientists as she was gathering data for Reaching the Animal Mind. She spoke of agreeing with Peter Holland; there is a great deal of difference between giving treats ("that's a good place to be") and using a conditioned reinforcer ("right there, do that again"). According to Holland, that difference is located in the brain pathways.
Again, I got it! Conditioned reinforcers are far more powerful than delivery of primary reinforcers.
Kathy Sdao drove home this conclusion even more when she coached me to teach my two Havanese dogs to each station on their respective little chairs while the other was being trained. I had tried to train this behavior the "old way," giving treats while the dog was in position (on the chair), without much progress. I would work on it a bit, then give it up for a few months, then go back and try again.
I wanted this behavior badly and finally had the chance at ClickerExpo to ask Kathy Sdao how she trained marine mammals to stay on station. Kathy described a Premack training session for me: Dog A on his chair was marked for being on his chair, and then immediately released to get off and be trained while Dog B was cued to get on his chair. It took only a few days of this kind of training before I e-mailed Kathy: "Kathy, that worked really great, but I have a problem. I now can't get either of the dogs off their chairs to be the working dog!"
Kathy replied, "Oh! I forgot to tell you that Premack will reverse itself!" [Editor's note: The Premack principle is described in more detail below.]
In Reaching the Animal Mind, Karen Pryor discusses tertiary reinforcers—cues for positively trained behaviors are even more powerful than clicks. By inserting a cue for another behavior as a "click" to mark duration criterion, I could make use of a cue as a reinforcer and use the Premack Principle. Once it is fully trained, the cue or signal for an opportunity to earn a click/treat can produce a spike of dopamine in the brain—the joy, the delight, the delicious anticipation.
Thanks to the adjustments that I made to my own training, I've learned to tweak the criteria I employ when I train duration and other behaviors. I've also learned about the many benefits of treating out of position, as opposed to intermittent treating in a stationary position. With this article and its accompanying videos, it is simple to learn to adjust criteria and make progress toward a behavior goal like "stay until released."
There is a difference between treating a dog in the desired position, "explaining" that the dog should remain until cued otherwise, and having good, clean training loops. As coined by Alexandra Kurland, a loop translates to this sequence: cue->behavior->criterion is met->click/treat.
If you click the dog for meeting the duration criterion of the moment successfully, then the dog has offered that level of duration. In the next repetition, if you click for the same (or a new) criterion of duration, the dog is still offering duration.
On the other hand, if you give a treat while the dog is "doing duration," then another treat, then another, and then finally a click, it is the trainer who is making that duration happen. The dog is no longer offering to extend duration. It's the difference between classical conditioning ("This is a very nice place to be.") and operant conditioning ("If I do that again, I'll get a click!").
Premack and other tools
I want the dogs that I work with to be in operant (offering) mode as much as possible. Once I focused on a specific duration criterion, clicked for the dog meeting that criterion, and treated in a location that would foster the success of the next behavioral "loop," my duration training soared!
I started adding a release cue ("OK!") as the new marker, followed by click/treat for a correct response to that release cue (jumping up from the position). Next, I used the Premack Principle. The Premack Principle states that a highly likely behavior can be used to reinforce an unlikely behavior. If it's a stretch for my puppy to stay for three full seconds, that's an unlikely behavior. On the other hand, jumping up out of the down position is a very likely behavior.
I used the Premack Principle by giving the release cue when the puppy met a reasonably certain duration criterion. I "Premacked" the duration with the best thing in my puppy's universe in that moment—permission to get up from the down!
The take-away principles here are:
- The click ends the behavior or, better stated, the click starts the reinforcement cycle.
- Use treat delivery that puts the dog in a position to be able to offer the behavior again quickly.
- Monitor your tactics constantly to ensure that you are setting up so that the dog can offer what you want, as opposed to having what you want created artificially by circumstances.
- Use a trainer's best friend, the Premack Principle, as often as you can.
- Duration training should be fast, fun, snappy—just like any other clicker training session!
Let Rosanne and I demonstrate
My Rosanne is the true star of the videos that accompany this article. A two-year-old Petit Basset Griffon Vendéen (PBGV – a French scent hound), she makes it easy to illustrate training points. In this series of videos where I train Rosanne, I demonstrate how to choose and change criteria to meet the varying circumstances encountered in training. The videos also demonstrate how to chain a release cue to keep duration training fast, fun, and loopy!
Rosanne was very excited and fidgety with her feet when we started. As a result, I chose a very short amount of time for her front feet to remain still—just a count of a-b. In the beginning, there were times I had to wait to start the count until her feet were actually still. By the end of the first session, she was getting into position with her front feet still. That was great progress for the first session!
Next, I chose to move up a notch in duration and go for a-b-c. I let go of any concern about Rosanne's fidgeting feet. There were times that a little fidget crept in, but I did not deem that important enough to address. (Had I seen more of it, then I would have addressed it.)
When there were a few successful reps with a-b-c, I immediately dropped back to give Rosanne some easy tasks. In any kind of training, be sure that there are plenty of requests for easy behaviors mixed in when you are stretching to a harder behavior.
In the next video, notice how carefully I chose my criteria. Rosanne doesn't make any mistakes because I chose very small increases in the amount of time I asked her to remain still. I was in no hurry to get to longer amounts of time. Rosanne is an excitable girl, and this early learning is the most important part! In this session, we achieved the count a-b-c-d-e-f.
More alphabet achievement!
The plan for the following session was to focus on counts e, f, and g. I decided that if I got to g successfully, for the next sessions I'd start using the release cue as the marker for meeting the current criterion successfully. Goal achieved! In this video I do get to "g" three or four times.
In this session, I introduced the release cue ("OK!") as the marker for meeting criterion correctly. I discovered once again how important mechanical skills are. It requires a new set of skills to give the verbal release cue and immediately "click" (in this video, whistle) to mark the dog getting up from the down. This coordination is awkward for me; I even miss a whistle once! Adding the release cue takes the place of the click for meeting criterion.
When Rosanne responds correctly to the release cue by jumping up out of position, that response is clicked (or tweeted as the case may be). The sequence looks like this:
- Dog meets current criterion.
- Trainer marks that with "OK!"
- Dog jumps up in response to the "OK!" cue.
- Trainer clicks, and then tosses the treat.
This sequence creates a very powerful two-piece behavior chain! This behavior chain is so powerful that one day your dog will surprise you. The day will come (it always has for me) when you open the crate door and forget to give the release cue. Maybe you are thinking about what you will prepare for dinner, or perhaps you aren't feeling well and really want to hurry things up. You'll then find yourself asking, "What's up with her? She's not dashing out of the crate." And then it dawns on you…
"I didn't give the release cue! What a good girl she is!"
At one point or another my dogs have each shown me how strong their understanding is of the concept. "You can't trick me! I didn't hear the release cue."
A changed course
In this session, I planned to focus on the longer time periods (increase the count to g-h–i) and also be more aware of my mechanical skills. As it turned out, Rosanne started the session off with frustration barking. I recognized immediately that I had to let go of the plan and work on only short measures of duration. I also dropped out the release cue to make the process simpler, all to aid Rosanne's recovery from frustration.
Learning to make these judgment calls quickly, on the fly, is an important lesson of this article and the accompanying videos. Thank you, Rosanne, for providing such a great example of why and how trainers have to be tuned in and think on their feet. Toward the end of the session captured on this video, we had recovered and I was able to add back in the release cue.
I switched to using the new i-Click™ Jewel for this training session. I felt that the verbal release cue should be followed by a timelier marker. The change from saying a word to tweeting the whistle was just not efficient enough. When I switched to using the clicker, we made great use of the release cue and I felt my timing was much better!
Working with Rosanne brought up an interesting and important point. Be cautious about when you give the release cue (or when you click). If you release or click at the moment the dog starts to get antsy, you'll reinforce being antsy—not at all what you want!
Stay with it!
With Rosanne's help via video, I was able to demonstrate how trainers must learn to lower or raise criterion on the fly based on current circumstances. Planning, flexibility, training loops, behavior chains, and the Premack Principle make it possible to progress toward teaching the behavior "stay until released" or other behaviors that are on the training wish list. Good luck!
Post new comment