A Logical Order of Training - Part 1
Now that we have the fundamental tools of clicker training, it is time to start applying them in sequence. Creating complete behaviors follows a flexible set of logical rules. For instance, the first thing you must do is shape the desired behavior. Since we don't really have a behavior yet, chanting a command word is most likely to be confusing. The logical sequence for adding a cue is after the behavior is shaped. But, you might ask, "How much longer?" Do we attach the cue before we vary the reinforcement, or after? The answer is, sometimes yes, sometimes no and sometimes, both.
The full answer to this question is that the sequence you use depends on the context of the behavior you are teaching. For instance, will the dog be expected to perform the behavior with complete attention on the handler? Will the dog be expected to focus on an obstacle or another task, with only periodic instruction from the handler? These two separate types of performance need two, slightly different approaches to the way they are assembled. Behaviors such as heeling and utility signals require the dog to pay close and almost constant attention to the handler. Agility obstacles, scent articles and directed jumps may require the animal to pay extremely close attention to the signal or command, and then perform a behavior independently. From the perspective of clicker training, these two fundamentally different types of performance are best achieved by applying slightly different sequences of the same basic processes. To really understand how this works, we will take these two topics in two separate columns. This month, we will look at training for formal, precise performance.
Precision with fixed attention:
If you wish to create "drill team" precision, you need to create consistent behavior patterns and stick with them. If you think about it, most single component obedience behaviors, such as a long-down or heeling, are actually tests of your ability to add duration and precision. These behaviors are not particularly complex and don't require much independent thinking from the dog. Here's is the best sequence I have found for teaching these types of behaviors.
1) Shape the behavior. This is pretty much self explanatory, but must be mentioned because shaping a behavior is a separate task from naming a behavior. i.e. During this initial phase of learning, you aren't chanting commands. The cues you are using at this stage are primarily "click", "wrong" and various forms of targets.
2) Attach a working cue. Many training programs use the same cue to trigger early, incorrect versions of the behavior that they wish to use later to trigger performance level behaviors. There is a major downside to this practice. If you use the same cue in shaping that you use in performance, every time you raise your standards, you are redefining the command. For instance, in the early stages, "Down" means "lie down for 5 seconds." Each time you increase the time unit, the old version of "Down" is supposed to be erased, while the new definition is supposed to be automatically adopted. Now, "Down" means "lie down for 10 seconds" and so on, until the behavior lasts for 5 minutes. With our human perception, we see "down" and "long-down" as minor modifications of the same behavior, but that point is not automatically obvious to most dogs. (That's why it takes so long to teach long sits and downs.) So, each time you add complexity or duration to the behavior, the dog may think you are creating a brand new behavior. The animal must then break the connection with the old set of rules and reconnect the command to the "new" set of rules. Over the many required levels of improvement of even simple behaviors, this process severely weakens the bond between the command and the correct version of the behavior. You can avoid this weakness by using a "working" cue for basic training and later attach a cue that is specifically connected with performance. This may seem like an unnecessary step, but precise performance requires a perfect connection between the cue and the behavior. If you leave weak points in the foundation, you can expect inaccuracy and confusion, later.
To create a working cue, use this pattern. First, reinforce the behavior until it is happening predictably, on a cycle of ten seconds or less. Do enough repetitions so that you can easily guess when the behavior is going to begin and end. Now, start saying the word or presenting the hand signal just before the behavior occurs. If the behavior happens on schedule, click and treat. If not, say "wrong" and go back to reinforcing the behavior silently for awhile.
3) Vary the reinforcement. Even though you want extreme precision, you are still going to have to go through the process of varying the reinforcement. While this causes the dog's behavior to "wiggle" it also allows you to create enthusiastic behavior that is resistant to extinction. Varying the reinforcement after creating a good looking version of the behavior takes nerves of steel. We tend to think of learning as a linear process and hate to do anything that looks like backsliding. The truth is that relaxing one set of standards while building another set is an integral part of clicker training. Once you get used to the fact that the behavior hasn't really disappeared and that you can get it back quickly, this will become far more comfortable with the process. The effort is well worth it.
4) Attach the performance cue. Once the behavior is 99% perfect, you'll want to attach the cue that you will use in performance. This process is no more difficult than attaching your working command in step #2. For instance, let's assume that you used the word "Drop" as a working cue. You have gradually taught the dog that "Drop" means "Whenever you hear the word "Drop", no matter where you are, lie down quickly, and wait until I release you." As we described in an earlier column, adding a new cue follows this simple pattern. First, give the new cue ("Down") - the one you want to use in performance. Next, give the cue that is currently triggering the behavior. ("Drop") After a few repetitions, the dog's natural tendency to anticipate will start causing the behavior before you have a chance to give the old cue. This will transfer control of the behavior to the word "Down." If you use this pattern correctly, the dog's natural ability to anticipate will kick in .
Now the importance of teaching a working cue and a performance cue may be easier to see. The word Drop was used during the phases when the dog was still making errors and wasn't really sure of all the criteria associated with the command. The new command, "Down", has never been associated with anything less than perfection. If the dog hears "Drop" there is the possibility that he might mistakenly give you a previous version of the behavior. When the dog hears "Down", however, the word itself conjures up the necessary standards of performance.After attaching a performance cue, it is inevitable that you will have to do some tweaking later. In order to polish a slightly fuzzy behavior, temporarily return to the use of your working cue. Once the behavior is again up to your standards, reattach the performance cue.
5) Integrate the behavior into the dog's repertoire / add consequences for failure. After spending a considerable amount of time reinforcing a new behavior, your dog will develop a bias toward that behavior. If you have been working on "Down", you may temporarily lose "Sit." To put the new behavior on par with all the other behaviors you have taught, merely start asking for behavior in unpredictable patterns. Use clicks and treats for correct decisions, and "wrong" + short timeouts for mistakes.
By definition, operant conditioning is "behavior that is determined by its consequences." To create a performance repertoire that is precise, crisp and unfailing, there must be consequences that maintain that level of performance. That means pleasant consequences for success, and unpleasant consequences for failure. While it is often suggested that "all positive" training can create such performance, I am not aware that anyone has ever actually done it with dogs in obedience competition. For performance animals, I include another step in my order of training - aversive control for failure. That topic requires more attention than I can give in this column, so we will discuss it in more depth, two months from now. Next month - an order of training for dynamic and independent performance.
Go to the next article | Go to Prior Article