Some years ago, Marian and I addressed this on a list. We also discuss this in some detail in our intro workshop. Here goes a brief synopsis of "getting" decreased latency.
First, rejoice! This is one of the few good thing in training that generalizes between behaviors. That is, get short latency on one behavior and, IF YOU DO IT RIGHT, it generalizes to the animals other trained behavior, at least to a considerable degree.
SO - pick ONE CUED BEHAVIOR that is short and sweet, unambiguous, and that the dog knows very well - and HEELING is NOT a short behavior. I am not talking about CHAINS of behavior. I am talking about a simple touch this or pull that (with a very low level of effort). You want this behavior to be VERY GOOD, actually, better than just very good.
Next - work on that ONE BEHAVIOR for at least two or three days. That's it - NO OTHER TRAINED BEHAVIORS. DON'T CHEAT. NO OTHER TRAINING OF ANY KIND. If you don't like this, well, all I can say is that this is the penalty for not getting short latency to begin with. You must pay the piper sometime; you are paying now <G>.
Why? Because the dog must learn that whatever you want, you want it NOW, NOT LATER. If you tried to enforce this with lots of behaviors, some of which might not be very strong to begin with, you are just creating needless problems for yourself, and the dog begins to learn that ONE behavior must be fast, but the other behaviors can be at the same old leisurely pace. So, the more choice you give the dog, the harder you make it on yourself, and the dog in the long run.
Now, work on this ONE BEHAVIOR for several sessions to be sure it is good and that it is strong. AND ALL TRAINING SHOULD BE CRF (continuous) - no ratios. I won't go into why not ratios in this instance. Most of you already know about my views on casual or unnecessary use of ratios.
As you near the point where the topography or shape of the behavior is just what you want, begin to measure by counting the duration between cue and the beginning of the dog's response. You will likely determine a pattern or range of time delay between command and response.
OK, now that the behavior has good shape, but there is a constant delay between cue and response, we are reading. As I said, this might take several sessions, maybe even a day or more. Whatever it takes, that's what it takes (and don't forget TIMING, CRITERIA, RATE).
From your observations (some of them preferably in writing), you figure out what time delay the animal offers about 50 to 60% of the time. Thus 4 or 5 trials out of 10, the dog will NOT be reinforced. This duration might be 1 sec, or it could be 3 sec. What ever it is, it is. Just be sure that the animal IS reinforced for every trial within specs - NO RATIOS. This procedure does not work well using ratios.
REMEMBER - YOU ARE TRAINING ONE AND ONLY ONE BEHAVIOR AT THIS TIME - NO HEELING, NO RETRIEVES, "NO NOTHIN'", ONLY THE SELECTED BEHAVIOR, and that behavior should not be a chain.
Now, in a short time, if your timing is precise, the animals latency will gradually reduce to the point where you get 8 out of 10 trials within specs. Now, reduce the target latency by 10 to 20 percent - don't push too fast. Be slow and easy at this time. Remember that you are learning as well as the dog.
Sometime in this regimen, the dog may BALK. Why? Because you likely are doing something totally different than you have in the past. Work on one behavior. Work on shorter latency. Stuck by your guns (the dog is used to manipulating your behavior by doing or not doing things). This does not take some special, higher order mentality to manipulate a trainer. All the animal has to do is to learn that its behavior causes environmental change, and that animal will do that over and over again. This manipulation has happened to me and every other trainer who has clicked a clicker or jerked a leash. This is another reason for selecting ONE AND ONLY ONE BEHAVIOR. No manipulation. Make the sessions VERY short. There should be only 10 or so trials per session, with several sessions per period and as many periods per day as you think you and the dog can take. Again, don't be in too much of a hurry at first. You have changed the game radically, and this change will take a while to sink in to the dog's thinking.
Every time the dog reaches the 80 percent level, reduce the time duration (called the contingency) required for reinforcement to reduce the dog to about 60 percent of correct responses. Now, it will be in here that you will begin to see the dog "get" the idea that faster is better. You will be tempted to really LUMP, and demand much faster response. While you might speed things up a bit - YOU CAN BLOW IT ALL BY ASKING TOO MUCH, TOO OFTEN, TOO SOON. Consistency will get you where you want to go. It should take just a few to several days, depending on the extent of the problem.
About now, you need to decide just how short a latency you really want for this response. Be realistic, but, if you want more, expect more. Keep at it until the behavior has the latency you want. Just keep reducing the allowed time, but don't raise the criteria so fast (LUMPING) that extinction begins to set in strongly. I know, I know, Gary Wilkes uses extinction to get stronger or faster (quicker response time) behavior. And, yes, I have done the same. Using EXTINCTION BURST behavior is very tricky, and not for the feint of heart. I suggest the afore-described systematic approach as a conservative and reliable method more suited for us "average" trainers.
By now, you should be seeing just how fast your dog can really move. Be sure that you are reasonably pleased with the speed, because we are now about ready to extend this short latency to ALL other behaviors.
Pick another behavior (again, a single behavior and not a long chain) that is VERY strong. DO NOT PICK A NEW OR RECENTLY TRAINED (probably weak) BEHAVIOR. Now, suddenly, in the midst of working whatever you have been using as the target behavior, introduce the new command. Do NOT be surprised if the animal gives you that behavior AT THE SAME, OR NEARLY THE SAME, AS THE TARGET BEHAVIOR. Be prepared to reinforce properly.
If the animal is still slower than you like, repeat the above procedure. I am sure you will find that the time required to get short latency will be much less than the first behavior. After you get that behavior to where you want it, introduce another behavior. I have seldom had animals fail to generalize this to all behaviors in a very short time. Just don't give the animal a choice between those behaviors that may be started slowly and those that must be done right away.
This is just a brief description, but I think every will get the general idea.
OF COURSE, WE SUGGEST THAT THE BEST WAY TO TRAIN IS TO CONSIDER LATENCY AS JUST ANOTHER PART OF THE TRAINED BEHAVIOR AND ALL LATENCIES ARE SHORT BEFORE THE BEHAVIOR IS CONSIDERED PERFECTED.
Copyright of all posts is the property of the original author. Please obtain permission from the original author before copying, quoting, or forwarding.
List and Site Owner: Melissa Alexander, mca @ clickersolutions.com