The Limited Hold

Dear Clicker Friends,

The limited hold is scientific terminology—laboratory slang, really—for a good way to use the marker and reinforcer to speed up response to a cue. We're all used to sluggish responses. You call folks for supper, and in due course, they come; meanwhile the soufflé falls or the soup gets cold. You ask the class members to be quiet, and some sit down and shut up, but it's quite a while before the last few stop talking. You call your dog to come in the house and it comes, grudgingly, finding half a dozen new things to sniff at before actually reaching the back door.

Suppose you are dealing with just such a behavior. You give the cue, and you get the response, but after a delay. Now you want to fix that. First, practice it a few times to judge the average length of the delay. You can count seconds to yourself, or actually use a stopwatch. Now give the cue, watch the time go by, and reinforce the response if it occurs within the average time. If the cue response occurs outside the time limit, call an end to the trial (with dogs, easily done by moving to a new location). Then give the cue again and start your countdown again.

In Don't Shoot the Dog I told this story about a limited hold. At Sea Life Park in the 1960s, one of our most effective show highlights was a group of six little spinner dolphins that did various leaps and whirls in response to underwater sound cues. The most spectacular behavior was the aerial spin for which they are named. Initially, when that cue went on, spins occurred raggedly and sporadically across a fifteen-second period. Using a stopwatch, we started turning the cue on only for twelve seconds, and marking and rewarding only spins that occurred during that time. When most of the animals were spinning within that period we cranked down the limited hold, shortening the time to ten seconds, then five, and eventually to two and a half seconds. It couldn't go much shorter, because the animals had to dive, first, to get up speed to jump fast enough to do a good spin. In any case, every animal learned that in order to get a fish it had to hit the air and perform the spin within two and a half seconds of the time the cue went on.

As a result, the animals poised themselves attentively near the underwater loudspeaker. When the spin cue went on, the pool erupted in an explosion of whirling bodies in the air; it was quite spectacular. One day while sitting among the audience I was amused to overhear a professorial type firmly informing his companions that the only way we could be getting that kind of response was by electric shock.

The length of time your learner takes to respond to a cue is called, in the lab, "latency." A long gap is called high latency, a quick response shows low latency. You can sometimes shape a low latency response, without bothering to measure the time, by asking for a lot of responses in quick succession and reinforcing only the quicker ones. For example, in demonstrating a shift in latency, I sometimes seek out a dog with a slow, lethargic sit: on cue, but high latency. I call him, back up a step or two (to get him moving forward), click and treat when he comes, and repeat a couple of times. When the dog is coming with me willingly I back up and say "sit," stop, and click just as his back legs begin to fold; treat, call him, back up again quickly, and cue "sit" again, clicking the act of sitting, not waiting for the full sit. Then I progress to backing up, cueing the sit, and clicking as the rump hits the ground; and then to clicking as he sits but only if he sits instantly when he hears the word. If he hesitates, I back up again, call him again, and say "sit" again. In about twenty clicks (and thirty seconds or so) the limited hold is down to practically zero and the dog is sitting like a champion. One would then repeat this facing the other direction, and then perhaps in another room, on another day, and finally outdoors, to generalize the low-latency response. And of course having embarked on this repair job you would also drop from your own repertoire the habit of reinforcing slow sits.

With the "come" cue, often a dawdly behavior, it can be hard to gauge what's faster and what's slower just by watching. In this case a methodical application of a limited hold can be useful. Think of the limited hold as a single criterion, like height of a jump, duration of a sit, strength of a push. You can train it in one situation, and then extend the criterion to other times and places. So, in the example of a dog that sniffs and dawdles over every blade of grass en route to the back door, you might train a low-latency recall indoors, first, and then gradually add speed as a criterion of the recall, in other circumstances. A long hallway is a great place to do this. Mark a chalk line across the floor at each end of the hall. Stand behind the line at one end. Enlist a helper to call or lead the dog back to the other end of the hall between each run, or to hold the dog while you move to the other end. Make the run very short (five feet or so) the first few times, clicking the dog's arrival at the chalk line and giving a highly preferred treat.

Now extend the run to the length of the hall. How are you going to tell which of two similar runs was faster? Most of us recite the alphabet at a pretty steady rate. You can use that to measure small increments of improvement. "Come," you say, and as the dog shambles toward you, you recite a-b-c-d-e-f-g-h-i-j-k-l until he crosses your chalk line, whereupon you click and treat. On the next run you'll know if he speeds up a little, because he will reach you by h or j instead of k. Good! Click/treat. And of course if he takes longer, you'll know that too.

The object is not to punish slower runs, but to pay for faster ones. You make you criterion roomy enough so that most, but not all, of the runs are within your chosen limit. The procedure itself may naturally speed up the dog. As he begins to pick up speed you can introduce the limited hold: choose the maximum letter you'll tolerate, and if he doesn't get there by that letter, change ends and call him back the other way. I have seen the world's slowest Newfie, who plodded all the way to l-m-n-o-p on his first try, end up by responding to the "come" cue at a nice canter, getting to me by b-c-d, after about fifteen clicks and treats.

Oh, of course this works with people. When I'm lecturing with a new group of people, for example, at the very beginning I ask for quiet with a gesture, a raised hand. Generally everyone goes on talking. Then I get down from the podium and go around the room for about thirty seconds, clicking and handing a treat (Hershey's Kisses) to any people who are being quiet. The next time I ask for quiet, usually when starting up again after a break, I stand still and look at my watch. When most of the people have stopped talking, I click into the microphone and say Good! And start talking.

By the third time it happens the audience falls silent, except for a giggle here and there ("She's training us!") as soon as I step up to the podium and raise my hand. I've established a cue, the raised hand, and I've also shaped a good low latency response.

As Bob and Marian Bailey put it, latencies are contagious. If you are attentive to selectively reinforcing brisk responses to cues in a few crucial responses—paying only for low latency responses—all of that learner's cue responses will tend to be brisker. That makes for a sharp-looking worker! On the other hand, if you generally accept and pay for any eventual response to your cue or request, even if it took the learner (the dog, the child, the teenager, the spouse) forever to actually do it, then high latencies and long waits is what you'll get in general. One example I've experienced personally is the difference between getting on one of the guest riding horses at a dude ranch, and getting on a working cowpony, a cutting or roping horse. The horse in the riding stable starts up into a slow walk after you've kicked it a few times, and requires several kicks and some urging to break into a trot. Steering it may also require some effort, and it stops slowly too, going from a canter down to a jog and finally a slow walk again. The cutting horse moves on a dime, in contrast, moves and changes speed and direction instantly, on the smallest of cues. What a pleasure; the latencies are so low it feels as if all you have to do is think what you want the horse to do, and it's already happening.

And all you need to do, to have your own learners respond that way, is to value and reinforce quick responses to cues, withhold reinforcement for slow responses, and, when the difference is hard to measure, reach for that useful tool, the limited hold.

Happy Clicking!

Sunshine Books, Inc.
49 River St., Suite 3
Waltham, MA 02453
1-800-47-CLICK(2-5425)

© 2006, Karen Pryor Clickertraining (KPCT)^TM