Distraction-proofing with Premack and P-
All training, whether clicker or conventional, tends to work a lot better in the living room than on the training field. The reason for this is that new and exciting out-of-door environments offer a lot of external reinforcers to dogs. Basically, they just get distracted, and proofing, or teaching the dog to continue to work with distractions around, is an essential part of every exercise. The presence of other dogs can be a particularly difficult distraction to overcome, but there is at least one surprisingly simple and fast approach to solving this problem. The article contains a couple of video clips so is best read via a fast connection.
Partridge, aka Puttle, my 8-month-old Dobermann pup, loves other dogs, and doesn’t get to see them often enough (he lives with several other Dobermanns but is very sociable and loves to meet and play with other dogs).
He’s the kind of pup who, when there are other dogs around, hangs out at the end of the lead, gasping with excitement and trying to get at them so he can play, and he ignores anything I say or do completely.
Or rather, that’s what he did at the last class we attended. The instructor’s solution was to make me walk him up and down against a wall, away from the other dogs, luring him every step of the way to keep him on the left, and body-blocking him when he tried to move towards where the fun was. It worked in that he moved along at heel, but he wasn’t enjoying it and neither was I, and I was clicking for sloppy heeling, which bugged me. Like a lot of clicker trainers, I regard heeling as perhaps one of the most difficult behaviours to train well (and guess where most competitors lose their points!). So there is a very strong case for leaving formal heelwork until late in the dog’s training, and merely teaching youngsters to walk on a loose lead.
So this afternoon we took a different approach to distraction-proofing. Here’s what we did:
The first concept we made use of was the Premack principle. David Premack theorised, correctly, that a more frequently occurring behaviour should reinforce a less frequently occurring behaviour. Now this sounds a bit abstruse, so let’s try to reduce it to more familiar terms.
If we give a dog a treat which he likes every time he sits, he will tend to sit more often (this is how clicker training works at the most basic level). So as time goes by, the dog sits more and more, and, as dogs are pretty basic creatures, we can assume that this is because they are starting to enjoy sitting because of its history of being rewarded.
Now suppose we ask the dog to walk at heel for two paces and then sit. The heeling is not currently a high-frequency behaviour (the dog isn’t doing a lot of heeling), but according to Premack, the sit, which happens often, will reinforce the heeling, which happens less often, and we will start getting more heeling.
This sounds very odd until, like Donald Griffin, we allow ourselves to use the heretical phrase "the dog thinks"! The dog does a bit of heeling, then he is asked to sit, and he thinks this is a good thing because he is used to getting rewarded for sitting, so he is more likely to heel some more. So we can use one behaviour to reinforce another behaviour, as long as the last one is consistently rewarded, and this is in fact how very complex exercises are trained.
Now in Puttle’s case, the behaviour he is trying to engage in the most is greeting and playing with other dogs, so I can use this to reinforce a behaviour I want, like responding to his name with attention.
And I can also punish non-attention by removing what he wants, i.e. access to another dog. This is negative punishment, the removal of something pleasurable to the dog, and is extremely powerful, especially with a dog who is primarily reward-trained. I already have a conditioned negative punisher (a sharply spoken, rather growly "aah-aah!") installed. This is a bit firmer than a non-reinforcement marker (NRM), which is a neutral "uh-uh". The NRM says: you won’t get clicked for that so try something else. The conditioned negative punisher says: I do not like that and something you do like is about to disappear as a consequence.
So I now have the building blocks of an exercise to teach Puttle to pay attention to me when there is another dog around. It goes like this:
We borrow another dog and handler, in this case Puttle’s auntie, Shady Lady, and her person Sandy. Sandy and Shady hang around us at the right distance, and Puttle, quite predictably, charges out to the end of the lead to talk to Shady.
I call Puttle gently, and depending on his response, one of two things happens. If he is attentive and turns towards me, I click. If he ignores me and continues to focus on Shady, I say "aah-aah!".
The instructions to Sandy, who is handling Shady, are explicit. If she hears a click, she stays hanging around where she is so that Puttle can choose his primary reinforcer: he can opt either for a treat or for more interaction with Shady. This is important: imposing our version of what the reinforcer should be on the dog can weaken the "charge" on the clicker, whereas allowing him to pick the reinforcer he wants (within practical limits – humping the pedigreed poodle down the road is not usually an option!) strengthens the reinforcing power of the clicker. And remember, by the time he gets clicked, he has already done what we want, so he is not being allowed to have his own way, simply to spend his "salary" as he chooses.
But if Sandy hears me say "aah-aah!", she immediately turns round and walks off, taking Shady with her, i.e. inflicting a negative punishment on Puttle. So Puttle loses not only the opportunity to earn a treat but also the opportunity to hobnob with another dog – a double whammy!
The videos are not great quality and there is a lot of wind noise, but you can just about hear the clicks. In the two clips, which are pretty much contiguous in time, I am on the left with Puttle and Sandy is on the right with Shady. The footage speaks for itself. After two negative punishers, Puttle got it right literally every time, and it took seconds to achieve. Even with Sandy calling him and nudging him, he continued to respond to his name with attentive behaviour and a sit. And he opted for the treat rather than for playing with Shady, which is a very good sign; it says he’s setlling down and enjoying work more than almost anything else. (He’s not quite old enough to have discovered poodles though………)
You may have noticed that on the first attempt to call Puttle in, there was what looked like a collar correction. In fact what you were seeing was an oppositional reflex called thigmotaxis, on the part of both the canine and the primate, i.e. me. When he pulled reflexively against the end of the lead, so did I, and this is one of the reasons loose lead walking is such a tricky thing for many people to train; when the dog pulls the owner, the owner pulls back, and generally doesn’t even realise that he or she is doing it!
And at this stage, that’s what I want from him. I don’t care how many commands he "knows". What I care about is being able to get his attention, improve his focus, and get him offering behaviour for me to shape, and to achieve that we’re doing a variety of apparently nonsensical exercises like Doggy Zen and Unrolling the Yoga Mat. But what we’re really doing is creating the groundwork for being able to teach him anything new, fast, accurately and reliably.
Leave a Reply
You must be logged in to post a comment.