Phoebe’s retrieve journey and the backchaining of complex behaviors

Some dogs are “instinctually” good at certain behaviors, and other dogs are not. For example, water dogs tend to be untiring and talented swimmers, herding breeds tend to have the proverbial herding instinct, and retrievers are, well, usually “natural” retrievers. Phoebe has many talents, but she is not a natural retriever.

I have been working on teaching Phoebe to retrieve to hand for a while now, and I was thrilled when, after almost six weeks of working on this behavior, I got her to put a piece of garden hose in my hand when I was sitting on the balcony steps in my living room. However, that Phoebe was able to put this specific object into my hand in this specific location while I was sitting did not mean that she had learned to put any object into my hand in any location, no matter what body position I assumed. Her learning experience only applied to this one behavior. She had acquired the behavior, but not generalized it yet.

Pamela Reid distinguishes four stages of learning: acquisition, fluency, generalization, and maintenance. At the point when Phoebe was able to put the hose into my hand after four weeks of training, she had mastered the first stage: acquisition. She wasn’t fluent in it yet – i.e. she still had to deliberately think about what she was doing -, and she hadn’t generalized it to all objects, all locations, and all body positions yet. For a dog who isn’t a natural retriever, retrieving to hand is a fairly complex behavior chain that can take quite some time to perfect. Even if I we split lit into very broad junks, the retrieve chain still consists of at least 4 links: walk towards object, pick up object, carry object towards handler, deliver object to handler’s hand.

However, the more times we explain a certain behavior in a new location, or in a new body position, or involving a new object, the faster our explanations will go, and at some point, the animal will generalize the behavior to all objects, all body positions, and all locations. However, depending on the dog and the task, this may take either only a few repetitions and little time or lots of repetitions and lots of time.

Since my goal is to have Phoebe retrieve anything in any location and no matter what body position I assume, I keep working on her retrieve to hand. I use shaping and backchaining in order to teach a retrieve, a time-tested approach to this behavior used by positive reinforcement trainers all over the world. Shaping is the reinforcement (in our case, click and treat) of successive approximations to the target behavior. We start small and gradually increase criteria, always surfing the extinction burst: we need to raise criteria slowly enough to set the animal up for success, but also fast enough to keep her from getting bored. Shaping is my favorite game, because it requires creativity, strategy, and patience on the trainer’s part, and thinking and creativity on the animal’s part, and it is a training approach that feels most like having a conversation with the animal: the animal asks a question, and we answer – either by means of a click (Yes!) or by not reacting (Try something else!). Sue Ailsby, one of my favorite trainers, says that shaping makes you recognize the unicorn in your dog: no two dogs are exactly alike; every dog you shape will have a different conversation with you … and this is the beauty of it. Another more pragmatic reason I love shaping is that 5 minutes of shaping tire Phoebe out as much as an hour-long walk.

If you have never shaped an animal, think of the children’s game of “hot or cold”: one person hides an object, and the other person moves through the room looking for it. The person who hid the object informs the seeker with “cold”, “warmer”, “warm”, “hot” etc. that he gets closer to or further away from the object in question. In shaping, the dog’s task is to figure out what we want her to do. An experienced shaper will offer all kinds of behaviors and make it easy for us to find something clickworthy. If our target behavior is something the animal is not likely to do by itself, we start with successive approximations – that is to say, we click for anything remotely resembling the target behavior, and then gradually narrow down our criteria. For example, in Phoebe’s retrieve, I started with the last behavior in the chain – the shared hold of an object – and shaped this behavior first. I presented a novel object in my outstretched hand. As Phoebe moved closer to sniff it, I’d click and reinforce her. Then, I’d wait for her to offer a nose-touch of the object. Next, I waited for ever-so-slightly touching the object with her teeth. Next, for putting her mouth around the object. Then, I built duration on the shared hold – in 0.5 second increments, I increased the time she had to keep her mouth locked around the object I was holding with her, playing 300-peck-pigeons (or, as known in Sue Ailsby circles, chutes and ladders). This way, I shaped a shared hold.

Next, I moved on to the last-but-one link in the retrieve chain. But before I go into details about this, let me explain to you why we are backchaining to begin with. Let’s start at the beginning. A behavior chain – such as the retrieve – is a number of behaviors that are performed in a certain sequence. Each behavior cues the respective next behavior, and is reinforced by it. Only in the very end, upon completing the chain, does the animal receive a primary reinforcer. In dog training, the primary reinforcer in the end of the chain is usually a treat.

I said that the retrieve is not one single behavior, but rather a behavior chain consisting of at least 4 links: walk towards object, pick up object, carry object to handler, deliver object to handler’s hand. I have explained the retrieve behavior to a number of dogs. Some of them needed only those 4 links to understand what I meant, others didn’t need an explanation at all, and yet others – among them, Phoebe has been the most challenging – need many, many more links. You always start the same way – at the last link in the chain – and then feel my way towards the beginning. Depending on the dog’s reactions I’ll arrive there within only a few sessions, or within lots of sessions.

When teaching a behavior chain, the commonsense approach is to start with the first link in the behavior (e.g. throwing the dumbbell) and work towards the last (e.g. shared hold of the dumbbell). However it turns out that the commonsense approach is not the smartest one. Behaviors are performed more reliably and are more stress-resistant if they are taught beginning with the last link in the chain. Let’s see: when we train with positive reinforcement, a behavior chain ends with a primary reinforcer. This is the goal; it is what the animal is working towards. The more often a behavior gets reinforced, the stronger it becomes. The stronger the reinforcement history of a behavior, the more likely the animal is to perform this very behavior. In fact, a behavior that has been taught by means of positive reinforcement will itself turn into a reinforcer. You have, so to speak, charged it with lots of positive reinforcement, and now it can in turn reinforce other behaviors. (This, of course, only applies if you train with positive reinforcement! A behavior taught by means of positive punishment will not acquire reinforcing qualities.) If we start with the last link in a behavior chain, this will eventually be the part of the chain the animal knows best – it will be the part that has been reinforced most often. Think of the dumbbell retrieve again: 1 walk towards object, 2 pick up object, 3 carry object to handler, 4 put object in handler’s hand.

If we start with the last link, our reinforcement history looks like this:

4 – primary reinforcer (PR)

3 – 4 – PR

2 – 3 – 4- PR

1 – 2 – 3 – 4 – PR

In this simplified backckchaining example, the fourth behavior (put object in handler’s hand) has been reinforced 4 times, while walking towards the object has only been reinforced once. The last link in the behavior (put object in handler’s hand) is the strongest link in the chain, because it has the strongest reinforcement history. It reinforces the link that comes before it. When it comes to behavior chains, we perform best when we are working towards something we know well – in this case, towards a shared hold. If we start with what we know well, but work towards something we are less sure about, we perform not es good – especially under stress. Susan M. Schneider uses an example most of us will have experienced ourselves in primary school: learning poems by heart, the nightmare of many schoolchildren. Even though the laws of backchaining have been well-known among behaviorists for a long time, they still have not made it into our schools – at least, they hadn’t made it to the classroom when I was in primary school: parents and teachers usually applied the commonsensical approach, telling children to start learning a poem from beginning to end. In the case of the retrieve, the reinforcement history of forward chaining would look like this:

1 – PR

1 – 2 PR

1 – 2 – 3 – PR

1 – 2 – 3 – 4 – PR

In this example, the first link in the chain (walk towards object) has been reinforced 4 times and is the strongest link in the chain with the most reinforcing qualities of all the links. However, since there is no behavior to precede it, its reinforcement power is wasted. The last link in the chain (deliver object to hand) has only been reinforced once, and has the least reinforcing qualities, because it is least well known.

In the case of the schoolchild learning a poem, the common approach is to start with the first line, rinse and repeat until you know it by heart, then first and second line, rinse and repeat until you know it by heart, then first, second and third line and so on. Let’s assume you want to learn Robert Frost’s The Road not Taken by heart and present it in front of your school class. You are nervous about speaking in public, and you don’t like to stand in front of the class with everyone staring at you. You could either start learning in the commonsense way – with the first line:

Two roads diverged in a yellow wood.

Two roads diverged in a yellow wood.

And sorry I could not travel both

Two roads diverged in a yellow wood.

And sorry I could not travel both

And be one traveler, long I stood

Two roads diverged in a yellow wood.

And sorry I could not travel both

And be one traveler, long I stood

And looked down one as far as I could

Two roads diverged in a yellow wood.

And sorry I could not travel both

And be one traveler, long I stood

And looked down one as far as I could

To where it bent in the undergrowth […].

By the time you get to the end of the first stanza, you have repeated the first line 5 times, the second line 4 times, the third line 3 times, the fourth line 2 times and the 5th line once. Which line do you know best? The first one, of course. When, during your classroom performance, will you have the most energy? In the beginning. So what should you start with – the part you know best, or the part you know least? The part you know least. You are most likely to make it to the end of the poem without stumbling over Frost’s iambic tetrameters if you work towards what you know best, not what you know least. As you spend your energy, you get to well-known terrain.

Ideally, then, you wouldn’t start learning at the beginning, but with the very last line of the last stanza:

And that has made all the difference.

I took the one less traveled by,

And that has made all the difference.

Two roads diverged in a wood, and I —

I took the one less traveled by,

And that has made all the difference.

Somewhere ages and ages hence:

Two roads diverged in a wood, and I —

I took the one less traveled by,

And that has made all the difference.

I shall be telling this with a sigh

Somewhere ages and ages hence:

Two roads diverged in a wood, and I —

I took the one less traveled by,

And that has made all the difference.

Try it yourself: pick two poems of equal length. Learn one from beginning to end, and one from end to beginning. Which way do you need less repetitions until you are able to fluently recite it?

This is Sam, a Weimaraner with no previous retrieve training. He learned to retrieve a beer can to hand in less than 10 2-minute sessions.

Phoebe, on the other hand, learned to retrieve her first object to hand in the course of six weeks, and in order for her to be successful, her individual sessions, spread out over the course of the day, needed to be 6 treats short rather than 2 minutes long. She needed to take a day off retrieve training every once in a while, and I needed to mix in other behaviors with the retrieve session in order to keep setting her up for success. In terms of shaping complex behaviors, Phoebe has been one of the most challenging dogs I have worked with. This also makes her one of the best teachers I’ve ever had: she has me to be a micro-splitter. Time and again, she lets me know that the slices of criteria I’ve come up with in my training plans are too big for her. Or that the training sessions are too long for her. Or that my mood is not calm and happy enough for her to be able to focus rather than worry. She has taught me to write training plans rather than wing it, and the importance of filming myself so I can then analyze the video and recognize the split second when things started going wrong, or what initiated her lightbulb moments. Phoebe also taught me how to work with dogs who aer extremly sensitive to my own body language, and how to adapt my own body language to help her become just a tiny little bit pushier rather than always being polite and keeping her distance. Anyways, back to the retrieve. After six weeks, Phoebe could do this and made me a very proud Poodle mama:

Here’s the 17 individual behaviors I had to split the hose retrieve chain into in the acquisition stage. Lumpier shaping approaches did not work for Phoebe:

  1. Sniff hose.
  2. Mouth hose.
  3. Mouth hose slightly longer.
  4. Introduce cue “Take it!”
  5. Get duration on the shared hold. – Fail. Even increasing duration in split seconds and playing the Chutes & Ladders game did not work. Get creative:

5.1 Teach chin target to open hand:

5.2 Get duration on the chin target.

5.3. Introduce cue “Chin!”

5.4. Combine Take it and Chin.

5.5 Get duration on the shared hold that resulted from this combination. – Success!

  1. Introduce cue “Halt fest!” (“Hold on to it!”)
  2. Lower the hose closer to the ground in 2-cm-increments and have Phoebe lift it together with me.
  3. Lower the hose closer to the ground in 2-cm-increments, let her lift it on her own and click after I grabbed it again.
  4. Lower the hose closer to the ground in 2-cm-increments, let her lift it on her own and then chin-target my hand while holding on to it.
  5. Gradually build duration on the shared hold after the lift.
  6. Eventually put the hose to the floor and have her lift it – fail: Phoebe would give up because getting her lower jar around it was too hard when the hose was on the ground. Be creative:

11.1: Put cardboard circles on both ends of the hose so it gets dumbbell-shaped and easier to lift off the floor. (Easier to put mouth around.) – Success!

  1. Tape 9 strips of duct tape on the floor, play chutes & ladders with it: put down on first strip, let her lift it and do a chin target. If successful, put down on second strip, have her lift it and do a chin target. If successful, put down on third strip, have her lift it and do a chin target. If not successful, return to strip one and start from scratch. (We returned to strip one lots of times.)

We run into problems here, since Phoebe did not know that it was possible to walk while holding the object. She’d lift it off the floor alright, but then stand there and look at me without bringing it. – Be creative, do some blending!

12.1 Take turns having her lift the object and put it into my hand from right in front of me, throwing the treat away from me and having her run towards me to do a chin target. Put object on strip no. 2, have her lift and put it in my hand. Throw treat away from me and have her run towards me in order to do a chin target. That way, I eventually got the first steps without dropping the object, followed by a chin target while holding the object. Success!

  1. Reduce distance. Place the object on the floor in different angles from me so she had to turn in order to bring it back to me.
  2. Start rolling the object a short distance with the help of the cardboard circles.
  3. Stop rolling; reduce the size of the cardboard circles so picking the object up got gradually harder, until she could lift it up without the circles that would help her get her lower jar under it.
  4. Introduce rolling again, this time without the circles.
  5. Introduce the first little object throws.
  6. Gradually build distance while throwing, have her run after it and bring it back to me.

Since achieving our first decent retrieve to hand with the help, inspiration and encouragement of the wonderful Donna Hill, I have worked on fluency and generalization, the next two stages of learning according to Pam Reid. The nice thing is that once she had the hose retrieve down in one position, I started my explanations from scratch in new locations and new body positions, but she got there much faster. By now, Phoebe can pick up and hand me the hose in the corridor and carry it up stairs (!) to hand to me, sitting on the top stair. She can also pick up the hose and hand it to me on two different outside locations while I’m sitting – both on pavement. And she can pick up the hose and hand it to me while I am standing on grass. However, we haven’t built distance in these new locations yet.

We have also started working on the next object – a rolled-up magazine. I chose this object next because I needed a novel object for a train-off with Tom. He wondered whether he could come up with a faster and more generalizable approach to teach the retrieve of a novel object. So we decided to test it. We would each use our own approaches to teach the retrieve of at least one novel object. Our rules excluded physical manipulation (such as holding the dog’s mouth shut or shoving an object into her mouth), harsh words and other types of positive punishment. Everything else was allowed, and how long, how often and with the help of what objects we trained was up to us. The person who first got Phoebe to retrieve a novel object to hand 4 out of 5 times from 1.5 meters distance would win.

This is not perfect yet (I still need to work on grabbing the magazine at different angles without dropping it), but I think it qualifies – it was all about getting there first, after all.

And some pretty awesome background reading:

Ailsby, Sue. Training Levels

Chance, Paul: Learning & Behavior

Hill, Donna: The Elusive Hand-Delivered Retrieve. (FDSA class & lecture notes)

Reid, Pamela J.: Excel-Erated Learning

Schneider, Susan M.: The Science of Consequences