How I Taught Bertie to Retrieve

Five years ago I began clicker training under the tutelage of Melinda Johnson and her Bird-Click list (Bird-Click@yahoogroups.com). At that time the recommended sequence of behaviours to train was targeting, three prop behaviours and then behaviours requiring a hand or verbal cue. I chose to teach Bertie to retrieve a whiffle ball to my hand as the second prop behaviour. What follows are three things you will need to do what I did, an understanding of the terminology, the steps I used, and the problems you might encounter.

The retrieve can be taught using a back chain or a forward chain.

Chain (def.): A chain is two or more behaviours joined together to form one behaviour.

For example, the one behaviour, the retrieve, as I visualized it, was comprised of four separate behaviours, walking to a ball, lifting and  holding a ball, carrying a ball to my hand and putting a ball in my hand. That also happened to be a forward chain.

Forward chain (def.): Training the first behaviour, walking to a ball, first, and working toward the end behaviour, putting the ball in my hand.

Back chain (def.): Training the end behaviour first and working backward.

I used shaping to teach each behaviour in the chain starting with whichever of Bertie's behaviours was closest to the one being trained.

Shaping (def.): Taking the behaviour offered that is closest to the chosen goal and reinforcing successive approximations or steps toward that  goal behaviour.

For example, in getting Bertie to walk toward the ball, I would first click and reward her for just looking toward the ball, then looking at the ball, then taking a step toward the ball and so on. When a step was fluent or reliable, I withheld the reinforcement to get an extinction burst in which I hoped Bertie would offer a behaviour a little closer to our goal. That new behaviour then became the criteria for reinforcement. Then I repeated the process.

Extinction burst (def.): When reinforcement is withheld for a behaviour that was previously reinforced, the behaviour is repeated, often with more intensity, frequency or with slight modifications.

So that first behaviour in my forward chain, walking to the whiffle ball, would look something like this on paper.

Bertie looks toward the ball. Click/treat.
Repeat until reliable.
Withhold the click/treat.
Bertie looks at the ball. Click treat.
Repeat until reliable. Withhold the click/treat.
Bertie steps toward the ball. Click/treat.
Repeat until reliable.
Withhold the click/treat.
Bertie takes two steps toward the ball, then three steps and so on until finally she just walks to the ball.

I found that the process moved much more quickly than the written word would suggest. The more clicker smart Bertie became, the larger were the successive approximations. Her first step in reality was walking to the ball.

Here is the actual plan, on paper, that I started with.

1. Walk to the ball:                a) Look toward the ball.
                                               b) Look at the ball.
                                               c) Step toward the ball.
                                               d) Walk to the ball.

2. Lift and hold the ball:        a) Touch the ball with beak.
                                               b) Grip the ball with beak.
                                               c) Lift the ball.
                                               d) Lift the ball and hold.
                                               e) Lift the ball and hold for successively longer periods.

3. Carry the ball to my hand: a) Lift ball, hold and take one step toward hand.
                                               b) Lift ball, hold and take one more step toward hand.
                                               c) Repeat (b) until hand is reached.

4. Place ball in my hand:       a) Bump  hand with ball.
                                               b) Place ball in hand.
                                               c) Gradually increase the distance between ball and hand.

As I mentioned before, Bertie worked more quickly than I anticipated, with one exception. I had to break step 2d into even smaller approximations. It was at this step that I  muddied the waters so to speak. A common problem,  at least for me, is having to decide, quickly, which of two behaviours offered almost simultaneously, to reinforce and continue with. For example, I am shaping Bertie to lift the ball to a specific height and the next step is to hold the ball at that height, say, for two seconds. Bertie lifts the ball slightly but holds it for two seconds. Do I stay with my plan and withhold the reinforcement because she didn't lift the ball high enough, or do I reinforce what I know will be the next criteria  and return to the lift after the hold is fluent? Although I am getting better at making that decision quickly, even a slight hesitation in clicking can reinforce an entirely different behaviour than any you might have in mind. In any event step 2d broke down to six smaller approximations to get Bertie and me back on track as follows:

1. Touch the ball with beak.
2. Touch the ball with beak and tongue.
3. Hook upper beak into the ball.
4. Grip the ball with upper and lower beak.
5. Grip and Hold or lift the ball.
6. Lift and hold the ball.

Today, I can put the whiffle ball anywhere on Bertie's cage top. She will go to the ball, pick it up, bring it back to me, often on the run, and put that ball into my hand.

In closing, has anyone suggested to you to include volunteer work on your resume because many skills learned in volunteer work are transferable? Well, once your parrot has learned to retrieve, the skills learned, are quickly and easily transferred to playing basketball, putting rings on a peg and putting coins in a piggy bank just like Bertie does.

Published in Good Bird Magazine June 2007
www.goodbirdinc.com