The Model-Rival Method
I mentioned “The Model-Rival Method” earlier when talking about training dogs to associate words with objects, and I thought it’d be fun to illustrate what I was talking about. The video at the bottom of the post is an example of this method, famously used by Dr. Irene Pepperberg to train Alex the African Grey Parrot to label a large number of objects, materials, colors, etc. It was originated by the European scientist Todt, in contrast to the “Skinner Box” kind of training in which a parrot got a food treat from a mechanized box for vocalizing something similar to the sounds being played by loud speaker. Using that method, American behaviorists had concluded that parrots “can’t be taught language,” but Todt noted that it had little relationship to how our own children learn language. He criticized such a model as being a less than conclusive test of the cognitive abilities of parrots and suggested what he called the “Model-Rival” method. Dr. Pepperberg was intrigued by his argument, and set up training sessions using this method.
The training included 2 people and Alex, with one person being the ‘trainer’ and one being the ‘model’ as well as the ‘rival.’ I’ll describe a sample session as including Irene, Alex the parrot and a woman named Julie (I made that name up.) The term model is used to describe the technique because one person models what they want to parrot to do. In Alex’s case, the goal is to get him to say the word. For example, Irene would hold up an object and say “Julie, what’s this?” and Julie would answer “Crayon.” Irene then handed the object to Julie, who got to handle and play with it (thus, the human trainee is also a rival of Alex’s, because they got to play with the object but Alex did not.)
Last year I tried using a standard operant conditioning paradigm with Willie to teach him to label objects, putting down 2 toys and clicking and treating if he went over and nosed the ‘correct’ one. Willie not only was NOT able to make the distinction between the sounds I was making and the 2 objects, he became so stressed (presumably) about not knowing what he was supposed to do, he shut down. I quit after a few weeks because we were getting nowhere and Willie looked too miserable to continue. He seemed to want very much to get the game, but he couldn’t figure it out.
After reading about Chaser and his 1022 labels for objects (see my post on January 11th, 2011) I decided to try to teach Willie the names of objects in a more natural way. For about a week Jim and I used the word Scorch for his scorpion toy (a favorite, and still, amazingly, in perfect shape after how many years?). We’d toss it around, hold it up, saying Scorch as often as seemed possible. But after awhile I started thinking about the Model-Rival system, and have starting giving it a try. I’d love to tell you, after about 10 days of training, that Willie has it down pat, but he doesn’t, at least not once we added in a second name for a second toy. After about 5 days he would reliably go get the scorpion when we said Scorch 100% of the time, even if there were several other toys around , but once we added in a second toy his response fell apart. That’s not surprising, and that’s what we’re working on now, understanding that Scorch means scorpion, but Pony means the stuffed pony. You can see where we are in the video below (in which Willie boy is extremely distracted by a parked car outside and the camera especially).
By the way, out of curiosity I checked online and there is a published paper that compares the model-rival method and operant conditioning method for training domestic dogs that came out in 2003 in Applied Animal Behavior Science (by McKinley and Young). There are some pretty significant issues with the methods: they ask the dog to label any of 3 red objects as “socks” and any of the 3 yellow objects as “cross.” I’d then argue that the test is not about labeling an ‘object’ but a color (and red is not a color dogs see well). That’s a very different exercise than matching a word to an object, but granted their intent was to compare the efficacy of training methods, not so much what was being trained. Their conclusion was that the model-rival method was as effective as operant conditioning, and their conclusion is that this is a method that might be used more in dog training. The sample was small and I think there are some confounding factors, but still, very interesting stuff.
Note: You’ll notice that once I introduce Pony I say this is a “test” but then switch to saying this is “training.” I’m trying to use the same kind of method we’d use with a child, so once Willie doesn’t do what is expected (and doesn’t ‘pass the test’), I help him to do the right thing. And I’ll talk later about another issue, the potential that I, or anyone, could be inadvertently cuing their dog to the correct object…., but for now, my objective was to illustrate Willie’s progress and demonstrate the Model-Rival Method.