How Dogs Learn
Permission to reprint - Ed Bailey
First printed GunDog 1996
At one time I thought there was something mystical involved in training a dog, some ethereal, secret technique known only to select trainers. Now I wonder if saying we “train” a dog is a bit like bragging, like taking all the credit for something we really had very little to do with. In reality, all we do is set up a situation so the dog learns [hopefully] what we want it to. Then we repeat it often enough so the learned task becomes second nature to the dog. We change the location, the background and general situation so the dog will perform as we want no matter what. If that is what training is all about, and I think it is, we can forget the mystique and the secret techniques, and define training as just giving the dog a controlled opportunity to learn. However, there are several types of learning and the type required will differ from one task or problem to another. Or, in many cases, a dog will need to draw on more than one learning type in order to solve a single problem. The only mystery, the only secret is to know how learning occurs and then turn it to our advantage so we produce a trained dog.
There are six basic types of learning. Some behaviorists might suggest more, some fewer, depending on how the variations are classified. For our purposes we can consider six types as the essential ones:
1) Primary socialization or socialization for short
2) Classical conditioning, also called type 1 learning, associative learning, Pavlovian conditioning, or just conditioning for short
3) Operant learning, also called type 2 learning, Skinnerian conditioning, or trial and error learning
4) Latent learning,
5) Insight learning, also called one trial learning
Of these types, only primary socialization is limited to a specific time or age in the dog’s life. All other forms of learning can, and do, occur throughout the life of the dog. Primary socialization occurs only during a relatively brief period early in the dog’s life. During this period the puppy learns how to be a dog, learns to read and understand dog communication signals. Dogs also learn how to relate to people during the socialization period. A puppy reared in a vacuum with no contact with its mother, siblings and with people during the critical age between three and 12 to 14 weeks cannot relate to either dogs or people. His fear of everything precludes any learning. It cannot be trained. The puppy reared with proper exposure to dogs and to people develops behaviour patterns, particularly those dealing with social behaviours, which will not appear until a later age, mostly after sexual maturity. Primary socialization is the foundation on which future learning is based.
Classical conditioning or associative learning is the pairing of an arbitrary stimulus, the conditioned stimulus or CS, with an unconditioned or primary stimulus or US which has reinforcing properties, and which is already associated with a specific response. The CS + US if closely paired in time, and repeated a few times causes the arbitrary CS to also become associated with the response, which had formerly been associated only with the primary stimulus (US), because of the reinforcing quality of the primary stimulus. After this secondary association is formed, the presentation of the CS alone will bring about the response. In a training situation we can use associative learning to teach a new command for an already learned task such as a whistle command for a response performed previously only to a vocal command. The essential point of associative learning is the conditioned stimulus always precedes the primary stimulus and both precede the response.
In Operant learning the arbitrary stimulus precedes the response which in turn is followed by the reinforcing stimulus. The arbitrary stimulus can be almost anything. For a dog in a training situation, it can be a command which is new to the dog and which on first presentation is meaningless to the dog. On hearing the command, the dog does something in response. If the response is correct, the dog is immediately reinforced or rewarded. If incorrect, the dog gets no reward. If not rewarded, dog will tend to respond differently on the next presentation of the arbitrary stimulus. He will not repeat his mistake if not rewarded. He will vary his response until the correct response is found and he gets the reinforcing reward. Any rewarded response will have an increased probability of occurring on the next presentation of the arbitrary stimulus because of the influence of the reward (reinforcement) acting retroactively to influence the response. Most training utilizes this type of learning.
Latent Learning is difficult to detect or to measure. It occurs when an animal is exposed to a situation which provides no immediate, obvious reward, but which does make a deposit in the memory bank. Many of the play behaviours such as the stalk and pounce game of a five-week old puppy trying to catch a butterfly provides a format for latent learning. Later, given the opportunity to pounce on something, the pounce is more accurate, more directed because the re-enactment of the butterfly pounce provides the learning to allow correction of mistakes made as a puppy. A dog can make a withdrawal from his memory bank and apply it to an investment on a pounce.
Similarly, any play-fight, play-hunt, play-sex, or any of the thousands of experiences throughout life can be drawn on to facilitate the solving of a new problem. There need be no immediate or obvious reinforcement for a memory trace to be formed. Every dog with a normal background of experiences has a huge memory bank account, with interest, to draw from.
In so-called Insight learning, a dog appears to size up a situation never before encountered and responds to it correctly with no trial and error attempts, apparently solving the problem intuitively. However, latent learning could be involved in what appears to be intuitive learning. Because we can never know all the experiences the dog has had or how these behaviours might chain together, we can never separate insight from latent learning.
The final learning type, Imitation learning, is as the name suggests, one dog sees another doing some relatively complex behavioural sequence. Then, by imitating what the other dog is doing, the imitator gets some favourable result. A goal is attained or at lest there is positive reinforcement of some kind. ON the next occurrence of the same situation or problem, the successful behaviour patterns used when imitating another dog are repeated. Each successful repet on the same or similar situations is reinforced and the behaviour becomes a fixed feature in the memory system. Probably most tracking hounds learn by imitating the dog that already knows what it’s all about. Obviously trail and error [operant learning’ plays a role to fine tune the tracking behaviour patterns.
Now that we know how learning can occur in a general sense, we can apply that knowledge to training a dog to perform any task within his physical limits. For example, we can easily teach [train] a dog to put holes in boot socks, but we can never train him to darn the hole we taught him to make. The dog just can’t get the rope thick darning wool through that tiny hole in the needle. It’s beyond his manipulatory skills, and I can relate to that.
However, in order to use the ability the dog does possess, we must control the situation. We cannot control latent learning and its related insight learning because there is no way of knowing that it is occurring, on outward signs. Similarly, imitation learning cannot be controlled in that we cannot be certain of what the dog perceives or whether the dog is capable of imitating any behaviour other than what another dog does. Also, we really spend relatively little time with our dogs to observe whether or not he is actually imitating or whether he is learning by some learning type other than imitation. Still we can postulate that these types of learning do go on and we can take advantage of them, and indeed we do. But because we don’t know how or when latent learning, insight learning and imitation learning occur, we are not consciously aware we are using them. We all do use them simply because we can’t help it. We could not avoid them if we tried. The memory traces, all the experiences a dog has from before birth right up to the present will make the dog appear as an extraordinary quick learn, average, or an apparent dunderhead.
Return to Main Page
Updated 29th December 2004