In a study published this week in Science and devoured by dog lovers worldwide, researchers at the Family Dog Project in Hungary concluded that pooches can understand the meaning and tone of human speech and that they process language in the same way humans do. To figure this out, the scientists analyzed scans of dogs’ brain activity when hearing words. And to get those scans, they needed their subjects — 13 family pets — to lie completely motionless in an fMRI scanner for eight whole minutes while wearing earphones and a radiofrequency coil on their heads.
Several readers wondered how on Earth this is possible. What sort of dog magician made this happen? The answer is Marta Gacsi, an ethologist at Eotvos Lorand University in Budapest, co-author of the study and longtime researcher of man’s best friend. She was the head trainer of the dogs — six border collies, five golden retrievers, a German shepherd and a Chinese crested. We spoke to her about how she got them to cooperate.
This interview has been edited for space and clarity.
Most of them were really young. Some of them were trained assistance dogs for disabled people; they need a dog that is eager to please, so that’s what we needed here. But we also had very, very untrained dogs. We had dogs that could only sit down and stay. It’s not really an important thing to be well-trained to be successful in this thing.
We need dogs that trust the owner. As in the case of a child [in an MRI exam], you say, maybe you won’t see me, but I will be there. Some were trained by me and handled by me. And it was okay, because the dog trusted me. That was a very important prerequisite of the task.
At the beginning, it seemed logical that we needed highly trained dogs with a calm temperament. None of these hypotheses turned out to be true. There were many trained dogs that turned out to be perfect and some highly trained dogs that failed, because they wanted to do something. They wanted to do their best, but they couldn’t understand that here the task is doing nothing. Usually these dogs were trained using many treats. When they couldn’t get their treat in half a minute or so, then they were frustrated and we couldn’t use them. Some other dogs didn’t understand that they were required to stay absolutely motionless. They always moved a little bit or they licked their mouth every third minute. And that’s enough to ruin the whole test.
[Successfully trained dogs], you can see in their eyes when a drop of water falls on their noses and they know, ‘I cannot lick it.’ It’s really … I don’t know what to say. They are not forced. They are asked. You can’t imagine how happy they are at the end. They bounce to the others like, ‘Okay, I did it! I did it!’ We are really seeing that they are proud.
We didn’t want to work with normal trainers, because, at least in Hungary, there are two different lines. The traditional way, they use too much punishment. The other way is based on positive reinforcement. That means treats — food, food, food, and nothing else. And that wouldn’t work in the scanner. We base our training on ethological research. So it’s a little bit different than normal training like obedience or Schutzhund. It’s rather like you want to persuade your child, your 5-year-old little one, to spend a minute motionless in the scanner. How would you do it? That was the major thing. It wasn’t, how would you train an animal? We shouldn’t handle them as infants, but we definitely shouldn’t handle them as wolves. They are dogs, and that’s a very special thing. So we use social learning and, of course, we use food rewards as well.
The first part was lying on the floor with the head between the paws. Then the same on the table, motionless. Then the same while the table is moving a little bit. And then the same wearing earphones, and their head is gently tied with some cloth. That’s to make them feel that something is on their head. But it was never fixed. That’s a very crucial point of the training, that dogs could always at every second leave the position and the place if they wanted to do so.
Because of our method, the real problem we had is that when we opened the door to the MRI scanner room, all the dogs wanted to rush in. The scanner is extremely noisy. It’s not fun to be there. Try to explain to your grandma why it is good to be in a disco. It’s very noisy, it smells like hell, there are too many people. It doesn’t make sense, and people even pay for it! It’s social learning. When you were a child, you could see that all the young adults love to go there, so you want to be there. You want to be part of the party. That’s what we wanted to do with the dogs, so we always had a big party in the scanner room. The dogs that were placed on the scanner, we praised them. All the little ones, all the novice dogs, they wanted to be there. They didn’t know why, but they wanted to be there. Even the owners of the [dogs in training] praised and petted the [already] trained dog on the scanner bed. So that dog was the focus, and we ignored the other ones. Everybody wanted to be the focus, they wanted to be praised, they wanted to be rewarded.
This is the model-rival training method. Some dogs acted partly as a model for the other dogs, showing what to do and where to be. And also as a rival – all the attention went toward this dog. That’s a social thing. It’s not about cheese or treats. Its about being socially involved in the social interaction with the owners. A good dog wants to be part of this.
How long it took
There were dogs that could only come to the lab once in a month. So that’s not a really efficient training, but we could do it. We needed to ask the owners to please come to the scanner on Sunday night. That’s when we had access, because otherwise it’s used for medical purposes for humans.
Usually the basic training, not with the scanner, that would need from five to 20 sessions. It depends on the dog. And training at the scanner, usually about 10 sessions. At the beginning we always used the dogs together for motivation and, of course, we used food reward as well. But we had to be very cautious, because we couldn’t give them treats during the test. So that’s at least eight minutes without any reward. And they even couldn’t expect food, because of drooling! If they drool, they lick their mouth and swallow, and that’s it. So they couldn’t even think of getting a food reward during the scan. So we had to explain to them that it’s a very long story, and at the end you will have a lot of very good food — but only at the end.
My major focus in this respect is to work out methods based on social learning techniques, instead of using this continuous food reward. Positive reinforcement-based methods are really nice compared to the punishment-based methods, but it seems that we tend to forget about social aspects. Positive reinforcement can be anything, like praise and social rewards. So we use our scientific experience. And we try to run experiments that would give some insights to the applied field as well, for trainers. Because there are so many training theories, but they don’t really have data. All of them have experience, and that’s good and nice, but it’s very difficult to tell what’s really the truth.