Chaser (34 page)

“Take ball to Frisbee” did not work well, however, because of Chaser's tendency to focus most intently on the last thing she hears. For Border collies the shepherd's last word or phrase contains the essence of the command. With “take ball to Frisbee” she tended to pick up the Frisbee rather than the ball.

Not long before the address and demonstrations at the APA, it occurred to me that the two halves of the sentence might be clearer for Chaser, and easy enough for me to say, with a structure of “to Frisbee take ball.” The sequence of grammatical elements then became
prepositional object, verb, direct object
.

This sentence structure made the last thing I said the first thing Chaser needed, and wanted, to do: pick up an object in her mouth. The structure also followed the format of herding commands, in which the last words a dog hears represent what he or she has to do next.

The first times I told Chaser “to Frisbee take ball” and “to ball take Frisbee,” with only those two objects on the floor, there seemed to be less hesitation in her responses. But I wasn't sure if the difference was real. Over the following few days, however, it was obvious that her confidence was increasing.

Alliston told me, “You're using the same syntax structure as in Spanish.” He is fluent in the language from his years of teaching in Mexico. It tickled me that I had stumbled onto a sentence format used in a real human language.

Finding that a Spanish-style sentence clicked for Chaser was a neat aha moment for both of us. The experiment could proceed and the play could continue.

I selected a hundred of Chaser's 1,022 named toys as a training group. Working with random pairs of toys, I frequently reversed the roles—prepositional object and direct object—that two toys had. I also reversed their positions on the floor, sometimes putting the direct object on the right and sometimes on the left.

At first I pointed to each toy as I spoke its name. Over time I phased out the pointing. I also complicated the task by putting two possible prepositional objects and two possible direct objects on the floor. Later I put the two prepositional objects and the two direct objects in different rooms.

An early training trial began with two toys five to ten feet apart on the floor. With Chaser close in front of me and both of us facing the toys, I said, for example, “Chaser! To Sugar take Decoy.” She had to pick up Decoy in her mouth, carry and drop it near Sugar, and then take Sugar in her mouth. Completing the trial brought Chaser praise and play with Sugar.

If Chaser headed toward the wrong object or got hung up in indecision at any point, I recalled her without correction and repeated, “To Sugar take Decoy” or whatever the initial instruction was. As always I wanted Chaser to start off with errorless learning and gain confidence with each success.

As we worked on “to Sugar take Decoy” and “to Decoy take Sugar” type sentences, there was less need to recall Chaser and repeat the instructions. If she hesitated, I said, “Do it, girl. Do it!” Enthusiastic encouragement usually emboldened her to make a choice, and usually it was the right one. If she kept hesitating I recalled her without correction and gave her the “to A take B” instruction again. Her comfort with that gave me the impression that hesitating became her way of saying, “Could you please repeat that sentence?”

It was fascinating to see Chaser try various strategies for responding to sentences like “to Santie Claus take Flipflopper” and “to Flipflopper take Santie Claus.” After being reinforced for picking the toy on the right a few times in a row, for example, Chaser apparently formed the hypothesis that I always wanted the toy on that side. Gradually she realized that this strategy was unreliable.

It also emerged that how I said “to A take B” mattered quite a lot to Chaser's comprehension. She was best able to process the words and hold them in working memory when I slowly but emphatically said, “to A, take B,” with a definite pause between “to A” and “take B” and rising energy as I completed the sentence. The more enthusiastic and encouraging I made “take B” sound, the better.

Early in 2012 the journal
Learning and Motivation
invited Alliston and me to submit a paper on Chaser for a special 2013 issue on animal learning. Alliston didn't have time, but he suggested I do it on my own. I decided to report my three-elements-of-grammar experiments, assuming the results were statistically significant.

In addition to my training-and-play sessions with Chaser, Sally and I incorporated three-elements-of-grammar sentences into our interactions with her, including our nighttime routine. She plays and snuggles on the bed with us, lying in between Sally and me and being petted by both of us, until we say, “That's enough, Chase.” At that point she curls up on the end of the bed for a while, before moving off to the living room couch and other favored sleeping spots. “To Nanny, take Sugar,” “to Pop-Pop, take Decoy,” and “to bedroom, take Crawdad” closed the day's training at home, as “to ball, take Frisbee” and “to Frisbee, take ball” trials began it at Wofford.

By fall 2012, Chaser became virtually perfect in responding to three-elements-of-grammar sentences in informal trials and play with Sally and me. She could complete the reversal trial (to B, take A) even if it did not follow the initial trial (to A, take B) until after intervening trials with other object pairs. It was another instance of relationships powering Chaser's learning.

Seeing Chaser's accuracy exceed 90 percent in informal trials and play convinced me that she was ready to work with other testers in formal trials under double-blind conditions. The double-blind tests would provide results for the paper for
Learning and Motivation
.

The tests were blind in that the testers, Wofford student volunteers furnished with instruction sheets with the commands for each trial, did not know the names of Chaser's toys. So there was no chance of Clever Hans cues. The tests were double blind in that other students, who were not present during the tests, evaluated audio-video recordings of them. During the tests, I operated a camcorder to capture audio and video.

We did the double-blind tests in two experimental scenarios with random pairings, varied placement of toys, and initial and reversal trials for each pairing. In experiment 1, using toys from the training group, Chaser got twenty-five of thirty-two right, or 78 percent. In experiment 2, using toys we had not used in training the sentences, she got eighteen of twenty-four right, or 75 percent. In both cases the probability that her correct answers resulted purely from chance was less than one in a thousand. This showed that her understanding of this kind of sentence could not be attributed to chance factors.

When Alliston watched the video of experiment 2, he noticed that Chaser looked briefly at the prepositional object and the direct object as she heard their names, and then went straight to the direct object. This suggested that Chaser's brain might not be processing a mental image of the direct object into working memory, because she did not take her eyes off it after hearing its name.

To test this possibility, I devised experiment 3 using six random object pairs. For each pair there was an initial trial and a reversal trial, with two possible indirect objects an inch apart on the living room floor and two possible direct objects an inch apart on a pillow at the head of Sally's and my bed.

Standing in front of Chaser as she lay on the foot of the bed facing me, so that she could not see the toys on the pillow, I gave her the “to A, take B” command. When she stood up and turned around to select one of the direct objects, I could not give her a visual cue because the objects were so close together. Having made her choice of a direct object, Chaser raced with it into the living room and raced back with her choice of a prepositional object, which we played with before starting the next trial.

Chaser got all the trials in experiment 3 correct, with a less than one in a million probability that her choices resulted from random chance. Together with the first two experiments, this indicates that her brain does process a mental image of the direct object into working memory when she hears its name.

The three experiments also demonstrate both that Chaser securely lodged the names of all the toys and the meanings of “to” and “take” into long-term memory, and that she successfully held two toy names, “to,” and “take” in short-term working memory while she made a semantic judgment about which toy to pick up first. The results provide strong evidence that Chaser can understand the syntax and semantics of sentences with three elements of grammar. And in showing that she can combine two cognitive abilities, long-term memory and working memory, to solve language tasks, the results raise the bar in terms of expectations for a dog's language learning.

Chaser's results suggest the possibility that she experiences a “phonological loop” when she completes “to A, take B” tasks. A phonological loop occurs when people repeat the directions for a task silently in their heads until the task is done. Researchers with access to brain scanning equipment might track a canine phonological loop by comparing activity patterns in a dog's brain to those in human brains when people silently repeat a sentence to themselves.

The dolphins Phoenix and Akeakamai respectively got 62.3 percent and 56 percent of their three elements of grammar sentences correct. In her equivalent trials Chaser got 78 percent and 75 percent correct. Taken together, these results indicate that an animal's brain can in some way operate like a language-learning toddler's brain and display implicit understanding of rudimentary grammar.

No one teaches little children the rules of grammar—they may never learn the formal rules of grammar—but somehow they acquire an implicit understanding of it. From the first babbling to the first sentences, from “Mama” and “Dada” to “More juice” to “I love you” to “Tell me a story,” they listen and respond in ways that show they are gaining a sense of how words go together to create sentences (syntax) and how the same words can have different meanings depending on their order and other relationships (semantics). “Mama, breakfast” means something different from “Mama's breakfast,” and so on.

Again, science has not yet solved the mystery of how this happens during a critical early developmental stage for toddlers. But we know that if either children's brains or their social relationships are insufficiently developed and nourished during the critical phase, language acquisition suffers. In the worst cases of either, including children growing up in abusive conditions of severe isolation from others, full language acquisition may not be possible. Extreme isolation from other people can be worse than a brain injury. The developing brain is so flexible in the way it organizes itself into networks of nerve cells that it is frequently possible for a child to gain full command of human language despite a severe brain injury—if the child is young enough and benefits from good language-based interactions with other people.

This picture suggests that dolphins and dogs, both highly social species, may make progress in language learning precisely because they can connect emotionally with people. The social character of these species and their cognitive evolution seem to be interdependent. Chaser's apparent edge on the dolphins in understanding sentences with three elements of grammar may result from the unique interspecies social relationship that has evolved between people and dogs. In any case I have no doubt that Chaser's focus on relating to people, a characteristic all domestic dogs share, is her strongest asset in learning language.

Support for that idea comes from language learning trials with nonhuman primates. The chimpanzee Washoe and the bonobo Kanzi, who were both raised and trained with lots of affectionate human contact, have far exceeded in language learning apes raised and trained without such contact.

Matching to sample is another way I am trying to use Chaser's emotional and social intelligence to advance her learning. When we do match-to-sample trials, I hide several objects—a newspaper, a washrag, a ball, a stuffed animal—in a corner of the living room, behind the couch, and in other places around the house. I'll show Chaser a duplicate of one of those objects and say, “Find this.”

The only way I identify “this” is by pointing to it. Chaser has to recognize the most important characteristics of the object I'm holding and find another object with the same characteristics. At some level, conscious or unconscious, she must make the mental inference that I want an object like the one I am holding. In addition, Chaser has to understand that the object I am holding is not absolutely unique. She has to imagine that there can be others like it, and then hold that abstract idea in her mind while she searches for something similar. This may sound like a simple task, but rats and pigeons require hundreds of trials to learn to complete it.

Chaser's performance in matching to sample differs at home and at Wofford. At home, she has become very reliable in matching to sample, with results similar to those Juliane Kaminski has obtained in asking dogs to fetch an object after showing them a replica or a picture of it. But at Wofford, Chaser can be quite erratic in matching to sample. On some days, she quickly matches six objects in a row. On other days, she gets only three or four right. This is the one area of Chaser's training where there is such a difference with regard to place, and I haven't yet figured out what might explain it. In any case, match-to-sample learning remains a challenge as I try to identify what problem-solving hypotheses, implicit or explicit, might be in her head.

Imitation learning has also intrigued me for a long time, because to imitate something you have to be able to imagine yourself doing it. Both the imitation and the imagination can be unconscious. Young children generally don't realize they are imitating when they reflect behaviors they see their parents or other people display. Throughout life we consciously and unconsciously model ourselves on people who have a strong influence on us. But whether we realize it or not, we must make a mental inference about the point of view and intention of any individual we imitate before we can attempt to do as they do.