Authors: John W. Pilley
Rico was only tested in fetching objects by name, and Paul Bloom's biggest question was whether a dog could show knowledge of words in relation to more than one action. As I've described, Chaser showed this in learning to take named objects in her mouth, or nose or paw them, on command, including objects she'd never been asked to do any of these things with. This demonstrated both that she understood independent meanings for two elements of syntax, an object name and an action verb, and that she understood novel combinations of objects and commands.
Bloom also questioned whether animals could learn common nouns. I was looking forward to documenting Chaser's understanding of “ball,” “Frisbee,” and “toy” as common nouns.
Markman and Abelev seconded Bloom's reservations and added two of their own. First they were skeptical that language-trained animals could progress beyond “mutual exclusivity,” the assumption that things have only one name. Markman and Abelev observed that in the early stages of language learning five-month-old infants “expect that a novel word will refer to a novel object rather than a familiar one,” but by seven months most babies learn that this “mutual exclusivity” does not always hold true and that a single thing can have multiple names. They learn that a set of common nouns such as “cat,” “kitty,” and “pet” can all refer to the same thing, and that a set of proper noun names and nicknames such as “Elizabeth,” “Lizzy,” and “Sweetie” can all refer to the same person.
When I read Markman and Abelev's article on its publication in November 2004, Chaser was seven and a half months old. At that point she knew that she was also called baby, Puppy, girl, and sweetie, as well as combinations of those names and nicknames. Simply by overhearing the various names, she knew that Sally was also Mom and Nanny, and that I was Pop-Pop, Dad, John, and Pilley. She knew that food had many names: breakfast, lunch, supper, treat, chew stick, and bone. She knew that the command to jump had three labels (“hoop,” “up,” and “over”) and that the command to come had four labels (“here,” “come,” “hurry,” and “right now”).
In her common noun learning, Chaser showed that she could distinguish balls and non-balls, Frisbees and non-Frisbees, toys and non-toys. She could also identify one of her balls both by its proper noun name and as a ball and a toy, or one of her Frisbees both by its proper noun name and as a Frisbee and a toy.
Thus Chaser demonstrated the one-to-many and many-to-one mapping of words and things that both Bloom and the team of Markman and Abelev doubted dogs could achieve.
Markman and Abelev's second big reservation about the Rico study had to do with learning by exclusion, also known as reasoning by exclusion. Rico demonstrated this when he retrieved a novel object from a group of familiar objects on hearing its name for the first time. In the absence of any external or physical cue, he apparently inferred that the novel name referred to the novel object. However, Markman and Abelev questioned whether Rico's success really showed that he was drawing an inference.
Markman and Abelev granted that the procedure for Rico's “reasoning-by-exclusion test trials . . . closely parallel[ed] those used to demonstrate word-learning by exclusion in young preschoolers.” But they raised three possible objections. First they noted that Rico's tested vocabulary only included proper nouns having a one-to-one correspondence with the objects they named. Based on this prior experience, Rico might assume that words are mutually exclusive and that thus a novel name must refer to a novel object.
The second possibility Markman and Abelev suggested was that Rico had a default assumption that words are mutually exclusive based on other prior learning or an innate propensity. This could also make him always pick a novel object on hearing a novel name.
Chaser's combined proper noun and common noun learning showed that she knew that words are not mutually exclusive. She understood that objects could have more than one common noun category label (“toy” as well as “ball” or “Frisbee”) plus a unique proper noun name. And she understood that the same individual could have more than one proper noun name. Her demonstration of many-to-one and one-to-many word-object mapping put these first two possibilities to rest, insofar as her own learning by exclusion was concerned.
The third possibility Markman and Abelev suggested was that Rico had a baseline preference for choosing novel objects. They pointed out that the Rico study said nothing about testing whether Rico had such a baseline preference. Studies of learning by exclusion in preschoolers did test for a baseline preference for novelty, and such a preference, rather than a correct mental inference, might be why Rico picked the novel object after hearing its name for the first time.
In response to this third possibility, the Rico study researchers replied that, before asking Rico to retrieve a novel object, they first asked him to retrieve two familiar objects. This demonstrated that his choices were not owing to a preference for novel objects.
I decided to replicate this procedure. In addition, I decided to measure Chaser's baseline preference for novelty before her first opportunity to learn a word by exclusion.
I did not attempt to train Chaser to learn by exclusion. But I felt confident she could do it, based on Rico's example and, more important, her own ability to grasp a series of abstract concepts. Chaser grasped that objects could have names, and then became able to learn a new object name on one trial. She learned that common nouns refer to categories, and then became able to distinguish categories based on shared physical characteristics (balls versus non-balls and Frisbees versus non-Frisbees) or more abstract shared functional characteristics (toys versus non-toys).
Before testing Chaser's ability to learn a word by exclusion, I tested her baseline preference for novelty. I took eight familiar objects and rehearsed Chaser in retrieving them by name. The next morning, with Chaser waiting in my study upstairs so that she could not see what I was doing in the living room, I put the eight familiar objects and two novel objects behind the couch. Then I called Chaser downstairs and asked her to retrieve each of the eight familiar objects by name. She did this without a hitch.
Over the next week I repeated this procedure seven times, each time with two different novel objects and eight different familiar objects. Because Chaser now knew about nine hundred proper nouns, there was no shortage of familiar objects.
In all eight trials, Chaser only retrieved the familiar objects I asked for by name. This established that she had a zero baseline preference for novelty.
Next I rehearsed Chaser on retrieving seven different familiar objects by name, in order to be sure that she still remembered their names. After that I had Chaser wait upstairs so that she could not see what I was doing, and I put the seven familiar objects behind the couch along with a novel object that I had not used in looking for a baseline preference for novelty. When I called Chaser to come downstairs into the living room and sit in front of the couch, I first followed the Rico researchers' procedure by asking her to retrieve two familiar objects.
And then I said, “Chaser, find Lounge. Find Lounge.”
Chaser padded behind the couch while I sat with my back to her. Several seconds passed without her bringing anything to me. I repeated, “Chaser, find Lounge. Find Lounge.” I waited ten more seconds, marking the time by the clock above the television, before turning around to see what Chaser was doing, or rather, not doing.
She was just standing over the novel object and the remaining five familiar objects. I turned back around quickly to make sure I didn't unknowingly give her an identifying cue. And then I repeated, “Chaser, find Lounge. Find Lounge.” Over the next minute I repeated those words several times, but she still did not return with an object.
I was wondering how much longer I should wait when Chaser slowly came around the couch. In her mouth was a plush miniature chair from a dollhouse set.
“Good dog! Good girl, Chaser! You found Lounge!” thundered out of my mouth in relief and excitement. The five minutes of play with Lounge were extra joyful for me.
The next day I repeated the procedure with seven different familiar objects and a different novel object she'd never seen before. She immediately retrieved the two familiar objects I asked her to get. And then I said, “Chaser, find Chipmunk. Find Chipmunk.”
Again there was quite a long pause, more than a minute altogether, during which I repeated “Find Chipmunk” several times. Again she finally came slowly out from behind the couch with the correct object, a little stuffed animal in the form of a chipmunk.
In subsequent trials, there was no delay. And Chaser made the correct inference every time. After hearing its name for the first time, she always found and brought me the novel object she'd never seen before.
I immediately followed these novel choice trials with retention tests. I placed a novel object whose name Chaser had learned by exclusion behind the couch in a group with three novel objects whose names she hadn't learned and four familiar objects. Immediately after the selection choice, the retention test showed that she remembered the object name. When I repeated the retention test after ten minutes and after twenty-four hours, the results were inconsistent, as they also were with human toddlers.
With this learning, Chaser matched Rico's demonstration of selecting an object on the basis of exclusion. She was in shouting distance of learning more than a thousand proper nouns. And she had met all the additional criteria for word learning proposed by Bloom and by Markman and Abelev. It was time to begin writing up the experiments for publication in a peer-reviewed journal. Given Chaser's unprecedented language learning, I assumed it would be no trouble at all getting a paper published in a good journal.
I should have known better.
12
B
Y THE TIME
Chaser turned three in the spring of 2007, she knew more than a thousand objects by their proper noun names. There were 800 stuffed animals, 116 balls, 26 Frisbees, and 100-plus plastic and rubber items. I began writing a paper that would share the impressive results of her learning. I hadn't written a peer-reviewed paper in a long time, but I knew the form well. And I thought I had an excellent recent model in the Rico paper, which was distinctive not only for its content but also for its fairly conversational style, though I intended to provide a more thorough explanation of my training and testing procedures.
I included the full spectrum of language learning I observed in Chaser, even if I didn't have extensive data on some aspects of it. I believed that the remarkable nature of the findings would justify publication in journal editors' eyes. I entitled the paper “Can a Dog Learn Nouns, Verbs, Adverbs, and Prepositions?”
My hope was that
Science
, the world's most prestigious scientific journal along with Britain's
Nature
, would publish the paper as a sequel of sorts to the Kaminski paper on Rico. And I hoped they'd ask Paul Bloom to contribute another “Perspective,” in which he would acknowledge that Chaser's learning met all the basic criteria for word learning. Toward the end of the summer I told Alliston Reid that I was going to send the paper to
Science
.
With a smile Alliston said, “They only give you one shot there, John.”
“I know,” I said. “But they published the Rico paper and I've gotta think this is a pretty good shot, given how much Chaser has learned and how she's met Bloom's and Markman and Abelev's major criteria for referential understanding. Here's hoping, anyhow.”
A few days after that, with my eyes blurry from reading and rereading for typos and grammatical errors, I sent the paper to
Science
. Several weeks later an editor at
Science
briefly e-mailed me to say that upon review they were rejecting the paper. There was little detail as to what might have been said about the paper during the review process or what its specific flaws might be.
I was stunned.
Rereading my paper, I confirmed that my experimental procedures in testing Chaser's understanding of proper noun words and her ability to learn by exclusion were identical to those in the Rico study. In addition I had sharpened the paradigm for testing Chaser's exclusion learning by establishing that she had no baseline preference for novel objects. Likewise, I had described rigorous procedures for teaching and testing the learning of common nouns.
But I had to admit that some critical details of my studies were missing. Unfortunately I had not presented the usual tables and figures to display my findings but had only described them in words and a few key numbers. I also recognized that much of the paper was too informal and did not say enough about how Chaser attained her language learning and how I tested it. So I optimistically set about rewriting the paper.
Although I felt my paper was comparable to that of Kaminski and her colleagues, I recognized that as a retired professor from a small liberal arts college and without a long list of peer-reviewed papers to my credit, I had nothing like their stature in the animal science community. At the time of the Rico study's publication Kaminski led a research group at Germany's Max Planck Institute for Evolutionary Anthropology (she has since moved to the University of Portsmouth in England), the equal of any research institute in the world in that field.
In evaluating scientific experiments there are two major errors: Type 1 errors and Type 2 errors. Type 1 errors occur when you read too much into experimental data. That means you see an effect that has not actually occurred. Type 2 errors occur when you read too little into experimental data. That means you miss a significant effect that has actually occurred. The event is accurately represented in the data, but you don't see it.
As an experimenter, I was more concerned about making a Type 2 error and failing to report Chaser's genuine achievements than I was about making a Type 1 error and overstating those achievements. But scientific journal editors and peer reviewers have to be more concerned about making Type 1 errors and publishing something that won't stand up to scrutiny by the scientific community at large. Those two perspectives constitute one of the constant push-pull dynamics in the course of scientific progress. And science needs both. It needs experimenters to try bold things, and it needs their peers to hold them to a tough standard.