Kanzi: The Ape at the Brink of the Human Mind

by Sue Savage-Rumbaugh

  Knowing how to use the symbol “banana” as a way of getting someone to give you a banana is not equivalent to knowing that “banana” represents a banana. (It is not obvious that this distinction had been seriously dealt with by other ape-language researchers.) Full communication would require that the chimp be able to use the symbol “banana” without expecting to receive one. We hoped we might simply reverse the request task to achieve this understanding on the part of Sherman and Austin. We held up a food item and lighted the “?What this” symbols. We didn’t expect them to know immediately the intent of what we were saying, but we tried to make it clear by nonverbal means. If the chimps replied correctly, we gave them another food item as a reward.

  Initially, both chimps selected the appropriate symbol when we held up specific food items, but they then reached out for the food, expecting to receive it. They were puzzled when they were given something else instead, and continued to gesture for the original food item. Eventually, they became upset at what was happening, not because they were being deprived of food—they weren’t—but because their expectations were being violated. Soon they stopped producing the appropriate symbol for the food item being displayed.

  We pondered how we might make the distinction between requesting things and naming things more obvious to the chimps, and decided to make the food items we displayed nonedible. We did this by coating the food with plastic resin (the sort you see in the windows of restaurants in Japan). We hoped that by removing the possibility of consuming the food, we would also remove the chimps’ expectation that they would receive it when they selected its name.

  This procedure proved to be futile as well. When Sherman first saw me hold up the plastic-coated M&M’s, he hurried to the keyboard, said “M&M,” and held his hand out to receive the candy. When I gave him one, he popped it into his mouth, got a strange look on his face, and spat it out. He then asked for another one. I gave him another one and he did the same thing. After a few more trials, he looked at me and said “banana” instead. When I gave him a piece of plastic-coated banana, he did not even bother to taste it before throwing it on the floor and stomping on it. This didn’t help because there was no way to “open” this banana. When it became apparent that none of these plastic-coated foods was going to be edible, Sherman refused to use his keyboard to name or request them. Austin deemed the task equally meaningless and both started pointing at the refrigerator and the door to communicate their wishes to me in a more direct manner.

  Sherman and Austin finally learned to name things independently of any event that followed naming by a procedure known as “fading.” It begins like a request regime, with a food item being held up and given to the chimp after he lights the correct symbol. Fading refers to the fact that the size of the food item given to the chimp gradually diminishes, while the size of the item he is being shown does not change. At the same time, we lavish great praise for correct answers, and give a sizeable portion of a reward food item. For example, if the food we wanted them to name was “sweet potato,” we gave them a smaller and smaller piece of sweet potato each time they did so, along with a large piece of some other food, to indicate we were still pleased. This additional food was the same whether the food they were asked to name was a sweet potato, an M&M, or soycake. The procedure worked. After 102 trials for Sherman and 201 for Austin, they both could reliably name these three food items without error through thirty trials. Thus if I held up sweet potato, M&M’s, or soycake, Sherman and Austin could easily select the appropriate lexigram, regardless of what happened afterward—even if I gave them nothing.

  Had they really gotten the idea? Could they now use their symbols to indicate the names of things without expecting something specific to happen each time? The only way to tell was to show them a number of items not used during the fading training to see if they could generalize this concept to the other symbols in their vocabulary. They did so easily. We tried 21 additional trials with all new items, and they were both 100 percent correct.

  It seemed like a giant bridge had been crossed. Sherman and Austin now really fully understood whether they were asking me for something or just telling me that a particular symbol was supposed to “go with” something. One was a communicative act, the other was merely a show of competence. Neither in itself was dramatic. What was important was that they could do both and they knew there was a difference. It meant that they recognized a distinction between communication for its own ends and communication about the properties of units of the system. I knew that linguists would dismiss it all as unimportant because syntax was not required and that some sort of “conditioning” explanation would be evoked by behaviorists. So I did not bother to write a paper saying that apes were able to understand the distinction between symbols as symbols and symbols as communicative devices—but I knew that a watershed had been crossed. Still, we were some distance away from something as simple as two-way communication between apes.

  Before Sherman and Austin could really talk to each other, they had to become competent listeners as well as speakers. Other ape-language projects had understandably taken the process of listening for granted, assuming that if an ape could use a word, surely it could understand the word as well. I had already learned that it wasn’t that simple. I had tried to get Sherman and Austin to ask each other for foods with the vocabulary they had. It was a disaster. Both chimps used symbols, but neither paid very close attention to the other, and certainly neither chimp picked up food and handed it to the other in response to a request—as I did when they asked me for food. Names they knew—but listening and cooperating seemed a world apart from the skills they currently possessed.

  With a distinction between requesting and naming apparently established in their minds, we now had to move to receptive competence, or comprehension. This new skill would require that the chimps be good listeners, something no chimp had been taught in any ape-language study. I knew that teaching and testing for comprehension wasn’t going to be easy, partly because of its “private” nature. But how could we expect Sherman and Austin to communicate with each other unless they knew what to do when they saw a word being used? Other apes like Lana and Washoe could respond to simple inquiries like “Are you hungry?” or “What is the name of this?” But learning to listen with your eyes and to cooperate when someone else says something like “Please pass the sweet potatoes, not the bananas” was an entirely different matter. It required more than just answering familiar questions with acceptable symbols or signs. It meant that they had to understand precisely what someone else was telling them to do, and they also had to be willing to actually do it. They could not get by just using words to indicate their own needs.

  In our case the task was made even more challenging, because Sherman and Austin’s vocabulary was small, thus limiting how we might interact. The simplest path would be to ask the chimps to give us something, such as a food item. Thirteen-month-old human children can perform such tests, but we knew that giving things—particularly food—was not among chimps’ favorite activities.

  My fear that Sherman and Austin would be reluctant to give me food items when I requested them proved unfounded, at least after a series of practice trials. But usually they failed to give the specific food I asked for. Instead, they gave their least favorite foods to me. Was decoding my request and searching for a specific food item too much to ask of the chimps? I needed another path to comprehension—one that would emphasize that in order to convey very specific information, I and others used words that they should try to comprehend and use to guide their ensuing actions.

  I decided to use a hiding paradigm, in which my communications would function to reveal the nature of a specific food hidden in a container. The chimps already knew that when I entered a room with food items, they were allowed to ask for them using the keyboard. I decided that I would enter the room, after giving enthusiastic food barks to catch their attention and interest, but with a single food item hidden in a container. I could then u
se the keyboard to tell them what food I had hidden. I assumed they would want to know, so they could ask me for the food.

  The first time I did this Sherman rushed to smell the container, but was unable to detect what was in it. He gestured for me to open the container, but I refused. Instead, I went to my keyboard, located just outside Sherman and Austin’s room, and stated this chow. When I used my keyboard, it made the symbol “chow” appear on projectors located just above Sherman and Austin’s keyboard. Sherman saw this information and apparently believed me because he immediately used his own keyboard to say open chow. On the next twenty trials of this novel situation, Sherman made just two errors, even though I used many different words. Sherman watched what I had to say each time and then asked for the food that I had indicated.

  One of the errors he made was quite curious. On trial five, I told Sherman that M&M’s, one of his favorite foods, were inside the container. Unlike the previous four trials, he did not ask for the food that I said was in the container. Instead, he asked for each of the four foods he had seen hidden in the container during the four previous trials! Did he think that I was simply going to repeat what I had done before and that he no longer needed to pay any attention? After trying four times to tell him that M&M’s were in the container, and being ignored, I finally decided just to open it up and show him. His mouth fell open and he seemed somewhat astounded that indeed there were M&M’s in the container, just as I had said. It seemed as though he could hardly believe his eyes. From that point on, he attended closely and seemed to believe everything I told him about hidden foods. I will never know, but perhaps Sherman was testing my veracity in saying I had M&M’s.

  Although the exchange, this X … give X might look like simple repetition on the chimp’s part, I knew it was not. Known as match-to-sample, this simple sequence was one that Sherman and Austin had previously been incapable of engaging in. If shown one lexigram and asked to select its match, they inevitably failed such tests until they reached seven years of age.

  If they could not match-to-sample, why could they then ask for the food that the teacher asserted was in the container? The communicative situation—made salient by desired food items—seemed to encourage comprehension of what was being said. It appeared that they understood I was telling them about the hidden M&M’s. Once they realized that M&M’s were in the container, they began to search their boards for this lexigram. It seemed that they were using the referential value of the lexigram to help them. That is, the symbol for M&M’s on the projector made them think of M&M’s hiding in the container. They knew what symbol to use to ask for M&M’s and so could find it on their keyboard. Without a communicative situation like this, however, they could not match a symbol to itself.

  My training as an experimental psychologist told me that this did not make any sense. Why could I not show Sherman and Austin the symbol for, say, banana and have them find it easily on their keyboard? Match-to-sample tasks were reportedly easy for Washoe, Lana, and Sarah. And why was it that when I told Sherman and Austin there was a banana hidden in a container, they could then readily find the banana symbol on their keyboard to ask me for it? If I analyzed only what they did, both were ostensibly reducible match-to-sample tasks. The principle of parsimony would require me to take the simplest possible interpretation—in other words, that they were matching a symbol to itself. Consequently they should have utilized this strategy regardless of how I interpreted their behaviors of interest in hidden objects and so forth.

  Yet only if I took into account what they seemed to think they were doing did it make any sense that they could do one task and not another. Matching symbols to ones flashed on their projectors made no communicative sense, so they paid little attention to what they saw. However, learning about the nature of a hidden food made a lot of sense, so they watched closely and remembered what they saw. Yet there was no framework in psychological theory at the time that attributed any sort of thought process to animals. Nor was there any theory that even discussed the fact that the understanding of another’s communicative intent might affect the way that the communication was perceived, or, indeed, if it was even perceived at all.

  Many psychologists have steadfastly maintained that there are no important differences between a match-to-sample procedure and a communicative one such as we used—both are really just forms of elaborate imitation. After all, in both cases, the teacher says X, then the ape says X. Of course, whether or not these procedures differ depends upon the state of mind of the participants. If the chimpanzee assumes his job is to imitate the teacher, he will approach the situation with a different frame of mind than if he assumes that the teacher is conveying a privileged bit of knowledge about a hidden item. We could not directly ascertain the contents of Sherman and Austin’s mind; nonetheless, it was clear that they assumed we were making a statement about the contents of the container.

  The idea that chimpanzees could possibly approach communicative situations with any sort of hypothesis about the intentions of the communicator was not one that was widely accepted at this point in time. Consequently, my explanations regarding the importance of the manner in which the apes internally characterized the experimenter’s goals met with general disbelief and occasionally outright anger. Sherman and Austin’s expectancies, however, made all the difference. If, for example, I mistakenly informed them about the container’s contents (which happened a few times), they would grab it from me and look persistently inside, as though they could not believe that I was wrong. Of course, they did not refuse to eat the food that was in the container—this would have been foolish—but they did persist in looking under it for the food that, according to my statement, should have been there.

  Once Sherman and Austin could respond to information about the kind of food that had been hidden, we were ready to take the next step in building communication between them by using the system we had just developed. I wanted them to tell each other what was hidden in the container. After all, if they could now listen to what I had to say about the container’s contents, perhaps they were ready to listen to each other. Moreover, since they knew the names of all the different foods I was using, they should have been able to tell each other what they had seen hidden as well, as long as one of them was permitted to see the container as the food was being placed inside.

  On the first trial Austin accompanied me to the refrigerator, where he saw me put some banana slices in the container. My colleague Sarah Boysen, who had not seen the transaction, then took Austin to his keyboard and encouraged him to use the keyboard to identify the contents of the container. Austin quickly commented “banana,” even though there were no bananas in front of him, only the closed container. Thus he had to recall the food that he had seen placed there. Meanwhile, Sherman had seen Austin go into the room with Sarah, with an evident interest in the container; and he saw Austin press the symbol for “banana.” Sherman quickly went to his keyboard and also lit “banana.” The chimps then shared their spoils. Again, this cannot be counted as simple delayed symbol-matching, because Sherman and Austin routinely failed at such tasks when we administered them in contexts devoid of communicative intent.

  While we had no reason to believe Austin recognized on the first trial that he was telling Sherman what was in the container, Sherman seemed to believe that Austin was describing the container’s contents. After all, he had seen Austin go into the room to get the food, so it would be reasonable to assume that Austin knew something he did not about the container’s contents. It is both interesting and important that no one had to show either Sherman or Austin that the chimp who saw the container being baited possessed knowledge that the other did not. The singular fact that the chimp who did not know what was in the container nearly always waited and watched until the other one revealed its contents was sufficient to show that the chimp recognized differing states of knowledge, based on observing, or not observing, the baiting process.

  While the chimp who had not seen the bait seemed to
know at once that he needed to watch what was said by the other, there did not exist an equally clear understanding on the part of the chimp who did know, that he needed to tell the other chimp. After all, if he knew, was it not sufficient for him simply to ask for the food for himself? Why should it matter if the other chimp saw what he said or not?

  This was why we imposed the experimental constraint that required both chimps to ask for the correct food before the container was opened. Only if both had correctly requested the food in the container were the contents of the container shared. This meant that the “informer” needed to be certain that the other chimp paid attention and got the answer right. For the first eight to ten trials the informer (who had seen the food being placed in the container) simply requested the food for himself. He did not put himself in the mind of the uninformed chimp—not initially, at any rate. After both chimps had been the informer and the uninformed several times they began to realize that the informer had knowledge that the other chimp needed, if both were to receive food.

  This understanding became obvious when one or the other of the chimps made a mistake. For instance, on one occasion Sherman (the informer) correctly indicated apple, but Austin hit the banana key. Perhaps he didn’t believe Sherman, or was hoping that he would be able to get a banana, which he preferred. When the container was opened to show Sherman and Austin its contents, it was clear who had made a mistake, and Austin tried to change his request to “apple.” Thereafter, Sherman monitored Austin’s behavior very closely, and if Austin looked hesitant he would urgently repeat his identification of the container’s contents.

  We could see, therefore, that through this procedure Sherman and Austin had learned two important features of humanlike communication: interindividual communication and cooperation. Not only had they learned how to communicate with each other, but they had also learned the value of communication—they achieved joint access to food items. Nevertheless, the presence of a teacher was important for the interchange to work. In the absence of the teacher, the chimps were unlikely to share the food voluntarily. In their natural habitat, chimps may collectively feed at a fruiting tree, but for the most part each individual eats alone, often turning its back on others. Only in the unusual circumstance of a captured monkey or other prey is food shared, and then in a rather grudging manner. If we were to establish humanlike communication between Sherman and Austin using food as a motivator, we would have to teach them to share food freely.