by Diana Reiss
In another laboratory I had heard examples of computer-generated sounds that were supposed to be akin to dolphin whistles, and for the most part, they sounded exactly that, computer-generated and tinny. I wanted something that at least sounded biological. A sound engineer from Mountain Music Systems in Silicon Valley, not far from the lab, helped me to create synthesized dolphin whistles. The computer whistles needed to be similar to natural dolphin whistles in frequency and time so they could easily be reproduced by the dolphins, yet they also needed to be different from the dolphins' own signals to show us that the dolphins could acquire new sounds. Through trial and error we found waveforms that generated sounds rich in harmonics that had the timbre of dolphin signals. But, of course, what sounded "dolphinish" to our ears might not have sounded the least bit so to dolphins. Could we create a shared code for humans and dolphins? Could we discover anything about dolphin learning?
Dolphins are adept at mimicking sounds. I therefore expected these dolphins to imitate our artificial whistles in some way, and I hoped to learn more about the process by which they learned these whistles. Dolphins produce a wide variety of whistle types and little was understood of their structure and function. I grandly hoped my keyboard could be a Rosetta stone for helping our decoding efforts.
I began the keyboard study on July 13, 1984, when Pan and Delphi were eleven months old, by simply lowering the keyboard into the pool and recording what happened. Terry and Circe were swimming with their respective offspring, as usual. Each dyad was in what I called a P2 swim formation—side by side within inches of each other. Eventually Terry and Pan approached the keyboard, apparently curious to investigate it. They got very close, and then Terry touched the triangle key with her rostrum. The ball whistle immediately sounded in the underwater speaker right next to them. But they turned and fled at high speed before I had a chance to give her the ball. They continued to swim rapidly around the pool for a while. This novel sound had obviously freaked them out, and they didn't come near the keyboard again during the rest of the half-hour session.
A few minutes after I put the keyboard in the pool the following day, Terry and Pan again approached it, this time much more cautiously. Then Terry pec-rubbed Pan, an act of reassurance, as if to say, Don't worry. It's okay. You try this time. He moved forward and touched a key, and it was like a light bulb went on for him; he quickly got into pressing the keys, hearing the whistles, and getting the objects or rubs. I had designed the protocol so that the positions of the keys changed every minute. The keyboard would be deactivated, and I'd hear in my headphones a mechanical voice saying, "Ball—position one; fish—position three; rub—position four," or whatever the preassigned pseudo-random set was. We changed the positions of the symbols from key to key at the end of each minute so the dolphins would have to learn which symbol was associated with what object or activity rather than erroneously thinking that the position of the key was the critical factor in obtaining what they wanted.
Delphi had been watching what Pan was up to with the keyboard, and two days after Pan's first tentative touch, he joined in too. (Pan was always first to do new things, and Delphi generally followed suit shortly thereafter.) Very soon we had the two boys enthusiastically using the keyboard, just as we had hoped. Terry and Circe, meanwhile, had no interest in it. Why, I'm not sure, but they seemed to be happy to use it as a baby-sitting device. Both mothers were constantly on the go with their calves, always attending closely to them, always swimming with them. Before the calves were born, Terry and Circe regularly rested or slept for ten-minute intervals. They would lie stationary at the surface or swim slowly with one eye closed; remember, dolphins are "specialized" sleepers. But once the boys were on the scene, there were no more breaks, which any human mother can understand. The only time Terry or Circe got to rest after that was when one of them took care of both calves for a while. This form of baby-sitting is called allomothering, and it is observed in both wild dolphins and dolphins in aquariums. Now that the keyboard was in the pool half an hour each day and the boys were enthusiastically using it, both mothers could get some rest. I empathized with them.
Both Pan and Delphi were still nursing at this point, as they would continue to do for another year or two, at least. Dolphins usually nurse for about three years, and sometimes longer, even though they are eating fish as well. Pan had already started to take an interest in fish; at first he'd simply mangled them and generally seemed to be playing at or practicing eating fish, but now he had started to actually eat them. It soon became clear that our inclusion of a fish key, with the promise of a real fish when it was touched, had been a mistake. In each session, Pan spent most of his time requesting and getting fish—small silver smelt—and used the other keys for ball and rub less frequently. (Rub was the least-favorite request throughout the study—perhaps, we thought, because they could get rubs from each other, while they depended on us for balls and fish.) We decided to remove the fish key after thirteen sessions and replace it with a symbol meaning "ring," a category of toy that the dolphins played with often.
At the beginning of the fourteenth session, Pan was the first to come to the keyboard, as usual. He normally pressed the fish key as soon as he got to the keyboard. This time he paused for a few seconds, scanning the keyboard. It was as if he were thinking, Okay, where's the fish key? He didn't try any of the other keys. Instead, he swam to the bottom of the pool, apparently looking for something. I had no idea what. He swam back to the keyboard with a silver smelt in his mouth, left over from the morning's feed. He held the fish up to a blank key and looked me in the eye expectantly. I could almost hear him asking for the fish. Both Pan and Delphi had hit blank keys a couple of times in the early sessions, but they'd quickly learned that those keys produced nothing, not a whistle, not an object, and they hadn't touched them since. The fact that Pan was using a blank key in the way he did was quite remarkable, an obviously deliberate act of attempted communication.
I was suddenly in an awful position. I knew in my heart and in my mind that he was trying to communicate with me in a very innovative way, and I would have loved to give him the fish. But that would have been against the rules of the game. The rules of the game were that he got objects when he touched the appropriate keys. And there was no fish key on the keyboard. I had no choice but to ignore him. He continued to look at me for a few seconds, as if he were thinking, What don't you understand about my message? Can't you get what I'm asking for? He soon gave up. A good example of scientific rigor working against fundamental decency and perhaps discovery. I really wanted to give him a fish.
***
On the nineteenth occasion that Pan hit the ball key, a few weeks into the project, I heard something in my headphones that made me start. I heard the computer-generated ball whistle, of course. But there was something else. Did Pan make something like a ball sound? I thought to myself. I couldn't tell. As I said earlier, unlike dolphins, we humans are not superb acoustic detectors, especially in these circumstances. Pan was pushing his newly acquired ball around. He pushed it against the wall near the hydrophone to my right, and I heard another very clear ball signal, repeated twice. He came back to the keyboard, and this time he clearly produced the ball whistle and then pressed the ball key; again I heard something in my headphones that sounded a lot like Pan making the ball whistle. The same thing happened a third and fourth time. I couldn't wait to get to the lab to print out a spectrograph (a sound picture) of what I'd been hearing. Then I would know if my ears had been fooling me, if it had been nothing but wishful thinking.
What I saw set my heart pounding. It was one of those moments a scientist sometimes experiences, when you know you have seen something new, something that no one has ever seen before. It was a delicious moment. The ball whistle is quite simple: it's a sinusoidal waveform with flat ends, an initial rise, followed by a fall, then ending with a rise, all of which takes about a second. What the spectrograph showed me was that on the first occasion I thought I'd heard Pan mak
e a ball sound, he had in fact produced just the end of the whistle, the final fall and rise, a U-shaped component with a flat tone at the end. On the second occasion he had made the beginning of the sound, a rise and fall. On the third occasion he had matched the timbre, the overall resonant quality of the model sound, by adding harmonics, and in his forth rendition, he'd brought it all together in a beautiful facsimile of the entire ball whistle. In each case, he matched the exact duration of the model sound, but he transposed the frequency a bit, whistling just a little higher. Pan had spontaneously mimicked a new sound, with no teaching, no training, no food reward. No reward at all, except what pleasure he might have derived from learning something new in an interactive, social environment. This was entirely novel.*
By repeating the last part, then the first part, then putting it all together, Pan had done what young children do when they learn new words. It's called segmentation. Kids often imitate the last part of a new word first, such as saying nana after hearing banana. In psychological terms, this is called the recency effect; it refers to the tendency to best recall what has been perceived most recently. Conversely, repeating the first part of a word is called the primacy effect. Then kids begin to put the two parts together.
When a child is able to put a whole word together, he or she often practices it. For instance, a child who has learned the word dolly can often be heard repeating to herself "Dolly, dolly, dolly" while playing with the doll. And this is what we saw with Pan and, subsequently, with Delphi too.
Both Pan and Delphi learned to produce the ball whistle after about nineteen or so exposures to the sound. They subsequently learned the rub sound, the ring sound, and the disk and float sounds that we later added, each one after fewer and fewer exposures. At first, Pan and Delphi imitated the sound immediately after hitting the key and hearing the computer-generated sound; we call this mimicry. But they made the sound at other times too, such as before pressing the key or later or while playing with the ball or being rubbed; we call this production. The difference between mimicry and production is that production is an indication that the dolphin has made a strong association between the sound and the object or activity it's associated with. When Pan or Delphi repeated the ball whistle when playing with a ball, it was the equivalent of the young girl saying "Dolly, dolly, dolly" when playing with the doll. Repetition serves to reinforce the word in the utterer's mind and perhaps forms associations in the dolphin's mind.
Whether these sounds represent to the dolphins what we might call a label, or even a word, is hard to say. And I was very cautious in my interpretation. I stressed that while the dolphins' productive and contextual use of their facsimiles clearly indicated that they had formed close learned associations among the sounds, visual symbols, and objects and activities, I couldn't conclude that the symbols or sounds represented words or labels to the dolphins. But it looks that way. Neither Pan nor Delphi ever whistled the sound for, say, "ball" and then hit the rub key or some other inappropriate key. They were 100 percent consistent in associating the appropriate sounds with the visual symbols.
On two occasions, Pan appeared to use the ball sound as a label while communicating with Delphi. The first time was at the end of the day. I wanted to clear the pool of toys, so I gave Pan the hand signal for "fetch." I thought there were a couple of balls left in the pool. Pan swam off looking for toys to bring to me, but there weren't any floating around. The only remaining toy was a ball that Delphi had in his mouth. Pan went head-to-head with Delphi. I heard the ball sound on my headset, Delphi released the ball, Pan took it in his mouth, and they both came to me at the side of the pool. Did Pan "ask" for the ball by whistling the ball sound? Did Delphi respond to the "request" by giving the ball to Pan? I can't say for sure, but that's what it looked like. These are the kinds of incidents that give you some real insight into how smart dolphins are. But because of the rigors of the scientific method, they are not suitable for reporting in a scientific paper. It happened on just a couple of occasions, and it was uncontrolled.
A year into the study, the results we were getting were exceeding my expectations. We had almost enough data to publish, but given my near-obsessive compulsion to get just that bit more in order to nail things down completely, not quite enough. I wanted more hours of observations, just to see what else Pan and Delphi might show us of their extraordinary minds. But suddenly the owner of Marine World, Mike Demetrios, announced that he was going to have to close the facility within the next few months and that all the animals had to be moved elsewhere. He'd found a new location for the park, in Vallejo, north of San Francisco, but there would be a two-year hiatus before it opened, from September 1985 to September 1987, and everything would be on hold until the new facilities were ready.
We moved the dolphins into their much larger pool complex at the new site in October 1985. All the animals—dolphins, elephants, tigers, otters, chimpanzees, water buffalo, sea lions, and many others—were safely moved into their new homes. But although the site was ready to house the animals, it was not yet ready to house humans. For two years, while the infrastructure and all the human facilities were built, we were unable to conduct research. The Marine World staff and our research team worked in temporary trailers on the fairgrounds across the road from the new site. Our team's primary concern was keeping the dolphins fed, healthy, and stimulated with enrichment activities and objects. The keyboard had to wait.
Finally, on September 28, 1987, we were able to begin again. I remember the date so precisely because as I began to lower the blank keyboard into the pool in preparation for a research session with Pan, Delphi, and their moms, we heard a beautiful production of the ball whistle. Who made it, Pan or Delphi, I wasn't sure. But here they were, not having heard the sound in the previous two years, and one had produced a near perfect rendition of the whistle. Had they been using it between themselves those past two years? Or did the sight of the keyboard elicit the response? Hard to say. But this told us something about their minds and their memory for whistle sounds.
***
The observations of their behavior this second year, during which I was joined by Brenda McCowan, a doctoral student at Harvard, were definitely worth the enforced wait. Pan and Delphi now produced whistles much more frequently than they mimicked them. They often whistled the facsimiles when they interacted with the corresponding objects; they weren't just mimicking the sounds after pressing the keys. This increase in productive use over mimicry suggested a strengthening of the association between the sounds and the objects and implied that, for them, there was some social or functional use of the sounds. It's what you would expect in animals that are learning. The dolphins' own versions of the computer whistles showed great fidelity to the model sounds of their earliest imitations. But during the first and second year, we found that in some of their versions, the duration and frequency (pitch) of their renditions was altered, compressed or expanded. It was the equivalent of saying a word more quickly or more slowly, or varying the highness or lowness of one's voice. When we speak, we vary our voices. What the dolphins' variations meant to them, we don't know.
When we looked at the sonograms (sound pictures) of their calls, we could see that their versions of "ball" and "ring" and other computer-generated whistles were interspersed with whistles from their own repertoire. In this second year of the study, the dolphins' productive use of their facsimiles of the computer whistles increased significantly over their mimicry of the signals after using the keys. But what did this increase in productive use mean? Brenda and I conducted a detailed vocal and behavioral analysis of the video- and audiotapes recorded in all the keyboard sessions, trying to figure out the behavioral contexts in which they used their newly acquired whistles. We discovered that the dolphins were using the whistle facsimiles in behaviorally appropriate contexts! Appropriate contexts were narrowly defined as those in which a dolphin had physical contact with the specific object, not just when the dolphin was approaching or near the object. So when a fa
csimile of, for example, a ball was produced, it was in the context of ball play or contact with a ball. We know next to nothing about the overall repertoire of their calls, what each of them mean, and we know even less about the possible semantics or function of their more complex whistle sequences. But the fact that they showed such a proclivity to learn and use our artificial whistles correctly in context encourages the speculation that their natural whistles might also mean something; perhaps they are even references to objects (or individuals) and activities. We do know that among the rich and varied repertoire of whistles, squawks, and other types of calls the dolphins produce, the most frequently used signal is a contact call that conveys signature information about the caller. Dolphins use these relatively stereotypical calls and can produce the calls of relatives and other members of their social group. It seems that dolphins may call to or refer to other dolphins in what may function as a referential call.
In one study, Brenda and I found that the contact calls of female dolphins from the same social groups shared certain acoustic features.4 I am now studying whether the acoustic features provide information about alliance, family, or social group membership. But as far as decoding the meaning or function of their whistles, we don't yet have our Rosetta stone. And although their whistles look structurally simple to us, this does not necessarily imply that they are simple to the dolphins or that their meanings are simple.
Pan and Delphi's biggest innovation in this second year of the study was the creation of a new type of toy play and the concurrent emergence of a new whistle. In this new type of play, Delphi would hit, for instance, the ball key, actively play with the ball for a few minutes, and then return to the keyboard and hit the ring key. He now had two toys. It isn't easy to hold an air-filled ball underwater, but he deftly managed it at the same time as he had the ring in his mouth, and then he and Pan would toss both back and forth. It took some coordination, but, as everyone knows, dolphins have excellent physical coordination. We called this new activity double toy play, showing how verbally inventive we scientists can be.
