Tusker Home



SETI NEWS
 The SETI Institute Newsletter



DO DOLPHINS TALK?
 Laurance R. Doyle
Principal Investigator at SETI Institute

Click Pic to Enlarge
Chief Seattle once said, "Mankind would die of a great loneliness if all the animals disappeared." Could it be that they have something to share with us? Do animals communicate and will we ever really be able to tell what they are saying to each other—and perhaps to us?
If one is willing to admit that all animals communicate (in some way or another) then the next question might simply be to ask how complex their communication systems are. A mathematical tool for actually calculating the complexity of a communication system was invented in the late 1940s by Claude Shannon of Bell Labs (who, incidentally, thanked our own Barney Oliver for helpful discussions in the acknowledgements of this seminal paper). The new field was dubbed "Information Theory," and has since been used to calculate the amount of information to be sent along, for example, computer lines. But we were the first to apply it to dolphin whistles.
I got interested in dolphin whistles years ago when I thought that some of the signal detection techniques I'd learned for astronomical applications (planet detection, for example) might have application to the identification and classification of dolphin whistle-vocalizations. Little did I know, when a colleague of mine, Jon Jenkins, and I visited Dr. Brenda McCowan and Sean Hanser at Marine World in Vallejo one afternoon, what we would be getting into!
Many people have, of course, intuited that dolphins are very bright "folks." Their trainability, learning, and playfulness all attest to this. Many scientists have embarrassed themselves by initially underestimating dolphins' intelligence. (It is a well-known fact that dolphins also have to think the experiment is fun or they won't do it.) But, we asked ourselves, would it ever be possible to measure the complexity of their whistle communication system? We wondered if one could put a number, so to speak, on just how much communication intelligence (really the meaning of the Drake's Equation term Fi) they express.
That day at Marine World the dolphins had had their fun with me by tossing a ball to one side of the pool, where I ran around and got it, and after tossing it back they tossed it to the other side, where I ran around to get it, and they kept this up until I caught on to the notion that I was being run around on purpose. Looking up at the natural smile on the profile of the dolphin who had been throwing the ball, it was difficult not to think that I had just fallen for a dolphin practical joke.
Heading back into the analysis room where Brenda, Sean, and Jon where discussing signals, I remember that Jon pointed out that there had recently been an article that discussed the "linguistic properties" of junk DNA, and how this result had been pretty controversial. In this paper the genetic scientists had made some plot of the frequency of occurrence of so-called "junk" DNA and had been able somehow to show that this distribution might imply linguistic properties. This was my introduction to information theory.
In the late 1940s a linguist named George Zipf had had graduate students count all the letters in some typical English texts—Ulysses, for example (before word processors!). He plotted the frequency of occurrence, FO, (actually the logarithm of FO) of the letters in decreasing (also log) order. The letter "e" for example, occurred the most frequently (10.1% of the time), followed by the letter "t" (about 8% of the time), followed by the letter "a" and so on down to the letter "q" (0.1% of the time). Drawing a line through these values gave a slope of -1, that is, a 45-degree line from top left to bottom right pretty much went through all the points on the graph.
He tried the same for Chinese letters and also got a -1 slope. The same for Russian phonemes, Arabic letters, and English words, and many other human languages, all giving this -1 slope line. However, when the same plot was done of baby babbling sounds the slope was much less steep, that is, the plot of baby babbling sounds gave a closer to horizontal line. This meant that the sounds babies were making were more random than adult language. (Babies make all sorts of sounds that humans can make before they start to specialize in one language—but the sounds don't have a distribution of FO that indicates communication complexity.)
A plot of adolescent baby talk, however, showed a line even more vertical than the -1 slope line. This also made sense since adolescents learning a specific language repeat a few sounds a lot (especially, in English, for example, "Mama"). As vocabulary is added—adolescents to adult speech, the Zipf plot approaches the -1 slope of adult languages in every case.
Armed with this idea, we decided to plot adult dolphin whistles, since Brenda and Sean had measured enough whistles to know their frequency of occurrence. It was indeed quite a surprise to find that the distribution of their signals gave a -1 slope—the same as human languages! It was decided next to record baby dolphins—less than one month old—and the plot of the FO of their whistles fell exactly on that of human infants! Next, the distribution of adolescent dolphin whistles showed more redundancy in a manner similar to adolescent human language learning (a more vertical slope), before approaching the mature language -1 slope again, just as in humans.
From this work we really had to conclude that dolphins learn their whistles, and statistically do so in a way similar to the way humans learn their languages. Also, this was mathematical proof that the dolphin whistle-vocalization system of communication contains a substantial degree of complexity. How complex is it? We don't know yet, but we know the way to begin finding out.
Human languages have syntax—we rely on the juxtaposition of certain words, for example, to other words for meaning (i.e. context). Human languages, for example, apply context to strings of up to about 8 words in a row. However, there is not much relationship of a given word to words more than 8 words farther away in typical spoken English; that is, we tend to lose the context after a string of about 8 words. In information theory we would say that English word does not have any additional complexity after the 8th-order Shannon entropy. (Mathematically we would say that the conditional probability of a given word more than 8 words away from a previous word is zero.)
For dolphins we have found that they have at least 2nd-order context to their signals and quite possibly 3rd or 4th-order as well (we are analyzing the data now). I wonder how many dolphins would be allowed to be caught in a fishing net if we find they have, for example, a very complex "language"? In addition, we have also applied these methods to Squirrel Monkeys, Humpback Whales, Elephant Seals, and we soon hope to be able to include Elephants, Hippos, Parrots, Bees (their waggle dance) and perhaps some of the simpler "critters" (those where we expect only a first-order entropy communication system) for comparison.
Someone asked me recently if we were taking the first steps toward a "Star Trek-type universal translator". I don't know. So far we can only quantify the complexity of the signaling system of other species; we may be quite a few steps away from "translation" yet. But if we do get a SETI signal some day, and can record the modulation and classify individual signals, we should be able to use the general rules of what we have learned about the communication systems of our fellow inhabitants of this planet, to at least tell the communication complexity of this extraterrestrial transmission. We should be able to tell immediately if it is consistent with a communication system, for example.
But what if such a SETI signal turns out to have 15th-order entropic complexity? That is, what if it turns out to be far more complex than any possible human communication system to date? If communications were ever to ensue, we might hope that they would look upon us with more respect than perhaps we have looked upon most other species of this planet to date. They may just consider an 8th-order entropic communicating species as approaching the level of intelligence necessary to recognize the necessity of taking care of every species on its planet - intelligent enough to avoid the great loneliness. I hope so.

SETI  Newsletter - Fourth Quarter 2001
Volume 10, Number 3