LANGUAGE OF ANIMALS AND INFORMATION

THEORY

by Zhanna REZNIKOVA, Dr. Sc. (Biol.), laboratory and department head involved with behavioral ecology and comparative psychology, Novosibirsk State University, Siberian Branch of the Russian Academy of Sciences, Novosibirsk, Russia; Boris RYABKO, Dr. Sc. (Tech.), Scientific Research Pro-Rector of Siberian State University of Telecommunications and Information Science, Leading Researcher of the Institute of Computer Technologies, RAS Siberian Branch, Novosibirsk, Russia

Experiments based on ideas and methods of the information theory (the general theory of communication) have revealed a sophisticated system of communication among ants that they are using to convey abstract messages. These insects can optimize their messages, make them terse and concise, and they are even capable of basic arithmetic operations. Research methods and approaches employed thereby can also be applied when studying the language and intelligence of social animals of different species.

Scientific cognition knows of certain turning points making us take a fresh look at the world we live in, and bid farewell to what seemed to us orthodox, self-evident truths. This puts us in mind of King Solomon's magic ring that made it possible to under-stand the language of beasts and fowls of the air. Konrad Lorentz of Austria (Nobel Prize, 1973), an outstanding zoologist and one of the founders of con-temporary ethology (behavioral science) had a dream of getting hold of a ring like that. Yet real avenues of research are quite matter-of-fact-through experi-ments fostering progress in natural studies. Our subject deals with one such avenue.

Do social animal species have a language of their own? Although human beings had been concerned with this question for thousands of years, scientific research into this matter began as late as the 19th century. A paradox-ical situation had obtained by the close of the 20th cen-tury: certain species of monkeys, dolphins and parrots were found to be capable of mastering go-between lan-guages thought up by man; but none of them spoke a nat-ural language of their own. Thus far only melliferous bees and African green monkeys have been found to have such kind of language. Decoded by scientists, these languages were found to be quite simple. In other cases, just short fragments could be interpreted. Bees, for instance, using their "dance language", signal data on the flight direc-tion toward a food source and on the distance from it; in turn, monkeys make use of definite sound signals to warn their fellow creatures against dangerous predators.

Zoologists have tried to compile every kind of dictio-naries-Elephantine-English, Lupine-German, and the like. Yet nature is none too generous in presenting us with a magic tool enabling us to pick out meaningful elements from a flow of signals. The oft-repeated com-binations of events in animals' natural life may give us a clue to deciphering such elements. The "language of shrieks" used by monkeys has already been decoded; however, just one "word", the signal of loneliness, has been interpreted in wolfish howling. A study into all the various sounds given off by dolphins get biologists and mathematicians to think that their intricate system of communication* possesses the characteristics of a lan-guage, though only proper names could be identified among the so many words uttered. This discovery-that dolphins call one another by name!-was made quite recently at St. Andrew University in Scotland. Looking into such fantastic results, we cannot escape the expres-sion that in a sea of signals experimentalists snatch out just the crests of a few waves slipping through the fingers. But to return to intermediate, go-between languages. Back in the 1960s, two American psychologists, Allan Gardner and Beatrice Gardner, taught chimpanzee apes the basics of the body language used by the deaf (surdolanguage). The results were sensational, and spurred many similar experiments on other animal species with the use of other systems of signs. That was a real revolution in our understanding of the mental and communicative potential of creatures living side by side with us. But these studies involved artificial lan-guages. The absence of data on natural languages of animals does not mean that such languages are not in

* See: M. lvanov, "Communication of Dolphins", Science in Russia, No. 2, 2009.-Ed.

existence-rather, it indicates that ethologists, have no adequate search methods. We have tried to find them.

IDEA OF EXPERIMENTS

The cover of the textbook Animal Intelligence: From Individual to Social Cognition (written by one of the authors of the present article, Zhanna Reznikova, and published by Cambridge University Press, Cambridge, Britain, in 2007) carries a photo: a group of ants mov-ing in a "binary tree" labyrinth in search of a feeding-trough filled with syrup. This picture illustrates our experiment that proved the existence of the language of ants (formic language). The task that these insects had to solve reflects the realities of their daily life. Hill ants (Formicae rufae) build large anthills and control large tracts of woodland, with the population of their fami-lies running into millions. Searching for food, they keep looking for colonies of plant lice, the suppliers of carbohydrate food. The information on the newfound colonies of plant lice is vitally important to ants. We proposed that ants should have a rather involved sys-tem of communication enabling them to trace their route in twigs and branches, pass on their findings, and do it fast. That is why these insects, as we see it, are a worthy object in applying our essentially new method to studies of the language of insects and animals.

The general theory of communication (information theory) from which we were proceeding was initially conceived as a mathematical theory of communication. It became clear, however, that its concepts should play an important role relative to technical and other systems of communication. Our experiments demon-strated that the principles of this theory hold for the signaling systems of animals and, more than that, can serve as a basis of further experimental studies eluci-dating the key question-wether we are dealing with "talking" creatures or not.

The substance of our information theory approach in studying the subject of communication among animals boils down to the following. A situation is created when test animals have to pass on a pre-assigned quantity of information. We measure the time that this process takes. This is how we can determine the rate of infor-mation transfer.

Many ethologists and linguists agree that the produc-tivity of communication, that is the ability of making up a large number (potentially infinite number) of mean-ingful texts (phrases) comes first. Only in this case can we evaluate a system of communication as a language. We believe it should have at least one essential charac-teristic: the length of a message (and, consequently, the time spent on its transmission) should correlate with the quantity of information contained in it. Let us explain. The basic ideas of the information theory were first set forth by Claude E. Shannon in 1948. One key concept was that of a "quantity of information". Accordingly, in many human languages the length of a message corre-lates with the quantity of information in it.

Now, what is information according to Shannon? If you toss up a coin, it would be either heads or tails. Only two options are possible here. Which one? The uncertainty factor will be equal to one bit (information unit). If an experiment has n equiprobable outcomes, and we learn the result, we get log₂ (n) bits of informa-tion.

Proceeding from these ideas, which are the basis of the present communication media, we studied a sys-tem of communication in animals just as a means of transmission of information as a measurable quantity.

ANTS AND ... BITS

To make ants pass on a preassigned quantity of infor-mation, we invited them to give of what they knew about the syrup feeding-trough placed on one of far-thest "branches" of the "binary tree" labyrinth. We tagged the insects with color labels and watched the carryings-on in transparent laboratory homes. Each family was of around two thousand. Laboratory arenas were divided into two parts: the smaller had a home, and the larger, a labyrinth in which the insects got their feed. They traveled there through a drawbridge. To prohibit them from making straight for the trough, an obstacle (a water-filled dish) stood in their way.

Hill ants know of such a thing as division of labor. They split into work groups comprising one scout (pathfinder) and from 3 to 8 foragers (providers). Each scout comes in touch with its group on locating the food. In our experiments ants could get the syrup only if they left messages on the number of zigzags leading to the trough. That what our "binary tree" was for. In the simplest case it had just one bifurcation point with two troughs at the ends-one empty, and the other with syrup. To find the dainty, ants had to inform one another which way to take-"left" or "right", that is transmit one bit of information. More bifurcations were used in other experiments. The bait was put on different end "leaves of the tree", while the number of bifurcations was the same. In each experiment the sequence of turns on the way to the bait was deter-mined at random-we tossed up a coin-"heads" or "tails". The highest number of bifurcations was six. Just one "leaf had the syrup trough, and the other five were empty. In such experiments ants were able to find the feed fast enough once they got information on the sequence of turns like LRLRRL (left, right, etc.). If there were four bifurcation points in the labyrinth, they had to signal 4 bits of information; if there were five, 5 bits had to be given off, and the like. All told our "binary tree" experiments involved as many as 335 successful scouts and their groups. The word "success-ful" is quite in place here: the number of pathfinders capable of memorizing the way to the trough was going down with greater task complexity.

Once a scout was back home after a successful sally, we measured the time (in seconds) of its contacts with for-agers. We took it as the time of data transmission. We counted such contacts from the moment as the scout and the first ant touched their feelers up until the first two foragers came out of the nest. As the "informer" kept contacting its group of foragers, we changed the labyrinth with another, identical one that bore no traces whatever. Instead of the sweet syrup, there was just water in the trough. So, no odor of the food and traces left by the scout. Meanwhile the foragers, on having been in touch with the "informer", had to fend for themselves after we had removed the scout with tweezers for a while (videoclips of our experiments are available on the site: http://www.reznikova.net/infotransf.html). During each session only one group of foragers was let in. If the group was moving in the right direction, and there were no lag-ging ants but one, the location of the target was consid-ered faultless. As soon as the insects were in, they got their feed. Statistical calculations and control experi-ments confirmed our supposition that ants make no use of orientation modes other than those they have learned from their scouts.

In our "binary tree" experiments the quantity of information (counted in bits) required for correct ori-entation within the labyrinth was found to be equal to the number of bifurcation points. We assumed that the time of the scout's contact with a group of foragers (t) should obey the equation t=ai+b, where i is the num-ber of bifurcations, a-the proportionality factor equal to t spent on transmitting one bit of information, and b-the constant. We introduced this value since ants can leave messages having a direct bearing on the assigned task, e.g. they can signal a "there-is-food" message. Using the data thus obtained, we evaluated parameters of the a and b linear regression and calculated the optional correlation factor r. The dependence between the time of the scout's contact with the foragers and the quantity of conveyed information (number of bifurcations) was close to a linear one. Ants are ten times as slow in information transmission (1 bit/min) as man. This is not so bad at all: the possibilities of the communicative system of insects prove to be astound-ing indeed.

To assess the potential productivity of the formic lan-guage we counted the minimal number of messages essential in target searching. A "tree" with two bifurca-tions offers 2 possible ways, one with three-2³, with six-2⁶; thus, the overall number of possible ways towards the target is equal to 2+2²+2³+...2⁶=126. Such is the minimum of messages that the scouts had to signal to enable access to the feeding-trough for the ant colony.

KOLMOGOROV COMPLEXITY, INTELLIGENCE, LANGUAGE

The "binary tree" labyrinth allowed us to look into yet another essential characteristic of the language and intelligence of ants-their ability to take quick notice of regular patterns and use that for the coding of informa-tion, its compression. The length of a message about some object or phenomenon should be laconic-the simpler this object or phenomenon, the shorter the message. It is much easier for a human being to memo-rize and convey the sequence of turns toward the goal LR-LR-LR-LR-LR-LR-LR (left-right, and so on seven times) than, say, the shorter but unordered sequence RLLRRRLR. The "binary tree" experiments show that language and intelligence enable ants to use the simple regularities of a "text" for its compression ("text" here is the sequence of turns toward the trough). The insects spent half as much time for sending the message LLLLL (five "turn-left" commands) than when dealing with a random sequence of the same length (say, LRRLR). Remarkably, ants start "com-pressing" information only confronted with rather long texts, four or five bifurcations, and more. The time spent by the scout to pass on information on "regular" sequences proved to be at least half as much as in the case of a "random text".

The regularity that we found in the formic language is tied in with the concept of the Kolmogorov complexity. The algorithmic definition of information and complex-

Experimental arena cum a "binary tree" labyrinth.

ity introduced in 1965 by Acad. Andrei Kolmogorov, an eminent Russian mathematician, applies to words (texts) made up of the alphabet's letters, e.g. two letters, L.R. In nonformal terms, the complexity (and uncertainty, or indeterminacy) of a word is equal to the length of its shortest code. For example, the word LLLLLLLL can be distilled to "8L". Its complexity and uncertainty are not high. As shown by our experiments, it takes ants more time to pass a message the more information "by Kolmogorov" it carries. In fact, the Kolmogorov com-plexity of a word is not computable; we can only say that it is easier for ants and humans alike to memorize and get the word LLLLLL across rather than LRLRLR, let alone LRRLRL.

Thus, our ants were found to be capable of leaving many different messages, with the communication ses-sion time correlating with the quantity of information contained in each communication. More than that, they can take account of certain regularities and "com-press" their messages, i.e. make them shorter. Dealing with such an advanced communicative system, we may call it a language.

The absolute majority of ants (and there are as many as 12 thousand formic species!) have no need of an advanced language at all. Many formic species live in small families, and their members cope well with the foraging job singly. Another, fairly large groups of formic species make use of smelly traces left by scouts, and thus get foragers moving. Only few species have reached a higher level of social organization, hill ants in particular. We have studied different ant species and found that members of this high-level social group alone were "talking" insects.

CAPABLE OF NOT COUNTING ONLY

Having learned about the mechanisms implicated in the communication of ants and making use of these mechanisms coupled with our information theory approach, we demonstrated hill ants to be capable of communicating data on the quantitative characteristics of different objects; we showed them capable of simple arithmetic operations, too. At first the results of our experiments seemed truly fantastic. Yet lately data obtained by other researchers have been coming in-about the counting monkeys, birds, bees and even meal worms! Be that as it may, ants are still keeping in the lead-not because they have extraspecial talents in this regard, but rather because of the absence of ade-quate experimental procedures with respect to other species of insects and animals.

We carried out experiments involving what we called the "counting labyrinths". We used them to teach ants to count. These labyrinths comprised thirty "combs", each one supplied with 40 "teeth" (we called them "trunks" and "branches", respectively); each "tooth" was 10 cm (4 inches) long. The "branches" carried troughs, one filled with syrup, and the others, with water. These experiments followed the "binary tree" pattern. On com-municating with their scout the foragers were supposed to strike out for themselves and find the right "branch" with the syrup trough. In most cases the foragers made straight for the trough (put on different "branches" in course of experiments-from the first to the thirtieth) without any wrong passes.

First, about the experiments showing the ability of ants to learn of the number of objects and signal this informa-tion. Taking part were as many as 32 groups of foragers, all told. Upon their contact with the scout they made 152 sorties to the troughs. In 117 cases they made no mistake and hit the right "branch", bypassing the empty troughs. But in 35 cases ants sallied forth to empty troughs and continued foraging for food by crawling onto the neigh-boring "branches". In all these cases the selfsame "lag-gard" scouts were at work. Identified in course of experi-ments, they were barred from further reconnoitering.

The length of formic communications invited the suggestion that the scouts informed the foragers on the number of a particular branch. We could demonstrate that with the use of statistical methods and control experiments. Hypothetically, the insects could have been passing other bits of information, say, about the distance to this or that "branch" or about some other parameters, e.g. about the number of steps to the trough. Still and all, ants are able to handle quantita-tive characteristics and keep one another in the know. To verify this conclusion we carried out many series of experiments by changing the very form and orientation of the lab setup (say, by putting the "comb" vertically, not horizontally, or by bending it to a circle; we also changed the length of the "branches" and the distance in between). In all these experiments, like it was in those involving the "binary tree", the time (t) spent on sending the message and the number of a particular "branch" could be well described by an empirical equation t=ai+b. The values a and b were close for all the variants. Therefore it is quite probable that ants inform one another about the number of a particular "branch". The time that the ants needed to say "No. 20" was about twice as long as that for No. 10, and ten times as much as for No. 2.

The situation is quite different in contemporary human languages. It takes about log₁₀ (i) to record the integral number i in the decimal number system (nota-tion). But people have not been using it always. We know that in some archaic languages this time (of recording and pronunciation alike) correlated with the length of a message, the same way as it is with ants. Thus correlating with No. 1 was the word "digit"; for No. 2 we have "digit, digit", and for No. 3-"digit, digit, digit". And so forth. The decimal system of nota-tion emerged in a long and complex evolutiona-ry process. But this is not to say that the formic lan-guage is primitive. Not at all. The thing is that in an "optimal" language the length of a word should agree with the frequency of its use. Therefore we can-not call the number system "perfect" unless we look into the occurrence (frequency) of different numerals (numbers).

Are ants capable of such arithmetic operations as addition and subtraction? To find out, we should be in the clear how the system of notation works in present-day human languages. Say, addition and subtraction. We can see that best in the example of Roman numer-als: VI=V+I, IX=X-I, and so on. In our experiments we taught ants a mode of notation similar to the Roman one. Although they do not use it in their dai-ly life, under our experimental conditions they had to learn how to add and take away numbers from one to five.

Let us get to our experiments showing the flexibility of the formic language and the ability of ants to per-form arithmetic operations. We were proceeding from this information theory fact: the time of message trans-mission (t) and the frequency of message occurrence (P) can be expressed by t=-log(P), with a letter, word, phrase, etc., regarded as a message. In natural human languages this equation describes situations when the length of the word encoding a particular message decreases with higher frequency of this message. Sundry jargonisms, abbreviations, pronouns and the like are eloquent proof of this trend.

Shorthand of counting labyrinths.

In our experiments some numbers were used much more frequently than others. Our ants were offered the same gadget ("comb") as in previous experiments. In the initial part of the experiment the number of a trough-carrying "branch" was selected using a table of random numbers up until 30. The time needed to pass the "rough-on-branch-i" message correlated with the i as it was in similar experiments carried out before. Then we stimulated the urgency of two messages- "trough on branch 10" and "trough on branch 20" putting the feed on each of them with a probability of 1/3, and doing the same on the other 27 "branches" with a probability of 1/84. In experiments conducted in different years we varied the numbers of "special branches"-10 and 20, 7 and 14; one variant offered just one "special" branch, No. 15.

By way of example let us take a situation when No. 10 and No. 20 "branches" were playing a "special" part. Outwardly these "branches" were in no way dif-ferent from the others. But the frequency of feed occurrence was different: the feed was much more fre-quent on these two "branches". Upon dozens of repeat experiments our ants cut the time of transmitting the "trough-on-branch-10" and "trough-on-branch-20" messages compared with the beginning of the experi-ment when the troughs were set on any of the thirty "branches" at equal probability. Thus, the insects must have altered their system of communication and cut the time for the two most frequent messages.

At the next, third stage of experiments the number of a food-carrying "branch" was in the same 1 to 30 range, and the chances of finding it were equiprobable. The dependence of the time (t) taken to pass the mes-sage that the rough was on the i "branch" proved to be quite different at the third stage of experiments than at the first. The time needed to convey information about the "branch" number was the shorter the closer this "branch" was to one of the "special' ones, No. 10 or No. 20, or to the base of the "comb". For instance, to pass the message that the feed was on No. 11 "branch", it took ants 70 to 82 seconds at the first stage, and only 5 to 15 seconds at the third (recall the Roman numerals: XI=X+I). We assume that in the lat-ter case the scout's message was in two parts: about the shortest way to the trough-carrying "branch", and the distance from the "special branch" to the feed. In all likelihood, the ants "named" the "special branch" closest to the trough, and the number that had to be added or taken away to locate the required "branch". This conclusion is confirmed statistically. Similar experiments followed the same pattern, though with other numbers attached to "special branches". The results concurred in all the cases.

Thus, our experiments based on the ideas and meth-ods of the information theory go to show: first, the lan-guage of ants is flexible enough and not at all primitive; and second, they can add and deduct small numbers (in our experiments one of the addends and subtra-hends varied from 1 to 5). Ants are pretty smart where it concerns vital information on food, its source and coordinates. Treading the hitherto unknown terrain of intelligence in creatures so much unlike us, we have garnered a considerable amount of information on certain general regularities pertinent to intelligence proper to different species of social animals.

We are hoping our results can work radical changes in mind sets concerning the place and role of humankind in the biosphere and universe.