Donald’s evolutionary theory, unlike many others, stresses the unique cognitive development of human beings, the importance of their culture, and the consequences of their inventiveness, but he does not invoke the concept of a second replicator. For him, the function of language is part of the wider function of symbolic representation, whose advantage is ultimately to the genes.
I have considered several popular theories of the function of language. All their authors realise there are serious problems, and have tried to explain why language would have given early hominids a selective advantage. I am not convinced that any of them really solves the mystery of human language origins. They need to explain why there is just one
species capable of communicating with complex grammatical language, why this one species has a brain so very much bigger than its nearest relatives, and why this one species goes around talking not only about sex, food, and fights, but also about mathematics, the advantages of Macintosh over Windows, and evolutionary biology. There are obviously some advantages to being able to communicate complicated things. When the environment changes, a species that can speak, and pass on new ways of copying, can adapt faster than one that can adapt only by genetic change. Could this be reason enough for all the expensive changes that evolution has brought about in order to give us speech? I do not know. I can only conclude, after this necessarily brief review of the existing theories, that there is no real consensus over the issue.
The situation can be summarised like this. Darwinian accounts of the evolution of human language have assumed that language provided a selective advantage to the genes, but despite many suggestions there is no unanimous agreement on what this selective advantage was. However, this argument assumes that Darwinian explanations must rest entirely on genetic advantage. If we add a second replicator the argument changes completely.
Language spreads memes
Memetics provides a new approach to the evolution of language in which we apply Darwinian thinking to two replicators, not one. On this theory, memetic selection, as well as genetic selection, does the work of creating language. In summary, the theory is this. The human language faculty primarily provided a selective advantage to memes, not genes. The memes then changed the environment in which the genes were selected, and so forced them to build better and better meme–spreading apparatus. In other words, the function of language is to spread memes.
This is a strong claim and I shall therefore take the argument slowly, building on our understanding of coevolution.
I have already explained how meme–gene coevolution could have produced the big brain. To summarise – once imitation has evolved, a second replicator comes into being which spreads much faster than the first. Because the skills that are initially copied are biologically useful, it pays individuals both to copy and to mate with the best imitators. This conjunction means that successful memes begin to dictate which genes are most successful: the genes responsible for improving the spread of those memes. The genes could not have predicted the effect of creating a
second replicator and cannot, as it were, take it back. They are now driven by the memes. This is the origin of the dramatic increase in brain size. This theory predicts not only an increasingly large brain but a brain that is specifically designed to be good at spreading the most successful kinds of memes. I shall argue that this is exactly what we have, and that this explains the evolution of language.
If successful memes drive the evolution of the brain, then we need to ask which memes these are. To some extent the success of memes is a matter of serendipity and accidents of history. In our long past it might have been the case that long hair or ringlets, painted faces or scarred legs, singing, worshipping the sun or drawing pictures of insects, came to be the favoured memes. These would then have exerted pressure on the genes to provide brains that were especially good at copying these particular things. If the forces of accident were the major pressures in memetic evolution we would have little hope of ever making sense of our past. However, I am going to assume that overwhelming these forces of serendipity are the fundamental principles of evolutionary theory. That is, there are some basic qualities that make for a successful replicator – in this case a meme.
Dawkins (1976) identifies three criteria for a successful replicator: fidelity, fecundity, and longevity. In other words, a good replicator must be copied accurately, many copies must be made, and the copies must last a long time – although there may be trade–offs between the three. We must always be careful of comparisons with genes, but we can usefully consider how they match up to these requirements.
Genes are high on all three. Their method of replication is extremely accurate. That is, genes have high fidelity in the sense that very few errors are made when long sequences of genetic information are copied. When errors are made there are elaborate chemical systems for repairing them. Of course, there are some remaining errors, and these contribute to the variation that is essential for evolution, but the errors are very few. Also, the process is digital, which makes for much higher fidelity, as we have already seen.
Genes, at least some of them, are extremely fecund, producing masses and masses of copies, though the fecundity varies with the kind of environment a species inhabits. Biologists distinguish two kinds of selection at the extremes of a continuum: r–selection and
K
-selection.
r
-Selection applies in unstable and unpredictable environments where it pays to be able to reproduce rapidly and opportunistically when resources allow. High fecundity, small size and long–distance dispersal are favoured, as in frogs, flies and rabbits.
K
-Selection operates in stable, predictable
environments where there is heavy competition for limited resources. Such conditions favour large size, long life and small numbers of well–cared-for offspring.
K
-Selected species include elephants and humans. These are the extremes, but even in the most
K
-selected species many copies of the genes are made.
Finally, genes are long lived. Individual molecules of DNA are well protected inside cells, and those that are passed down through the germ line can sometimes survive as long as the lifetime of the organism. Depending on the size of unit you count as the gene, its lifetime varies, but in some sense genes are immortal, since they are passed on from generation to generation to generation. Genes are extremely high–quality replicators.
Were they always that way? Presumably not, although we do not know much about the early history of DNA. However, it is reasonable to assume that the first replicators were simpler chemicals than present–day DNA, and were not packaged efficiently in chromosomes inside cell nuclei and with a complex cellular machinery devoted to their maintenance and duplication. They may, for example, have been simple autocatalytic systems that give rise to two identical molecules, followed by polynucleotide–like molecules, and then RNA (Maynard Smith and Szathmáry 1995). But why should these chemicals have evolved to produce the high–quality replicating system that we have today?
Imagine the competition between various forms of early replicator in their primeval soup. If a low–fidelity replicator and a high–fidelity replicator existed at the same time, the high–fidelity one would win out. As Dennett (1995) puts it, successful evolution is all about the discovery of ‘good tricks’. A replicator that makes too many mistakes in copying soon loses any good tricks that it stumbles across. A high–fidelity replicator would not stumble upon them any quicker (and arguably could be slower) but at least it would keep any it found – and thus outperform the competition. Similarly a highly fecund replicator would, simply by virtue of making more copies, swamp its rivals. Finally, a long–lasting replicator would still be around when its competitors had fizzled out. It is obvious really. In this early environment there would have been selection pressure for better and better replicators, and this could ultimately have resulted in the exquisite cellular machinery for copying DNA.
The same principle can be applied to memes. Imagine early hominids who have discovered the biologically ‘good trick’ of imitation. Initially, this good trick allowed some individuals to profit by stealing the discoveries of others, and these individuals therefore passed on the genes that made them imitators until imitation became widespread.
Then a new replicator was born and, using the copying machine of the brain, began to make copies – copies of actions, copies of behaviours, copies of gestures and facial expressions, and copies of sounds. This world of early memes is the memetic equivalent of the primeval soup. Which of these potentially copyable actions will be more successful as a replicator? The answer is those with high fidelity, high fecundity, and longevity.
Now we can see the relevance of language. Language certainly improves meme fecundity. How many copies of an action can you spread at once? As many copies as there are people watching. But not many people can watch one person perform at once, and also the people nearby may just not be looking, or may get bored and look at something else. On the other hand, if you make a sound, many people can potentially hear it at once and they do not need to be looking – they can even hear it in the dark. This advantage is obvious in the difference between sign languages and speech. They may both be effective for private conversations but you cannot shout to the masses ‘Hey, you must listen to this’ in sign language. The masses have to be looking first. Also, sound can travel over fairly long distances and round corners. A lot more copies can be made by shouting out your news than by demonstrating it with hand signs, facial expressions, bodily movements or any of the other available signals.
This means that vocalisation is a good candidate for increasing fecundity and thereby winning the battle to become the better replicator. How, then, could the fidelity of the copies of the sounds be increased? One obvious strategy is to make the sounds digital. As we have seen, digital copying is far more accurate than analogue, and genes have certainly adopted the ‘get digital’ strategy. I suggest that language has done the same. By making discrete words instead of a continuum of sound, copying becomes more accurate.
We might imagine many versions of early verbal language going on simultaneously as people began imitating each other. Any which divided speech up into discrete, easily copyable sounds would have higher fidelity and hence outperform the others in the race to get copied. The problem with copying always lies in deciding which aspects of the stimulus are the important ones to copy. Language is a system that makes these decisions clear by, for example, breaking up the sounds, and adopting norms of pronunciation, while ignoring overall pitch. Note that other forms of communication, such as the warning cries of monkeys, can become gradually more and more distinct by genetic selection, but the process described here works much faster because it spreads from person to person within one generation. Because higher fidelity copies spread more effectively, they tend to predominate, and the language improves.
What about longevity? No individual behaviour has much longevity in itself, but longevity inside the brain is important. Some actions are hard to remember and therefore hard to copy, especially after a delay. We would expect the successful memes to depend on behaviours that are easily remembered so that they can be reproduced even after long delays. Language has very efficiently improved memorability, remembering dance steps can be troublesome, but remembering ‘slow, slow, quick–quick slow’ is easy. We find it impossible to reproduce a long series of meaningless noises, but easy to repeat back a sentence of a few dozen words. Without too much trouble we can repeat whole stories and conversations. Indeed, many cultures have depended entirely on rote learning of long stories and myths to pass on their history. By structuring the meanings of sounds, language makes them far more memorable.
We can look to technology for another kind of longevity – as when the invention of pots creates long–lasting models for new pots and more pot–making behaviour, or when the building of bridges spreads the idea of a bridge to everyone who crosses one. The longevity of language took a dramatic turn with the invention of writing – committing words to clay, papyrus or floppy disk – but I shall consider these further steps in longevity later.
I have described the appearance of words as a process of digitising. The real problem for understanding language origins is not so much the words, which at least in principle can be learned by simple associative learning, but the grammar. However, grammar also improves replication. How many things can you say with a given set of words? Not very many, unless you have some way of specifying different meanings if you combine the words in different ways. Adding prefixes and suffixes, inflecting them in different ways, and specifying word, order would all increase the number of possible separate utterances that could be produced and copied. In this sense, grammar might be seen as a new way of increasing fecundity as well as fidelity. The more precisely the copies are made, the more effective they will be. Then, as more and more possible things can be said, more memes can be created to continue driving the process.
Remember that all that is going on here is selection, with no need for conscious foresight or deliberate design on the part of either the memes themselves or the people who are copying them. We need only imagine groups of people who all tend to copy each other, and they copy some sounds more than others. Whether a particular sound is copied because it is easy to remember, easy to produce, conveys a pleasant emotion, or provides useful information, does not matter as much as the general
principle, that when lots of sounds are in competition to get copied, the successful ones will be those of high fidelity, high fecundity, and longevity. This is the selection pressure that produced grammatical language.