The Co-evolution of Speech and Speakers

Forbidden Planet's Robby the Robot. What sort of language would a race of them evolve?

Two kinds of evolution tangle thinking about the origins of speech. The first is biological evolution that adapted our bodies to the task of speaking. Then there is the linguistic evolution that shaped the languages we speak. If we define evolution vaguely as change over time, we can agree that both bodies and languages evolve. But if we speak more precisely and say that evolution refers to the Darwinian processes that produce change, can we also say that language really evolves? A recent report in the journal Cognitive Processing says yes, linguistic evolution is more than just analogous to biological evolution; it uses Darwinian processes. If the report’s authors are correct, two evolutionary processes were triggered when people began to speak. One adapted our bodies to speech, the other adapted speech to our community. The dual evolution is very much like the co-evolution of brain and language that Terrence Deacon proposed in his book The Symbolic Species.

Just what evolves in language? Biologically speaking, we know that the genes as encoded in DNA molecules evolve. Biological evolution, when reduced to its minimalist core, is the change in the genome over time. What changes over time when language evolves?

As a minimum, sounds evolve. Whan that Aprille has become When April with the a (ah) in whan becoming e (eh), while the second a went from ah to ay. Then a bunch of sounds disappeared. The e at the end of Aprille is gone, and what happened to the that sound? In this context, it’s gone altogether

The article, “Language Evolution as a Darwinian Process” by Pierre-Yves Oudeyer and Frédéric Kaplan (available here) examines the ways soun might evolve. A word of warning, the authors’ examination takes the form of computer simulations and although they call the simulations “experiments,” a more accurate term is “calculations.” The researchers were performing calculations in the modern style about how certain things might work. Science has used calculations for centuries. It is a proven technique for generating ideas. Sometimes, as in the cases of Newton and Einstein, the results can seem uncanny, they are so accurate, but every time the results have to be tested against reality. The report, therefore, presents a series of promising calculations about how selection might work to produce a community of speakers who share a common set of sounds and how those sounds could change over time.

Their calculations cannot bring us all the way back to the origins of speech. Just as biological evolution takes the existence of life for granted, the authors take the existence of speech for granted. But once you get past that hump, the many calculations reported tell an interesting story that suggests explanations for why there are so many languages and variations on individual languages. Their work also reinforces Deacon’s idea about why children learn to speak so easily.

Put simply, Terrence Deacon suggested that, given two possible versions of a language, children will learn the easier one. Over time, this process results in languages that are very easy for children to learn. The orthodox theory holds that children learn a language easily because the rules of language are specially built into their brains. In other words, humans adapted biologically to the task of learning to speak. Deacon’s idea flips that order and says language adapted to the task of being easy to learn. The Oudeyer-Kaplan team does not include Deacon in their list of references, but their work supports his thesis. Most importantly, their work favors linguistic evolution right from the beginning without need for innate guidance.

Their method is to suggest possible ways a sound might be selected, and then run simulations to see if the method makes computational sense.

One proposition that survived simulation says that a language is easier to learn if you can pronounce it. The idea is hardly startling. Who can learn what cannot be spoken? The authors ran simulations based on the idea that people with no language (that would be infants) can only learn to use sounds they have already made, or that are at least close to sounds they have already made. Since babies of eight months or so babble a wide range of syllables, this demand is not as hard as it might seem at first. In their simulation others nearby hear the babbling. If the syllable heard is close to one the listener has already spoken, the listeners voice an approximate repetition of the sound heard. Over time groups of individuals in close proximity will develop a group of shared syllables.

Interestingly, if one runs two simulations, the populations of agents will always end up with their own particular repertoire of syllables. [p. 31]

A second calculation used these syllable repertoires developed under the first simulations. A new generation coming along and, “hearing” these existing sounds, the newcomers quickly and easily learned them too. In an alternate simulation, a new generation was placed among “speakers” who used syllable repertoires that had not been evolved but simply generated randomly. It turned out that the new generation had much difficulty learning this random system.

In other terms, the syllable systems developed culturally by agents were adapted to their cognitive biases, and the random systems were not. [p. 32]

Here we can see linguistic evolution in action. Just by following a common sense method of selection—imitate those sounds you already know how to make—a set of sounds will be shared that are easy for all the members of the group to make, and this ease of use will be passed on through the generations.

Notice, by the way, that sentence structure: “the syllable systems … were adapted to….”  It was the syllables, not the “agents” doing the adapting. To use the non-Darwinian jargon of our times, the speech that survives is user-friendly speech.

This ease of speech also figures in the selection of “words.” Suppose something new comes along. An example might be Internet-based, personal, journalism/commentary/meditation. The phenomenon appeared in the late 1990s and spread rapidly in the new century. Obviously we cannot keep calling it Internet-based, personal, journalism/commentary/meditation, but how is a name selected. You can call them Internet diaries, IDs, web diaries, web logs, net journalism, and on and on. Many different names are possible and did appear. So how did we settle on blogs?

The Cognitive Processes authors found that, of the selection methods they examined, the only one that quickly gets an entire population to use a single name is when speakers use the name they hear most frequently. In our global era, with communications world wide, this selection process seems to have resulted in a very widespread use of the word blog. A couple of generations back, when televisions entered the conversation, populations were limited to more local communications so teevee evolved in the United States while telly emerged in Britain.

The authors report that hard to pronounce words fair less well in competition because they are pronounced in so many different ways that speakers do not hear a consistent sound. So once again, the mechanical processes of evolution are enough to ensure the appearance of an easy to pronounce word. Getting from web log to blog required more than dropping of the we. The o sound went from aw to ah. (Does anybody know how the word blog has fared in languages that have trouble with the bl sound?)

Thus sounds and even whole words can evolve through a selection process that is hardly more conscious or deliberate than biological natural selection. Of course, the calculations do not settle anything. They still need serious, systematic experimentation in the real world before their computations become scientific facts. Their general point, that the sounds of a language are arbitrary is one of the first axioms of linguistics, and it would be nice to see that axiom turned into something more precise and scientific. In this case, a better word than arbitrary might be “contingent,” the sounds of a language are the result of selective chances that could have gone some other way, but not just any old other way. They had to evolve in a direction that supported their survival and stability.

If the underpinnings of language evolve so easily and inevitably, we can wonder again why they have not appeared among other primates, particularly great apes. The processes reported are simple and natural enough to explain the existence of communities of bird songs. Why didn’t something similar arise among the gestures of chimpanzees and bonobos?

We come back to the basics of this blog: meaning, attention, caring. Speech is more than sounds, it means something. The meaning of speech is discovered in the way it directs attention. An individual is willing to let another direct their attention, only if the individual cares about the thoughts of another. Time and again, that process of coming to care about others reveals itself as the fundamental question of how we came to be what we are today. Once we came to care, we were smart enough, and the basics of evolution, both biological and linguistic, were simple enough, to ensure that in the fullness of time we would become the species that could not shut up.

Comments

[jess tauber]

Yet there are many harder phonemes in the languages of the world- witness the very complex phones in Northwest Caucasian languages, involving secondary and tertiary articulations, or words in Khoisan languages- clicks are so unnatural that speakers end up with characteristic scars on the back of their throats from wear and tear. And what about glottalized/laryngealized vowels, and so on.

Historical change often bring together features and contrasts that the baby/simple motivation doesn't explain. How can this be?

The answer is COMPETING motivations- simplification in one area may result in complexification in another. Often the competition is between what is paradigmatic (vertical axis/same-time slot) versus syntagmatic (horizontal axis/across-time). Getting rid of a vowel intervening between two consonants simplifies things on the syntagmatic side, but if it ends up wholly or partially merging the consonants into an even more complex single unit, then the paradigmatic complexity increases, as phonemes phonologically are strings or stacks of features.

Comparative work has also shown that such interactions often go hand in hand with changes in the stress marking system, syllable type, word order, etc., though all the rules are not well understood, enough so that these are still typological statistical generalizations. Still there is somthing going on beneath the surface.

Ultimately, if easy/simple was the only motivation, unlinked to anything else, we would end up with no segments at all.

Jess Tauber

[Simon]

In Chinese blog is 博克 (bókè), in Korean it's 블로그 (bŭl-lo-gŭ) and in Japanese it's ブログ(burogu).
-----------------
BLOGGER: Thank you. I was pretty sure that eventually a visitor would be able to fill in that blank.

[Michael Santomauro]

Fantastic essay!