As for language evolution, these facts are undeniable: Chomsky dismissed it for a long time, his dismissal was treated as an irrefutable argument, and now language evolution has taken on a life of its own. Probably the truth is that the boom in language evolution has occurred both because of and in spite of him. Chomsky brought the attention of the world to the complexity of language and its innateness. Whether his version of complexity and innateness will endure is another matter.
The overriding outcome of the language evolution debate kicked off by Chomsky’s 2002 paper was that it became abundantly clear to everyone in the field that, as Jackendoff put it, one’s theory of language evolution depended on one’s theory of language. And even though Chomsky’s contributions set the agenda for linguistics and cognitive science for the latter part of the twentieth century, many researchers rejected the way that that paper attempted to rein in all the evidence and set the crucial questions for language evolution in the coming century. There’s no doubt that Hauser’s and Fitch’s experimental work will be central to the ongoing language evolution dialogue, but the specifics of the Chomskyan framework may not last as long.
In some ways, it bodes well for the study of language evolution that it can’t yet be compressed into a neat framework. It has always been a quirky field, and it retains much of its oddness. For example, the energetic back-and-forth between Pinker, Jackendoff, Chomsky, Hauser, and Fitch belies the fact that all five subscribe to a basic model in which language is somehow generated from the human brain.
Lieberman, on the other hand, is antigenerativist, and yet both he and Pinker agree on a first and fundamental principle—that you have to start with evolution in order to really get at the true nature of language. Jackendoff, who has been a Chomskyan linguist from the very start of the Chomskyan era, now proposes that formal grammars should be constructed so that they are consistent with an exploration of language evolution.
Within language evolution, computational modeling has been an enormous hit. In fact, Simon Kirby’s success with modeling has led him back to an interesting place. Now he’s trying to run generations of language learning through the minds of real people. He recently conducted a pilot study where he put individuals in a room and presented them with a small-world, talking-heads-style experiment. The subjects looked at a screen that contained a number of objects that were distinguished along a few dimensions, like color, shape, and movement. Across the bottom of the screen ran a series of invented words, an “alien” language that described what was pictured on the screen. The subjects were asked to try to learn the alien language. They were then tested on a series of pictures, which included some they hadn’t seen before (hardly any of the participants noticed this fact). Inevitably the subjects did not feed back only the language elements that they had been given. There were mistakes, modifications, and elaborations.
The study was intergenerational, because Kirby ran the subjects one after the other, and each time the alien language was, in fact, the answers the previous subject gave to the test pictures. Except for the initial random language given to the first subject, there was no alien language, only the contributions of each individual, which were culturally transmitted from generation to generation. Each subject in the experiment believed that he was simply giving back what he had learned, but instead the language was evolving. “It’s the same as the modeling,” Kirby explained, “in that it gets easier to speak the language with each generation.” He had originally thought that speakers might generate different elements to mark each of the features and then combine them in a precise kind of way. But that’s not how they did it. “People take whatever elements of structure they are given,” said Kirby, “and they go with it.”
Kirby’s first foray into modeling language evolution with human agents bears out what his digital models have predicted. “Structure organically emerges in the alien language, and it does it in a cumulative way. No single individual has created structured language, but it emerges after several generations from the accretion of lots of individuals’ contributions.” Darwin alluded to the emergent properties of language when he wrote in The Descent of Man that language is a cultural invention, though not a conscious one. As he and others have put it, the appearance of design does not necessitate the work of a designer. Kirby said, “It is real cultural evolution, steered by the biology of our experimental participants, but with an evolutionary dynamic and adaptive logic of its own. Features of the evolving languages in our experiments are there for their own selfish reasons (they are better at surviving to the next generation), not because of our desire to invent them.”
Luc Steels’s work heads in ever more creative directions. Steels, Vittorio Loreto (a physicist at the Università di Roma), and other colleagues are investigating ways to integrate what is known about the dynamics of semiotic systems with technology. The researchers are intrigued by the way that Web sites such as del.icio.us and flickr.com enable users to tag online resources, sharing commentary and other data with users. “Tagging sites glue online social communities by pushing thousands of people to take part in a collective effort to attach metadata,” said Loreto. With these sites, the popularity of a tag will typically begin to spread slowly; however, there is a phenomenon where one tag may suddenly become significantly more popular than all the rest. Steels and Loreto’s new experiments with autonomous agents engaged in language games (such as Steels’s “talking heads”) are showing that in the same way that widespread agreement about a tag may suddenly emerge in a social networking site, there can also be dramatic transitions in a network of digital agents, where a shared set of conventions suddenly replaces a phase of chaotic disagreement. The dynamics of meaning can help explain a similar phenomenon in human communication—how large populations of speakers suddenly converge on the use of a new word or grammatical construct.
This has obvious implication for stages of language evolution, where a new level of complexity replaces a previous level without any conscious agreement by protospeakers. Loreto and his colleagues suggest some interesting ways to exploit semiotic dynamics. For example, scientists could deploy groups of robots with such capabilities in situations where contact with humans is unreliable or impossible. Such robots might explore distant planets or deep seas, creating a way to communicate about, and respond to, events that were completely unforeseeable by their human programmers.
The involvement of a physicist like Loreto in a project connected to language evolution is a striking sign of just how many tentacles this problem has. Another language evolution researcher with a surprising background is Ramon Ferrer i Cancho. He is a former computer scientist who now works in the Department de Física Fonamental at Universidad de Barcelona. Ferrer i Cancho uses Zipf’s law to model language, exploring the trade-offs between speakers and hearers during communication.
Speakers must make an effort in order to be understood. For a speaker to be as clear as possible and avoid ambiguous meanings, greater effort is required. Listeners, on the other hand, must make an effort to interpret the correct meaning of an utterance, and they must work harder to decipher the intent of a speaker who has devoted less effort to clarity. Accordingly, Ferrer i Cancho’s models explore what happens when there are small shifts in the balance between the effort of the speaker and the hearer. In fact, a tiny change in the balance between the two can dramatically alter the properties of a communication system. Says Ferrer i Cancho, it’s possible that similarly small changes may underlie a dramatic shift from a communication system with a simple vocabulary made up of a few precise words to a larger vocabulary with varying levels of semantic precision.
The history of animal language research has been a turbulent one, but that may also be changing. Of language evolution conferences, Heidi Lyn said, “If I go to talks by some of the more established people, it tends to be either that they don’t mention the ape language research at all or they dismiss it. And there are people who consistently stand up and get things wrong. For example, an older
linguist at the Harvard language evolution conference in 2002 who was asked about Kanzi dismissed him. ‘Kanzi’s an aberration,’ he said. ‘He is the only example that we’ve ever seen of this.’” At the same conference, Herb Terrace stood up and asked Lyn if Kanzi was trained with food rewards. Lyn explained that they didn’t do this, yet Terrace persisted with that line of questioning. “It’s different with scholars my age or younger,” Lyn observed. The next generation gives a lot more credence to ape language research, and to work like Sue Savage-Rumbaugh’s. “They are willing to look at the data,” said Lyn. “It’s not just a matter of age. It’s the difference between people who lived through the Terrace criticism and the people who didn’t.”
For more than two decades Savage-Rumbaugh herself has been working closely with scholars from a language research program in Atlanta to apply the picture keyboards and other techniques she has used for communicating with the bonobos to communication with mentally retarded individuals whose levels of language skills have reached only those of small children. They have had great success with some individuals, equipping them with an ability to connect with other human beings that they wouldn’t have otherwise had.13
Other applications of language evolution research are completely futuristic but, at the same time, surprisingly practical. Philip Lieberman’s experiments on Everest not only illuminate the path that language evolution took but are serving as a model for NASA to monitor the well-being of astronauts on their way to Mars. The brain damage that Everest climbers suffer when they experience oxygen deprivation is similar to the kind of damage that a Mars-bound astronaut would incur from exposure to cosmic rays. If scientists back on Earth are able to detect subtle or profound neural damage in astronauts simply by listening to how they pronounce certain vowels and consonants, they’ll be able to react, and, it is hoped, treat them accordingly. This same project is also promising to improve the early diagnosis of Parkinson’s disease, not to mention help the mountain climbers of the world.
The way that evolutionary research is redefining language has social consequences as well. Lieberman argues that if language were a true instinct, if it simply flowed from every single one of us regardless of the environment into which we were born, then our governments would have very little responsibility to promote its expression. Because language is a skill, and one that is closely connected to thinking, he says, it is improved by practice and training and environments that are conducive to learning. This creates a civic responsibility to help all students hone their language skills.
At the Evolution of Language conference in Rome in 2006, Tecumseh Fitch listed the many ways in which the field had made progress since the 1866 ban on the subject. He started by noting that for the first time at the language evolution meeting, no one had mentioned the ban.
16. The future of language and evolution
Five years after Pinker and Bloom wrote about the evolution of the eye and its lessons for language evolution, Dan-Eric Nilsson and Susanne Pelger of Lund University in Sweden published a paper called “A Pessimistic Estimate of the Time Required for an Eye to Evolve.” Nilsson and Pelger digitally modeled the trajectory of the eye, beginning with a flat light-sensitive patch of cells—the kind of simple eye that we know some creatures have—and inflated it over time into a fully functioning mammalian eye.1
The scientists worked out a sequence of very small changes that had to occur if the light-detecting cells were to evolve into the separate specialized parts that interact with one another in an eye. For their model to be realistic, each small evolutionary step had to confer some survival advantage and therefore improvement in vision. Even though the changes were extremely tiny (no more than 1 percent change at any one time), each slightly modified eye was able to detect more and more spatial information. As the title of the paper suggests, Nilsson and Pelger erred on the side of pessimism, always assuming that it would take more generations for the eye to evolve rather than fewer. Given this, they calculated that it would take about 1,829 separate evolutionary steps for the flat-patch eye to evolve into a stereo-vision globe. That amounts to less than 364,000 years, not long at all from an evolutionary perspective.
We know from the fossil record that animals with modern eyes lived as early as the Cambrian period, 550 million years ago, which means there has been time for eyes to evolve more than fifteen hundred times since then. As perfect and wondrously complicated as our eyes seem to us, they are not irreducibly perfect from an evolutionary perspective.
To extend Pinker and Bloom’s analogy to language: this means that abilities and organs that seem wildly complicated from our perspective may be able to come together relatively rapidly as functioning, complex wholes. In addition to this biological potential, we know from the work of people like Deacon, Kirby, and Christiansen that language itself may also evolve and that linguistic evolution occurs even more rapidly than biological evolution. Language may have appeared very recently in the human lineage, but that doesn’t mean it was the product of a single, crucial event. No one mutation of genes or social order caused language to erupt from the mouths of our ancestors.
Even if researchers can’t pinpoint every evolutionary event that led to the language we have today, and even though we don’t know exactly what all the bends in the historical road looked like, the principles for further illuminating the path of language evolution are now self-evident. Fundamentally, the appearance of design in biology and in language can be taken as a sign of evolution, not of a designer. Additionally, where complex design does exist, it makes sense not to treat the whole as a monolith that simply developed from nothing to something in one or two quick steps. Finally, the most likely scenario is that both evolutionary novelty and derivation played a significant role in the evolution of a phenomenon as complex as language.
What does it mean that we are getting closer to the answer of how language evolved? The implications are as diverse and varied as the story of evolution itself. First, from a research perspective, it means that good data lead to better data, and there is still a great deal of data to be gathered before the big picture can be filled out. “People have been arguing about Neanderthal speech for the last thirty-five years and whether chimp sign language is really language,” said Tecumseh Fitch, “yet nobody even thought to ask what chimps do when they vocalize. We still don’t know—nobody’s put a chimp in an X-ray setup and watched it vocalize. It’s amazing how much data is out there that hasn’t been collected, like taping birdsong and whale song and doing linguistic analysis of that. We could apply this huge theoretical apparatus that phonologists have developed to birdsong. It’s not even that hard, and it’s an obvious thing to do.” Fitch added, “What amazes me coming into this field is how many things you can answer that no one even thought to look at.”
One of the biggest questions yet to be answered is posed by Ray Jackendoff: How do neurons do it? Magnetic resonance imaging and other ways of seeing the brain in action have taught us a lot about how our brains function. Overall, imaging has shown that for many higher-level activities, like language, neural activity is distributed across the brain. There are no specific areas that light up for language and language alone. Still, there’s no doubt that scientists fifty years from now will find the wonders of our neuroscience to be fairly crude. Although we can now map the brain as it works, we still have no actual idea how it works. How do the neurons do what they do? How do they process, store, and produce language? There is no predetermined meaning inside our heads. Neurons don’t contain symbols, but mainly pass on (or don’t pass on) activation signals to one another. So how can the patterned flare of electrical charge across our brains mean that we recognize the word “cat,” even when it is spoken by one hundred different speakers with their one hundred unique voices? How can we tell the difference between a p and a b when there is no tiny prototype of these sounds deposited in our neurons?
“We know we can’t think of the brain as a digital computer anymore,” said Jackendoff. “It�
��s sort of a parallel, semi-analog computer. But how does it do these digital things?” Discovering how neurons work should allow us to determine once and for all which of these frameworks for analysis—from the prototypical p to the syntax of English—are real and which are mirages.
It’s clear by now that the problem of language evolution is completely intractable when you approach it from the perspective of a single discipline. For all the salient questions to be answered, the multidisciplinary nature of the field will have to become even more so. So far, it has taken years for individuals in different departments to start talking, to develop research questions that make sense for more than one narrow line of inquiry, and to start to understand one another’s points of view. The field of language evolution needs students who can synthesize information from neuroscience, psychology, computer modeling, genetics, and linguistics. The more this happens, the richer and wider the field will become, instead of devolving around one or two theoretical issues.
The First Word: The Search for the Origins of Language Page 32