YOU can tell a lot from the way people speak. , for example, signals he is from New Zealand by saying “rite” instead of “rate”. Sometimes that is enough to confuse. A few years ago, during a talk Gray gave while visiting the University of Oxford, Richard Dawkins interrupted to ask what he meant by “evolutionary rite”. New Zealanders can only afford one vowel, Gray jokes.
As a trained biologist, Gray notes these differences with the same eagle-eyed curiosity that he has used to study the evolution of bird behaviour. “If you’re looking at courtship displays in birds and how their differences are produced by descent with modification, it doesn’t seem like a huge leap to think about languages in that way,” he says. Living among the Pacific Islands – a hotspot for language diversity – Gray just has to listen to the sounds around him to hear the way that languages can mutate, splinter and proliferate like separate species.
But whereas biologists have long tried to track differences between species to reconstruct our evolutionary history, most linguists thought that language change was simply too unpredictable to use it to dig deeply into our past with any kind of accuracy.
Advertisement
Gray and his colleague Quentin Atkinson, both at the University of Auckland, beg to differ. It is possible, they say, to excavate thousands of years of human history from the words we speak, if only we treat them like evolving creatures. The trick is to plunder the same tool kit that allows biologists to find sense in the messy genetic variations between species. Besides reconstructing the origins of languages, the linguistic changes they find can show us the way different groups migrated and moved, potentially settling a 200-year-old argument over the origins of European culture. The technique can even tell us about our ancestors’ day-to-day lives – what kind of marriage system they had, and even what type of gods they believed in.
“We can even use these ‘language trees’ to find out what kinds of gods our ancestors believed in”
Darwin’s inspiration
Comparing languages to living, breathing organisms certainly has a striking historical precedent. In his 1871 book The Descent of Man, Charles Darwin notes the connection explicitly: “The formation of different languages and of distinct species, and the proofs that both have been developed through a gradual process, are curiously parallel.” Some have suggested that Darwin’s model of speciation may have been influenced by the concept of language families put forward by the linguists of his day.
“Evolution doesn’t have to be about DNA,” says evolutionary linguist at the Max Planck Institute for Psycholinguistics in Nijmegen, the Netherlands. “Language has variation, replication and selection. So, as far as I’m concerned, language is something that evolves. That’s not a metaphor.”
The comparison doesn’t end there. In the same way that related organisms often have highly similar genes, words in different languages can reflect a shared origin from a common ancestor. Such words are called “cognates”: the English word “mother” is recognisably similar to mutter in German, madre in Spanish, mater in Latin and maatru in Sanskrit, for example, suggesting that all five evolved from the same ancient tongue. “They are instantiations of the same word,” says at Yale University. As the mutations amass, some words will drop out of use and be replaced by entirely new ones, meaning that over time the language builds a new dictionary that may be very different from its relatives.
By comparing these changes, you can draw up family trees showing the relationships between different languages, but problems arise when you then put an age on the different tongues. One approach, known as glottochronology, is based on the idea that words are replaced at a fixed rate. To calculate when two languages diverged, researchers simply counted how many cognates they share from a list of 200 universal concepts – such as “one”, “two”, “heart”, “mouth”, “dog”, “louse”, “sleep” and “sun” – and plug it into a formula. Essentially, the more cognates two languages share, the more recently they split.
“The basic idea isn’t too far-fetched,” says Bowern. “Words change over time. If we can work out how quickly they change, we can work out the common ancestor.” The trouble comes with assuming that all words change at the same rate. In practice, different words and languages evolve at very different speeds – so any timeline given by glottochronology is imprecise, particularly when you try to delve far back in time.
As a consequence, historical linguists also look for other clues to estimate a time frame. By comparing cognates across many languages you might be able to infer the first appearance of the word “wheel”, say, in their common ancestor. If the archaeological evidence gives you a date for the invention of the wheel, you should then be able to place a figure on the age of that particular language.
Unfortunately, language evolution has laid many tripwires in the way of such easy conclusions. One major problem is that you don’t know whether the word in question was coined for its current purpose, or if our ancestors had simply co-opted another, older term – the earliest form of wheel might have originally meant “anything that rolls”, for instance, predating the technology by centuries. So, for the most part, researchers adopted the slogan “linguists don’t do dates”.
Then Gray and Atkinson came along. Their brainwave was to feed linguistic data into software originally designed to trace genetic lineages. Treating cognate words like common genes, it creates a forest of possible family trees that each might fit the relationships between the different vocabularies. Importantly, unlike the methods of glottochronology, the rates of change between languages are allowed to vary – essentially, the software builds each tree with slightly different branch lengths. It then calculates the probability that each family tree fits the original data, according to known trends in language evolution, allowing the researchers to pick out the dates and relationships with the highest likelihood of being true.
Marauding horsemen
So that’s how it works in theory. In 2003, with this new tool in hand, Gray and Atkinson set to testing it out on a 200-year-old linguistic dispute. Indo-European is by far the largest language family in terms of both number of languages and number of speakers. It includes those spoken across Europe – such as English, Spanish, French and Russian – as well as many from South Asia, such as Bengali, Hindi and Urdu. These languages must have diverged from a single ancestor, known as proto-Indo-European: but who were these speakers, and when and where did they live?
By the time Gray and Atkinson tackled the problem, two hypotheses vied for dominance. One was that our first words took root with the expansion of farming out of Anatolia in modern-day Turkey around 8000 to 9000 years ago (see map). But the more popular idea was that it came 3000 years later, galloping out of modern-day Ukraine with the spread of nomadic horse-riders. This hypothesis was based, in part, on those attempts to match the origins of words with archaeological finds, using examples from different locations and languages to bolster the theory.
Yet when Gray and Atkinson fed vocabulary from Europe and Asia’s ancient and modern languages into their evolutionary software, it estimated that proto-Indo-European must have been spoken around 8000 to 10,000 years ago – a date that fitted perfectly with the Anatolian hypothesis ().
Unfortunately, that put them at odds with many colleagues, who thought they were repeating the mistakes of glottochronology. “The answers fall where the answers fall,” Gray says. “But life would have been a lot easier if we’d got the orthodox answer.” Linguists would walk out of his conference talks and one colleague even branded his findings “worse than creationism”. “I have the scars to show for many a long-fought battle,” Gray says.
“One colleague branded the technique ‘worse than creationism’ – but no one has yet proven it wrong”
But after 10 years, no one has yet proven them wrong, and Gray and Atkinson have managed to douse some of the criticisms with a more refined analysis. In a collaboration with Dunn, for example, they recently married their family trees with a geographic technique that was originally used to pin down the origin of virus outbreaks, based on the appearance of viral mutations at different locations. The resulting maps placed the origin over Anatolia, at the time predicted previously ().
Looking to test their method closer to home, Gray and Atkinson have also studied the Austronesian language family to understand the peopling of the Pacific Islands. This time the results matched the prevailing archaeological theory. They show that people first spread from Taiwan into the Philippines, then into Borneo and Papua New Guinea. From there they moved east through Micronesia, island hopping in a handful of distinct migrational pulses before finally reaching land as remote as Hawaii, Easter Island and New Zealand. All of this happened between nearly 4000 and 800 years ago ().
Slowly, other researchers are starting to take an interest in the computational approach. “In terms of chronology and geography, it’s better than the techniques that linguists themselves had been using,” says , a linguist at the Max Planck Institute for Evolutionary Anthropology in Leipzig, Germany, who specialises in cross-disciplinary methods.
As interest grows, some researchers are beginning to wonder whether the technique might help us write a dictionary of the words our ancestors spoke. Linguists used to do this by hand, working backwards from common trends in language evolution, but at the University of British Columbia in Vancouver, Canada, wondered whether a computer could do it automatically. To find out, he created an algorithm that works through the linguistic family tree constructed by Gray, using the same kinds of probabilistic analyses to predict the most likely sound shift at each stage. Earlier this year, his team set the software the task of reconstructing the proto-Austronesian languages. In most cases, the words matched the predictions of trained historical linguists – with the advantage that the software could reconstruct far more words in far less time ().
Atkinson is trying to delve even further back in time. The difficulty, which had deterred most other linguists, is that this requires finding cognates between languages as apparently unrelated as English and Japanese. His tactic is to look at the most frequently used words, which seem to evolve less quickly – perhaps because it is harder for new versions to sail against the tide of convention. Using his software to uncover hidden connections between these “ultra-conserved” terms, he found that Asian, Inuit and Indo-European languages may be traced to a single mother tongue that existed 15,000 years ago. But many are sceptical of using statistics to find cognates where no obvious similarity exists. Even Gray suspects that we are stuck at a “time barrier” of around 10,000 years.
Whatever the limitations, the techniques are now indispensable. “These methods let you ask a whole lot of questions that you haven’t been able to ask before,” says Dunn.
But the real worth of these family trees may go far beyond language. “You can take a language tree and use it as a proxy for human history,” says Gray. After all, cultures are likely to transmit more than just their words as they split and migrate. The idea is that once you pin the known practices of modern cultures to the family trees, you can use software to work backwards and infer how they might have evolved and been passed down over history.
One wife or two?
For example, at the Santa Fe Institute in New Mexico used the method to examine marriage systems in early Indo-European cultures. Currently, in modern Indo-European cultures there are two marriage systems: monogamy, seen throughout Europe and much of South Asia, and polygyny, where men can take more than one wife, which can be found in some Asian countries. So which came first? Fortunato’s suggest the early Anatolian farmers who spoke our first words were monogamous, with a later cultural split leading to the polygyny now found in some areas. The result is still up for debate, but the finding does have some support from the remains of families buried together. Further work may also tell us what caused the split – whether it accompanied some change in religious belief, for example.
Gray has even greater ambitions. He is now using the trees to explore questions about the kinds of gods these people believed in. One theory is that complex social structures with hierarchies of political power could not have developed without a belief in a moralising god or gods to keep cheats in check. By tracing the co-evolution of complex societies and religious beliefs backwards through the trees, Gray hopes to get a clearer picture of the connections between the two. The work has yet to be published but initial results suggest the moralising theory might be wrong, as the two do not appear to evolve hand in hand. Instead, religions can be taken up by societies whatever their social complexity.
This is just the start. With the linguistic framework in place we can expect many more questions about our past to be tackled this way. The rigorous statistical approach also makes it easy to correlate the results with the work of geneticists and archaeologists addressing the same questions. Historical linguistics has become a science of human prehistory. “History is hard,” says Gray. “You need to bring all the powerful methods you can to bear.”
