How languages spread
A big question for prehistory has been the role that agriculture has played in human dispersals. Agriculture is comparatively recent in human history. While humans have probably used food management techniques (such as fire-stick farming, reseeding, and the like) for tens of thousands of years, farming is only about 10,000 years old, and intensive agriculture is even younger. Moreover, while there are about 7,000 languages spoken in the world currently, most of them belong to just a few big language families. For example, the Austronesian family has spread from Madagascar to Taiwan to Hawaii, and contains over 1000 languages. Niger-Congo covers most of Sub-Saharan Africa and contains at least another 1000. Sino-Tibetan and Indo-European are smaller in terms of number of languages (around 400) but larger in terms of numbers of speakers, each with over 1 billion. Other families with 300-500 languages include Afro-Asiatic (including the Semitic family), Pama-Nyungan, and Trans-New Guinea. Together, over half the world’s languages belong to a handful of language families, while the remaining languages are distributed between another few hundred, much smaller, language families (with some containing just a single language).
Given that the majority of these large language families are spoken today by groups who get their food from farming, it is a logical assumption that farming technology facilitates (if not ’causes’ in the strict sense) language and population expansions. But in that case, the Pama-Nyungan family is an exception, since agriculture wasn’t used in Australia until the 18th century. Some have dealt with this by suggesting that speakers of Pama-Nyungan languages must have spread with initial colonisation of Australia, 55,000 years ago (our Hypothesis 4). But given that Pama-Nyungan languages are fairly closely related, this would make the processes of language change unparalleled elsewhere in the world. Our research showed clearly that the ‘original colonisation’ hypothesis simply does not fit plausible and flexible assumptions about how languages change. By refuting Hypothesis 4, our research shows that agricultural expansions are not the only trigger for language families to spread widely.
The researchers who have proposed a mid-Holocene (that is, 4,000-6,000 year old) expansion of Pama-Nyungan have posited a variety of triggers for the expansion. For example, some note that the approximate time of expansion correlates (at least approximately) with the inferred entry of dingos into Australia (see Pugach et al 2016 for one version of this theory). Others, such as Nicholas Evans and Rhys Jones, assume not a material advantage, but a cultural one. While most of these theories posit the origin of the Pama-Nyungan expansion in the North, around the Gulf of Carpentaria, many are not very precise about the time of dispersal.
Like other work, we note correlations with major changes in the climatic record. But correlation is not causation: just because two patterns appear related, it does not mean that one caused the other. For example, per capita chocolate consumption correlates with the number of serial killers in a country, but the amount of chocolate consumed in a country doesn’t make people murder, or vice versa. Likewise, just because two events happened around the same time, it doesn’t mean that one caused the other. In this case, however, our research provides evidence for interaction between language spread and environmental change.
First, an assumption of Hypotheses 2 and 3 in the literature on Australian languages is that people spread when the climate is favourable, and contract when times are hard. This makes intuitive sense: when times are good, there is more food available, which means more people, which means more people (and language) to spread. However, our results tell a different story. What we appear to see is faster movement when conditions are less good. This also makes sense. Worsening conditions place more stress on communities and their food resources, which mean they need to split and move so that everyone gets enough water and enough food to eat. Our dating of the parent language of Pama-Nyungan is consistent with a break-up as conditions worsen.
We also see this speed-up/slow-down in our modeling of speed of movement of groups since the Pama-Nyungan break-up. We found a difference in speed of movement when groups were near water (big rivers or the coast). In our model, groups moved twice as slow near water compared to elsewhere. In these resource-rich areas, more people can be supported. It’s good country, there’s no need to move elsewhere to find more food.
Languages and genes
Languages can tell us a lot about our past. Because they change regularly, we can use information from contemporary languages to work out who groups were talking to in the past (through loan words), where they lived, who they are related to, and where they’ve moved. We can do this even in the absense of a written record and in the absense of archaeological materials. We can answer many of the same questions with data from genetics. Language gives us fine-grained information at time-scales over the last few thousand years, whereas genetic information usually gives us information at much longer timescales.
An outstanding problem for human prehistory has been working out the ways we can combine data from linguistics and genetics (along with archaeology and anthropology) to understand human prehistory. It’s particularly interesting when different data gives different results. A recent paper by Anna-Sapfo Malaspinas and colleagues (paper here, report here) suggested that the ancestors of the Pama-Nyungans could have begun spreading across Australia as far in the past as 30,000 years ago; later than the original settlement of the continent, but much earlier than our results. Another recent genetics paper, based on Mitochondrial DNA (paper here, report here), also emphasized stability and continuity of Aboriginal settlement for more than 30,000 years. How then can we say that Pama-Nyungan is only 6,000 years old?
There are a couple of ways we might reconcile these two dates:
- Perhaps methodological issues have made our dates too young, or dates from genetics too old. We talk about some of the ways we tested our methods in the paper. We are confident in our dates. They are of the same general scale as other language families; they are internally consistent, and reflect plausible rates of change in vocabulary. We also tested for the type of problem that would make our dates seem artificially young, such as undetected borrowing. We note that the genetic samples from both papers are patchy, whereas we sampled almost every Pama-Nyungan language.
- Languages can spread independently of people. Therefore, perhaps the genetic dates reflect the spread of a particular group of people, but the languages we call Pama-Nyungan spread later, This is posslbe, and indeed likely in some parts of the country. We know that parts of Victoria, Queensland, and Western Australia were continuously inhabited from 20,000 years ago or more. We know of many cases where groups of people have shifted language. Indeed, the massive rates of language endangerment we see across the world are examples of extensive language shift. But that’s not the whole story for Pama-Nyungan, since we also have cases where the archaeological record shows recent resettlement, which correlates with the separation time of Pama-Nyungan subgroups in our tree. We discuss a number of these cases in the supplementary materials.