When we explore the greatest factors influencing biodiversity, we often see factors like ecological niches, geographical barriers to migration, and various ecological and environmental factors come to mind. As researchers prescribe ecological means to explain linguistic diversity, drawing the parallels biodiversity shares with linguistic diversity, similar theories to linguistic diversification also arise. Mountains and large water bodies can pose as obstacles to migration and interaction between two populations, leading to allopatric diversification (and in ecology, allopatric speciation).

But in the attempt to explain linguistic diversity, we see one factor that is not covered by the factors explaining biodiversity — that of ecological risk. In fact, this factor has been implicated as a major driver of linguistic diversity in studies like the one we covered previously here, which tries to explain the latitudinal gradient of linguistic diversity. Additionally, this factor has been used in assessing global patterns of linguistic diversity as well as the linguistic diversity of certain regions. By extension, it could also be used to attempt to explain the linguistic diversity of the Caucasus, driven by the practice of endogamy. So given its prominence, what exactly is the ecological risk hypothesis, and how did this arise?

The ecological risk hypothesis was first coined and pioneered by the researcher Daniel Nettle in the mid to late 1990s. It was developed not as a biological concept, but more rather, an economic one developed from observations made in the decades leading up to Nettle’s studies. It is built upon the observations of economic units in agrarian societies, where the ecology of the area imposes many restrictions on the lives of these societies. It was initially tested to evaluate the impact of this ecological risk on the diversity of West African languages, a region many would associate with being heavy on subsistence agriculture.

With such ecological constraints, subsistence economies could have great fluctuations in crop yield from a given area of land. To ensure some sort of “insurance” or a “buffer” against the risks subsistence agriculture carries, such as a year of bad harvest leading to food insecurity for the affected people, the basic economic units in these societies, which are usually extended families or compounds, would have to form and maintain a network of social exchange. In other words, how communities adapt to this type of risk, according to the interactions between these societies and their surrounding ecology, would be along some sort of social kind.

The main direct effect of this social exchange network would be a gift economy — where households exchange their local products with one another. According to economists, this minimises the fluctuations or variability seen in subsistence agriculture.

However, this social exchange network carries an immense cost to build and maintain. If some connections involve some sort of travel to maintain, this is usually done by foot, which is expensive in both time and energy. In addition to the amount of resources produced and allocated for such maintenance, this would mean that having large resource surpluses at the end of a year would be rare. As such, in many cases, social networks are not incentivised to be larger than they need to be.

But how does this relate to linguistic diversity as a whole? Well, languages may be spread through social networks, which can occur extensively as a means to reduce this ecological risk, to maintain a reliable level of food supply. Thus, for societies with a greater ecological risk, communities tend to be large, and vice versa, to facilitate exchange of general commodities like grain. This exchange is called generalised exchange in Nettle’s publications. As he noted, if this exchange is generalised enough such that people willingly share a common social identity, the languages spoken by these people would converge to a common language. Thus, within these communities, the languages spoken would tend to be more homogeneous, meaning that they would tend to converse in only one language.

Nettle also postulated that communities will occupy an area proportional to the ecological risk they face, and the population of these communities will increase proportionally to the ecological risk they face. His third hypothesis was that more difficult terrain would decrease the geographical area occupied by language groups. From the first hypothesis, this could imply that in a given area, if there is less ecological risk, there could be a greater density of languages spoken in that area compared to an area of greater ecological risk. This means that the diversity of languages in a given area could be inversely proportional to the ecological risk of the area.

So, how is ecological risk measured?

As discussed, ecological risk is affected by a number of environmental variables, but the main factors assessed by Nettle in his 1996 study included climatic factors, measures of relief (or elevation), vegetation type, and duration of growing seasons. The selection of the two most significant factors assessed was done by a method called ‘principal component analysis’. This is a technique used to reduce the number of dimensions in the data (i.e. weed out the more ‘unnecessary’ factors in the analysis), but preserving the meaningful properties of these data. This makes complex data substantially easier to interpret.

From this analysis, Nettle found that growing season duration and average altitude to be the most significant factors affecting ecological risk. These factors were then used to test the hypotheses he laid out, in the form of linear regression. He found that the duration of growing seasons was inversely proportional to the area occupied by a linguistic group and number of speakers in those respective groups (p<0.001 for both dependent variables). Average altitude was only significantly negatively correlated with number of speakers in a linguistic group (p<0.05) when assessed alone on its own. It seemed to suggest that the duration of growing seasons played a dominant role in influencing ecological risk.

There are certain outliers noted by Nettle, however. For example, he noted some regions of the Sahel region, that is the savannah just south of the Sahara, were more more linguistically diverse than expected from his analysis. This included the various Chadic peoples, amongst whom the Buduma people rely on fishing in Lake Chad to reduce the ecological risk imposed by a notable seasonal variation which would otherwise normally drastically affect more agrarian or pastoral cultures.

But how robust is this analysis? Nettle noted two aspects he thought would violate statistical assumptions made in the regression analysis. Both of these pertain to the violation of the assumption of independence of the data points. This means that Nettles has to demonstrate that an observation of a certain data point has no impact on the observation of the other data points, to ensure that the results and interpretations obtained were not muddied with statistical artefacts as a result of these violations.

The first was to show that large languages (in both geographical size and speaking population) were independent samples of ethnolinguistic group size, and not just having multiple samples taken from the same linguistic group. This was most pronounced in the northern regions in the area assessed, where one would find languages such as Bedouin Arabic, the Songhay languages, and Hausa. To demonstrate independence of these data points, Nettle coalesced these groups into supersectors, which used mean values for environmental variables and geographical area size, and the median values for number of speakers. Using these supersector data, Nettle ran the linear regressions again, and reported similar significant results between growing season durations and average altitude, and geographical area size and number of speakers.

Next, Nettle had to demonstrate that the association between ecological risk and linguistic group size was independent from underlying cultural practices which would prefer or facilitate certain group sizes. Some people groups may be culturally dependent or related to one another, which could potentially stem from an evolutionary process, as these cultural practices could have once been adopted from a common ancestor of these cultures. To show this independence, Nettle had to demonstrate that linguistic groups of a large geographical or demographic size had to evolve independently on a significant number of occasions “as a response to ecological risk”.

To show this, Nettle used a ranked correlation analysis between the duration of growing seasons and the median number of speakers of the languages in a certain language family or a large linguistic affiliation. This assesses how similar the two rankings are, and to show that these two rankings are independent of each other, the rank correlation would need to be significant at a 95% confidence level. Basically p<0.05. The main linguistic affiliations he tested and reported were the Niger-Congo languages and the Chadic Afro-Asiatic languages, assessed separately, within which he found significant rank correlations between these languages and growing season lengths. There were also languages from the Nilo-Saharan language family spoken in the region, although the test statistic was not reported in the main manuscript of the paper. Nevertheless, Nettle concluded that the observations made were not just evolutionary artefacts in linguistic history.

The findings of the 1996 paper were further built upon in Nettle’s 1998 paper, in which he tested the hypothesis that mean growing season duration is associated with language diversity of a country with a certain area, and mean growing season is also associated with language diversity of a country with a certain population. This time though, instead of assessing the languages of West Africa, Nettle analysed this on a global scale, but only included countries of over 50000 square kilometres in size, to reduce the likelihood of languages being spoken across national borders, which could affect language diversities of smaller countries.

One of the most notable regions where the growing season hypothesis does not seem to exert a major influence on linguistic diversity is the islands of the Pacific. As noted in the 2012 paper by Gavin and Sibanda, this is likely due to the differences in mean growing seasons in the Pacific islands compared to West Africa and perhaps, a global scale. Crops could potentially grow year-round in a majority of the Pacific islands studied, meaning that there is considerably less variation in mean growing seasons in the Pacific islands. This lessens the impact of growing seasons as a determinant of ecological risk. Instead, other environmental determinants such as soil nutrients would exert a greater impact on ecological risk, even though these factors together still do not play as major a role in influencing Pacific language diversity compared to more biogeographical factors such as island area and isolation.

Nettle suggested that the ecological risk hypothesis could also have applied further into antiquity, when human societies were more dominated by hunter-gatherer societies. In hunter-gatherer societies, ecological risk would be minimised by moving from depleted areas to areas with more resources to forage, thus large communities were not necessary. The more sedentary farming societies were less able to move about to alternative resources or territories, and thus would have to form larger communities for generalised exchange. From this expansion in communities, it would have brought about greater linguistic homogeneity within these communities. That means that one large farming community would be speaking one language. This could imply that there would have been substantially more languages spoken globally before the agricultural revolution, as hunter-gatherer societies, being in small and less socially connected groups, would have been greater in number, with each group speaking what could be a distinct language.

Overall, the ecological risk hypothesis offers an interesting perspective in explaining the patterns of language diversity in certain regions, and even on a global scale. However, the interactions between environmental factors and ecological risk are a rather complex one, as it involves a people group’s interactions with the environment, the ecology of the environment, and other people groups. While principal component analysis pointed towards duration of growing seasons to be the primary driver of ecological risk, this process alone does not explain ecological risk in its entirety. Nettle also has a 1999 book called Linguistic Diversity, which covers additional processes that try to explain the pattern of linguistic diversity we see today. I hope you have learned a thing or two from today’s essay on a rather intriguing aspect of language diversity, and we will continue to look into publications covering the biogeography of language diversity when I chance upon them or when recommended.

Further reading

Gavin, M.C. and Sibanda, N. (2012) The island biogeography of languages, Global Ecology and Biogeography, 21(10), pp958-967. https://doi.org/10.1111/j.1466-8238.2011.00744.x.

Nettle, D. (1996) Language Diversity in West Africa: An Ecological Approach, Journal of Anthropological Archaeology, 15(4), pp403-428. https://doi.org/10.1006/jaar.1996.0015.

Nettle, D. (1998) Explaining Global Patterns of Language Diversity, Journal of Anthropological Archaeology, 17(4), pp354-374. https://doi.org/10.1006/jaar.1998.0328.

Nettle, D., Linguistic Diversity (Oxford, 1999; online edn, Oxford Academic, 31 Oct 2023), https://doi.org/10.1093/oso/9780198238584.001.0001.