The Return of the Phoneme Inventories

Right, I already referred to Atkinson's paper in a previous post, and much of the work he's presented is essentially part of a potential PhD project I'm hoping to do. Much of this stems back to last summer, where I mentioned how the phoneme inventory size correlates with certain demographic features, such as population size and population density. Using the the UPSID data I generated a generalised additive model to demonstrate how area and population size interact in determining the phoneme inventory size:

Interestingly, Atkinson seems to derive much of his thinking, at least in his choice of demographic variables, from work into the transmission of cultural artefacts (see here and here). For me, there are clear uses for these demographic models in testing hypotheses for linguistic transmission and change, as I see language as a cultural product. It appears Atkinson reached the same conclusion. Where we depart, however, is in our overall explanations of the data. My major problem with the claim is theoretical: he hasn't ruled out other historical-evolutionary explanations for these patterns.

Before we get into the bulk of my criticism, I'll provide a very brief overview of the paper.

N.B. In all honesty there are numerous other avenues we could explore with this paper. For instance, there are numerous ways to measure the capacity of a language's phonological resource (Ke, 2006), with there being plenty of alternative positions to phonemes: distinctive features (Chomsky & Halle, 1968), syllables (Levelt & Wheeldon, 1994), exemplars (Johnson, 2007) and rich, high-dimensional memory products (Port, 2007). Furthermore, there is a case to be made for phonemes being an emergent aspect of a speech community, rather than having explicit psychological forms (Port, 2010). Another point concerns the reliability of the data collection: for each language, or dialect, the study is reliant on the choices of potentially one researcher, at a very specific point in time, and with only a finite amount of resources.

Phonemic Diversity and Geographic Distance

Atkinson's central point is that the typological patterns observed for the world's phoneme inventories hints at a serial founder effect expansion from Africa. What does this mean? Well, following on from work by Hay & Bauer (2007), suggesting the phoneme inventory size positively correlates with population size, Atkinson shows that geographic distance from Africa is a good predictor of phoneme inventory size: that is, the further you get from Africa the smaller the inventory. As mentioned, the inference made is that correlations of geographic distance and phoneme diversity are indicative of a serial founder effect: where variation is lost through repeated migrations from an original source population. Proponents of the out of Africa model often appeal to serial founder effects as an explanatory scenario for human genetic differentiation, which increases in accordance with geographic distance (see isolation by distance). It's essentially a special form of repeated population bottlenecks, resulting in a clinal distribution of traits via geography. Such is the theoretical basis of Atkinson's paper:

The loss of phonemic diversity following a population bottleneck is predicted under a range of cultural transmission models... This effect is expected to operate on top of any immediate reduction of phonemic diversity due to sampling only part of the dialect variation from the parent population.

To test his claim, Atkinson compares the phoneme inventories of language's from regions throughout the world, and discovers those in Africa tend to have more diversity than those outside of Africa. He then uses these findings to support a parallel history of genetic and linguistic diversity, where modern human languages follow a single origin model. In summary, distance from the best-fit origin in Africa and population size explain approximately 30% of global phonemic diversity, although Atikinson is quick to note that "other sociolinguistic processes... and more recent population movements clearly also play a role". Below is a boxplot giving the distribution of diversity scores according to region:

There is a lot more to the paper, and a discussion of the stats certainly warrants an entire post, but for now I'm going to keep the emphasis on theoretical thrust of Atkinson's arguments, namely:

Is the serial founder effect a plausible model?

There are several assumptions made in the paper that I've already considered investigating in my own work. Most notable of these assumptions is that the serial founder effect model is the only explanation available. As in population genetics, where numerous papers have supposedly verified the out of Africa model, Atkinson doesn't really test any other competing hypotheses. It therefore makes it hard for me to accept he's really shown that geographic distance from Africa is concomitant with a series of population bottlenecks for phoneme inventories. Indeed, with bolstered support for Neanderthal admixture in some human populations, it is becoming increasingly likely that the serial founder effect model is unlikely to hold true in relation to genetic diversity, as John Hawks recently noted:

The "southern African origins" conclusion of the paper comes out of a simple analysis that assumes that the best-fit maximum for genetic diversity (as assessed by linkage) is the most likely point of origin of the population. That would be true if the African population emerged by a series of founder effects from a single small ancestral population -- the "serial founder effect" model that I have criticized here before. But of course in 2011, we know that model is false, because it is predicated on a lack of ancient mixture with Neandertals or other populations. If the serial founder model can't work outside Africa, it certainly can't work inside Africa, where populations were larger and regionally diversified during by the beginning of the Late Pleistocene. Without that false assumption, the "southern African origin" evaporates. The primary observation, a cline of linkage disequilibrium within sub-Saharan Africa, can be explained with reference to mixture of populations without assuming an origin and expansion from one geographic location [my emphasis].

Similar problems exist for phoneme inventories. I'm not saying that Atkinson thinks that serial founder effects are solely responsible for the observed patterning (as noted in the quote above)... But I do think the model's flawed on several theoretical grounds. Specifically, Atkinson hasn't ruled out the possibility of selection as having shaped these languages according to different socio-cultural niches.

Alternative Proposal: The Linguistic Niche Hypothesis

Much of the research I'm doing builds upon a paper by Lupyan & Dale (2010) where they put forward the Linguistic Niche Hypothesis: "that language structures are subjected to different evolutionary pressures in different social environments. Just as biological organisms are shaped by ecological niches, language structures appear to adapt to the environment (niche) in which they are being learned and used".

Building upon previous theoretical work (e.g. Wray & Grace, 2007), the authors propose that exoteric niches (large number of speakers, large area, many linguistic neighbours) and esoteric niches (small number of speakers, small area, few linguistics neighbours) will produce languages adapted to these respective socio-demographic conditions. One of these pressures is learnability: that languages adapt to become increasingly learnable for their speaker-hearer community. So, for example, a language that's increasingly being learned by adults will be under a pressure to become tailored for adult-learners. Lupyan & Dale's claim is that demography can amplify these learnability pressures: "With the increased geographic spread and an increased speaker population, a language is more likely to be subjected to learnability biases and limitations of adult learners". To test their claim the structural properties (e.g. case systems and the complexity of conjugations) of more than 2000 languages were compared with a language's speaker population, its geographic spread, and the number of linguistic neighbours. They found that exoteric niches tend to produce languages with simpler inflectional morphology, and an increased reliance on lexical strategies to encode certain linguistic features (e.g. evidentiality), than those in esoteric niches, where languages are more morphologically rich. The salient point: certain differences in structural complexity partially pattern with the underlying demographic conditions of a speaker community.

Lupyan & Dale's findings beg the question of whether large-scale changes in the linguistic niche also influence the phoneme inventory through the distribution of phonetic variation and a shifting of the selection pressures. In this context, we should also remember the obvious fact that phonemes are part of a wider linguistic environment: they are not easily divorced from the constructions being used and learned by individuals within a speech community. If phoneme inventories, as Atkinson argues, are largely a product of demographic dynamics, then we should not expect to see any adaptive patterning between phonemic diversity and other linguistic levels of organisation (e.g. morpho-syntactic features). My argument is that there is evidence for language adapting to its socio-cultural niche -- and the phoneme inventory size is a big part of the story.

With this in mind, my proposal is that the size of a language's phoneme inventory is contingent on numerous factors, and cannot be explained via the serial founder effect model. In fact, we need to consider all of these factors together, and how different demographic variables feed into each other in defining the linguistic niche, before putting forward one general historical model for the data. It's all pretty sporadic and vague at the moment, largely because I'm going to wait until I've got some results to show before giving up my hypotheses to the web 😉 . For more good blog coverage on the topic, see:


Atkinson QD (2011). Phonemic diversity supports a serial founder effect model of language expansion from Africa. Science (New York, N.Y.), 332 (6027), 346-9 PMID: 21493858