Origins of human language in Southern Africa?

[Bill Croft]

I have just read Atkinson's article, including the supplementary 
materials - the supplementary materials for a Science or Nature 
article are essential reading, because the actual article is too 
short to be more than just a long abstract for the real paper.

A number of linguists, here on Funknet just now but also in the 
comments section of the NY Times article by Nicholas Wade, have 
pointed out languages that are at a substantial distance from Africa 
but have large phoneme inventories as evidence against the 
hypothesis. It is worth putting this in context of what the paper 
actually says:

"We expect the number of phonemes present in a language today to reflect past
phoneme inventory size, combined with complex group dynamic processes driving
relative rates of merging, splitting and borrowing of phonemes. Many
factors are likely to influence the rates at which these processes occur, and
their relative rates will determine the trajectory of phonemic 
diversity in a language
through time." (supplementary materials, p. 8)

"It is worth noting that fitting a serial founder effect model to 
phoneme inventory data
describes an inherently stochastic (probabilistic) process and does 
not entail that
phonemic diversity is entirely determined by population size via a 
serial founder
effect. Distance from the best-fit origin in Africa and population 
size are shown to be
significant predictors of phonemic diversity, explaining 
approximately 30% of global
variation, but other sociolinguistic processes and more recent population
movements clearly also play a role. Neither of these factors are expected to
systematically bias results to produce the observed global cline in 
phonemic diversity."
(supplementary materials, p. 11)

"In a general linear model,
language family as a factor explains 50% of the variance in phonemic diversity
(adjusted r-squared=0.502, df=49, p<0.001) and 48% of the variance in phonemic
diversity across the largest 10 families (adjusted r-squared=0.476, 
df=9, p<0.001).
This level of conservation within major language families indicates that robust
statistical patterns in global phonemic diversity can persist for 
many millennia and
could plausibly reflect a time scale on the order of the African 
exodus." (supplementary materials, p. 7)

In other words, Atkinson argues that distance from Africa is only one 
of many factors accounting for phoneme inventory size, and explains 
only part of the variance in phoneme inventory size. The conclusion 
of the main article states that distance from Africa explains 19% of 
the variation in phonemic diversity (p. 348). Population size (also 
documented by Hay & Bauer, Language 2007) explains another part, and 
language family explains yet another, quite large, part of variation 
in phoneme inventory size. These statistical models are examples of 
the competing motivation models that many functionalists argue for. 
The point of the article is that there is still a signal of an 
African phylogenetic origin of modern human language in the 
geographical distribution of this typological trait.

Atkinson offers an explanation based on the small size of founder 
populations leading to the reduction of phoneme inventories, in 
turned based on the correlation between population size and phoneme 
inventory. So the explanation is in turn based on whatever 
explanation is offered for the latter correlation. That is the most 
interesting and most problematic part of the whole story, in my 
opinion. Atkinson presents an implausible explanation on p. 3 of the 
supplementary materials, but the more extended discussion on pp. 8-10 
is better. Hay and Bauer do not endorse any specific explanation, but 
suggest that in small social networks context allows more ambiguity 
to exist (hence fewer phonemes are necessary), and exposure to a 
larger number of interlocutors may enhance the creation and 
maintenance of a larger number of phonemic distinctions, citing 
respectively social network theories and frequency- and 
exemplar-based theories of phonology.

Atkinson's conclusion seems reasonable to me. The statistical signal 
seems robust, even if we have difficulty in explaining it. I 
encourage you to read the article and supplementary materials and 
judge for yourself.

Bill