Atkinson on phoneme inventories in Science

Søren Wichmann wichmann at EVA.MPG.DE
Wed Jun 1 12:13:32 UTC 2011

Dear Typologists,

prompted by the Atkinson paper colleagues and I set out to check the 
out-of-Africa correlation on a different dataset. We also checked the 
correlation between phoneme inventory and population sizes. In addition, 
we checked a third correlation which is also relevant, although it is 
neither mentioned by Hay & Bauer nor by Atkinson, namely the inverse 
relation between average word length and phoneme inventory sizes claimed 
to hold by Nettle.

The data we used were more than 3000 word lists of the ASJP project: We first made sure 
that the number of different segments contained in a 40-item word list 
can meaningfully be used as a proxy for segment inventory sizes.

All the correlations proposed were supported, including the reduction in 
segment inventory sizes as the distance from Africa increases 
(controlling for population size, which also decreases).

To account for the out-of-Africa correlation we prefer a different 
explanation which does not rely on population sizes at all, however. 
There are two reasons for this. One is that the relation between 
population size and segment inventory size seems to actually be a 
secondary effect of a relationship between population size and mean word 
length (shorter words in larger populations). So if population size is 
the ultimate driving force mean word length should also correlate with 
the distance from Africa, but it doesn't. Another reason is that 
populations in prehistory would mostly have been small, whether or not 
they went through bottlenecks. We believe that a viable different 
explanation could be successive events of language contact whose effects 
would accumulate with the distance from Africa. But that doesn't detract 
from the value of Atkinson's basic observation.

We intend to submit the paper to a special issue of Advances in Complex 
Systems, and invite your comments. I've attached the paper, but in case 
the attachment gets stripped you can also get it here:

The data are here:

If you would like to have them in Excel you can download them from here:


-------------- next part --------------
A non-text attachment was scrubbed...
Name: Rama-ACS.pdf
Type: application/pdf
Size: 553532 bytes
Desc: not available
URL: <>

More information about the Lingtyp mailing list