Proceedings of the Royal Society, London, Biological Sciences 267, 947-952. (2000)

The size distribution of language populations in New Guinea, representing over 15% of the world’s languages, is analysed, using models analogous to the resource division models of species abundance distribution in ecological communities. A model distribution of resource segments, reflecting population size, is created by repeated selection of an existing resource segment and its division into two. We found that any dependency of the selection probability on the size of the segment generated negatively skewed abundance distributions after log transformation. Asymmetric segment division further exacerbated negative skewness. Size-independent selection produced log-normal abundance distributions, irrespective of the segment division method. Size-dependent selection and asymmetric division were deemed reasonable assumptions since large language populations are more likely to generate isolates, developing into new populations, than the small ones, and these isolates are likely to be small, relatively to the progenitor population. A negatively skewed distribution of log-transformed population sizes was therefore expected. The observed distributions were however log-normal, scale-invariant for areas containing between a hundred and over a thousand language populations. The dynamics of language differentiation, as reflected by the models, may be therefore unimportant relative to the effect of variable growth rates among populations. All log-normal distributions from resource division models had a higher variance than the observed one, where half of the 1,053 populations had between 350 and 3,000 individuals. Possible mechanisms maintaining such a low variance around a modal population size of 1,000 are discussed.