The island biogeography of languages

Aim To examine the degree to which area, isolation, environmental conditions and time since first settlement explain variation in language richness among islands. Location Pacific islands ranging eastwest from Rapa Nui to Indonesia and northsouth from Hawaii to New Zealand. Methods We constructed a dataset of 264 Pacific islands that support 1640 languages (c. 24% of the world's languages). We examined possible predictors of language richness using three different types of models: linear regression models, linear mixed models that included random effects for language phylogeny and simultaneous autoregressive models. We tested whether the following variables, alone or in combination, predict language richness: island area and isolation, climate (rainfall, temperature), mean growing season, soil fertility, habitat heterogeneity (elevation, number of ecoregions), time since first human settlement. Results We identified two optimal models (delta Akaike information criterion < 2). One (R2= 0.52) included area, with 86% of remaining variation accounted for by random effects for phylogeny. The other (R2= 0.56) included a spatial component, area and a suite of other variables (of which isolation and settlement scale were significant). Of the hypotheses tested (mean growing season, ecological risk, habitat heterogeneity, climate, time since settlement, areaisolation theory), areaisolation performed best, alone explaining 44% of variation in language richness. Main conclusions Language diversity relates strongly to island area, and, after controlling for area, with variables linked to isolation (e.g. distance to continent, time since settlement). The influence of environmental productivity may be scale and context dependent. Although environmental productivity may shape language diversity patterns at a global scale, it plays little role on Pacific islands. Approximately half the variance in language richness remains unexplained. Unlike other taxa, for which area, isolation and environmental conditions explain up to 90% of variation in richness, human diversity patterns appear to also be influenced by other variables (e.g. economic, political and social factors).