Commentary: The role of language contact in creating correlations between humidity and tone

Everett et al. (2015) find a that complex tonal languages tend to be found in humid environments a correlation that holds up within different families and parts of the world. Despite the impressive statistical and experimental support for this causal claim, evidence is needed from natural language use, such as Chinese speakers changing their use of tone depending on humidity, before the claim can be considered well supported. There is otherwise a risk that this correlation could be an artifact of history of language families and language contact. To illustrate this, I show in a series of simulations that random selection of languages followed by language contact can create a positive global correlation between tone and humidity with as much as a 83 per cent probability, and a 47 per cent probability of holding within at least two different macro-areas. Language contact is additionally responsible for these correlations holding up when controlling for language relatedness, as I show that when using the random independent samples test employed by Everett et al., their result is still expected by chance as much as 60 per cent to 80 per cent of the time. I further show how contact can create correlations within families by a phylogenetic analysis of the evolution of tone in Niger-Congo and Sino-Tibetan. The number of tones that languages use correlates with humidity within five different global areas (Africa, Eurasia, South America, North America, and the Pacific), and within four different language families (Sino-Tibetan, Austro-Asiatic, Afro-Asiatic, and Niger-Congo). This is better statistical support than even for word order universals, which despite having some support when sampling from different macro-areas (Dryer 1992) do not seem to hold consistently within large language families (Dunn et al. 2011). In addition, the experimental evidence that they cite showing that dry air has …