Inferring the origin of linguistic features from an atlas: a case study of Swiss-German dialects

A linguistic atlas is a set of maps which visualize the geographical variation of linguistic features in a single language. We present a novel model for Bayesian statistics which infers when and where the variants of a linguistic feature were invented based on the geographical distribution shown in a linguistic atlas. Based on a spatial network representing the rate of diffusion between locations, our model evaluates the probability (likelihood) that the observed geographical pattern is realized by considering the genealogical relationship between variants at different locations. We apply our model to a linguistic atlas of Swiss-German dialects and infer the origin of three forms of the High-German word "nein" meaning "no".