Euclide, the crow, the wolf and the pedestrian: distance metrics for linguistic typology.

It is common for people working on linguistic geography, language contact and typology to make use of some type of distance metric between lects. However, most work so far has either used Euclidean distances, or geodesic distance, both of which do not represent the real separation between communities very accurately. This paper presents two datasets: one on walking distances and one on topographic distances between over 8700 lects across all macro-areas. We calculated walking distances using Open Street Maps data, and topographic distances using digital elevation data. We evaluate these distance metrics on three case studies and show that from the four distances, the topographic and geodesic distances showed the most consistent performance across datasets, and would be likely to be reasonable first choices. At the same time, in most cases, the Euclidean distances were not much worse than the other distances, and might be a good enough approximation in cases for which performance is critical, or the dataset cover very large areas, and the point-location information is not very precise.