Toponyms and the Geographic Extension of Languages: A computational-visual approach

Part of abstract (Hammarström och Segerer; for the full abstract and figures, contact the contact person):

Centre-point coordinates for the approximately 7,000 languages in the world are available via glottolog.org, but there is no open database with the full geospatial extension, i.e., polygon data, for languages globally. However, the Geonames database http://geonames.org gathers more than 11 million place names with coordinates.

[…]

Many of these toponyms contain a formative element, such as, for example, -ville in Libreville, Lastourville etc, or -dougou in Ouagadougou, Bélédougou etc. We propose that such formatives can be detected automatically by measuring the geographic concentration C(s) of potential sub-strings s as the average distance d (as the crow flies) between all pairs of toponym p1, p2 pertaining to s:

C(s) = (∑ₚ₁,ₚ₂∈ₛ d(p1, p2))/(#ₚ₁,ₚ₂∈ₛ)

[…]

Using this measure, we can calculate the probability P(s) that the concentration C(s) is due to chance. Each toponym can then be segmented according to its most non-randomly concentrated (if any) formative.

[…]

Further, since each formative belongs to a language and overlapping formatives arguably belong to the same language, we may try to associate each language with a set of formatives. This yields the desired spatial extensions.

The approach will be illustrated and evaluated in a case study of West African toponyms and languages.