|
Language migration rates Wichmann et al. (2010) introduced a method for determining the homelands of language groups and Holman et al. (2011) suggested a method for measuring language divergence times based on word similarities using the ASJP database (Wichmann et al. 2016). Putting these two methods together allows us to determine both the geographical and temporal distance between a language represented by a node in a family tree and its immediate ancestor. For instance, if we know when and where proto-Sino-Tibetan was spoken and when and where proto-Chinese was spoken we can infer how fast speakers moved from the homeland of proto-Sino-Tibetan to the homeland of proto-Chinese. This will give us one data-point for studying prehistorical language migration rates. The ASJP database yields around a thousand such data-points from across the world’s language families, allowing us to make generalizations concerning the spread of languages during different periods and in different world areas. For instance, we can infer that the average language migration rate is around 1/3 km for the whole period from 6000 to 2000 before present, but that it then increases rapidly to finally exceed 10 km/year some 500 years ago. In this paper I will focus less on the empirical results on language migration rates than on checking methods, particularly the method of inferring homelands of language groups. Since we rarely know where a given proto-language was spoken, the best way to check methods for inferring homelands is through computer simulations. To test the method of Wichmann et al. (2010) it is necessary to simulate language phylogenies and lexical evolution, since the method uses the information from a family tree and lexical distances among the languages in the family. Such simulations were presented in Holman and Wichmann (2017). It is also necessary to simulate how languages spread, and this was also done recently (Wichmann 2017). Given a simulation of a language family whose languages spread out in space and undergo lexical evolution it can be checked whether the original homeland can, indeed, be inferred from the location of the languages and their words at the end of a simulation. It was tested which of a set of methods worked best. For instance, one method of inferring a homeland might be to simply take the center of the current extension of a language family as the homeland; this method might be called the ‘centroid method’. As it turns out, the best performing method is that of Wichmann et al. (2010) and the ‘centroid method’ is the worst. However, even the best method can give wrong results when the distribution of languages in a family is non-continuous, as, for instance, when the proto-language was spoken on an island and there was an early spread of speakers from the island to a mainland. Thus, this test indicates the types of cases that are problematical for the method of Wichmann et al. (2010), and such cases could then be removed from the sample used to make generalizations concerning prehistorical migration rates. (责任编辑：admin) |