Best practices in justifying calibrations for dating language families

Abstract

The use of computational methods to assign absolute datings to language divergence is receiving renewed interest, as modern approaches based on Bayesian statistics offer alternatives to the discredited techniques of glottochronology. The datings provided by these new analyses depend crucially on the use of calibration, but the methodological issues surrounding calibration have received comparatively little attention. Especially, underappreciated is the extent to which traditional historical linguistic scholarship can contribute to the calibration process via loanword analysis. Aiming at a wide audience, we provide a detailed discussion of calibration theory and practice, evaluate previously used calibrations, recommend best practices for justifying calibrations, and provide a concrete example of these practices via a detailed derivation of calibrations for the Uralic language family. This article aims to inspire a higher quality of scholarship surrounding all statistical approaches to language dating, and especially closer engagement between practitioners of statistical methods and traditional historical linguists, with the former thinking more carefully about the arguments underlying their calibrations and the latter more clearly identifying results of their work which are relevant to calibration, or even suggesting calibrations directly.

Morpheme Ordering Across Languages Reflects Optimization for Processing Efficiency

“…The six languages under consideration represent five language families (Korean, Japonic, Uralic, Turkic, Niger-Congo). Finnish and Hungarian both belong to the Uralic family, sharing a common ancestor about 5,000 years ago (Maurits et al, 2020). The other languages in the sample are not genetically related according to their generally accepted classification (Hammarström et al, 2021).…”

A test of Generalized Bayesian dating: A new linguistic dating method

“…Out of the 52 calibration points, 8 points are excluded because they are not represented in our data by three or more doculects with different ISO 639-3 codes (Cham, Ga-Dangme, Goidelic, Sorbian, Northern Roglai Tsat, Ket-Yugh, Southwest Tungusic, Czech-Slovak); Khoe-Kwadi was taken out because the only member of one of the two major branches, Kwadi, is excluded for being extinct. A language group corresponding to Swahili could not be included since it is a subgroup of the larger Benue-Congo group, selected here because it has a greater number of languages than Swahili (23). By criteria (1-4) we arrived at the 30 core groups given in Table 1.…”

“…In the case of the Sino-Tibetan language family, a comparison of two recent papers [16,17] shows that the choice of cognate sets, language sample, and calibration points produces root age estimates differing by 1500 years. Finally, different tree calibrations can yield different root ages, as illustrated by a case study of Uralic [23], where the calibrations used in a previous paper [24] were revised, yielding a 850 year (15+%) increase in the age estimate and a 3000 year widening of the credible interval (i.e., the 95% highest posterior density [HPD] interval).…”

The uses and abuses of tree thinking in cultural evolution

“…[41]). In phylogenetic analyses, an increasing number of solutions are being offered by Bayesian approaches that allow intra-taxon variation [11] and ambiguity in the expression of traits [42] to be incorporated in analyses (for more information on why Bayesian methods are preferred, and further information on choosing models and priors, see [43][44][45]). In addition to considering emerging methods for accounting for intra-taxon variation, we encourage researchers to consider the sensitivity of their inferences to both traitmeasure inaccuracies and interference resulting from trait lability across varying sampling time windows (see [46] for an example).…”