Ancestry-constrained phylogenetic analysis supports the Indo-European steppe hypothesis: Supplementary materials

Abstract

The supplementary materials contain details of our BEAST analyses. Note that in addition to the four blocks of analyses (A-D) mentioned in Experiments ( §6), there is a fifth block of eight analyses (E1-8) in which each ancestral language in A1, in turn, has its ancestry constraint removed. This document contains the following: S.1 A synopsis of elements that vary between analyses. S.2 A description of accompanying electronic files. S.3 Summary trees and summary statistics for each phylogenetic analysis. S.1 Analyses names and descriptorsEach BEAST analysis has a name (e.g. A1) and a descriptor (e.g. a1-c0-d0-g1-l2-s1-t1-z8) whose elements denote the following: a Ancestry constraints. 0 No. 1 Yes. c A set of clade and time constraints, and a set of languages ( §4.1). 0 Narrow dataset, no time constraints on splits. 1 Medium dataset, no time constraints on splits. 2 Broad dataset, no time constraints on splits. 3 Narrow dataset, time constraints on splits in Table 12. 4 From Bouckaert et al. (2013). 5 Same as c4, but excluding six sparsely-attested languages. s Trait model. 0 SDC (no among-trait rate variation). 1 RSC with gamma-variate among-trait rate variation. 2 RSC without among-trait rate variation. 4 Covarion without among-trait rate variation. t Tree prior. 1 Generalized skyline coalescent. 2 Constant population coalescent.u If given, the adjoining eight bit number specifies, for each ancestral language, whether an ancestry constraint is used (1) or not used (0). x If given, the adjoining eight bit number specifies, for each ancestry-constrained language, whether its time constraint is ignored (1) or used (0). z If given, the adjoining number specifies the MCMC chain length, in multiples of 2 × 10 7 ; or else the chain length is 2 × 10 7 . S.2 Electronic filesThe datasets d0 and d2 are in files sup/ielex-130421-ag-cc.txt and sup/ielex-betal.txt, respectively. The files associated with each analysis can be found in the directory sup/runs-post/ descriptor . Each such directory contains: ieo.xml XML configuration file. info.txt Information about the configuration. ieo.log.sum Statistics summarizing posterior. mcc.trees Summary tree.As described in §5.3, we also run BEAST with the data removed in order to find the prior distribution of the root age. The associated files can be found in the directory sup/runs-prior/ descriptor . The descriptor in this case lacks the d, g, l, and s elements; e.g. the directory for A1 is sup/runs-prior/ a1-c0-t1-z8.The XML configurations must be run using a customized version of BEAST, available at https: //github.com/whdc/ieo-beast. After compiling BEAST, run beast ieo.xml from a UNIX shell. S.3 Summary trees and statisticsFor each analysis we give a plot of the MCC tree and summaries of other parameters in the posterior sample. Plots show the branch rate multipliers (width of horizontal lines), time constraints on ancient and medieval languages (bright red bars), clade constraints (vertical black bars), and posterior clade probabilities that round to less than 100%. There are 3...

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

“…Manually produced cognate sets have nearly always constituted the input data to the Bayesian approaches, but judgments about cognacy may vary among scholars and these judgments directly influence the structure and quality of the phylogeny inferred. For instance, correction of the lexical data [12] for both lexical errors and cognate judgments showed that the corrected datasets can lead to ages that are different-both younger and significantly different in terms of the Bayes Factor criterion [22]-from the original ages inferred based on the original dataset [14]. This result has been confirmed in [13], a paper that used a different set of tree priors but nevertheless reached similar conclusions as [12].…”

“…The choice of working with the short, 40-item word lists of the ASJP database [25] rather than longer word lists is a motivated one. We considered using publicly available datasets that have cognate sets for longer lists of meanings, including databases of Indo-European [12] (210 items), Bantu [26] (100 items), Austronesian [10] (210 items), and Pama-Nyungan languages [27] (200 items). However, these datasets are limited to a few language families, do not cover all calibration points listed in Table 1, and do not share the same set of meanings.…”

“…Bayesian methods for dating groups of related languages are popular, but the results are often debated, and the debates are rarely resolved-for instance, see [9,[12][13][14]18] for debates concerning the Indo-European family. While the literature points to some degree of consensus on the fruitfulness of Bayesian approaches to the dating of language groups, the results of these approaches are less prone to consensus.…”

“…For instance, correction of the lexical data [12] for both lexical errors and cognate judgments showed that the corrected datasets can lead to ages that are different-both younger and significantly different in terms of the Bayes Factor criterion [22]-from the original ages inferred based on the original dataset [14]. This result has been confirmed in [13], a paper that used a different set of tree priors but nevertheless reached similar conclusions as [12]. 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.…”

“…The next generation of Bayesian linguistic phylogenetic studies employed tip dating or node dating [11] techniques to jointly infer divergence times and phylogenies. These approaches have been applied to the Indo-European [12][13][14], Dravidian [15], and Sino-Tibetan [16,17] language families. The success of Bayesian approaches as applied to language data has traditionally been dependent on the availability of information regarding an ancestral language (e.g., Old English), a language pertaining to a branch no longer represented by extant languages (e.g., Hittite), or an internal node which can be associated with archaeological (e.g., Iranian) or historical (e.g., Romance) evidence for the diversification of the corresponding group of languages.…”