Towards identifying the optimal datasize for lexically-based Bayesian inference of linguistic phylogenies

Proceedings of the 27th International Conference on Computational Linguistics, pages 1578–1590 Santa Fe, New Mexico, USA, August 20-26, 2018. 1578 Towards identifying the optimal datasize for lexically-based Bayesian inference of linguistic phylogenies Taraka Rama♠ Søren Wichmann♦,♥ ♠Department of Informatics, University of Oslo, Norway ♦Leiden University Centre for Linguistics, Leiden University, Netherlands ♥Laboratory of Quantitative Linguistics, Kazan Federal University, Russia tarakark@ifi.uio.no, wichmannsoeren@gmail.com Abstract Bayesian linguistic phylogenies are standardly based on cognate matrices for words referring to a fix set of meanings—typically around 100-200. To this day there has not been any empirical investigation into which datasize is optimal. Here we determine, across a set of language families, the optimal number of meanings required for the best performance in Bayesian phylogenetic inference. We rank meanings by stability, infer phylogenetic trees using first the most stable meaning, then the two most stable meanings, and so on, computing the quartet distance of the resulting tree to the tree proposed by language family experts at each step of datasize increase. When a gold standard tree is not available we propose to instead compute the quartet distance between the tree based on the n-most stable meaning and the one based on the n + 1-most stable meanings, increasing n from 1 to N − 1, where N is the total number of meanings. The assumption here is that the value of n for which the quartet distance begins to stabilize is also the value at which the quality of the tree ceases to improve. We show that this assumption is borne out. The results of the two methods vary across families, and the optimal number of meanings appears to correlate with the number of languages under consideration. 1 Introduction Phylogenetic methods—both distance-based and character-based methods (specifically Bayesian phy- logenetic inference)—from computational biology are being widely used in historical linguistics and typology, for phylogenetic inference, for triangulating homelands of proto-languages (Wichmann et al., 2010b), for investigating rates of lexical change (Greenhill et al., 2017), dating and spread of lan- guage families (Gray et al., 2009; Holman et al., 2011; Bouckaert et al., 2012), reconstruction of proto- languages (Bouchard-Cˆot´e et al., 2013), and for investigating typological universals (Dunn et al., 2011). Most of the above studies require cognate matrices based on wordlists annotated by experts for cog- nacy. Each column in such a matrix represents a cognate class, with 1’s and 0’s indicating whether or not a form in the shared ancestral language has a reflex in the descendant languages with a given pre- defined meaning pertaining to the fixed list of meanings. Such cognate matrices are fed to Bayesian phylogenetic software such as MrBayes (Ronquist et al., 2012b), BEAST (Drummond et al., 2012) or BayesPhylogenies (Pagel and Meade, 2004). An alternative approach is to apply distance-based methods (Wichmann et al., 2010a; J¨ager, 2013; Rama and Borin, 2015). Such methods have the advantage that they do not necessarily require labor-intensive cognate identification or the assumption which goes with cognate identification, namely that the languages analyzed are actually related. …

Content truncated to 500 words