The lexicostatistical dilemma: is there a way out of the ‘measure-by-substitution’-method for lexical data?

Activity: Talk or presentationInvited talk


Title The lexicostatistical dilemma: is there a way out of the ‘measure-by-substitution’-method for lexical data?
Person and role
Description The most commonly used method in phylogenetic linguistics, the basic vocabulary method, follows a model designed already in the 1950s by the lexicostatistical approach: from a list of universal concepts, a concepticon (Poornima and Good 2010), normally a Swadesh or Leipzig-Jakarta list, i.e., a list of concepts of inherently low borrowability (Tadmor and Haspelmath 2010), a list of words, targeting the ‘common, everyday equivalent’ of the concepts, is compiled for various languages. These lists are then coded for mutual cognacy, and the datasets could be measured for lexical substitution, using a wide range of methods. Already from its first appearance, the lexicostatistical method has been very unsatisfactory to comparative linguists, who are used to reconstruct etymological trees of lexemes by a rich variety of possible prehistoric events, such as morphological derivations, meaning changes, loans, parallel evolutions, and analogical changes (Chang et al. 2015). However, the comparative method of lexical reconstruction is not capable of reducing and standardizing data sets into a format that could be analyzed by statistical methods. Currently, researchers discuss how the gap between comparative and phylogenetic methods can be bridged over (List 2017).
The current presentation will demonstrate a pilot project, targeting an Indo-European lexical dataset of about 100 cultural core concepts of high age and stability (such as BULL, GRAIN, AXE), where data have been coded in the form of bottom-up etymological trees which include semantic changes and derivations inside trees, but which include trees only as long as a majority of meanings of lexical concepts stick to the meaning of the targeted core concept. Since the data set is just about to be uploaded onto the database (DiACL (Carling 2017), no computational tests have yet been performed on the data. The presentation will discuss what possible impact a dataset such as this might have, using computational methods.

2017 Oct 20

Related organisations
2017 Oct 20

External organisation (University)

NameUniversity of Zurich