Back to listing

Seminar: Historical linguistics in the age of Bayes, Mark Ellison, 9 Mar

Australian National University, Wellsprings

Date: 19 February 2018

Seminar: Historical linguistics in the age of Bayes

Speaker: Mark Ellison

Venue: Engma Room (5019) HC Coombs Building, ANU

When: 3.30pm-5.00pm, 9 March 2018


We are currently living in the early days of the Age of Bayes. Bayesian algorithms protect our email inboxes from spam, help driverless cars recognise obstructions, and suggest corrections to typos in our word processors. Bayesian thinking is remaking statistics. It is also offering new conceptual foundations in physics, cognitive science and the philosophy of science, by reunderstanding existing learning or analytical techniques as Bayesian. I start the talk by introducing Bayes’ Theorem, why I think we are in an Age of Bayes, and survey how Bayesian methods are being applied in a range of scientific fields.

Historical linguistics has not been left untouched by this wave of Bayesian methods propagating through the sciences. The parallels between family trees of species and languages, and the individual-transcending dynamic processes apparent in both, have lead to the application of Bayesian analytic software, developed for biological systems, to linguistic data. Many historical linguists, however, are concerned that these methods ignore crucial insights gained in more than two centuries of exploration of language change within the field. In the second part of the talk, I précis the model(s) implicit in the Bayesian phylogenetic approaches most frequently applied to language data. I explore the disconnect between these models and traditional focus and methods of historical linguistics.

In fields such as statistics, cognitive science and biology, we see Bayesian reanalyses of existing analytical tools. For example, many statistical methods can be understood as Bayes’ theorem plus a particular model. It is likewise possible to reexpress a number of the analytical tools of historical linguistics in Bayesian terms. Thus recent computational tools are arguably not the first Bayesian methods in the field, only the first explicitly Bayesian methods. In the third part of this talk, I explore Bayesian recastings of the comparative method, some forms of internal reconstruction, and phonological reconstruction.

In my conclusion, I imagine a historical linguistics in the Age of Bayes which encompasses its overlaps with related disciplines while maintaining the conceptual tools its has built up over the centuries.

  • Australian Government
  • The University of Queensland
  • Australian National University
  • The University of Melbourne
  • Western Sydney University