Seminar Abstract

Wednesday, August 20, 2008 • 12:15 PM • Medium Conference Room, SFI

Eric Smith SFI Professor

Behavior Discussion Group - Opportunities in Computational Historical Linguistics

The combination of diverse features (vocabulary, phonology, morphology, syntax) and intense system-level integration that characterize all human languages make possible rather sophisticated inference of language relations and histories, from the divergences among the languages that are spoken today. At least a large part of this inference problem has been formalized as the classical "Comparative Method" of historical linguistics. For well more than a decade, SFI has drawn together the small community of historical linguists seeking to use the classical comparative method to uncover very deep temporal relations among languages that are currently widely distributed over the globe, potentially reaching back to an ancestor for all currently existing languages. However, the comparative method was formulated for manual application in an age of very limited computational power, and in the more-than-century since its origination, much has been learned both about regularities in language structure and the processes of language change, and in the probabilistic representation and inference of geneological processes, of which language descent is one. Thus many new opportunities exist for the elaboration of historical linguistic methods. I plan, informally, to first review major distinctions that currently exist in the way the comparative method is used (which were explained to me by George Starostin); then to list a few of the new data types and method-integrations that become possible with modern computational and phylogenetic methods; and finally, to suggest some conceptual issues that may alter how we think about language change. In particular, it is probably an uncontroversial, but is nonetheless still an unformalized, belief that long-term changes in regular usage originate in variations at a single time, which are distributed among speakers or speech events within a single language. Thus we may aim at a "modern synthesis" of historical linguistics, structurally akin to the modern synthesis of molecular biology, and driven by the same logic. Much of what I will present comes from work of Murray Gell-Mann and our visiting linguistic collaborators, and particularly from Dan Hruschka, Tanmoy Bhattacharya, and Jon Wilkins, and I hope they will freely add to and correct what I say.

Hosts: Willemien Kets and Dan Hurshka