The Weekly Newsletter of MIT Linguistics

LingLunch 11/21 - Damian Blasi (Harvard Radcliffe Institute for Advanced Study)

Speaker: Damian Blasi (Harvard Radcliffe Institute for Advanced Study)
Title: The case for next-gen models of language change
Time: Thursday, November 21st, 12:30pm – 1:50pm
Location: 32-D461

Abstract: Drawing robust generalizations from language change data has been, with a few exceptions, a challenging task riddled with concerns about generalizability. A number of models of language change (inspired on diverse dynamics and frameworks from evolutionary dynamics to Bayesian learning) have opened the door to a particularly elegant solution to this problem, which I refer to as the bias-to-structure model. This general framework consists in detecting instantaneous biases in humans when learning, using or transmitting language or language-like behavior, and using the direction of the bias jointly with the worldwide distribution of the relevant linguistic structure as a way of arguing for robust history-independent pathways for language change. This popular approach has been deployed for explaining patterns involving trade-offs between morphological and syntactic marking of grammatical functions, the linear order of NP modifiers and the emergence of compositionality and regular morphological paradigms, among others. In this presentation I will summarize a number of challenges associated with this approach, ranging from the empirical adequacy of these language change models, to the generalizability of linguistic biases in the laboratory and the reliability of cross-linguistic frequency as an indicator of species-wide preferences. I conclude that, in spite of the fact that these approaches have helped moving forward our discussions and have yielded a plethora of interesting observations, concerns about ecological validity merit a re-examination of alternative models of language change.