Abstract for Bert Vaux's talk

The Subset Principle vs. Power Maximisation in Phonological Acquisition
Bert Vaux (University of Cambridge)
September 18, 2008 (Thursday)
4:15 PM - ; Room 6417, The CUNY Graduate Center

Are the phonological generalisations formed by language learners upon exposure to underdetermined data sets (i) maximally broad (dictated by maximal representational efficiency, as in rule-based phonology of the SPE tradition), (ii) maximally specific (dictated by the Subset Principle, as in Hale and Reiss's 2008 "The Phonological Enterprise" and perhaps Tesar and Smolensky's 2000 "Learnability in Optimality Theory"), or (iii) variable in scope, depending on (a) minimisation of uncertainty about future events (Gallistel) or (b) Bayesian comparison of competing hypotheses (Tenenbaum)? In this talk I present evidence that a hybrid of (iiia-b) best accounts for what we know about phonological acquisition (and in fact animal learning as a whole), and circumvents problems encountered by theory (i) with scope variation and theory (ii) with trajectories of diachronic change and flaws in applying Gold's subset reasoning to phonology.
You can download Bert's handout and PowerPoint presentation, in PDF format, here.
You can listen to Bert's talk here.