Speaker: Sarah Payne (Stony Brook University)
Title: Marginal Sequences as a Window into Phonotactic Acquisition
Time: Thursday, April 18th, 12:30pm – 2pm
Location: 32-D461
Abstract: Most current theories of phonotactic learning (e.g., Hayes & Wilson 2008, Chandlee et al. 2019) assume a close relationship between attestation and licitness. Under such accounts, a sequence is licit only if its subcomponents (e.g., sequences of phones or feature bundles) are all attested in the input; illicit sequences are thus those that contain some unattested subcomponent. Under such theories, however, what is the status of marginal sequences (e.g., English ?[#sf])? Constraint-based views posit that marginal sequences are illicit but attested, making them an exceptional subclass of illicit sequences. However, marginal sequences pattern much more closely with licit sequences than illicit ones in terms of repairs in borrowings and in terms of production and perception errors, suggesting that they may instead be an exceptional subclass of licit forms. I argue for a theory of the phonotactic grammar in which attested sequences are divided into productive/licit ones and unproductive/marginal ones. I present a syllable-based computational learning model that learns a binary classification of attested forms into marginal or licit. When evaluated on English complex onsets, I show that this model matches well with human judgments, outperforming the model of Hayes & Wilson (2008) while accounting for the unique behavior of marginal sound sequences.