Speaker: Logan Swanson (Stony Brook)
Title: Phonotactic Learning with Abductive Principles
Time: Monday, April 22nd, 5pm – 6:30pm
Location: 32-D831
Abstract: A foundational learning problem in phonology is phonotactics: discovering the constraints which govern what configurations of sounds are licit in a particular language. These constraints are generally formulated over some kind of substructure, which could be segmental, feature-based, or autosegmental. These substructures form a partial order, with simpler structures contained within more complex ones.
In this talk I will present the Bottom-Up Factor Inference Algorithm (BUFIA), an algorithm which leverages this underlying structure to learn surface-true phonotactic constraints from positive data (Chandlee et al., 2019). In this sense, BUFIA offers a general form for phonotactic learning, since it can operate over any choice of representation and use any abductive principle to decide when to add constraints. While Rawski (2021) examined several possible constraint selection principles, my work has shown additional kinds of abductive pressures that can guide the search, including execution of the search path itself in addition to other constraint selection criteria.
I will discuss how the BUFIA framework provides insight into the impact of these factors on learning, and give a demo of the software implementation.