This paper introduces a new algorithm illustrating how it is possible to learn long-distance phonotactics without prior knowledge of a tier. The algorithm, whose target is the Tier-based Strictly 2-Local (TSL) class of formal languages (Heinz et al., 2011), provably induces both a tier and a grammar from positive data exhibiting a long-distance phonotactic pattern. This result shows that the phonological concepts of a tier and locality are sufficient to induce a particular tier and grammar from positive data. The paper presents demonstrations on natural language data from Latin liquid dissimilation and Finnish vowel harmony, and discusses challenges for the algorithm, such as noisy data.
Proceedings of the 6th Conference on Generative Approaches to Language Acquisition North America (GALANA 2015)
edited by Laurel Perkins, Rachel Dudley, Juliana Gerard, and Kasia Hitczenko
Table of contents