Learnable vs. unlearnable harmony patterns

Abstract

Phonological patterns have been characterized as regular, and regular patterns are those that are accepted by a finite state machine. However, being regular is only a necessary condition, but not a sufficient condition for phonology. Two subregular classes which further restrict the computational properties of phonological patterns have been identified: strictly piecewise (SP), and strictly local (SL). In this study, the learnability of a SP pattern and a pattern from the regular class (but not SP/SL) were tested by using the artificial language learning paradigm, and the results suggest that the identified computational boundaries are psychologically real.