Learning Opaque and Transparent Interactions in Harmonic Serialism

Gaja Jarosz


This paper presents initial modeling results comparing the relative learnability of four basic types of process interactions: bleeding, feeding, counterfeeding and counterbleeding. The principle finding is that the learning model does not predict an absolute hierarchy of learning difficulty based on the type of process interaction involved. Rather, relative learning difficulty depends on numerous interacting factors, including how evidence for the target language is quantitatively distributed in the learning data, how distant the target language is from the initial grammar, and what alternative hypotheses the learner must rule out during learning. The model predicts that under some circumstances transparent orderings (Kiparsky 1971) are easier to learn but that under different conditions, orderings that maximize process utilization (Kiparsky 1968) are learned more easily. The model also predicts that different aspects of interactions can be difficult to learn under different circumstances: in some cases it is the orderings themselves that are difficult to learn, while in others, the model is slow to learn one of the processes involved in an interaction. The discussion relates these learning difficulties to Kiparsky's (1968, 1971) discussion of diachronic rule reordering and rule loss and highlights directions for further research.


Computational Modeling; Opacity; Learning; Diachrony; Harmonic Serialism; Statistical Learning

Full Text:


DOI: https://doi.org/10.3765/amp.v3i0.3671

Copyright (c) 2016 Gaja Jarosz

License URL: https://creativecommons.org/licenses/by/3.0/