Emergent feature structures: harmony systems in exemplar models of phonology

Finished · Usage-Based Theories

Summary

This paper argues that generative models of phonology that are not equipped with mechanisms to explain why phonetically natural patterns are asymmetrically more prevalent than phonetically unnatural patterns fail to account for the typology of vowel harmony systems. Specifically back harmony is more prevalent than height harmony, and Cole suggests that this is because one articulator is responsible for frontness vs backness (some muscle), while vowel height could have several articulatory correlates. As such, without an enriched formal grammar to account for this (as is actually typical these days), formal grammars fail.

Cole then suggests that exemplar models account for harmony better because the phonetic features are built into the representations. She argues that exemplar theory doesn't explain why harmony arose diachronically. Instead, there are phonetic precursors that caused harmony to slowly arise, and exemplar theory explains why it has become a phonologically emergent pattern in the languages in which it occurs. Specifically, it arises because harmony initially occurs in some forms to facilitate production (less articulator movement) and perception (easier to perceive what word is being produced). Once some exemplars enter the lexicon with harmony, they can gradually build up a significant percentage in the cloud, resulting in further productions with harmony and eventually entrenchment and emergence of a seemingly categorical phonological pattern.

My main objection to the argument is that it shows that exemplar theory can account for harmony. However, it doesn't explain why Generative theories that assume phonetically-grounded constraints cannot account for it equally well.

There is also the computational issue of learning and generalizing patterns across large window sizes that harmony requires.