Feedback and regularity in the lexicon

Finished · Usage-Based Theories

Summary

Wedel argues in this paper that regular phonological patterns can arise in a 'memory-rich model' like usage-based theories due to positive feedback loops. Specifically, if there is an analogical bias that prompts future productions to be similar to previously heard and produced forms, this will eventually result in a categorical realization of a specific sound, rather than increased variation. In addition, external biases such a contextual physiological bias to pronounce a form in a specific way in a specific context, can result in allophonic, categorical realization depending on the context. The interaction of these biases will result in different states, depending on their relative strengths. What is different from my model of pre-N nasalization in my dissertation is that the model here employs the bias as a percentage chance to flip the feature value between one of two values. Thus, there isn't a gradient realization of the feature, but a dual option. This prevents incomplete neutralization as a stable state.

The main import of the model for me is that it shows that biases are what drive the emergence of phonological patterns. Wedel doesn't mention here, however, that the knowledge of the learner that drives the stable state is the weight of the biases primarily. This is linguistic knowledge, and no amount of changes or injection of certain patterns into the lexicon will shift the end state.