Exemplar dynamics

Finished · Usage-Based Theories

Summary

This paper shows several important results that an exemplar-based model can capture. First, categories become stable as successful exemplars are added to the cloud due to entrenchment. Pierrehumbert does a good job working through the complexities of entrenchment in this paper. As I understand it, entrenchment is the idea that new exemplars adopt the most prominent phonetic properties of a category, thereby reducing variance in the realization of a form over time. So if there is a category with phonetic property A having a value of 1 as its center of gravity, new exemplars will be drawn toward that value. Another important fact is that phonemes of different languages which are ostensibly the same (e.g., /p/ in English and German) are actually realized differently in a consistent fashion across languages and dialects. This can be explained in exemplar theory due to unique exemplar phonetic values for a category /p/ in different languages. Additionally, the effect of frequency on reduction and phonetic realization is a central component of exemplar models that is not easily captured in generative theories. Another important concept modeled in the paper is that memories decay over time, resulting in more peripheral and older memories ceasing to effect the properties of a category's cloud. Additionally random noise is explained by inaccurate target execution of motor control gestures. The model's main benefit in the paper is it's explanation of how lenition of a certain property can lead to the shift of a category over time. As exemplars accumulate that are influenced by some production bias, the category can shift over time until it stabilizes at some point. Finally, the neutralization of marked categories into unmarked categories is explained because the production bias can cause a shift into the unmarked gesture until the two categories completely merge.