Unnatural Phonology: A Synchrony-Diachrony Interface Approach (dissertation)
Gasper Begus
May 2018
 

This dissertation addresses one of the most contested topics in phonology: which factors influence phonological typology and how to disambiguate these factors. I propose a new framework for modeling the influences of Analytic Bias and Channel Bias on phonological typology. The focus of the dissertation is unnatural alternations — those that operate in precisely the opposite direction from some universal and articulatorily- or perceptually-motivated phonetic tendency. Based on a typological survey of unnatural alternations and gradient phonotactic restrictions, I propose a diachronic device for explaining unnatural processes called the Blurring Process and argue that minimally three sound changes are required for an unnatural segmental process to arise (Minimal Sound Change Requirement; MSCR). Based on the Blurring Process and MSCR, I propose a new model of deriving typology within Channel Bias. I introduce the concept of Historical Probabilities of Alternations (Pχ) and propose a method of estimating Historical Probabilities based on the statistical technique bootstrapping. The proposed framework has theoretical implications. The existence of unnatural gradient phonotactic restrictions reveals that both categorical Optimality Theory and weighted-constraint frameworks with restricted Con undergenerate. To address this shortcoming, I propose a formal model of phonological typology that combines estimates of Historical Probabilities with results from the artificial grammar learning experiments. The dissertation adopts the Maximum Entropy model and introduces prior Historical Weights (wχ), which are derived from the Historical Probabilities. Prior variance and Historical Weights allow for a disambiguation between Analytic and Channel Bias influences on typology: both metrics are compared to the observed typology, which yields a quantitative comparison between the two factors. To estimate the contribu- tion of Analytic Bias, I conduct an artificial grammar learning experiment that tests learning of a complex and an unnatural alternation. By combining statistical modeling of diachronic developments with experimental work, the proposed framework allows controlling for Channel Bias factors when testing the Analytic Bias influences and vice-versa and, in turn, provides quantitative means for disambiguating Analytic and Channel Bias influences on typology.
Format: [ pdf ]
Reference: lingbuzz/004041
(please use that when you cite this article)
Published in: Dissertation, Harvard University
keywords: phonology, typology, experimental phonology, phonological learning, artificial grammar learning experiments, bootstrapping, maxent, sound change, unnatural processes, phonotactics, gradient phenomena, phonology
Downloaded:286 times

 

[ edit this article | back to article list ]