This dissertation investigates the relation between the complexity of phonological patterns, their learnability by human learners, and their typological distributions.
The research presented in this book is divided into three parts. The first part is dedicated to computational evidence and aims to show how the building blocks of phonological patterns emerge in a neural network as it is exposed to auditory and lexical distributions. Taking a diachronic approach, I show that sound systems evolve towards stable states, regardless of their initial distribution.
The second part describes two experiments with human learners, establishing the learnability of a number of phonological patterns of various degrees of complexity. The results suggest that more complex patterns are more difficult to learn, and that learners reduce the complexity of their input if possible.
The third part assesses attested sound changes and plosive inventories in terms of their complexity. The typological data show that sound change does not necessarily reduce complexity, and that plosive inventories differ greatly in terms of their complexity. The tension between the experimental and typological data sheds light on the interaction of different forces – cognitive, auditory and articulatory – that shape phonological typology.