From Santa Fe Institute Events Wiki
Statistical Inference for Complex Networks Workshop, December 3-5, 2008, Santa Fe NM
Network models of sound change
Over the past two centuries, historical linguists have amassed vast amounts of data on the sound structure of words in the world's languages and have used such data to derive models of sound change and to infer historical relationships between languages. There are many ways to specify models of sound change all with varying assumptions about both what is changing (e.g. features, segments) and which changes are more or less likely. Models based on features (e.g. voicing) make strong assumptions about the uniformity of feature changes across segments and the sequential nature of such changes. Models based on sound classes make strong assumptions that sound changes occur within but only rarely between classes. A network approach to studying sound change, which considers sounds as nodes and the probabilities of sound changes as links, provides a common framework for comparing these different models and for assessing these assumptions against existing data. In this paper, we describe preliminary comparisons of these models of sound change using data from 29 Turkic languages.