Elliott Moreton

From the University of North Carolina. Phonology. SBS S207.

Modularity Bias in Human and Artificial Learners

Phonological dependencies in natural language tend to relate elements which are phonetically similar. Two main factors may be at the root of this typological asymmetry. One is channel bias, errors in transmission between speaker and hearer which introduce phonetically-systematic biases into the corpus perceived by the learner (Ohala, 1994; Barnes, 2002; Blevins, 2004). The other is analytic bias, cognitive predispositions which render some patterns inherently harder to learn than others (Wilson, 2003). This talk presents evidence that there is a typologically effective analytic bias favoring dependencies between two instances of the same feature over dependencies between two different features, and shows that this bias can emerge in a learner without being explicitly hard-wired in.

Full abstract below.