Friday, Feb. 29, 3:30 PM
32D-141
Paul Boersma
University of Amsterdam
“Emergent ranking of faithfulness explains markedness and licensing by cue.”
I show computer simulations of an Optimality-Theoretic learner who starts out with a constraint set without any bias towards the natural; that is, the set itself has no preference (e.g. it has *Onset, *Coda, Onset, and Coda), and the initial ranking of every constraint is the same. The language environment of this learner, however, does have some biases: there are frequency biases (e.g. final coronals are more frequent than final labials) as well as transmission biases (e.g. the value of the feature [place] is harder to hear for nasals than for plosives). The simulations (assuming a parallel three-level model of phonology and phonetics) show that when the learner listens to this language environment, she will automatically come to rank her (place) faithfulness constraints according to both frequency (higher for labials than for coronals) and cue reliability (higher for plosives than for nasals). When subsequently using these rankings in her own productions, she will automatically exhibit phenomena traditionally ascribed to “markedness” and to “licensing by cue”.