Being stressed by stress: competence, performance and learning of metrical stress in Optimality Theory. Invited talk given at: a departement seminar, March 13, 2012, Ben Gurion University of the Negev, Beer Sheva. Slides.
Linguistic theories traditionally concern themselves with representing linguistic knowledge in the human mind (competence). Therefore, they introduce a high number of abstract concepts. For instance, in order to explain the stress system of various languages, contemporary phonological theories refer to syllables within feet within prosodic words. Yet, we do not have trees in our brain, but billions of neurons. Moreover, these neurons form a computing device that too frequently makes mistakes (performance errors). In the first third of my talk, I shall explain how Optimality Theory can bridge the contradiction between these two approaches. My Simulated Annealing for Optimality Theory (SA-OT) Algorithm is a model of linguistic performance, built on top of traditional Optimality Theory, a model of linguistic competence. In the second third of my talk, using computer demonstrations, I will illustrate how SA-OT can account for the Dutch stress pattern in normal and fast speech. It will turn out that SA-OT, similarly to the human mind, can trade precision for speed. In the last third of my talk, I turn to language learning. Contemporary theories of stress make heavy use of the notion of "metrical feet", which, however, cannot be observed in speech. Does this lack of available information hinder the task of the language learner? Certainly not, I shall argue, while presenting the basic idea underlying a recent revision of standard OT learning algorithms.