Discovery procedures for sound structure
Aleksei Nazarov (University of Huddersfield)
Some aspects of phonological encoding (like phonological features, rule ordering, stress, and exceptionality) cannot be established from phonetics alone and must instead be inferred by the learner. This talk will illustrate the use of computers to simulate the acquisition of these aspects through discovery procedures (Harris 1946). I will particularly focus on the problem of finding words that are exceptions to a sound rule when the sound rule itself produces variation. The existing Optimality Theory literature can find exceptions (Pater 2010, Becker 2009, Coetzee 2009) or deal with variation (Boersma 1998, Goldwater & Johnson 2003, Coetzee & Pater 2011, Jarosz 2006, 2015) but not both. I propose to make the discovery procedure from Pater (2010) compatible with models that can handle variation. Essentially, I hypothesize that a word is marked as an exception when two different phonological constraints (for instance *[Vowel-Stop] and Faithful) have a ranking tendency for that word (e.g., tendency towards *[Vowel-Stop] >> Faithful) that is opposite to the ranking tendency for the entire lexicon (e.g., tendency towards Faithful >> *[Vowel-Stop]). I will show on the basis of simulations with data from Modern Hebrew that this proposal can indeed learn exceptionality in the face of variability.