A closer look at Boersma and Hayes’ Ilokano metathesis test case

An Error-driven learner maintains a current grammar, which represents its current hypothesis of the target adult grammar. The learner is exposed to a stream of data, one piece of data at the time. Whenever the current grammar is found to be inconsistent with the current piece of data, the current grammar is slightly updated, in a way that takes into account the nature of the failure on the current piece of data. Boersma (1998) develops an implementation of error-driven learning within the framework of stochastic OT, called the Gradual Learning Algorithm (henceforth: GLA). Boersma & Hayes (2001) (henceforth: BH) report that the GLA succeeds at learning variation in three complex realistic test cases. Furthermore, they report that variants of the GLA (which differ only for small details of the rule used to update the current grammar) instead fail. The success of the GLA implementation of error-driven learning on BH’s test cases is surprising, as nothing is built into the error-driven learning scheme to guide the learner towards probability matching. Indeed, it is not hard to construct artificial cases of variation where the GLA fails.1 We thus submit that the proper interpretation of BH’s successful simulation results is the following: the patterns of variation in BH’s test cases have some special structure (hopefully shared by other cases of variation in Natural Language) which the GLA (but not variants thereof which adopt slightly different update rules) is crucially able to exploit. Thus interpreted, BH’s simulation results of course raise the following question: what is this special structure displayed by BH’s test cases (and allegedly shared by other cases of variation in Natural Language) which allows the GLA to succeed? In Magri & Storme (in prep.) (henceforth: MS), we address this question through detailed analyses of the behavior of the GLA (and variants thereof) on BH’s three test cases. Here, we offer a preview of our analyses, focusing on BH’s Ilokano metathesis test case.