The Error-driven Ranking Model of the Acquisition of Phonotactics: How to Keep the Faithfulness Constraints at Bay

A problem which arises in the theory of the error-driven ranking model of the acquisition of phonotactics is that the faithfulness constraints need to be promoted but should not be promoted too high. This paper motivates this technical problem and shows how to tune the promotion component of the re-ranking rule so as to keep the faithfulness constraints at bay. Sections 1-2 introduce the algorithmic framework considered in the paper, namely the errordriven ranking model of the acquisition of phonotactics. Section 3 motivates a specific problem which arises in the design and analysis of this model, namely the problem of controlling the height reached by the faithfulness (F) constraints. Sections 4-6 sketch the theory of F-controlling. Magri (2014a) presents the theory in more detail. 1 The acquisition of phonotactics Generative linguistics assumes that the learner is provided with a typology of grammars G1, G2... The language-learning problem thus consists of individuating the target adult grammar G∗ within the typology, on the basis of a finite set of data generated by that grammar. Various formulations of this problem differ for the structural assumptions about the underlying typology, for the type of data fed to the learner, and for the criteria of success used to evaluate the grammar Ĝ chosen by the learner relative to the target grammar G∗. In this paper, I focus on the following specific formulation of this general language learning problem. The typology consists of the phonological grammars defined in Optimality Theoretic (OT) terms through the rankings of a given set of constraints (Prince and Smolensky, 2004). The data fed to the learner consist of surface forms sampled from the language L∗ generated by the target OT grammar G∗, namely the set of surface forms which are the phonological realizations of some underlying forms according to G∗. The criteria for success is that the OT grammar Ĝ chosen by the learner generates a language L̂ which coincides with the target one: L̂ = L∗. This specific formulation is called the problem of the acquisition of phonotactics. In fact, phonotactics is the knowledge of the distinction between licit and illicit forms. Assuming that the distinction is categorical (Gorman, 2013), knowledge of phonotactics reduces to knowledge of the set of licit forms (the set of illicit forms is just the complement). And the set of licit forms relative to an OT grammarG is the corresponding language LG.