Learnability of indexed constraint analyses of phonological opacity

Abstract

This paper explores the learnability of indexed constraint (Pater, 2000) analyses of opacity based on the case study of raising in Canadian English (Chomsky, 1964; Chambers, 1973). Such analyses, while avoiding multiple levels of derivation or representation, require the learner to induce indexed constraints, connect these constraints to particular segments in the lexicon, and rank these constraints. An implementation of Round’s (2017) learner for indexed constraints, which is an extension of Biased Constraint Demotion (Prince and Tesar, 2004), is used here to test whether a simple learner can rise to this challenge and learn a restrictive analysis of the opaque pattern (i.e., one that restricts raising to its proper phonological context). Three different datasets are used with decreasing evidence for a restrictive analysis, as well as three underlying form hypotheses (two of which entail entertaining multiple underlying forms for the same surface form simultaneously), with decreasing evidence for the phonotactic patterns in the data (cf. Jarosz, 2006). It is found that the learner can find a restrictive analysis of opaque raising in Canadian English, provided that the most informative dataset is used and multiple underlying forms are considered for those data points that contain [t, d, ɾ] after a diphthong.