Learning alternations from surface forms with sublexical phonology
Blake Allen, Michael Becker
May 2015

We present a constraint-based multiple-grammar approach to learning and productively applying morphophonological alternations, with a focus on non-local and non-concatenative phenomena, based on the sublexical approach (Becker & Gouskova to appear). Success of the model is demonstrated on the English plural and English past tense data sets, among a variety of others. We compare our model to the rule-based Minimal Generalization Learner, showing our model’s ability to handle a wider range of morphological patterns. Our model contrasts with constraint-based proposals that use abstract underlying representations to generate surface forms in a single grammar; we show that shifting the locus of complexity from the representations to the grammar allows learning from realistically large and noisy data.
Format: [ pdf ]
Reference: lingbuzz/002503
(please use that when you cite this article)
keywords: morphophonology, exceptions, variation, grammatical vs. representational, underlying representations, sublexical phonology, morphology, phonology
previous versions: v1 [May 2015]
Downloaded:1807 times


[ edit this article | back to article list ]