Radical substance-free phonology and feature learning

Abstract This article argues that phonological features have no substantive properties, instead, segments are assigned by learning strategies set to the task of devising a computational system for phonology is consistent with requirements UG. I address two problems such substance-free model. The first Card-Grammar problem, which has been suggested argue universal features, on premise that, otherwise, language data cannot be stored in fashion necessary correct errors. Card Grammar problem disappears, suitably modular theory mind learned interfaces, where still can retain information not parsed particular grammar. second need demonstration, just an assertion, reasonable grammar and access phonetic substance yield coherent feature assignments. accomplished modeling Kerewe.