Gradient Symbolic Representations and the Typology of Ghost Segments

Abstract

Ghost segments are best analysed as weakly active elements under the assumption of Gradient Symbolic Representations (Smolensky & Goldrick, 2016; Rosen, 2016). This assumption allows to predict the attested interactions between phonological markedness constraints and the (non)appearance of ghost segments we find in the languages of the world: first, the co-existence of different types of ghost segments that differ in whether they appear to resolve a markedness problem or whether they disappear to avoid a markedness problem, and, second, the weak contribution to markedness of ghosts. The assumption that all underlying phonological elements have a certain activation that can gradiently differ and might persist into the output structure predicts these two phenomena straightforwardly that are challenging under alternative accounts to ghost segments