Processing pronunciation variants: Rules and representations

Abstract

Both representational and inference rule mechanisms have been proposed for recognizing pronunciation variants. In our work, we have advanced a view for recognizing pronunciation variants in which multiple forms are represented in the lexicon, with non-canonical forms represented based on their frequency of occurrence and canonical forms represented in a privileged (immune to frequency of occurrence) status due to their congruence with orthography. These investigations have focused on variants in which the relevant alternation was word internal (e.g. schwa vowel deletion, flapping and nasal flaps). Other classes of pronunciation variants are formed due to interactions with segmental properties of surrounding words (e.g. place assimilation, fricative assimilation); the processing explanation advanced for such variants has focused on phonological inference rules that recover underlying representations. The current project investigated the relative role of inferential processes and representation in processing variants formed due to interaction at word boundaries (e.g. fricative assimilation).