Listeners are initially flexible in updating phonetic beliefs over time