In a prototypicality task in the laboratory, the preferred auditory token of the vowel /i/ is much more peripheral, i.e. has a much lower F1, than the most common token of /i/ in the listener's language environment or the token that the listener herself would most frequently produce (Johnson, Flemming & Wright 1993). This can be explained within a parallel model of phonology and phonetics (Boersma 2005), under the single additional assumption that representations that lie outside the minimal requirements of the task are not activated. The difference between prototype and modal produced token is then explained by the activity of articulatory constraints in production and their inactivity in the prototypicality task. The difference between prototype and modal environmental token follows automatically from the constraint ranking that results from lexicon-driven perceptual learning.