In normal (casual) speaking modes, speakers often modify, or seemingly delete, segments that are produced in citation forms of the same words. This paper discusses three examples of how attention to the acoustic detail of spoken language can reveal aspects of the articulation which are perhaps not readily apparent in more cursory examinations of the speech signal. A lexical access model which is sensitive to such acoustic detail will find a better match between both normal and citation spoken forms and their shared abstract representation, than one which is not.