How abstract are phonological representations? Evidence from distributional perceptual learning

1 Introduction Phonological representations are typically thought of as abstract: for example, phonemes are argued to be comprised of abstract subphonemic units of some sort, whether distinctive features (e.g., Chomsky & Halle 1968), articulatory gestures (e.g., Browman & Goldstein 1989), or acoustic-phonetic dimensions (e.g., Pierre-humbert 2000). There is direct evidence that people are sensitive to these phonemic and subphemic units. A good example comes from studying speech errors: as shown in (1) and (2), the phrase " big and fat " can sometimes be mispronounced by transposing whole segments (as in (1), where two initial stops are switched), but also just parts of segments (as in (2), where only the voicing appears to be switched between the initial stops; Fromkin 1973).