Finding phonological features in perception

This thesis investigated whether phonological features have perceptual bases. The search for features in perception was approached from several angles: from that of an adult listener who has a fully acquired language system in place, from that of a linguist who aims to uncover feature structures in languages, and from that of a learner who acquires the phonological representations for her native speech sounds. We employed vowel discrimination and identification experiments to determine whether phonological features are the categories through which listeners process speech. Subsequently, we assessed listeners’ pre-attentive sensitivity to a particular phonetic dimension in order to find out whether they encode that dimension into a phonological feature. Finally, we carried out simulations of perception-driven learning to reveal whether humans learn to represent the sounds of their language in terms of phonological features. Our results indicate that adult listeners map the auditory properties of speech sounds onto phonological feature categories, and that the mappings between sound and features can be redefined when a sound change occurs. Interestingly, the findings suggest that a phonetic dimension that is used contrastively in one’s language is not necessarily encoded into a phonological feature, but can be associated with a specific phoneme instead. Our learning simulations confirmed the above findings by showing that virtual infants learn to represent the sounds of their language in terms of both features and phonemes. The simulations demonstrated that feature representations are emergent, i.e. created on the basis of the phonetic and morphophonological input that learners are exposed to.