Feature, phoneme, syllable or word: how is speech mentally represented?

Four experimental approaches frequently used in speech perception research are discussed with respect to their impact on word recognition models and their implicit assumptions on the mental representation of speech. These approaches are (1) reaction time experiments; (2) the procedure of click localisation; (3) the method of selective adaptation, and (4) the assessment of word similarities. The results of the studies vary as a function of the experimental procedure chosen. Phonetic features, single sounds, syllables and words as entities are alternatively favoured as primary perceptual units. A critical evaluation and an attempt at integrating the data lead to the assumption that the adult speaker/listener has different kinds of mental representation of speech at his/her disposal. Depending on the focus of perception, units of different sizes are primarily focused in the recognition process. This implies that the listener is able to modify his/her temporal analysis window to a certain extent. Nonetheless, as a default case, the syllable serves as the primary perceptual unit.