Item

Abstract

Humans can rapidly extract sequential regularities from artificial speech. It is much debated if one single associative learning mechanisms accounts for such learning or if an additional rule-extraction mechanism needs to be assumed. A previous study by Pena et al. (2002) highlighted the importance of acoustic cues for triggering rule extraction. In the present study we recorded ERPs for correct and incorrect sequences of synthetic speech after short familiarization phases during which participants listened to speech strings which could be rule-based or not and contain an additional phrase-bracketing cue (500ms pause) or not. Correct sequences contained the structure AXC. Incorrect sequences were violated on the last element and had the structure AXX. Behavioural data suggested that rules were reliably extracted only if the cue was present. In the cued condition, incorrect sequences elicited an early anterior negativity and a later positivity. In the non-cued condition, only the negativity was observed.We interpret the early negativity as correlate of the automatic detection of the incorrect phoneme categories while the positivity is seen as a reflection of a later, more controlled process during which the critical element is related to the preceding elements of the sequence. The results indicate that acoustic cues may be crucial for the emergence of conscious rule-integration processes, while more automatic mechanisms of violation detection emerge even when only distributional information is present during familiarization.