Phonemic restoration in a sentence context: Evidence from early and late ERP effects

When a particular speech sound is obliterated and replaced by a non-speech sound in continuous speech the listener may not notice any disturbance in speech or have difficulties in understanding the word. The present study examined for the first time neurophysiological correlates of the perception of words with an obliterated initial phoneme. Behavioral responses and event-related potentials (ERPs) were measured while participants listened to naturally spoken sentences which had a highly or less expected final word. Half of the sentences were manipulated to have a cough replacing the beginning of the final word, thus, reducing the initial phonetic information available for the word recognition. An N1-P2 complex indicated an automatic registration of the cough's onset. An early negativity to less expected relative to highly expected words was observed for phonetically intact words but not for manipulated words. Although the N400 effect to manipulated words was elicited later than to intact words, after the fragment onset, its amplitude was not enhanced. Further, no significant enhancement of the N400 was found for the manipulated highly expected words. This finding, together with behavioral results, indicated an easier integration of the manipulated highly expected words with the sentence context than of intact but less expected words. Taken together, the study demonstrates an efficient usage of both a context-driven expectancy of the suitable word as well as a stimulus-driven processing of the phonetic information during online perception of speech. The present ERP results support the earlier behavioral research in showing that phonemic restoration is not a bottom-up phenomenon but rather reflects a top-down repair process.