Word-to-word variation in ERP component latencies: spoken words.

Data were collected shedding light on the brain electrical activity underlying word recognition. Subjects listened to a list of 48 spoken words in six random orders under two instructional sets: first to "think about the meanings of the words," and second, to learn the list. The scalp EEG associated with hearing and identifying the words was recorded at F3, F4, Cz, P3, P4, Pz, and Oz. Standard within-subjects time-locked averaging across words showed a late negative-positive complex with N2-P3 topography, the negative component peaking around 480 msec, the positive component peaking around 830 msec. Averaging within words across subjects uncovered considerable latency variability in both components. Within-word N2 and P3 component latencies covaried with word durations and with the "recognition points" predicted for the words by the "cohort theory" of word recognition. N2 latencies corresponded closely to the "N400" effect elicited with semantically incongruous sentence-final spoken words. Implications for ERP investigations of language processing are discussed.