Brain responses to semantic anomaly in natural, continuous speech.

In the present study, brain responses were recorded during the presentation of naturally spoken sentences. In two separate experiments, the same set of stimulus sentences was presented with subjects either being asked to pay close attention in order to answer content questions following the run ('memory instruction' - MI) or to press one of two buttons to indicate a normal or unusual ending to a sentence ('response instruction' - RI). Brain event-related potentials were not averaged across the exact same acoustic information but across 49 different words spoken in natural, uninterrupted sentences. There was no attempt to standardize the acoustic features of stimulus words by electronic means. Rather than splicing stimulus words (and trigger pulse needed for computer averaging) onto sentence stems, consonant-vowel-consonant (CVC) monosyllablic words were selected with voiceless stop consonants in the word initial position. This not only avoids acoustic overlap with the preceding word of the sentence but also allows the point of stimulus word onset to be precisely located. In the MI group, brain responses to the semantically anomalous endings were distinguished by the presence of a late negative wave (N300) followed by a sustained positive wave (P650). Responses to anomally in the RI group data was not consistently differentiated from normal in the 650-1000 ms range. Within conditions, the MI and RI waveforms were differentiated by the presence of an augmented positive-going slow wave in the RI condition which may reflect an augmented CNV release. The feasibility of averaging brain electrical responses across non-isolated words which differed acoustically but were of similar phonemic structure was demonstrated. This paradigm provides a means of studying speech-activated neurolinguistic processes in the stream of speech and may make complex spoken language contexts available for event-related potential investigations of brain and language functions.