The processing of good-fit semantic anomalies: an ERP investigation.

The failure-to-detect good-fit semantic anomalies is taken as evidence for shallow semantic processing, however the cognitive mechanisms involved are not well understood. To investigate this we recorded event-related potentials (ERPs) to sentences that contained good and poor-fit semantic anomalies and non-anomalous controls. Detected good-fit anomalies elicited an N400 effect when detection accuracy was stressed, indicating the registration of the anomaly. ERP analyses further ruled out that anomaly non-/detection is due to differences in initial word encoding or in processing prior contextual information. In addition, starting in the P2 interval, the ERP waveform was less positive for non-detected than detected anomalies and non-anomalous controls, presumably reflecting a language-driven modulation of visual input processing. And finally, detection of good-fit anomalies may also depend on the integration of sentential information into the discourse model at the end of the critical sentence. Overall, present findings support the shallow processing account of anomaly detection failure.