Readers routinely draw inferences with remarkable efficiency and seemingly little cognitive effort. The present study was designed to explore different types of inferences during the course of reading, and the potential effects of differing levels of working memory capacity on the likelihood that inferences would be made. The electroencephalogram (EEG) was recorded from five scalp sites while participants read 90 paragraphs, composed of 60 experimental paragraphs and 30 filler paragraphs. Each experimental paragraph was four sentences long, and the final sentence stated explicitly the inference that readers did or did not make. There were four types of experimental paragraphs: (1) Bridging inference, (2) Elaborative inference, (3) Word-Based Priming control, and (4) No Inference control. Participants were tested using the Daneman and Carpenter (1980) Reading Span Task and categorized as having low or high working memory capacity. The average peaks of the N400 component of the event-related brain potential (EM) were used as a measure of semantic priming and integration, such that the lower the N400 was in response to the explicitly stated inference concept, the more likely it was that the reader made the inference. Results indicate that readers with high working memory capacity made both bridging (necessary) and elaborative (optional) inferences during reading, whereas readers with low working memory capacity made only bridging inferences during reading. We interpret the findings within the framework of the Capacity Constrained Comprehension model of Just and Carpenter (1992).

Previous research on the N400 component of the event-related brain potential (ERP) has dealt primarily with measuring the degree of expectancy on the part of the reader as a result of the context within a sentence. Research has shown that when the final word in a sentence is unexpected or incoherent, a greater N400 amplitude is elicited than if the final word is expected or coherent within the context of the sentence. The present study investigated whether the N400 component is sensitive to global, as well as local, semantic expectancy. Global coherence refers to the ease with which subjects can relate the current proposition they are reading with theme-related ideas. In the present study, the effect of global coherence on event-related brain potentials was tested using four titled and untitled paragraphs (Bransford & Johnson, 1972; Dooling & Lachman, 1971), presented one word at a time. These paragraphs are noncoherent, and are made coherent only with the presentation of a title. The EEG was recorded in response to every word in all four paragraphs. We found an increase in N400 amplitude in response to the words in the Untitled paragraphs relative to the Titled paragraphs, indicating that global coherence does affect the N400. In addition, subjects in the Titled group showed an enhanced P1-N1 component relative to the Untitled group suggesting that the presence of global coherence allows greater attention to be allocated to early visual processing of words.