The combined temporal and spatial resolution of MEG (magnetoencephalography) was used to study whether the same brain areas are similarly engaged in reading comprehension in normal and developmentally dyslexic adults. To extract a semantically sensitive stage of brain activation we manipulated the appropriateness of sentence-ending words to the preceding sentence context. Sentences, presented visually one word at a time, either ended with a word that was (1) expected, (2) semantically appropriate but unexpected, (3) semantically anomalous but sharing the initial letters with the expected word, or (4) both semantically and orthographically inappropriate to the sentence context. In both subject groups all but the highly expected sentence endings evoked strong cortical responses, localized most consistently in the left superior temporal cortex, although additional sources were occasionally found in more posterior parietal and temporal areas and in the right hemisphere. Thus, no significant differences were found in the spatial distribution of brain areas involved in semantic processing between fluent and dyslexic readers. However, both timing and strength of activation clearly differed between the two groups. First, activation sensitivity to word meaning within a sentence context began about 100 msec later in dyslexic than in control subjects. This is likely to result from affected presemantic processing stages in dyslexic readers. Second, the neural responses were significantly weaker in dyslexic than in control subjects, indicating involvement of a smaller or less-synchronous neural population in reading comprehension. Third, in contrast to control subjects, the dyslexic readers showed significantly weaker activation to semantically inappropriate words that began with the same letters as the most expected word than to both orthographically and semantically inappropriate sentence-ending words. Thus, word recognition by the dyslexic group seemed to be qualitatively different: Whereas control subjects perceived words as wholes, dyslexic subjects may have relied on sublexical word recognition and occasionally mistook a correctly beginning word for the one they had expected.

An event-related brain potential (ERP) reflecting the acoustic-phonetic process in the phonological stage of word processing was recorded to the terminal words of spoken sentences. The peak latency of this negative-going response occurred between 270 and 300 msec after the onset of the terminal word. The independence of this response (the phonological mismatch negativity, PMN) from the ERP component known to be sensitive to semantic violations (N400) was demonstrated by manipulating sentence endings so that phonemic and semantic violations occurred together or separately. Four conditions used sentences that ended with (1) the highest Cloze probability word (e.g., “The piano was out of tune .”), (2) a word having the same initial phoneme of the highest Cloze probability word but that was, in fact, semantically anomalous (e.g., “The gambler had a streak of bad luggage. ”), (3) a word having an initial phoneme different from that of the highest Cloze probability word but that was, in fact, semantically appropriate (e.g., “Don caught the ball with his glove. ”), or (4) a word that was semantically anomalous and, therefore, had an initial phoneme that was totally unexpected given the sentence's context (e.g., “The dog chased our cat up the queen ”). Neither the PMN nor the N400 was found in the first condition. Only an N400 was observed in the second condition while only a PMN was seen in the third. Both responses were elicited in the last condition. Finally, a delayed N400 occurred to semantic violations in the second condition where the initial phoneme was identical to that of the highest Cloze probability ending. Results are discussed with regard to the Cohort model of word recognition.