Abstract
When two masked targets are presented in rapid succession, correct identification of the first (T1) leads to impaired identification of the second (T2). Behavioral and event-related potential (ERP) studies have demonstrated that semantic information about T2 can survive this ‘attentional blink’ (AB) even though discrimination accuracy is impaired. Here we use neural decoding methods applied to ERP data to investigate whether T2 information inherent in neural activity recorded during the AB can be used to discriminate the T2 stimulus independent of behavior. Twelve observers performed an AB task in which T1 required the discrimination of the direction of a central arrow that was flanked by arrows pointing in the same direction (easy) or in a different direction (hard). The second task involved discriminating whether T2 was related or unrelated to a context word presented at the very beginning of the trial. Each subjects' electroencephalogram was recorded using 32 electrodes and the ERP time-locked to the T2 word was extracted. Consistent with previous studies (Giesbrecht et al., 2007), ERP indices of semantic processing were robust during the AB when T1 was easy, but not when T1 was hard. We applied a linear pattern classifier to the T2-evoked ERP data and evaluated performance of the classifier for separate post-T2 time points using a k-fold cross-validation scheme. This analysis revealed that outside the AB the classifier accuracy was above chance from 400-600 ms post-T2 regardless of T1-difficulty; during the AB, however, classifier accuracy was greater than chance only when T1 was easy. These results indicate that there is information inherent in the neural activity evoked during this task that can be used to discriminate T2 independent of behavior, but that the utility of this information is constrained by task demands.