items is facilitated. Strong evidence for predictive processing in language came from ERP studies that examined responses to words preceding a predicted target – for example, function words or adjectives that, while perfectly compatible with the accrued context information, matched or mismatched in gender (e.g., van Berkum et al 2005; Wicha et al 2003) or form (e.g., English “a” versus “an”: DeLong et al. 2005) with a predicted (but not yet presented) upcoming word (e.g., “On windy days, the boy liked to go outside and fly a/an … [where kite is predicted]). Because, in the absence of prediction, these modifying words constitute equally good fits to the accrued contextual information, N400 reductions when the words matched as opposed to mismatched the predicted target showed clearly that information about likely upcoming words has shaped the system in advance. Furthermore, at least in the auditory modality where information about a word accrues over time, N400 data (Van Petten et al 1999) made clear that the processing of predictable and unpredictable words diverges prior to a word’s recognition point – essentially as soon as the system detects contextually mismatching perceptual information. ERP data have also suggested, however, a shift toward more passive, bottom-up
