Previous studies on lexical prediction typically address cases where a
lexical prediction is disconfirmed, and focus on the effects of another,
unexpected word found in its place (Delong et al., 2014; Frisson et al.,
2017). In contrast, we address the status of the originally predicted word
once the prediction is disconfirmed in a memory probe task and self-paced
reading task. We find evidence that words predicted from locally and
globally constraining contexts are not suppressed, as disconfirmed
predictions continue to influence memory and sentence comprehension,
suggesting that words gradually decay or are actively maintained in memory.