Repetition blindness (RB) refers to the failure to report both occurrences of a repeated item in a series of rapidly presented visual stimuli. For instance, the sentence When she spilled the ink there was ink all over, presented word by word at a rate of around 100 msec/word, may be reported as "When she spilled the ink there was all over" (Kanwisher, 1987), even though the sequence is nongrammatical. RB also occurs for repeated items in lists of words and pictures (see Coltheart, 1999, for reviews), but the phenomenon is particularly striking in sentences, where its occurrence generally fails to respect the constraints of coherence or syntax. This has contributed to accounts that locate the phenomenon at the lexical level, rather than at the later stages of memory and recall. In the experiments reported here, novel variations of the standard RB sentence-processing paradigm were used to shed further light on the relative contribution of bottom-up and top-down processes to RB and, more generally, on the role of lexical activation in sentence processing. Kanwisher's (1987) original interpretation of RB drew upon the distinction between types and tokens. Reporting the identity of an item in a stimulus sequence requires not only activation of the existing representation of the item's type representation in memory, but also creation of a token representing the occurrence of that type in the processing episode (Kahneman & Treisman, 1984). For example, the conceptual processing required to build a coherent structure from the words of a sentence requires more than activation of word types in lexical memory; comprehension and recall depend on establishing tokens of these words and the sentential relationships between them in that specific sentence. In this framework, RB is the consequence of a perceptual limitation that impairs the rapid creation of separate tokens for two occurrences of the type.
Type-Token Accounts of RBThere are two major variants of this type-token account that propose different sources for RB. According to the type refractoriness hypothesis, RB is due to the activation dynamics of type nodes, rather than to a problem with token formation or binding. The representation of a particular stimulus undergoes a period of reduced sensitivity immediately after firing, by analogy with the refractoriness of neurons. Accordingly, if the stimulus appears twice in close succession, the second occurrence may be unable to increase the node's activation enough to be recognized or induce a response. The second class of models attributes RB to problems in token individuation. Even though the two separate occurrences are both "recognized" at the type level, formation of a second token from a single type is briefly inhibited after the first token is individuated (e.g., Kanwisher, 1987).The type refractoriness hypothesis has been formalized in dynamical models based on a signal detection approach that implement a "rudimentary form of adaptation" (Bavelier & Jordan, 1992, p. 883). Bavelier and Jordan's model assumes that ...