Practice talk for the LSA.
Short abstract: Grammatical illusions have provided valuable insights into how speakers encode and navigate linguistic representations. A parade case involves illusory negative polarity item (NPI) licensing, where comprehenders temporarily accept sentences with an illicit NPI on-line, but judge those same sentences as unacceptable off-line. We show that the illusion is highly selective, and that the position of the NPI strongly modulates susceptibility to the illusion. These results are not predicted by existing accounts. We argue that the selective success reflects a qualitative shift over time in the representational format of sentences in memory. We discuss the relation to parallel findings in vision.