Poster C33, Wednesday, August 21, 2019, 10:45 am – 12:30 pm, Restaurant Hall
Memory benefits of expectation violations
Laura Giglio1,2, Peter Hagoort1,2, Kara D. Federmeier3, Joost Rommers1,2;1Max Planck Institute for Psycholinguistics, 2Donders Institute for Brain, Cognition and Behaviour, Radboud University, 3Beckman Institute, University of Illinois
Expectations can facilitate the processing of predictable input, but it is less clear whether they have downstream consequences, in particular when the input violates those expectations. On the one hand, expectation violations could be beneficial because the resulting prediction error signals may promote learning and drive memory encoding of the input. On the other hand, expectations may be so strong that they interfere with processing of the input, impairing encoding and yielding worse memory. In the present EEG study, we presented expectation violations in sentences, followed by a surprise memory test that allowed us to look at the consequences for memory and to link these back to electrophysiological signatures of sentence comprehension. Forty-two participants were presented with unexpected but plausible words following either a strongly constraining context wherein they violated a likely expectation (“Be careful, the stove is very dirty”, where “hot” was expected) or a weakly constraining control context that did not afford a strong expectation (“He is surprised, because the second object is very dirty”). Additional filler sentences ended in predictable words. Participants then provided old/new judgements on the two types of unexpected words and matched new words. Constraint affected sentence reading: relative to the weakly constraining condition, expectation violations were characterized by a larger late parietal positivity, as well as beta power decreases prior to and following the expectation violation. Behaviourally, we found better recognition memory for expectation violations than weak constraint controls, suggesting that expectation violations promote memory for the input. In addition, we found downstream effects of prediction disconfirmation during the memory test, with a smaller (more positive) N400 and a larger late positive complex (LPC) for expectation violations than new items. The N400 has previously been associated with priming, and the LPC with explicit recollection; both of these memory processes seemed to be influenced here. Finally, back-sorting items based on subsequent memory responses showed that the late parietal positivity observed during reading was larger in subsequently remembered than subsequently forgotten items, although there was no interaction with constraint. Beneficial effects of expectation violations on subsequent memory may in part originate from processes indexed by this positivity, as it was larger in successfully remembered items as well as in the condition characterised by better memory. Overall, these results show that, although expectation violations are likely costly during on-line processing, they can ultimately be beneficial for memory of the input.
Themes: Meaning: Lexical Semantics, Reading
Method: Electrophysiology (MEG/EEG/ECOG)