Slide Slam M2
The N400 ERP component reflects a learning signal during language comprehension
Alice Hodapp1, Milena Rabovsky1; 1University of Potsdam
The functional significance of the N400 is still actively debated. Recently, theories and computational models of the N400 have increasingly focused on a prediction error perspective. Error-based learning accounts state that when the input deviates from expectations, prediction errors occur and allow for an adaptation of the current internal model to make better predictions in the future. This offers a straightforward hypothesis: if larger prediction errors (i.e., N400s) lead to greater adaptation, this should be reflected in enhanced implicit memory. Here, we investigated this prediction by experimentally manipulating target word predictability (cloze probability) in a sentence reading task, while recording participants’ EEG. After a short break to avoid explicit memory effects, participants were presented with an implicit memory task that allowed the recording of reaction times corresponding to the identification of the previous target words (perceptual identification task). We analyzed our data using linear mixed effects models with random by participants and by items intercepts and slopes and included word frequency as an additional fixed effect. We hypothesized that words that elicit larger N400s during sentence reading (low cloze completions) would show enhanced implicit memory as indicated by faster responses in the subsequent perceptual identification task. As expected, the manipulation of cloze probability in the sentence reading task influenced N400 amplitude in response to the critical word. Crucially, as hypothesized, previously low-cloze (unexpected) words did not only elicit larger N400 amplitudes (β = 1.32 µV, SE = .375, t = -3.511, χ2 = 10.475, p = .001) than high-cloze (expected) words but were also recognized faster in the subsequent implicit memory task (β = -0.037, SE = 0.008, t = -4.557, χ2 = 17.764, p < .001). To investigate the relationship between N400 amplitude and the implicit memory effect more directly, we computed the correlation between participants’ N400 differences in the expected minus unexpected condition during sentence reading and the respective reaction time differences in the subsequent implicit memory task. Results demonstrated that participants with greater N400 amplitude difference also showed a greater implicit memory benefit for previously unexpected words (r = 0.46 [95 % CI: 0.14, 0.70], p = .007). To examine the specificity of the observed link between N400 amplitudes and subsequent implicit memory, we additionally investigated post N400 positivities related to unexpected sentence continuations. While there was a significant late frontal positive ERP effect in our EEG data (β = 0.635, SE = 0.304, t = 2.091, χ2 = 4.217, p = .04), no support for a correlation between its amplitude and implicit memory differences was found (r = -0.015 [95 % CI: -0.33, 0.36], p = .936). We conclude from the results that larger N400 amplitudes lead to increased implicit memory formation. This is in line with the interpretation of the N400 as a prediction error that drives adaptation and learning. The correlation between within-participant differences makes it unlikely that expectancy independently influenced N400 amplitude and implicit memory. Rather our findings support the idea that the N400 ERP component reflects a learning signal during language comprehension.