Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Enhanced and left-skewed N400 amplitudes for individuals engaging more in predictive processing
Poster Session A, Wednesday, September 30, 11:00 am - 1:00 pm, Wangari Maathai
Anna Hjortdal1, Patricia León-Cabrera2, Mikael Roll1; 1Centre for Languages and Literature, Lund University, 2Department of Cognition, Development and Educational Psychology, Universitat de Barcelona
While predictive processing is increasingly understood as a fundamental part of language processing, its prominence varies across population groups, individuals and communicative contexts (Huettig & Mani, 2016). We investigated how differential engagement in prediction across the same task influenced electrophysiological responses. The degree of predictive engagement (reliance on prediction) was estimated in a lexical decision response time study. The stimuli were 252 adjective-noun pairs and for each word pair, lexical surprisal was estimated based on a > 1 billion token Swedish language corpus. Half the nouns were replaced with pseudowords and participants (n = 55) read the pairs and carried out a lexical decision task. All participants were right-handed and L1 speakers of Swedish. In line with previous studies, most participants responded faster to less surprising stimuli, but not to the same degree. Reliance on prediction was operationalised as the correlation strength of surprisal and log-transformed response times. While most individuals responded faster to more predictable stimuli, others were not influenced by predictability; correlation coefficients ranged from −0.06 to 0.41, M = 0.11, SD = 0.1. Those in the lower end, the wait-and-seers, were barely influenced by the noun predictability whereas those in the higher end, the predictors, responded quite a bit faster. Reliance on prediction did not correlate significantly with response times or response accuracy. EEG was recorded at 64 electrodes while participants performed the task. Linear mixed-effects regression models were fitted over every electrode and time point with surprisal, reliance on prediction and their interaction as predictors. Noun frequency and length were included as covariates, and the baseline was included as a predictor (Alday, 2019). In line with previous studies, N400 effects were larger for more surprising nouns (Kutas & Federmeier, 2011). The surprisal effect peaked 300-450 ms after noun onset and was modulated by reliance on prediction as indicated by an interaction: As engagement in predictive processing increased, the effect of surprisal on N400 amplitudes did so too, meaning that predictable and unpredictable nouns showed larger voltage differences. The interaction effect was left-lateralised and peaked late, roughly 450 ms after noun onset. These effects remained significant after controlling the false discovery rate. The findings are in line with the N400 as indexing prediction error, i.e. the deviation between prediction and sensory input (Rabovsky et al., 2018). N400 effects were enhanced when language users engaged more in prediction – at least in a local context. The left-lateralised interaction effect could align with hemispheric differences in connectivity with feedback connections playing a larger role in the left hemisphere, as suggested by Federmeier (2007). More broadly, these findings contribute to improving our understanding of the electrophysiological correlates of language prediction (León-Cabrera et al., 2024 for a recent review).
Topic Areas: Control, Selection, and Executive Processes, Computational Approaches