Herbay, A. 1, 3 , Steinhauer, K. 1, 3 & Royle, P. 2, 3
1 McGill University
2 Université de Montréal
3 Center for Research on Brain, Language, and Music
Neurocognitive processes modulating the N400 ERP component typically associated with lexical semantics are still controversial. Several mechanisms have been proposed to account for N400 amplitude reductions associated with semantic priming: (1) automatic spreading activation (ASA), (2) prediction-based priming and (3) post-lexical integration processes in working memory (WM). Recent proposals support (1) and (2) and question (3) (e.g., Lau et al., 2008). However, Steinhauer et al.'s (2017) priming experiment showed that related word pairs of a given semantic relationship (e.g., part-whole) showed stronger priming effects when embedded in a list with other prime-target pairs of the same (part-whole) relationship (consistent list, CON) than a different relationship (e.g., antonyms; inconsistent list, INC). Moreover, the consistency N400 difference (INC-CON) had a later onset (>400ms) than the traditional relatedness effect (unrelated word pairs vs. INC) starting at 300ms. The consistency effect is compatible with WM-based post-lexical integration ('relational priming') but not with ASA, as semantic networks are arguably not organized according to abstract types of semantic relationship. Alternatively, the short 250ms Stimulus Onset Asynchrony (SOA) may have led to delayed prediction-based priming effects. Using the same experimental design with nine lists of distinct semantic relationship types (antonyms, hyponyms, metonyms, etc.; 80 word pairs each) and a longer 450ms SOA, we hypothesized that N400 consistency effects tied to prediction-based priming should now occur earlier (N250 and early N400 effects, respectively reflecting prediction of orthographic and semantic features) and eliminate late-onset effects. In contrast, WM-based post-lexical effects should still occur after 400ms. We found no N250 effect, and relational priming effects still started late (after 390ms). We argue that relational priming is not compatible with either ASA (see above), or with predictive processes, as relational priming effects onset is not significantly modulated by different SOAs. Thus, relational priming effects are best explained by post-lexical integration processes.