Reproduce “Contextual Semantic Relevance and Word Surprisal Predict N400 and P600 Dynamics During Naturalistic Reading” on OpenResearch

About this paper

Kun Sun, Rong Wang

Sun and Wang at Tongji University and the University of Tuebingen demonstrated that both word surprisal and a refined contextual semantic relevance metric predict N400 and P600 brain activity during naturalistic reading. Their research shows these metrics offer distinct, complementary insights into neural responses, with semantic relevance providing additional explanatory power for later P600 effects.

Key insights

• Word surprisal and contextual semantic relevance, despite a weak negative correlation (r = -0.10), capture distinct aspects of word-level processing, each providing unique explanatory value for neural responses during reading.
• Both N400 and P600 ERP components are modulated by both lexical expectation (surprisal) and local semantic integration (relevance), suggesting an interactive processing architecture rather than strictly sequential stages.
• The 'attention-aware' contextual semantic relevance metric is validated as a computationally interpretable predictor of EEG dynamics, demonstrating broad effects on N400 and P600 amplitudes even after accounting for surprisal and other lexical factors.

Summary by alphaXiv.