Contextual Semantic Relevance and Word Surprisal Predict N400 and P600 Dynamics During Naturalistic Reading

2026-07-08 · Source: cs.CL updates on arXiv.org · Field: Science & Research — Life Sciences & Biology, Artificial Intelligence & Machine Learning, Mathematics & Computational Sciences · Depth: Expert, extended

Summary

A study utilizing the Dublin EEG-based Reading Experiment Corpus (DERCo) investigated how contextual semantic relevance and GPT-based word surprisal predict N400 and P600 EEG dynamics during naturalistic reading. Across 22 participants and 32 EEG channels, researchers employed regression-based ERP (rERP) analyses and Generalized Additive Mixed Models (GAMMs), controlling for word frequency and length. Findings indicate both predictors reliably associate with EEG responses, but with distinct temporal and scalp patterns. Word surprisal, quantifying lexical expectation, showed robust effects, particularly later in the P600 window. Contextual semantic relevance, an attention-aware measure of semantic fit to recent discourse, demonstrated strong explanatory power across both N400 (300-500 ms) and P600 (500-800 ms) windows, contributing unique variance beyond surprisal. Specifically, semantic relevance was significant in 25 of 32 channels for N400 and 29 of 32 for P600, suggesting naturalistic reading involves both lexical expectation and local semantic integration.

Key takeaway

For AI Scientists and Research Scientists designing language comprehension models, you should integrate both lexical expectation and contextual semantic relevance. This study demonstrates that human naturalistic reading relies on both, with semantic relevance providing unique explanatory power beyond surprisal for N400 and P600 dynamics. Incorporating an "attention-aware" semantic relevance metric alongside probabilistic surprisal will yield more robust and human-aligned models of language processing, improving the interpretability of neural responses.

Key insights

Naturalistic reading simultaneously processes lexical expectation and semantic integration, each with distinct neural correlates.

Principles

• Lexical expectation and semantic integration uniquely shape comprehension.
• N400 indexes early lexical-semantic access; P600 reflects later integration.
• Computational models link linguistic data to neural activity.

Method

An "attention-aware" method calculates contextual semantic relevance using fastText embeddings, distance-based weighting for three preceding words, and pairwise context-context similarities.

In practice

• Apply attention-aware semantic relevance for discourse-level semantic fit.
• Utilize GPT-based surprisal to quantify lexical expectation.
• Combine rERP and GAMM for comprehensive EEG analysis.

Topics

• Naturalistic Reading
• EEG
• N400 Component
• P600 Component
• Word Surprisal
• Contextual Semantic Relevance

Best for: AI Scientist, Research Scientist

Related on AIssential

• Modeling semantic association in self-paced reading with language model embeddings — Methodological choices, especially embedding model type, critically impact semantic association quantification in reading comprehension.
• Probing for Reading Times — Early language model layers predict initial human reading effort better than scalar surprisal, which excels at later stages.
• The Dynamics of Human and AI-Generated Language: How Semantics Fluctuates across Different Timescales — The ACW-0 semantic timescale analysis effectively distinguishes temporal organization of generic versus specific content in human and AI speech.
• Do Language Models Know Theo Has a Wife? Investigating the Proviso Problem — LLMs align with human judgments on the proviso problem but use shallow pattern matching, not deep pragmatic reasoning.
• N-Grams and Markov Assumptions: The First Predictive Models of Language — N-grams were the first predictive language models, using statistical counts over short word sequences to anticipate the next word.
• Semantic Chunking and the Entropy of Natural Language — A new statistical model explains natural language redundancy through hierarchical semantic chunking and predicts variable entropy rates.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CL updates on arXiv.org.