EEG Decoding Reveals How Brain Predicts Words During Reading

A new study from researchers Boi Mai Quach, Binh T. Nguyen, Cathal Gurrin, and Graham Healy explores how the human brain predicts the next word while reading, using electroencephalography (EEG) to capture brain responses at millisecond resolution. The research, submitted to arXiv (ID: 2607.18321), focuses on the N400 time window (300-500 ms after stimulus presentation) and how predictability—measured by cloze probability—affects neural signals across different lexical and grammatical categories. Previous work often examined either N400 effects of predictability or lexical categories separately, but this study combines both, leveraging a decoding technique that extracts more nuanced cognitive representations than traditional event-related potential (ERP) analysis.

The results show that N400 differences between high and low predictability words are much larger for content words (like nouns and verbs) than for function words (like prepositions or articles). Among content words, verbs elicited greater N400 differences than nouns, yet nouns carried more distinct information about their own predictability. This suggests the brain processes predictability differently depending on word type, with top-down (predictive) and bottom-up (linguistic structure) mechanisms interacting during comprehension. The decoding approach proved more effective at capturing these dynamics over time, offering a powerful tool for future cognitive neuroscience research and potential applications in brain-computer interfaces and adaptive language models.