Language comprehension is actively predictive: comprehenders anticipate upcoming content based on semantic and syntactic constraints. When reading 'The detective examined the evidence,' readers predictively activate typical continuations. Eye-tracking and neuroimaging show that failed predictions ('The detective examined the... potatoes') cause processing difficulty, demonstrating that comprehension proceeds through prediction and confirmation, not purely reactive analysis.
From your study of working memory and sentence comprehension, you know that processing language in real time requires holding incomplete structures in memory while integrating new words. Prediction in language processing builds directly on this: rather than passively holding open slots until they are filled, the comprehension system actively generates expectations about what will come next. Comprehension, on this view, is not a purely bottom-up process (reading each word and integrating it) but a predictive coding process in which incoming input is checked against prior expectations.
The evidence for prediction comes from several converging paradigms. In eye-tracking during reading, readers slow down at unexpected words — the "garden path" effect you may already know — but prediction goes beyond structural garden paths. Even when syntax is unambiguous, words that are semantically unexpected ("The detective examined the potatoes") produce longer fixation times than semantically expected continuations ("The detective examined the evidence"). This happens even though both sentences are perfectly grammatical. The processing cost reflects the effort of revising a prediction that had already been committed to. In visual world eye-tracking (where participants look at pictures while hearing speech), listeners move their eyes to the likely referent before the word naming it is complete — demonstrating that prediction is operating on partial phonological input, not just after words are recognized.
The N400 effect in EEG is perhaps the most direct neural signature of prediction. The N400 is a negative deflection in the EEG signal occurring roughly 400ms after an unexpected word; it scales with how semantically anomalous the word is in context. Crucially, the N400 is not just a "surprise" signal — its amplitude is reduced when a word is *more predictable* than average, even for words that are perfectly acceptable. This means the brain is not simply processing unexpected items harder; it is continuously generating predictions and showing reduced effort when those predictions are confirmed. The signal is graded, not binary.
What drives prediction? Both syntactic constraints and semantic/pragmatic knowledge contribute. Syntactic structure creates strong predictions: after "The girl was chased by the," a noun phrase is predicted, and a transitive agent is likely. World knowledge contributes independently: after "The detective examined the," *clues*, *evidence*, and *witness* are highly activated regardless of syntactic structure. Context at the discourse level modulates both: a story about cooking will raise the probability of food-related words throughout, even in semantically neutral sentence frames. The comprehension system integrates all available constraints simultaneously, which is why prediction feels effortless — it is the normal mode of operation, not an occasional strategy.
The theoretical implication is significant: comprehension and production share more machinery than a purely modular view would predict. To predict the next word, the comprehension system must be running something like a production simulation — activating words based on what would be contextually appropriate to say. This convergence of comprehension and production processes, mediated by prediction, is one of the most active research frontiers in psycholinguistics.