Prediction in language comprehension
Linguistic prediction is when your brain starts to activate or guess upcoming words before you actually hear or read them. Researchers study this with eyetracking, brain signals, and other methods. Prediction tends to happen most when the sentence context strongly limits what could come next. For example, in “In the summer it is hot, and in the winter it is…” many people anticipate the word “cold” before hearing it.
Prediction also shows up in lexical priming, where a related word makes processing easier. So, a word can become easier to process if it is preceded by another related word.
Eye-tracking experiments: people listen to sentences while looking at pictures. Their eye movements reveal how language guides attention to pictures related to upcoming words. For instance, while hearing “The boy will eat…,” listeners often look at “cake” rather than “ball.” The subject of the sentence can also influence which object is predicted, as in “The little girl will ride…” leading to pictures like a merry-go-round rather than a motorcycle. Overall, listeners use sentence context to predict upcoming word meanings.
Reading experiments with eye-tracking show similar effects. Readers fixate on a word for a shorter time when it appears in a moderately or highly constraining context, and they are more likely to skip it in a highly constraining context. This pattern occurs regardless of the word’s frequency or length. Chinese reading shows similar prediction effects, though readers may skip more words in moderately constraining contexts.
Computational models of eye movements during reading include the E-Z Reader model and the SWIFT model, which use predictability to explain reading behavior.
Brain signals and prediction: the M100 is a brain response linked to visual processing and prediction in MEG/ERP studies. Predictable visual forms can change the M100 amplitude. There is debate about how this relates to early language structure processing (eLAN). The P2 component often grows for words in highly constraining contexts and seems tied to attention and prediction of visual word forms, sometimes showing a left-hemisphere pattern.
The N400 is a key ERP marker for how predictable a word is in a context. Its amplitude is smaller (more positive) when a word is highly predictable. Two factors shape predictability: cloze probability (how often people would supply a given word to finish a sentence) and sentential constraint (how tightly the context limits possible continuations). If a word isn’t predicted, semantically related words can still cause a smaller N400 than unrelated ones, and strong context can boost this facilitation.
A classic study by DeLong and colleagues found that people predict whether the next word begins with a consonant or a vowel based on the article “a” or “an.” That suggested prediction at both semantic and lexical levels. However, that study failed to replicate in later large studies, so the exact word-form prediction is still debated.
Other ERP components: the P300 (P3b) reflects updating context when something unexpected happens. The P600 is linked to processing syntactic violations and similar restructuring when a sentence is hard to interpret. A late positivity after the N400, called post-N400 positivity (PNP), has two forms: a parietal PNP seen with incongruent endings and a frontal PNP that may reflect unexpected lexical items rather than unexpected meanings.
Imaging with fMRI: fMRI has high spatial detail but poor timing. It’s not as good as EEG or eyetracking for studying prediction, but it can show which brain areas are involved. Some fMRI studies suggest that when people predict words at longer time intervals, there is more activity in areas like the anterior cingulate gyrus and Broca’s area, indicating strategic prediction and control.
A theory called PARLO (Production Affects Reception in Left Only) proposes that language prediction mainly involves the left hemisphere, tied to language production and top-down processing, while the right hemisphere supports bottom-up integration. Both streams contribute, but their roles differ across hemispheres.
Surprisal theory explains prediction in information-theory terms. The cost of processing a word depends on how self-informative it is given the context—the more predictable a word, the lower the processing cost. Across many studies, processing effort matches how surprising a word would be in context.
In short, people often predict upcoming language based on context, and this prediction shows up in eye movements, reading times, brain signals, and brain imaging. The exact mechanisms and where prediction happens in the brain are active areas of research, with ongoing debates and new findings shaping our understanding.
This page was last edited on 2 February 2026, at 13:59 (CET).