Probing Prediction-Related Processes in Language Using an EEG Word Stem Completion Paradigm

Humans comprehend language rapidly, with active prediction of upcoming words as a key mechanism. Studies using sentences have documented facilitations in behavior and brain activity (N400) when people encounter predictable words, and brain responses (anterior positivity) specific to prediction violations. However, it has been challenging to investigate prediction formation mechanisms, given the complex, incremental context provided by sentences. Here, we examined prediction by measuring EEG in a simple word stem completion task. Participants saw three initial letters (e.g., bro___) varying in constraint/entropy of possible completions and were asked to generate a word completion. They then saw a target that was either a probable completion (brother), an improbable completion (bronze), or a pseudoword (*brom). In Experiment 1, participants self-reported whether the target matched their prediction; in Experiment 2 they made a lexical decision to the target and then typed in their prediction. Consistent with findings from sentences, N400s were reduced for probable completions, primarily driven by a greater proportion of match responses with highly facilitated N400 amplitudes. No anterior positivity was found to improbable words in Experiment 1, but it was present in Experiment 2. Given these task differences, the anterior positivity may therefore reflect engagement of processes to manage activation of lexico-semantic information when there is competition, as from a strong prediction. Crucially, neural responses following the word stem (i.e., during prediction formation) showed a sustained centro-posterior positivity that was graded with entropy (more negative for high-entropy stems). This effect reveals that prediction-formation processes are sensitive to the distribution of possible completions based on the context.