Goals and strategies influence lexical prediction during sentence comprehension

• It is unclear if lexical predictions during comprehension are generated automatically.
• ERP context effects (N400 and PNP) were affected by top-down comprehension goals.
• Readers were sensitive to the validity of predictive cues across environments.
• When global validity was low, participants no longer showed contextual priming.
• Lexical prediction can be modulated by top-down goals and strategies.

Predictive processing is a critical component of language comprehension, but exactly how and why comprehenders generate lexical predictions remains to be determined. Here, we present two experiments suggesting that lexical prediction is influenced by top-down comprehension strategies, and that lexical predictions are not always generated automatically as a function of the preceding context. In Experiment 1 (N = 24), participants read predictable and unpredictable sentence-final words while EEG was recorded from the scalp. When comparing two different sets of task instructions, the neural effects of cloze probability were enhanced when predictive processing was emphasized. In Experiment 2 (N = 252), participants read predictable and unpredictable sentence continuations in a self-paced reading task, and the overall validity of predictive cues was manipulated across groups using a separate set of filler sentences. There was a linear relationship between the benefits of a constraining sentence context and the global validity of predictive cues. Critically, no reading time benefits were observed as prediction validity approached zero. These results provide important constraints for theories of anticipatory language processing, while calling into question prior assumptions about the automaticity of lexical prediction.