Contextual support and predictive inferences: What do readers generate and keep available for use?

This study investigates whether readers can generate predictive inferences that remain available to guide comprehension after a number of intervening sentences. The nature of the inference (detailed versus general) was also examined. In four experiments, participants read stories that suggested an inference. Reading time was measured to target sentences that later mentioned the inference. In Experiment 1, reading times were equivalent among inference and explicit versions. In Experiment 2, however, reading times to inference versions were slower than either explicit or general versions, suggesting that the participants only generated general state change outcomes. Experiment 3 supports this conclusion because changing the target sentences to include mention of general outcomes did not change the data pattern. Experiment 4 weakened the causal support for the inferences and found that generation of the outcome was disrupted. The results suggest that readers do not routinely generate predictive inferences that remain available very long.