The role of relevance in activation and instantiation of predictive inferences

Abstract

Previous studies have demonstrated that activation of predictive inferences is often the product of a convergence of activation produced by the prior context, general world knowledge and the current input. This study investigated whether predictive inference activation is contingent upon the prior context being currently relevant to or consistent with the on-going discourse, an issue which is critical with respect to theories about updating of discourse representations. In Experiments 1 and 2, naming times to inference concepts were faster following an inference-evoking sentence than a baseline sentence, regardless of whether the inference was consistent or inconsistent with the prior context. In Experiment 3, reading times on a sentence that reinstated the inference were facilitated compared to a baseline condition, but only when the inference was consistent with the prior context. These results have implications for the process via which inferences are activated as well as the process governing inference instantiation. The results are discussed within the context of current models of discourse comprehension that encompass both activation and integration processes.