The generation of predictive inferences and the availability of contextual information

Five experiments were conducted to determine whether readers generate predictive inferences when biasing context was available at either a local level (i.e., in short-term memory) or global level (i.e., in long-term memory). Subjects read passages that described two characters; one description contained contextual information supporting a predictive inference and the other description did not. Experiments 1 and 2 examined whether subjects generated a predictive inference when biasing context was locally available. Reading time differences in Experiment 1 demonstrated that subjects experienced comprehension difficulty when critical sentences were inconsistent with biasing context. Experiment 2 showed that subjects had generated a predictive inference when accompanied by biasing context. Experiments 3 and 4 replicated the results of Experiments 1 and 2 respectively, when biasing context was available at a global level. Reading time differences in Experiment 3 demonstrated that subjects experienced comprehension difficulty when critical sentences were inconsistent with the biasing context. Experiment 4 showed that subjects had generated a predictive inference when accompanied by biasing context. Experiment 5 confirmed that biasing context was available at a global level in Experiments 3 and 4. Results are discussed in terms of several recent discourse comprehension models.