Superordinate Goal Inferences: Are they Automatically Generated During Comprehension?

Long, Golding, and Graesser (1992) and Long, Golding, Graesser, and Clark (1990) have reported evidence that readers spontaneously generate superordinate goal inferences as they read action statements in stories when they have sufficient time to do so (i.e., a long delay between presentation of the inference-eliciting sentence and the test probe). The purpose of the present study was to determine whether readers generate these inferences under relatively demanding time constraints. We used a rapid serial visual presentation (RSVP) procedure, a 250-ms stimulus onset asynchrony (SOA), and a lexical decision task to test the prediction that superordinate goal inferences are more likely to be automatically generated during comprehension than are subordinate goal inferences. In addition, we had subjects answer simple comprehension questions in order to assess their memory for episodes in the stories. The data indicated that subjects who scored well on the comprehension test exhibited a pattern of decision latencies such that latencies to test words from superordinate goal inferences were facilitated relative to test words from subordinate goal inferences within 250 ms, whereas subjects who performed poorly on the comprehension test exhibited no reliable differences in their latencies. These results provide support for a global-coherence model of inference generation which argues for the importance of causal information in constructing a coherent text representation.