This study explores whether readers automatically generate superordinate goal inferences during comprehension under demanding time constraints. The results support a global-coherence model of inference generation which predicts on-line generation of superordinate goal inferences, as they place actions in a relevant causal structure context.