Inferences during Reading: Converging Evidence from Discourse Analysis, Talk-Aloud Protocols, and Recognition Priming

Abstract

The conditions under which inferences that connect goals to other goals and actions in narrative texts were investigated following a three-pronged approach. (1) Contrasting texts that allowed or prevented inferences were constructed according to the expectations of a causal network, discourse model. Inferences depended upon whether a subgoal or action directly fit the plan of a superordinate goal, and upon which goal was the most recent unsatisfied goal in a hierarchical plan. (2) The making of predicted goal inferences was evaluated and supported by subjects talking aloud as they read each sentence. (3) Convergent evidence for goal inferences during on-line processing was found by recognition priming of goal statement information at those locations predicted as likely by the discourse analysis and talk-aloud data. The convergent methods and data are discussed with respect to theoretical positions and problems in revealing inferences in on-line processing.