Detecting Causal Inconsistencies in Scientific Text
Murray Singer, Nathalie Gagnon · 1999
This study was designed to inspect readers’ ability to infer causal connections in expository text. These inferences, termed causal bridging inferences, have predominantly been studied with reference to narratives (e.g., Keenan, Baillet, & Brown, 1984; Myers, Shinjo, & Duffy, 1987; Potts, Keenan, & Golding, 1988; Singer & Ferreira, 1983; Trabasso & Sperry, 1985). Evidence stemming from both online and memory measures of narrative comprehension has converged on the conclusion that causal bridging inferences routinely accompany comprehension. Our own studies of these issues have been guided by two levels of theoretical analysis. The higher level analysis is the constructionist framework of Graesser, Singer, and Trabasso (1994). Two central principles of constructionism bear on causal bridging inferences. First, readers are proposed to strive to preserve text coherence at the surface, textbase, and situation model levels of text representation. As a corollary, connections are proposed to be identified among the elements of the causal situation model derived from text. Second, according to the explanation assumption, readers routinely access the causes of physical outcomes and the reasons for intentional actions described in text. Causal outcomes and intentional actions both signal changes of state (Cheng & Holyoak, 1985; Graesser, 1981; Trabasso, van den Broek, & Suh, 1989), a feature that we posit to contribute to the initiation of causal search.