Inferences during Story Comprehension: Cortical Recruitment Affected by Predictability of Events and Working Memory Capacity Sandra Virtue 1 , Todd Parrish 2 , and Mark Jung-Beeman 2 Abstract & Although it has been consistently shown that readers gen- erate bridging inferences during story comprehension, little is currently known about the neural substrates involved when people generate inferences and how these substrates shift with factors that facilitate or impede inferences, such as whether inferences are highly predictable or unpredictable. In the cur- rent study, functional magnetic resonance imaging (fMRI) signal increased for highly predictable inferences (relative to events that were previously explicitly stated) bilaterally in both the superior temporal gyri and the inferior frontal gyri. Interestingly, high working memory capacity comprehenders, who are most likely to generate inferences during story comprehension, showed greater signal increases than did low working memory capacity comprehenders in the right superior temporal gyrus and right inferior frontal gyrus. When comprehenders needed to draw unpredictable inferences in a story, fMRI signal increased relative to explicitly stated events in the left inferior gyrus and in the middle frontal gyrus, irrespective of working memory capacity. These results suggest that the predictability of a text (i.e., the causal constraint) and the working memory capacity of the comprehender influence the different neural substrates involved during the generation of bridging inferences. & INTRODUCTION To successfully comprehend a story, individuals often need to make connections between information that is currently stated in a text with information that was pre- viously mentioned in a text. Consider the following example: ‘‘From the gate, Walter could see his grand- mother coming towards him. After she walked away, he knew that his cheeks would be sore for days.’’ To understand the last sentence, individuals must generate the inference that Walter’s grandmother pinched his cheeks. When comprehenders encounter such a gap in understanding (i.e., a coherence break), they need to connect information about the causes and consequences of text events. In other words, comprehenders need to generate bridging inferences (also known as backward inferences) to establish a coherent representation of a text (e.g., Graesser, Singer, & Trabasso, 1994; van den Broek, 1990). Although behavioral research has revealed various factors that influence the likelihood of people drawing bridging inferences (Linderholm & van den Broek, 2002; Just & Carpenter, 1992; McKoon & Ratcliff, 1992; Singer, Andrusiak, Reisdorf, & Black, 1992), rela- tively little research has examined the component pro- cesses or neural substrates involved in generating such inferences. The goal of this program of research is to use functional magnetic resonance imaging (fMRI) to ulti- mately provide insight into the specific brain areas and cognitive mechanisms that are involved during bridging inference generation. One factor that influences how easily comprehenders generate bridging inferences is the causal structure of a text. It is well known that people draw inferences when necessary to maintain story coherence (e.g., Graesser et al., 1994; van den Broek, 1990), as in the example above. In addition, the more the text constrains or limits the number of possible inferences readers can generate during story comprehension, the more likely people are to draw these types of inferences (van den Broek, 1990). For example, when reading ‘‘From the gate, Walter could see his grandmother coming towards him,’’ indi- viduals are more likely to generate the inference ‘‘pinch’’ after reading ‘‘Walter could see her hands reaching out for his cheeks’’ than after ‘‘Walter could see her hands reaching out for him.’’ Among other effects, the level of causal constraint in a text appears to differently affect how the right and left hemispheres contribute to bridging inferences. Specifically, weakly constrained text shows greater semantic priming (i.e., speeded responses to related words, indicating greater ‘‘semantic activation’’ of inference-related information) for bridging inferences in the right hemisphere than in the left hemisphere (Virtue, van den Broek, & Linderholm, 2006). Manipulat- ing causal constraint then could reveal specific neural components involved in bridging inference generation. 1 DePaul University, Chicago, IL, 2 Northwestern University, Evanston, IL D 2008 Massachusetts Institute of Technology Journal of Cognitive Neuroscience 20:12, pp. 2274–2284