ABSTRACT
Consistent with the theoretical approach that relies on a causality-based search for coherence, the Causal Network Theory provides strong predictions as to what types of inferences readers generate online, and what mechanisms underlie inference generation (see chapter 5). In particular, van den Broek (1990a) developed a model of inferences called the Causal Inference Maker (CIM), which proposes a classification of the inferences readers are assumed to make, regardless of whether those inferences are necessary for comprehension. In this chapter, I describe this model, and then focus on one particular type of inference, predictive or forward inferences, assumed to be drawn only under very specific conditions. My attempt here is to highlight the fact that, although it has been widely acknowledged that predictive inference generation is constrained by a strong, immediately preceding context, other more distant information may have an effect on the inferences made. Some recent findings (Cook, Limber, & O’Brien, 2001; Peracchi & O’Brien, 2004) have shown that the process of generating predictive inferences combines two types of context information as readers process a focus statement: the immediately preceding context, and more distant knowledge (episodic or semantic). It has also been demonstrated (Guéraud, Tapiero, & O’Brien, submitted) that readers’ episodic and semantic knowledge may override the immediately preceding context during predictive inferencing. I describe some of the studies that have underlined these aspects of the inferential process, and I also show that the nature of contextual information located far from the immediately preceding context influences this process (Galletti & Tapiero, 2004).
