Inferences and text comprehension

Two types of inferences are produced during text comprehension : inferences that are necessary to construct the coherence of the representation and optional inferences. Optional inferences are elaborative and depend on world knowledge. They are regrouped in three categories: particularization, prediction and superordinate information. Globalist (Singer, Graesser and Trabasso, 1994) and minimalist theories (McKoon and Ratcliff, 1992) predict that optional inferences are rarely generated while comprehending a text. A review of experimental studies shows that their generation, is favored by the following factors : the semantic associations provided by the text, the inference's typicality, and the superordinate status of the inference. Other factors do not only favor the generation of optional inferences, but also contribute to building the coherence of the representation : explicit mention of an instrument, text information causally linked to the inference, focalization of readers attention on information related to the inference, causal explanatory status of the superordinate inference. Recent studies suggest that the generation process begins with a transitory, and sometimes partial, activation of the inference content and ends with the integration of this activated information at different levels of the representation built from the text. Furthermore, the generation process could only reach the first activation stages for the optional inferences. This partial generation would facilitate complete generation, when additional factors intervene. Thus, beyond the controversy that opposed the globalist and minimalist views of inference generation, a more gradual view of inference processes in text comprehension is proposed.