The role of working memory in inference generation during reading comprehension: Retention, (re)activation, or suppression of verbal information?

The present study explored the role of working memory (WM) in online activation of bridging and predictive inferences during reading comprehension. Using short narratives and a probe-naming procedure, five hypotheses were examined: Text retention, text reactivation, inference retention, inference activation, and text and inference suppression. In addition, three types of WM span tests—listening-, operation- and symmetry-span tests—were used to examine whether the role of WM in inference generation is domain-specific for discourse items, domain-specific for verbal items, or domain-general, respectively. Different patterns of results were observed for high- and low-span groups only when participants were divided based on the listening-span test. High-span participants generated predictive inferences faster than low-span participants, and then quickly inhibited them when they became less relevant in the following sentence. Moreover, high-span participants generated more bridging inferences than low-span participants, possibly due to enhanced retention and reactivation of inference-evoking textual information. These findings support the inference activation, inference inhibition, text retention, text reactivation, and discourse-domain-specific hypotheses of WM's role in inference generation. The unique contribution of this study to the field is discussed in relation to existing findings and theories of WM.