ABSTRACT
There is general agreement within the neuroscience and cognitive-science communities that complex multidimensional stimuli are represented and processed in at least partially independent subsystems or “modules.” However, real-time processing typically results in local indeterminacy or ambiguity within a module because each subsystem is operating with a limited amount of information. Two general approaches for resolving these indeterminacies have been explored in depth within the cognition and perception literature. One approach is to incorporate domain-specific decision principles into each module. This preserves encapsulation (processing modularity) at the cost of inconsistent solutions across modules that later have to be reconciled. The other approach, which we argue for, is to make use of correlated information from within and across domains, without appealing to domain-specific principles. Although this approach violates the more traditional views of processing encapsulation, consistent solutions across domains can be rapidly coordinated. Within language comprehension, these approaches have been reflected in two-stage parsing models and in interactive or constraint-based models.
