| This dissertation provides a coherence-driven psycholinguistic model of discourse production and comprehension. From the perspective of production, I examine several factors that narrators rely on to produce coherent discourse. These factors, which I call scene salience, are tested in a series of discourse production experiments. These experiments reveal how differences in scene salience influence what narrators decide to mention next during narrative production. Two dimensions of scene salience are tested: expected temporariness in states and event complexity. My production experiments additionally show that patterns of temporal update observed in discourse are side-effects of deeper discourse-level processes, contrary to what most theories of temporal interpretation assume.;From a comprehension perspective, I examine how scene salience affects processes of integration in narrative comprehension. The experimental results support a more active view of discourse comprehension than what is commonly assumed. My experiments support the view that in constructing a mental representation of the unfolding narrative, readers activate prior knowledge associated with the situation being described, and generate expectations about how the narrative will unfold. Readers generate expectations about 1) what may happen next, and 2) when it may happen. I also explore the relationship between scene salience and prior knowledge, as well as their effects on discourse integration.;Overall, this dissertation integrates psycholinguistic research on discourse processing with the coherence-based approach taken in artificial intelligence and formal semantics. The scene-salience-driven model of discourse production and comprehension that I introduce here provides a way of explaining how narrators make sure the narratives they produce are felicitous and not random, as well as how comprehenders construct and update their mental representations of the unfolding narrative. |
