Higher level language processes in the brain : Inference and Comprehension Processes

s 2 FMRI signal when people generate causal inferences during stories. Although the majority of psycholinguistic research concerns how people process words or sentences, the bulk of human communication occurs at the level of discourse. Some language comprehension processes may be far more evident as people comprehend complex discourse than when they comprehend words or sentences. For instance, when people comprehend stories, they generate causal inferences as necessary to fill coherence gaps. Studies with neurological patients and with normal participants demonstrating inference priming of lateralized target words suggest that multiple, bilateral, components of semantic processing are necessary for optimally drawing coherence inferences. I will discuss evidence of such processing f rom several methodologies (brain damage, visual field studies with healthy subjects, and neuroimaging) and types of stimuli (multi-word prime-target pairs, stories that promote inferences, and insight-like verbal problems). Results are interpreted in a framework positing bilateral activation, integration, and selection of semantic representations, each hemisphere performing slightly different computations within each semantic process. Specifically, I will argue that the left hemisphere performs relatively f iner coding of semantic information, which is extremely efficient for most straightforward language comprehension; whereas the right hemisphere performs relatively coarser coding of semantic information, which is especially useful when people understand certain discourse constructions and when semantic integration is difficult, and adds richness to discourse representations generally. The ability to extract propositional information from text is a fundamental prerequisite for models of sentence processing and text comprehension. That people encode sentences in something like propositions, and not simply as lists of words, is well established (see Kintsch 1998 for a summary of the evidence). However, the exact nature of these representations and how the cognitive system goes about the task of transforming a string of words into these propositional representations remains an open question. Current systems require substantive designer intervention to build grammars and conceptual schemas that specify the way in which information is structured (e.g. Lenat 1995). However, as yet no generic account of how these templates could be acquired by people has emerged. The Syntagmatic Paradigmatic (SP) model offers a solution to this problem. Like approaches such as Latent Semantic Analysis (LSA, Landauer & Dumais 1997), the SP model is completely data driven. Rather than propose pre-existing syntactic and conceptual rules, the SP model presumes that people store large numbers of sentence traces. Processing a sentence then becomes a matter of retrieving appropriate …