Neural and Behavioral Indicators of Integration Processes across Sentence Boundaries

A fundamental characteristic of a text is that its sentences are not unrelated but cohere. To understand a text, a reader must therefore cognitively establish specific relations between a new statement and the previously read text. The coherence between sentences can be established by different kinds of integration processes: anaphora resolution, memory processes that resonate for words with related meanings (O’Brien, Rizzella, Albrecht, & Halleran, 1998), and more effortful inference processes that are driven by a search for meaning (Graesser, Singer, & Trabasso, 1994). These processes may work at different levels of a text representation (cf. van Dijk & Kintsch, 1983; Fletcher, 1994). Anaphora resolution may occur at a linguistic level creating argument overlap (Kintsch & van Dijk, 1978), resonance processes may be strongly memorybased, and the more effortful inference processes may occur at the situational level, as suggested by Schmalhofer, McDaniel, and Keefe (2002). Readers process a text sequentially word by word. We can thus investigate how the proposed integration processes unfold at one word or another or at two subsequent words (e.g., a noun followed by a verb) that reference a proposition. Such two-word combinations may also be employed in verification tasks (cf. Griesel, Friese, & Schmalhofer, 2003). So far, research on word-level effects across sentence boundaries is relatively sparse. In one example of “on-line” word comprehension research, Van Berkum, Zwitserlood, Haagort, and Brown (2003) investigated when and how listeners bring the knowledge from the prior discourse to bear on the processing of the final word in a new sentence. They presented sentences like Jane told the brother that he was exceptionally slow and measured ERPs on the