Cohesion-Establishment in Text Comprehension: A Case-Study of Anomaly Detection

This thesis develops the view that establishing a cohesive mental representation from a text demands elaboration by inference. Such cohesion-based inferential activity is necessarily incomplete, since there is no limit to information potentially relevant to coherence tests. Hence, cohesion is only more or less complete, possibly varying as a function of textual and contextual factors. Text-based anomalies which pass undetected during reading act as signals for shallow underlying cohesion. The rate of detection for any particular anomaly is an index of processing depth. The detection rates for several manipulations of one anomaly are investigated. Cross condition comparison of the obtained rates allows evaluation of manipulation effects. Over a series of eight experiments, considerable non-detection is observed, supporting the above views. The results suggest that the contribution any word makes to overall coherence is not a simple function of its own properties. Rather, these interact with more global constraints such as the background interpretative scenario and the pragmatic status of the message. These factors have a controlling effect on on-line processing. Since pragmatic status, for example, may only emerge gradually over message presentation, selective influences on prior information can be delayed, even across sentence boundaries. These effects suggest that the time-course of full interpretation may be longer than has previously been thought to be the case. Implications for inference measurement and models of text processing are discussed