Using Think-Aloud methods to understand text comprehension

Most theories of text comprehension assume that there are observable properties of text that influence how the text is understood, encoded into memory, and later recalled. Discourse analysis of the text is used to identify particular properties that predict how a text is understood and remembered. However, discourse analysis leaves out the reader's contribution because not all readers process the same text in the same way. The central purpose of this thesis is to use both discourse analysis and thinking aloud procedures during reading in order to account for the contributions of both the properties of the text and the online understanding of the reader.; Participants performed a modified Think-Aloud procedure by reporting what they understood while reading a pair of causally related sentences. The degree of relationship between the sentences was systematically varied so that it would require more inferences to bridge the gap between less related sentence pairs. We applied discourse analysis to the Think-Aloud protocols to yield causal networks of the inferences and connections that readers actually made. We integrated and quantified these networks with a connectionist model that simulates online understanding. We then used a measure of connection strength between the sentence pairs from the model to predict recall in two control groups that did not perform the Think-Aloud task.; The success of discourse analysis and Think-Aloud procedures in predicting recall indicates that the readers' mental representations of the text are influenced by both the text and readers' individual understandings of it. The findings also indicate that the readers' contributions accounted for more of the variance in recall than did the text properties alone.