Optimizing LSA Measures of Cohesion

One important application of LSAhas been to measure cohesion in text and to predict the effects of cohesion on comprehension (see Foltz, chap. 9 in this volume; Foltz, W. Kintsch, and Landauer, 1998; Louwerse, 2004; Shapiro and McNamara, 2000). We use the term cohesion to refer to the elements or cues in the text, and coherence to refer to how these cues are used in the construction of a mental representation (Graesser, McNamara, and Louwerse, 2003; Louwerse, 2002; Louwerse and Graesser, 2005). Many studies have shown how cohesive cues facilitate the coherence of readers’ comprehension, as evidenced by measures such as summarization, recall, question answering, and keyword sorting (Britton and Gulgoz, 1991; Gernsbacher, 1990; Zwaan and Radvansky, 1998). For example, referential cohesion, or argument overlap, between two sentences can help the reader to make links between the sentences and better understand the relationship between them. Cohesion can arise from a variety of sources, including explicit argument overlap and causal relationships, and can operate between sentences, groups of sentences, paragraphs, and chapters. Cohesion between sentences is generally identified as local cohesion, whereas cohesion between larger chunks of texts is generally identified as global cohesion (Givón, 1995; W. Kintsch, 1995; Louwerse, 2005). A cohesion gap in a text forces the reader to make inferences to fill in the gaps, which if successful can improve comprehension (McNamara, 2001; McNamara and W. Kintsch, 1996; McNamara, E. Kintsch, Songer, and W. Kintsch, 1996). However, the reader often lacks either the knowledge or skill to generate the inferences, resulting in poor comprehension of the text. Thus, for the most part, cohesion facilitates comprehension and coherence.