Content area
Full Text
Comprehension: a paradigm for
cognition
WALTER KINTSCH
Boulder, CO: Cambridge University
Press, 1998
xvi + 461 pp.
ISBN 0-52-1583608 (alk. pp.);
0-52-1629861 (pbk, alk. pp.)
Kintsch's overarching goal is to "work out a constraint-satisfaction theory of comprehension and show how this kind of cognitive architecture can serve as a paradigm for much of cognition" (p. 2). "Comprehension" refers, first and foremost, to the comprehension of text. Text comprehension is widely taken to involve the construction of a mental representation of a text's meaning. There are numerous factors constraining this construction process-including surface structure (namely, morphology and syntax) and the knowledge base of the reader (namely, knowledge for word meanings and facts about the world). According to Kintsch, a particular computational approach, which he terms the "Construction-Integration (CI) model," best accounts for how these diverse constraints are satisfied. At the heart of the CI model is a connectionist system (similar to McClelland & Rumelhart's (1988) iac architecture) utilizing excitatory and inhibitory interactions amongst units and a settling activation process. According to this approach, "Comprehension occurs when and if the elements that enter into the process achieve a stable state in which the majority of elements are meaningfully related to one another and other elements that do not fit the pattern of the majority are suppressed" (p. 4).
To take a specific example, consider how an ambiguous word like "bank" is understood in different contexts. In the context of (1) "bank" carries a certain meaning and in the context of (2) it carries a very different meaning:
1. Fred went to the bank and emptied his savings account.
2. The boat slid down the bank and into the water.
There are different ways that this disambiguation might be effected by the cognitive system. Kintsch's view is that both meanings of "bank" are accessed early in processing and the more contextually appropriate one comes to suppress the one that is inappropriate. Details aside, an interactive process of settling activation seems a possible candidate for effecting this sort of processing.
Kintsch takes the CI model to account for various aspects of text comprehension and to extend quite naturally to domains such as problem solving and memory. What Kintsch presents, then, is "a collection of specific models, all employing...