• DOI:10.1598/0710.29
• Corpus ID: 60958200

Toward an interactive model of reading.

@inproceedings{Rumelhart1994TowardAI,
  title={Toward an interactive model of reading.},
  author={David E. Rumelhart},
  year={1994}
}

• D. Rumelhart
• Published 1994
• Psychology

R eading is the process of understanding written language. It begins with a flutter of patterns on the retina and ends (when successful) with a definite idea about the author’s intended message. Thus, reading is at once a “perceptual” and a “cognitive” process. It is a process that bridges and blurs these two traditional distinctions. Moreover, a skilled reader must be able to make use of sensory, syntactic, semantic, and pragmatic information to accomplish his task. These various sources of information appear to interact in many complex ways during the process of reading. A theorist faced with the task of accounting for reading must devise a formalism rich enough to represent all of these different kinds of information and their interactions. The study of reading was a central concern of early psychologists (see Huey, 1908). Now, after years of dormancy, reading has again become a central concern for many psychologists. It would seem that the advent of the information-processing approach to psychology has given both experimentalists and theorists paradigms within which to study the reading process. The formalisms of information processing, the flowcharts, notions of information flow, and so forth have served as useful vehicles for the development of first approximation models of the reading process. Unfortunately, the most familiar information-processing formalisms apply most naturally to models assuming a series of noninteracting stages of processing or (at best) a set of independent parallel processing units. There are many results in the reading literature that appear to call for highly interactive parallel processing units. It is my suspicion that the serial, noninteracting models have been developed not so much because of an abiding belief that interactions do not take place, but rather because the appropriate formalisms have not been available. It is the purpose of this chapter to adapt a formalism developed in the context of parallel computation to the specification of a model for reading and then show that such a model can account in a convenient way for those aspects of reading that appear puzzling in the context of more linear stage-oriented models. No claim is made about the adequacy of the particular model developed. The primary claim is that this richer formalism will allow for the specification of more detailed models. These will be able to characterize aspects of the reading process that are difficult or impossible to characterize within the more familiar information-processing formulations. First, I will review two recent models of the reading process. Then, I will discuss some of the empirical evidence that is not conveniently accounted for