ABSTRACT
The purpose of this chapter is to develop a formalism within which psychologists can develop detailed information processing models of the reading process. I argue that such a formalism is necessary because the usual formalisms tend to lead most naturally to bottom-up, serial, stage-by-stage models of reading. Moreover, I argue that there is a good deal of evidence suggesting that reading is best characterized as a process of applying simultaneous constraints at all levels and thereby coming up with the most probable interpretation of the input string. Although it is probably not impossible to use the usual flow chart formalisms to represent such models (have arrows pointing back from higher levels to lower levels) it is not especially natural and when carried to the extreme of a completely interacting system is not very informative (two way arrows between every pair of levels). I suggest that the formalisms designed for parallel computing applications are the best substitutions. Finally, I develop a model based on HEARSAY II and GSP and argue that such a model has many very promising features.
