ABSTRACT
According to Zenon Pylyshyn, Allen Newell took the high road in psychology by working on a theory of the mind intended to cover a wide range of cognitive behavior. Both Newell and Pylyshyn have put forth convincing arguments about the merits of such a research path, and the central role that cognitive architecture plays in it (Newell, 1973a, 1973b, 1990; Pylyshyn, 1984, and Chapter 2, this volume). Theories of architecture allow one to develop detailed models of local phenomena while addressing global issues about the mind. One such issue that Pylyshyn has raised with respect to Soar is modularity (Pylyshyn, 1991, 1993). The challenge is clear: How can a uniform theory such as Soar be right in the face of evidence that the mind is, in at least some respects, modular? As Pylyshyn points out, Newell (1990) began to answer this in his discussion of Fodor's 1983 monograph. There are two major parts to the answer. First, Soar can indeed admit additional processing modules because it is a bus-oriented system (the working memory is the bus); Newell (1990) speculated that imagery, for example, might require such a special system. Second, the present set of mechanisms underlying central cognition in Soar already exhibit some features of modularity. The purpose of this chapter is to further explore modularity and Soar along this path by considering in some detail one area of psycholinguistics—sentence processing—where modularity has been a central concern, and asking how Soar could account for the pattern of results that has accumulated.
