Control and Communication in Mental Computation

Frawley is correct to stress the importance of control in cognitive processing, and the essence of his proposal—that disorders of cognition can result from either within-module or between-module breakdown—is an interesting and potentially important contribution to cognitive science. However, Frawley’s commitment to an overly-literal interpretation of the computational metaphor, combined with his confusion over the possible relationship between any putative language of thought and modularity in peripheral versus central processes, distracts and detracts from his work. In particular, mental processes can employ a language of thought without that language being in any way similar to standard functional or procedural programming languages. In addition, communication between (peripheral) modules does not need to be bound by a single language. Furthermore, the absence of a plausible computational system within which to demonstrate the logic/control distinction and its consequences for the language disorders under discussion weakens the within-module/between-module argument. The computational metaphor—that the brain is like a computer—may be accepted at any one of many different levels. At one extreme, one may accept merely that the functions computed by the neural hardware are subject to the constraints of computability theory. Thus, as Frawley notes, this view is consistent even with anti-representationalist dynamic systems theory (Port and van Gelder 1995). At the other extreme, one might conceive of the neural hardware as a glorified von Neumann machine that executes a program of instructions, and that may crash if any of those instructions embodies a false assumption (e.g., by referring to non-existent data or by attempting arithmetic operations with invalid operands). Frawley appears to opt for this second interpretation, but intermediate interpretations are possible, and arguably more desirable. Several difficulties with Frawley’s interpretation of the computational metaphor are apparent. First, it suggests that the neural hardware rigidly adheres to a predetermined instruction sequence. (Although that instruction sequence may contain conditional ‘‘if-thenelse’’ statements, it remains a rigid instruction sequence.) Where does this sequence come from? Second, it begs the question of how an operating system complex enough to perform the necessary instruction following and allocation of cognitive and effective resources might have evolved within the neural hardware. Third, it raises the possibility of both program and operating system crashes, and the question of how the system might be rebooted. Finally, it limits our imagination with respect to non-standard operating systems and control mechanisms. In particular, it prevents us from considering operating systems and control mechanisms that are not subject to the previous three difficulties. A limited imagination is apparent in the view, implicit in Section 2.1, that all programming languages are necessarily either functional or procedural. Plausible control mechanisms that fall into neither of these categories are conceivable, and such control mechanisms have been proposed within higher cognitive domains. Thus, production systems (Newell and Simon 1972) have proved to be highly adequate in the modelling of problem solving behaviour. Their potential in other cognitive domains has been illustrated through the development of general cognitive architectures such as Soar (Laird, Newell, and Rosenbloom 1987; Newell 1990) and ACT-R (Anderson 1993; Anderson and Lebiere 1998).