Summary of Discussion of Architectures

Prepared by Robert Armengol (additions by Frawley)

Using as a basis the third chapter of Green, we discussed the possibility that our mental architecture is modular. By "mental architecture" we mean the structure of the human mind; by "modular" we mean compartmentalized into particular task-solving units.

When studying the architecture of the mind, we attempt to delineate and analyze those features of thought that are constrained by our nature, that is, processes or tendencies that are already "hard-wired" into the mental system at birth. Such processes are said to be "cognitively impenetrable" because they are constant and generally unaffected by a change in beliefs, knowledge, or motives. The constraints we attempt to discern are different, then, from those placed on us by learning or habit. The ideal standard for cognitively impenetrable processes is "cognitive reflexes," which are pre-specified, automatic, and involuntary mental reactions like visual and auditory illusions.

Our modular hypothesis about the mind is based largely on Fodor's argument that mental systems are highly specialized. This refutes the behaviorist position that the mind is associative and domain-general (one all-purpose problem-solving machine). The primary characteristics of modules:

* domain-specific -- They receive a distilled amount of input and compute limited conclusions based on their specified task.

* encapsulated -- Their information is restricted and cannot readily draw upon knowledge from other modules.

* mandatory -- They cannot avoid functioning as specified.

* innate -- Built-in. Pre-specification down-plays the importance of variations in learning.

* subunits -- Modules are further encapsulated within themselves.

From those, several other characteristics follow:

* ontogeny -- Growth and development of mental modules proceeds similarly across the species.

* breakdowns -- Each encapsulated system must have characteristic failures.

* the brain -- Modules are easy to connect to the neural architecture,

The language module:

Even at very early ages, infants show innate linguistic knowledge that is at least partially unlearned (reactions to certain linguistic sounds, etc.). Facial movements and non-linguistic sounds seem not to affect their speech perception (domain-specificity). In adults, "word parsing" of sentences seems relexive. The process has been shown to be, at least initially, disconnected from contextual interpretations.

The central system problem:

Fodor suggests that while our input and output systems appear to reflect a modular kind of specialization, our "central processing" (the so-called higher level thought) is heterogeneous and can be influenced by almost any source. Such a scheme would imply a mixed mental architecture. We discussed, however, strong suggestions that higher-order computations may also be modular and that mental associations may arise by making key links across domains. It is possible, in other words, that "central processing" itself may be distributed across different domains. Evolutionary theory supports complete modularity of the mind since it allows for separate development of particular adaptations. It also seems advantageous to survival that thought be organized in domains because (a) this grants them reliability (for example, visual interpretation is not affected by what our beliefs assume should be there); and (b) a change to one system based on newly acquired knowledge would not drastically alter the functioning of another -- the mind is easy to "debug."