The Architecture of the Mind: Modularity and Modularizations
by Robyn Carston

Relevance To Lectures

The relationship between this chapter and the lectures is pretty straightforward. We've been talking about the need for the functional architecture in functional analysis, and about cognitive penetrability. Here, we see a rather detailed account of one fashion in which the architecture might be organized -- modularity. Many of the characteristics that are putatively true of modules must also be true of functional primitives. So, while we've been focussing on the theoretical need for modularity in the lectures, here in our reading we spend a little more time considering the practical side of modularity -- what are modules, what are their properties, etc.

Margin Notes


There is a vast range of mental activities that we engage in daily. "We will consider, on the one hand, whether some of these different metnal capacities are specialized and independent of others and, on the other hand, whether some fall together as the achievements of a more general unitary intelligence." This quote reflects a basic question -- the nature of the cognitive architecture. Criterion for determining if something is part of the architecture: cognitive penetrability

Perceptual Processes: Computational Reflexes

Reflexes, such as eye blinks, are belief independent. Many intelligent activities seem opposite to reflexes in this respect. But could there be a cognitive reflex? "It would be a computational response which is fast, automatic, innately specified, and immune to the beliefs and goals of the organism. In a series of writings, Fodor has proposed that there are indeed such reflexive (as opposed to reflective) mental processes; they are the perceptual processes, including language perception."

For example, perceptual illusions are isolated from beliefs. "Perceptual processes are similar to reflexes. They are also fast, automatic, and innately specified, like reflexes, according to Fodor. However, they are computational and to that extent are like the thoughtful processes of decision making and planning." (NB: making the case that these are computational is not easy!).

Poverty of the stimulus is used to argue for pre-existing perceptual structure. "The persistence and unshakability of perceptual illusions, such as those mentioned above, are strong indications that, in one respect at least, perception is more like a dumb reflex than an intelligent thoughtful process." In other words, reflexes can be computational.

Mental Modules

Fodor's view of the architecture -- it consists of transducers, input systems, and central systems. Transducers convert stimuli into information. Input systems deliver representations. Central processors operate on representations to fix beliefs. Input systems are presumed to be modular.

What are the main properties of modules?

  1. Domain specific
  2. Informationally encapsulated
  3. Mandatory in their operation
  4. Fast
  5. Generate relatively shallow output representations
  6. Universal ontological development
  7. Characteristic and specific breakdowns
  8. Associated with fixed neural architecture

Indeed, it is this last point that is used to create independent systems that exhibit the other seven characteristics.

Language Module

There is lots of evidence for perceptual modularity. It is much more difficult to make a case for the modularity of language. Language perception is fast and automatic. Speech perception system seems to be innately specified. "There is now a large body of work which indicates that at least the core elements of the lexical and syntactic structure of a person's native tongue are acquired in an orderly and rapid way which cannot be accounted for by the variable, and inevitably 'underdetermining' evidence provided by the environment." This is the "poverty of the stimulus"argument. There is also lots of evidence for double dissociations involving language, including dissociation of language from general cognition as evidenced in the case of a linguistic savant. Language is domain specific. Main issue in the language debate isn't the modularity of language perception, or nonmodularity of higher-order language processes, but the modularity of what lies in between.

What is the highest level of language processing that is cognitively impenetrable? Fodor argues that parsing is impenetrable. Word recognition might be...but this is controversial. Contextual effects on word recognition are evident, but they might be based on simple associations.

The Central Systems and the Frame Problem

Fodor's problem: little is known, and little can be known, about nonmodular processes. "We do not know, says Fodor, and mot likely cannot know, precisely because of the global, unencapsulated nature of these processes, that is, the absence of any architectural constraints on their function." Related to this problem is the frame problem in AI research -- "the problem of getting the robot to revise or update its beliefs (knowledge representations) so as to appropriately accommodate the changes brought about in the world by its own actions." The problem is to constrain this updating so that robot does not get lost in an infinite regress of belief updating. But humans don't seem to be bothered by this problem so, "the problem, then, for cognitive science is why the frame problem is, n fact, not a problem in practice for human cognition; it is, in short, the problem of describing and explaining how unencapsulated domain-general processes work."

How Modular Should The Mind Be?

Why should the mind be at least partially modular? Evolution depends on the arrival of intermediate forms of organisms. So, modularity makes sense from an evolutionary perspective -- Marr's principle of modular design. But this seems to stand against the notion of a uniform cognitive architecture.

Modularity dates back to the phrenology of Gall. Then why have nonmodular systems? Fodor's idea is that there still is survival value in having slow, contemplative systems too. But evolution may not be able to produce such systems!

Progressive Modularization and Going Beyond Modularity

What is the architecture of the human neonate? Is it modular, or is it homogenous? Karmiloff-Smith has argued for a compromise view -- modularity, but "representational redescription which repackages existing info so that it becomes available to general cognition. The idea here is to reflect the plasticity of the infant mind.

Pearl Street | "An Invitation To Cognitive Science" Home Page | Dawson Home Page |