The basic outlook of modularityis that we should not conceiveof the f-mind as a giantmaximally interconnectednet,
such that, for instance, one's language processing can potentiallybe affected by what one ate for breakfast or the color
of the speaker's hair or millions of other ridiculous things. Fodor intends his book as an argument for the
specialization of mental functional architecture, and as an argument against generalized association, general-purpose
proble msolvers, general connectionis m, heavily top-down AI, and top-down“New Look”psychology.
Fodor's realization of this agenda is the hypothesis that input and output systems are organized into faculty-sized
modules; the putative module devoted to language perception is the main topic of the book. The general idea is that
modules are almost reflex-like: they are fast in operation and mandatory, so that, for instance, one cannot helphearing
speech-like signals as speech. They are also reflex-like in that they are relatively stupid—in two important respects.
First, they are domain-specific; for instance the language perception module deals only with linguistic structures, not
with visual or social or historical information. Second, they are informationally encapsulated, which means that no
other cognitive processes such as general inference or contextual understanding can“break into them”to affect their
operation. In addition, Fodor claims, modules have characteristic brain localization and an innate basis; however, we
observed in section 4.8 that many overlearned abilities behave like modular processors but are by no means innate.
Fodor contrasts the modularity of input and output systems with the process of“belieffixation,”a central capacity
which he claims is not modular in his sense. Fodor's chief example of belieffixation is scientific theorizing and
discovery. But in general, belieffixation is less grandiose: it is the process of deciding whether you think that what
someone tells you is true or not. Fodor's basic argument is that, unlike the process of determining what sentence you
have heard, deciding whether it is true can take a long time and can draw on all sorts of disparate knowledge. More
precisely, Fodor characterizes belieffixation as relativelyslow, non-domain-specific,“isotropic”(i.e. potentiallydrawing
on any kind of mental information), and“Quinean”(i.e. potentially resulting in the revision of any kind of mental
information).
The view of processing advocated here leads to a variant of Fodor's position that I'll call“structure-constrained
modularity”(I'll callFodor's position“Fodorian modularity”). We can see already fro mFig. 7.2 that,like Fodor,we are
regarding the brain as a collection of specialists rather than an all-purpose cognizer. However, the locus of modularity
in Fig. 7.2 is not large-scale faculties such as language perception, but the smaller scale of individual integrative,
interface, and inferential processors. There is no extrinsic border around modules.