units out of which they are built (the substantive universals), the repertoire of rule types that govern them, and the
overall architecture. Therefore, in order to see what aspects of languagecanbe reduced to more general principles of
the f-mind, we cannot just observe, for instance, that hierarchical structures are found in motor control (e.g. tool
construction), and thereby clai mthat these are directly related to the hierarchical structures of language (such an
argumentseemstobeoffered byCorballis1991; seethediscussionbyBloom 1994a). Rather,itis necessary todevelop
a functional description of the cognitive structures in question, parallel to that for language exemplified in Fig. 1.1, so
we can look forfiner-scale commonalities.
I take David Marr (1982) to have been developing such a functional description for the visual system; unfortunately
this goal seems to have receded since Marr's death, shortly before publication of his book. Lerdahl and Jackendoff
(1983) develop a functional description for musical cognition. Within this system one can see detailed similarities to
both vision and language. In particular, the similarities between musical rhythm and linguistic prosody are
striking—although neither is reducible to the other. The two are related more or less likefingers and toes.
At a different scale, connectionist neural network modeling (Rumelhart and McClelland 1986a) has suggested some
very general principlesoff-mental computation and learning thatdemonstrablyextendover many capacities. These are
interestingly echoed in Lakoff's (1987) and Jackendoff's (1983) arguments at the levelof functional description for the
existence offlexible, violable (and therefore nonalgorithmic) constraints involved in many domains of cognition and
perception, including language, and they re-emerge in contemporary linguistics to some degree in Optimality Theory.
These general principles may be conceived of as formal universals of the f-mind. They undoubtedly constrain and
shape formal universals specific to language. But they do not determine the particular content of the linguistic
universals, precisely to the extent that language is a distinct specialization.
On stillanotherscale, theonearea wherea great deal is known aboutneural instantiation is thelow-levelvisual system
(e.g. Hubel and Wiesel 1968). This syste mand how it maps to higher visual areas or to visuo- motor coordination are
sometimes invoked (e.g. Churchland 1986) as a model for how the rest of cognition works. The trouble is that low-
level vision shares very little in the way of functional properties with any aspect of language. For example, as pointed
out in Chapter 3, low-level vision does not require freely combinatorial structure mediated by typed variables. Hence
we cannot expect arguments about low-level vision to carry over very decisively to language.
On the whole, linguists have taken more interest in establishing universals than