that such links, which get concept learning off the ground, should necessarily be abandoned.
When the subordinate is another kind rather than an instance, the strength of a downward link may correlate with the
judged“typicality”ofthesubordinateas a subcategory (Roschand Mervis 1975; Smithand Medin1981). For instance,
fruitevokesapplemore quickly and reliably than it evokespomegranateorprickly pear. A category such asfurniturehas no
independent perceptual characterization (what does a piece of furniture look like?), so the collection of its prominent
subcategories—chairs, tables, sofas, and so on—comes more prominently to the fore.
There are other categories for which taxonomic information seems to play little role. What larger category includes
puddle? It doesn't reallyfitintobodies of wateralong with lakes and rivers. And it doesn't have subcategories or
permanentlyremembered prominentinstances. The same goes forjunk. So simpletaxonomicstructureis oftenuseful,
but not always. As will be seen, this is typical of every kind of lexical information wefind.
11.5 Contributions fro mperceptual modalities
Next let us ask how perceptual features are structured. In order to learn a category fro mpresented instances (“This is
a dog”), one must make use of levels of cognitive structure in the visual system that encode what the instances look
like.Memories stored in terms of these levelsof structure are used to identifynewinstances of a category on the basis
oftheirappearance(“Is THIS a dog?”), and torecognize knownindividuals (people,objects, and scenes). Structures in
other sensory modalities must be stored as well: the sound of aflute or of a particular individual's voice, the taste of
curry, the feel of the key in your front door lock, the sensation of hunger.
If such structures are stored, there is no reason they should not be linked up in long-term memory with lexical items.
One might wonder: Are they still part of“meaning,” or are they somehow“encyclopedic”information? Such a
questiongrows out oftheusual assumptionthat thedomain ofmeaning is a uniform, homogeneous levelof structure,
say some form of logic.^177 But this is an oversimplification. After all, phonology and syntax are built out of formally
distinct interacting subcomponents. Why shouldn't“meaning”also subdivide into natural subcomponents?
A major division in the structure of meaning appears to lie between what I
LEXICAL SEMANTICS 345
(^177) Fodor (1975; 1983) shares this assumption, in his claim that the Language of Thought is homogeneous and non-modular.