Alternatively, one could conceive ofalltaxonomic structure appearing in eachlexical entry. ThenPOODLE would
containKIND-OF[DOG],KINDlOF [ANIMAL],KIND-OF[PHYSICAL OBJECT], and so on. This makes it a one-step
process to compute an inference based on any taxonomic relation, and we do not have to decide exactly how many
layers of taxonomy go into a concept. Of course, the price is crushing redundancy in lexical concepts.
A parallel proble moccurs in dealing with knowledge of individuals. Should we include in the lexical entry forDan
Dennettjust that he is a male human (or a male philosopher?), and then should we derive the fact that he can move
under his own power by going up the taxonomy toANIMAL? Or should we include in his entry all the superordinate
categories he belongs to, and accept all the attendant redundancy?
This dilemma is reminiscent of the one we faced in the encoding of morphological and constructional structure in
Chapter 6. There I suggested that the proper solution probably lies in how long-term memory hierarchies are
instantiated in the brain, not in some choice of notation in the competence theory. Similarly, Chapter 7 invoked
taxonomic relations in discussing lexical priming, without saying how the brain actually encodes these relations. For
present purposes, I suggest that we leave our options open.
It has often been noted that inferences based on taxonomy are not altogether strict. For example, one would want to
characterize birds as animals that fly—but then there are ostriches and penguins. This too resembles the
semiproductiverelationsfound inmorphology(section6.2.) and especiallyinheritancehierarchies;Pinker (1999) draws
an explicit parallel betwen the two phenomena, suggesting that morphological and conceptual taxonomies are two
manifestations of the same brain process.
More controversialperhaps is thequestionof whether a lexicalconcept carries structure relatingittolowermembers in
thetaxonomy. Forinstance,doesTREEcarrya listofitssubkinds, includingPALM, PINE, andPLUM? TO carrythings to
an extreme, doesPHYSICAL OBJECTcarry within it a list of all its known subkinds? Implausible. On the other hand,
people do use information derived by going down the hierarchy in order to draw inferences: this is“case-based
reasoning.”Of course, case-based reasoning is notoriously unreliable (“Welfare should be eliminated, because let me
tell you about this woman who took advantage of the system”), but people use it all the time nevertheless.
Itseems clear thatatleastsomedownward linksfrom kinds totheirinstances are readilyaccessible.For mostpeoplein
cognitive science,linguistevokesChomskyandbonoboinvokesKanzi. For dog owners,doglikely evokes their own as an
exemplar. Moreover, concept learning is often based on the presentation of exemplars (“This is a kangaroo”), and
there is no reason