language. In fact, there are without question general processes such as metonymy and metaphor, as well as the
processes illustrated in (5), which apply productively to any conceptually appropriate expression, so that the derived
meanings don't have to be listed in the lexicon. I have taken pains to show that the cases discussed in this section are
not like this: we still need lexical polysemy of various sorts.
The distinction between productive processes of meaning extension (such as (5)) and semiproductive instances of
polysemy such asopenandsmokeis strongly reminiscent of the distinction between productive and semiproductive
morphologydiscussed inChapter 6. My senseisthat thesamesolution isapplicable:thesemi-regular cases are listed in
the lexicon, related by inheritance hierarchies, whereas the regular relations are extracted as independent rules, which
(with luck) can themselvesbe formulated as lexical items. A couple of thelatter cases willbe addressed in section 12.2.
11.4 Taxonomic structure
One sort of structure universally acknowledged as part of lexical semantics is taxonomic structure: poodles are kinds
of dog, dogs are kinds of animal, animals are kinds of livingthing, livingthings are kinds of physical object. The point
ofsuchstructureis thatitprovides moregeneralaccesstorulesofinference: anyinferencerulethatpertainstodogs or
to animals in general will automatically apply to poodles.
However, there are some problems about how taxonomies are to be arranged formally. Suppose we list under each
category only its immediate antecedent in the hierarchy, so thatPOODLEonly says it isKIND-OF[DOG]. Now suppose
we need to draw an inference based on a poodle's being a physical object, say that it continues to exist over time: we
havetoconstructa chainofinferencesfromPOODLEtoDOG, fromDOGtoANIMAL, and so ontillwereachPHYSICAL
OBJECT. This seems like an inefficient way to arrange matters.^175 In addition, there is the proble mof how many layers
thetaxonomyshould contain.For example,shouldMAMMALbeinsertedbetweenDOGandANIMAL? IfSO, thiswould
appear to further lengthen chains of inference. If not, where doesMAMMALfit in?^176
LEXICAL SEMANTICS 343
(^175) Well, maybe this is just a competence theory, and nothing follows about performance. But, as urged in previous chapters, such a move is dangerous.
(^176) The same problemscropup if weconceiveof thetaxonomic structure as an arrangementof nodes in a semanticnetwork,linked byIS-Arelations:fromPOODLEwe have
to go all the way up the taxonomy in order tofind out what can happen to it by virtue of being aPHYSICAL OBJECT. An approach based on meaning postulates (e.g.
POODLE →DOG) faces parallel difficulties. I encourage advocates of these approaches to recast all the arguments of this section in their own terms.