words, theymust havesome resourcesfromwhichthedifferences can beconstructed. Therefore itoughttobea long-
range goal for conceptualist semantics to characterize these resources.
A particularly austere approach to primitives would be the thoroughly empiricist claim that meaning is built entirely
fro mpri mitives provided by sensation or perception, and that all else is built by association or statistical inference. I
take it that this is a non-starter. For instance, it was shown as long ago as David Hume'sTreatise on Human Naturethat
the notion of causation, hereCAUSE, cannot be built up by these means (see discussion in Macnamara 2000). And
words likefunction, any, andneverthelesssee maltogether unapproachable perceptually. In the ter ms of Chapter 10,
perceptual descriptive features alone are not sufficient for characterizing meaning; and there is no way to construct
inferential descriptive features fro mperceptual pri mitives. Thus there is no question that we are going to have to
accept some abstract primitives.^173
In view of these considerations, we will adopt as a practical policy the necessity for lexical conceptual decomposition.
And we will note as we go the prospects, good and bad, forfinding primitives.
11.3 Polysemy
Another much-discussed question of lexical semantics is how the theory should deal with polysemy. How should we
treat apparently different senses of a lexical ite mthat bear so me intuitive relationship, for instance the two uses of
break, open, androllin (1) and (2)? Let me lay out some cases, mostly not new, that illustrate the complexity of the
problem; others will appear in the course of the chapter.
First, a baselinefor how rigorous it is possible to be.Gooseberryandstrawberryhave nothing to do withgeese and straw.
Yet given the chance to create folk etymologies, people will invent a story that relates them. Overcoming the strong
intuition that theremustbe a relation is not easy, and one must not insist on closer relations than there really are. But
how can we tell when to give up and
LEXICAL SEMANTICS 339
(^173) For an interestingcase thatalso bears onnon-definitionaldecomposition, ithas oftenbeennoticed (Talmy 1978; Jackendoff1991) thatthesubstance–object(mass-count)
distinction in nouns (water vs.bowl) parallels the process–event distinction in verbs (run vs.arrive), and that there are often strong interactions between the two
distinctions in the meanings of sentences. This parallelism motivates a semantic feature[±bounded] that contributes to both domains: objects and events have inherent
boundaries, substances and processes do not. As shown in Jackendoff (1991) , it is in part this common distinction that allows us to say that both a table(object) and a
speech (event) can have anend. But there is no perceptual similarity between the end of a table and the end of a speech.