How does the lexicon capture this redundancy, this“borrowing”? One approach, which might be associated with
Chomsky's viewofthelexiconas a repositoryofalland only theexceptionsofthelanguage, wouldattempttobleedall
the redundancy out of (34) and encode it as a minimal skeleton from which everything else could be predicted. The
difficultywith thisapproach is in determining whatcould possiblybe lefttoconnect toCRITICIZE, such thatonecould
predict it would be pronounced“take to task.”Jackendoff (1997a: section 5.6) shows, for a different case in the same
spirit, that it is impossible to take all the redundancy out and still retain a sufficiently coherent structure. There is no
place one can draw the line and say“These redundant parts we will take out, but these we will leave in in the interests
of coherence.”I conclude that this approach cannot be carried out technically, whatever its intuitive appeal.
An alternative in the literature (Pollard and Sag 1994; Goldberg 1995; Michaelis and Lambrecht 1996) appeals to
“inheritance,”whereby a more highly specified or highly structured item“inherits”structure fro mless specified or less
structured items. Under thisapproach, wewouldsay that(34)inherits mostofitsstructurefrom (35a–e);but notallits
structure is inherited, since the syntax–semantics links in (35c–e) are overridden. Similarly,wentinherits its structure
fromgoandpastbut overrides theexpected phonology–syntaxconnection.(Thiswouldalso betrue ofsang, drank, etc.)
A non-idiomaticstoredphrasesuch ashappy birthdayand a storedregular verbsuchasdreamed(whichmustbestoredin
order to compete withdreamt) of course inherit everything. The intuitive idea, then, is that the more an item inherits
fro mother stored ite ms, the si mpler it is to store.
Inheritance is actuallya specialcaseoftaxonomiccategorization, oftenexpressed in terms ofa semanticnetworkalong
familiar lines like (36).
Each elementin thehierarchyinheritsbydefault allpropertiesoftheelementsthat dominateit. Theinheritance is only
a default because, for instance, although birds typicallyfly, penguins don't. Semantic networks in the literature usually
expressa stricttaxonomylike(36).But itisalsopossibletosetupnetworkswithmultipleinheritance,for instance(37),
wherepetis an alternative classification ofanimal, orthogonal to themammal-bird-fishdimension.