6.2.3 The necessity of a heterogeneous theory
The conclusion is that the term“lexical rule”has been applied to twodistinct phenomena. The principles that express
semiproductivegeneralizations are indeed lexicalrules: they relatelexicalitems tooneanother,but speakers must learn
onebyonethelexical items that therulerelates. Thatis, theruleexpresses thepossibilityofrelated forms butnottheir
existence or their details. By contrast, productive generalizations result fro mthe free co mbination of individual
morphemes that are stored as separate lexical items. (This does not preclude high-frequency regular combinations
being stored as well, and there is some evidence that they are, e.g. Baayen et al. 1992; Baayen etal. 1993; Booij 1999;
Baayen et al. 2000).
On this proposal, then,devouris both a word and a lexical item; -dis a lexical ite mbut not a word; the regular for m
devouredis a word but not a lexical item, being made out of the lexical itemsdevourand -d; and the irregularateis both a
word and a lexical item.
We can formalize thistreatment of affixationalong thelines used for wordsin section 5.9.3. (5) shows theregular past
tense.
Like an ordinary word, this is an interface rule with feet in all three components: it is a piece of phonology linked to a
piece of syntax and a piece of semantics. (On the other hand, some affixes, such as nominative and accusative case in
German, have no semantic effects, so theyare“defective”in the same way as expletiveitand defaultof.) Just as a verb
has variablepositions for itsarguments, thepast tense affix has a variable positioninphonology, syntax, and semantics
for the verb to which it attaches.
(5) combines with a regular verb such asdevour(Ch. 5, (51a)) to for mthe co mposite structure (6). (If phrasal syntax
and morphosyntax are separate subcomponents, as suggested in section 5.6, this combination is accomplished by the
latter.)