the lexicon and discovers that it is linked to [V+past] in syntax and [PAST [EAT]]in semantics. (See Chapter 7) for
extended discussion of lexical look-up.) The relation betweeneatandate, then, is a relation between two stored lexical
items, not a derivational process.
Consider now the relationships among semiregular past tense forms. These are analogies among pairs of items, for
instancesingis tosangasdrinkis todrank. (8) makes this more explicit.
That is, the regularity is in the relation between the items, not among the items themselves. As described in section
3.3.2, this sort of regularity has been called alexical redundancy rule.
A rule that expresses this regularity might be stated very approximately as (9).
This can be regarded as a relation between two lexical entries; notice that it has variables in phonology (indicated by
“...”) and in semantics (“x”).^82 And indeed people have written lexical rules in such a format (including Jackendoff
1975; 1997a).
However, if Pinker and the connectionists are correct about semiproductive regularities, no rule like (9) is stored
explicitlyin the speaker's head. Rather, the relations in (8) are merely implicit for the speaker. They are a consequence
of the way the brain stores information associatively, capitalizing on similarity. In any
166 ARCHITECTURAL FOUNDATIONS
(^82) If (9) said“can be converted to”rather than“can be related to,”the rule could be regarded as a derivationalrule instead. Aside fro mthis, all the conditions would be the
same.