event, the pattern suggested by (9) is not a strict regularity; it is shot through with holes. Some of the -ing/-inkverbs
have -ungl-unkpasts, some have regular pasts, and then there isbrought; moreover the /I/-/æ/ alternation occurs also
withswim, which isn't covered by (9).
Where does this leave a rule like (9)? Recall fro msection 3.4 one of the possible views of the rules of linguists'
grammars: that they are just epiphenomenal descriptions of implicit regularities in the language user's head, without
explicit psychological reality. Generative grammar, of course, has always rejected such an interpretation, claiming that
the rules we write are explicitly represented in the brain either as a knowledge base or in the for mof the language
processor. But such an interpretation can no longer be sustained for semiproductive rules such as (9): they now fall
back into the despised epiphenomenal status. They are not“real”rules of grammar, but only descriptions of semi-
regularities.
Always ready to turn an apparent dilemma to advantage, I would urge the view that this highlights the psychological
reality of the productive rules. In the present approach, productive morphological affixes have emerged as lexical
items. Thus“productiverules”are explicit, evenifnotexactlyintheway envisionedbytraditional generative grammar.
By contrast, there is nothing corresponding to lexical redundancy rules in the language user's head—there are only
lexical redundancies. However, this does not absolve linguists fro mlooking for lexical redundancies; it does not make
lexical redundancies “less interesting.” After all, attempts to be as precise as possible about the character of
semiproductive regularities have yielded valuable evidence about the design space of human linguistic competence.
6.5 Idioms
So far we have been investigating lexical items smaller than words, i.e. stems and affixes. As suggested at the outset of
the chapter, the lexicon also embraces itemslargerthan words, with idioms as a prime case, but also including longer
stretches of memorized text. We now take a somewhat closer look at idioms (more detail appears in Jackendoff 1997a:
ch. 7).
Itisclear thatsomething aboutidioms mustbestoredinlong-term memory. Thereisnowaytoconstructthemeaning
‘die’by combining the morphemeskick, the, andbucket. Moreover, despite a tendency among grammarians to treat
idioms as a relatively marginal phenomenon, there are in fact thousands of them—probably as many as there are
adjectives. So theories of grammatical structure and of processing ignore idioms at their own risk.
Idioms have been problematic to all the successive architectures adopted by Chomsky sinceAspects. The difficulty lies
in the character of lexical insertion: