contrast, when a pattern with a variable develops, the relation among the stored items“goes productive”: the pattern
can participate in free combination with other lexical items that satisfy its typed variables, and new combinations need
no longer be stored.
As we can see fro mthe range of pheno mena treated in this chapter, such an approach can apply to all scales of the
grammar, from syllablestructure to phrase structure, withmorphology in between. It applies as wellto the learning of
idioms and constructions.^93 That is, as Culicover (1999) puts it, rule learning is accomplished by the same process as
word learning—because both are types of lexical item. The main difference is the extent to which specific material is
replaced by variables.
The process of extracting a pattern that can“go productive”is information-theoretically equivalent to the traditional
conceptionof learning a rule.Thesame questions about thedistribution oftheinput arise, thesame constraintson the
child apply. The major difference is in the formal apparatus. Here, stored items and rules are of the same formal
character; there is no episte mological divide between the for mof stored ite ms and the for mof rules. The only
innovation necessary in the learning theory is a way to learn variables fro minstances—which is needed in any event.
There is, however, an interesting substantive difference between this approach and the traditional approach to rule
learning. Up to now I have described the difference between a regular and a semi-regular pattern as a binary choice:
either a pattern with a variable in it is present in the lexicon (regular) or it is not (semi-regular). Actually, the existence
ofan 1-ruleinthebrainismorelikelya matterofdegree.Theeaseor speedwithwhichan 1-ruleisactivated relativeto
stored forms undoubtedly plays a role in how freely productive it is in performance. Individual words are only
gradually established in the child's vocabulary,first in passive command, then active. We should expect the same of
productive patterns. And this is indeed what we observe: although the stereotype in the literature is of a
LEXICAL STORAGE VS. ONLINE CONSTRUCTION 189
(^93) The learning ofconstructional idioms may callfor some comment. Section4.7 spoke oftheoverwhelming problems posed bytheacquisitionof wordmeanings. Everyone
wouldacknow-ledge thatidioms havemeanings justaboutas idiosyncraticand complexas thoseofwords, so thesame problemsariseinlearningthem.Inthepresentstory,
it is only a short step from the problem of learning idioms to the problem of learning constructionalidioms. And the meanings of constructional idiomsare also about as
complex as the meanings of words. Consider for instance thetime-away construction, which means roughly‘spend time wastefully or frivolously doing some-thing.’This
meaning is comparable in complexity and subtlety to, say,procrastinate,‘spend time wastefully in order to avoid doing something’, and so it should not be more difficult to
learn (howeverchildrenmanage to do it!).On theother hand, thegrammaticalstructure ofa constructionmightbemoredifficulttolearn thanthatofa wordbecauseithas
more variables in it and less overt phonology to mark its presence.