Another apparently recent class involvingthe particleoutis totally productive. For example, if I have been knitting or
programming for the last six hours straight, I may say (40).
(40) I' m(all) knitted/progra m med out.But this odd combination is not confined to verbs. If I've drunk fourteen cups of coffee in the course of a morning I
might utter (41a), and if I've watched fourteen Edward G. Robinson movies in a row I might even utter (41b).
(41) a.I' m(all) coffeed out.
b. I' mEdward G. Robinsoned out.The model for this construction is rather clear: it is an extension oftired out, worn out, burned out, and so forth. But as in
other productive constructions, one does not need to learn which verbs and nouns can be substituted into it. What
would see mto be stored in the lexicon for this case is an idio matic construction, an 1-rule containing variables; its
syntactic for mis roughly (42a) and its meaning (42b).
(42) a.[ApV/N +−d[Prtout]]
b. ‘worn out from too much V-ing/too much N’So the question is: how does the language learner distinguish between the productivity of this case and the altogether
similar class in(39), whichremains only semi-productive?Noticethatneitherofthese patterns is especiallyfrequent;my
guess is that if anything the semi-regular one is the more frequent of the two. How does it come about that only the
pattern in (40)–(41) acquires a schema with a variable?
At the outset, of course, the child has no idea what any of the 1 the the)-rules are, so we must suppose that the child
stores everything. Taking the connectionist approach to heart, we may further suppose that, due to the characteristics
of memory, implicit relations of similarity develop spontaneously in the course of storing everything. That is, nothing
further need be said about semi-regular patterns^92
Something extra has to happen for the regular patterns. They require a further sort of learning process, one that
attempts to generalize stored items and extract explicitpatterns containing typed variables. Without such a pattern, the
language user is restricted to using stored forms and word-by-word analogy. By
188 ARCHITECTURAL FOUNDATIONS
(^92) A reminder, though: an unadorned network model is not sufficient. As Pinker and colleagues have observed, we need at the very least the notion of a discrete lexical
entry—not just a string of sounds—to which syntactic category and meaning can be associated.