mostly) conform to rules of grammar. Therefore they invoke the language faculty in their production and
comprehension. Like individual words, there is no other faculty that they necessarily invoke. Moreover, there is no
principled dividing line between little clichés such ashappy birthdayandsigned, sealed, and delivered, larger units such asthe
bigger they are, the harder they fall, stilllarger unitssuch as song lyrics, and monsters such asHamlet. Themaindifference is
that the units we call clichés will tend to be in everyone's long-term memory, but linguistic examples, poems, song
lyrics, and plays arelikelytobeknownonlybysomesmallproportionofspeakers. Inanyevent,thelonger passagescan
be constructed online (say when hearing the mfor thefirst time) but need not be. (For documentation of the number
and variety of clichés and otherfixed expressions, see Jackendoff 1997a: ch. 7.)
The fact that memorized texts must be stored as linguistic structure is largely overlooked, thanks to a widespread
stereotype in linguistic theory (and philosophy and the popular conception of language) to the effect that the
memorized units of languagearewords. Infact,theterms“word”and“lexical item”are oftenused interchangeably. For
instance, Chomskyintroducestheterm“lexicon”inAspectsroughlytomeantherepositoryofallthewordsthespeaker
knows. Following Bloomfield (1933), he points out that the lexicon must contain all the exceptional, non-predictable
features of words. But there is scant attention to larger stored units.
Chomsky further stipulates, without argument, that the lexicon containsonlynon-predictable features, that is, that it
contains no redundancy. (It is not clear to me that this was Bloomfield's intention, though.) Here Chomsky likely
deviates fro mpsychological reality: there is no reason to think that the brain stores infor mation non-redundantly. We
return to this issue in sections 6.5 and 6.8.
To make thedistinctionclearer,I willuse theterm“lexicalitem”exclusivelytodenotean itemstored inthelexicon,i.e.
in long-ter m me mory. I will use the ter m“word”for quite a different notion, based in grammatical theory. In
phonology, a phonological word is a constituent such as those dominated by“Wd”in Fig. 1.1; it is a domain over
which certain segmental and prosodic constraints are defined. In syntax, a roughly corresponding notion is that of an
X°(or so-called lexical) category such as N, V, A, or P. Di Sciullo and Williams (1987) point out the distinction
between grammatical words and what are here called lexical items—they call them“listemes”—and they declare from
theoutset(p. 1) thatthetheory of listemes is“of no interestto thegrammarian.”By contrast, I wish totake thetheory
of listemes quite seriously here: it corresponds precisely to our question of storage vs. online construction. We willsee
that it is indeed of considerable interest to the grammarian, despite being on the face of it an issue of“performance.”