Thestored irregularatehas thesame syntacticand semantic structures as (6), and thesame connectionsbetweenthem,
but a different connection to the phonology: only the unit as a whole is coindexed with a phonological constituent.
This is exactly what it should mean to say the form is irregular.
What may seem unintuitive here is the inhomogeneity of the solution. Two forms with the exact same syntactic
properties, for instancedevouredandate, may havedifferentsources: oneis a productof free combination and theother
comes from the lexicon.^78 In fact, most morphological theories have tacitly assumed that theremustbe a uniform
solution for both cases. The result has been a tension between“morpheme-based” or “item-and-arrangement”
theories such as Chomsky and Halle (1968) and Di Sciullo and Williams (1987) and“lexeme-based”or“word-and-
paradigm”theories such as Halle (1973), Jackendoff (1975), Aronoff (1976), Beard (1987), and Anderson (1992).
Morpheme-based theories are perspicuous for productivemorphology and spend most oftheir effort tryingtoexplain
semiproductivephenomena. Lexeme-based theories are good at semiproductivephenomena, wherewhole wordshave
to be listed, but theyforce one to believe,for instance, that the entire Turkish verb paradig mis stored for every verb.^79
LEXICAL STORAGE VS. ONLINE CONSTRUCTION 161
(^78) Theydohoweverhavedifferentcombinatorialpropertiesinatleastonerespect.Pinker (1999 and referencestherein)pointsoutthatirregular pluralforms,whicharelexical
words, enter more readily into compounds than do composed plurals, whichare not lexical words. For instance, children spontaneously produce bothmouse-eater andmice-
eater, but produce onlyrat-eater, not ??:rats-eater. Evidently morphosyntax is sensitive to the distinction, but phrasal syntax is not.
(^79) The discussion here has implied that all morphology can be analyzed as affixation. To keep myself honest, I have to mention two kinds of exotica here.Anderson (1992)
argues against the view that regular affixes can be treated as independent lexicallistemes. His reason is that many morphologicalchanges cannot be treated as adding parts.
The clearest sort of case he presents is truncation, when a morphologicalprocessremovesmaterial from the stem, as in pairs like Englishvirus/viral (virusal), charity/charitable
(charityable). These examples are drawn from semi-regular patterns; however there are also cases of productive truncation morphology (some citations in Blevins 1999
Blevins Blevins). Within the present approach, I suggest that we take a chapter out of the analysis of visually presented shapes, in particular entities like holes and cracks.
These see mto be conceptualized (Landau and Jackendoff 1993) as“negative parts”of objects—instead of adding material to the object (like a lump or a handle) we have
scooped material out. Whatever innovation must be added to the theory of visual form for such cases might be adapted to phonological structure too: we could treat a
truncation process as, say, a suffix with the morphophonological form“negative consonant.”In the course of correspondence to phonetic form, the combination of this
with the preceding word would result in the“scooping out”of thefinal consonant of the word to which it is affixed. This is only a preliminary suggestion, and serious
consideration is required for these cases, as well as for the more complex cases such as morphological metathesis that Anderson (1992) offers as counterexamples to the
additivetreatmentofmorphology.Another“exotic”and heavilystudiedcaseisreduplication, inwhichaffixationresultsina copyor partialcopyofpartofthewordtowhich
it is affixed (Marantz 1982); Moravcsik 1978); Spencer 1991). An example is Mandarinjang (‘sheet’)/jang-jattg(‘every sheet’),ren (‘man’)lren-ren(‘every man’), etc.Reduplicative
morphology is widespread in the languages of the world, associated in different cases with different meanings. For example, reduplication on nouns may be used to mean
plurality, universal quantification, diminutives (‘little X’), augmentatives (‘big X’), and‘Xs and things like that’on verbs, to mean past tense and continuative aspect (‘keep
doingX’); on adjectives, to mean‘extremely.’An interestingcase from contemporary colloquial English (Ghomeshi et al. 2001) attachesto any syntactic category and means
roughly‘mostprominentinstanceofthecategory’:a hostesssays,Wouldyou like wine, or would you like a DRINK-drink?Sincereduplicationis productiveinmanyofthesecases,
thepresent theory says there must bea“reduplicativemorpheme”listedin thelexiconthatcancombinefreely withother words. Whatphonologicalcontentcanpossiblybe
giventosucha morpheme? One possibility, roughlyfollowingMarantz (1982) and McCarthyand Prince (1986), is a“metaphonological”instruction“REPEAT”thatcanbe
part ofan item's stored structure, accompaniedbya pointer towhat is toberepeated.(Inpartialreduplication, themorpheme wouldhavesome specified contentthatcould
not be overlaid by the realization of REPEAT (Alderete et al. 1999).)