What disturbs me more is that, within the competence theory of the lexicon developed in the past two chapters, I
cannotfind a way to cleanly divide lexical items up into separable lemma and word form portions. Suppose, for
instance, a lemma is a pairing of conceptual features with syntactic features, and a word form is a pairing of
phonological features with syntactic features. The proble mis that syntactic features alone are far too coarse for
matching lemmas with word forms. For instance, as we have noted in previous chapters, the wordsdog, cat, kangaroo,
airplane, andcalendarall have the same syntactic features: [singular count noun]. Thus a match of lemmas to word
forms based on their common features will be unable to distinguish these words from each other.^105 That is, the
semantics is necessary to individuate the phonology; this is possible only in the standard tripartite organization of
lexical entries. Other possible ways of dividing the lexicon into lemmas and word forms either suffer from similar
problems or make it altogether impossible to establish linking (I leave the verification of this point to the interested
reader).
Letus seehowclosewecangettotheconclusions oftheLeveltetal.traditionwithinthepresenttheory, whoselexicon
is notdifferentiated intolemmas and word forms. At the momentin time when a lexical item is initiallyactivated, only
its semantics is of interest. The conceptual integrative processor, which binds the lexical item to part of the thought
beingexpressed, sees nothing but meaning. The next step is to use thesemantics-syntax interfaceto put some material
on the syntax blackboard. For this process, it is crucial to activate and bind the item's syntax. Then the syntactic
integrative processor can begin to work this ite minto the syntax of the utterance being built. None of the processors
invokedso far can see the item's phonology—so for all intents and pur-poses theyare workingwiththe item's lemma.
And whether or not the item's phonology is activated in long-term memory at this point is irrelevant to these
processors. Similarly, the later step of phonological integration is carried out by the syntax–phonology interface
processor and the phonology integrative processor. These can see only the relation of the item's syntax to its
phonology, and are oblivious to its meaning. Hence they are in effect working with the item's word form.
Whatthismeansis thatit may notbenecessary toreify thelemma and wordform as componentsofthelexical itemin
long-term memory. Rather, lemmas and word forms emerge naturally as functional realizations of what the various
processors can see. The fact that lemmas are invoked ahead of word forms in
214 ARCHITECTURAL FOUNDATIONS
(^105) Most treatments in theliterature evade thispointby having a“syntacticnode”[cat] thatcan serve as thelinkbetweensemanticCATand phonetic/kæt/. But, as stressed in
Ch. 5, there should be no such syntacticnode. It is onlythe custo mof drawing syntactic trees with infor mal phonologicalinfor mationat the botto mthat leads us to think
there is one.