speaking is a natural consequence of the logic of the processor. This is very close to the Leveltstory; it remains to be
seen if the differences are more than rhetorical.^106
7.4.4 Speech errors and tip-of-the-tongue states
As early as Lashley (1951) wefind a basic accountof the logic of speecherrors. Consider Reverend Spooner'sour queer
old deanin place ofour dear old queen. Lashley points out that, in order for thekwsound to be inserted in the intended
worddear, it must already be present in working memory, waiting its turn for phonological integration. Lashley uses
thistoarguethat speechcannotbe producedbysimplechaining ofwordsonetoanother; rathersomeoverallplanning
must be going on in advance.
An observation by Victoria Fromkin (1971) sets up speech errors as a paradigm for studying modularity in language
processing. Fromkin points out that when morphemes are exchanged, they are treated phonologically as if they are
completely normal in their new environment. Consider an example likeI'd hear that if I knew it, uttered when what was
intended was I'd know that if I heard it. The exchanged verbs appear in the appropriate for mfor their new
environment—includingknowtaking onitscharacteristicirregular past tenseform. This suggeststhattheexchangehas
taken place in the course of syntactic integration, where the tense is still independent fro mthe verbs. It is only at the
stage of phonological integration that the tense is combined with the verbs into the
IMPLICATIONS FOR PROCESSING 215
(^106) I should address one important experimental result. Van Turennout et al. (1998) show that in a picture-naming task, syntactic information about a noun (its grammatical
gender) is available a tiny amount of time (40 milliseconds) ahead of phonological information. This is taken as evidence for successive lemma and word-for mstages in
production. In the present story there are twopossible interpretations. Thefirst invokesthelogic of processing. A spoken response cannot be produced withoutintegrating
the lexicalitem in questionintoworkingmemory. And perhaps syntactic integration must precede phonological integration evenin a picture-namingtask, where the output
is a single word.A second interpretation invokes the neural instantiation of processing. Consider what it takes to activate an item's syntax and phonology in long-term
memory. We havebeenassumingthatwhenan item'ssemanticsisactivated,itssyntaxand phonologyfollowsuitimmediately;letus nowgo beyondthisfirstapproximation.
A lexicalitem's syntacticstructure is far sparser than its phonologicalstructure: a part of speech plus a few inflectional features and perhaps a syntactic argument structure.
The phonological structure, by contrast, has to spell out all the distinctive features of every segment. This might well have a consequence for activation:the syntax, having
fewerfeatures, in effecthas less“inertia,”and thereforemightwellrise abovethreshold activationfaster thanthephonology.Eitherofthese interpretationsmakes it possible
to imagine the Van Turennout et al. results without invoking a lemma/word form split in the lexicon itself. Sorting the possibilities out experimentally is a challenge for
future research.