Abstract: . . .  argument structure of the sentence to form what Garrett refers to as the 'functional level representation'. Speech errors such as This spring has a seat in it (for This seat has a spring in it ) where the exchanged words are of the same grammatical category can be interpreted as errors in the assignment of words to slots in the functional level representation. In the second stage, syntactic encoding procedures generate a syntactic planning frame which contains slots for the content words specified in the functional representation. These slots are also carry diacritic markers for tense and number, and so on. The phonological forms of the relevant lemmas are then inserted into the relevant slots in the planning frame. This explains why when words exchange, they are appropriately . . .
. . .  support for the idea that syntactic information is associated with the lemma. Levelt et al. (1999) provide further arguments for positing a lemma level of representation. However, this assumption has been contested by Caramazza & Miozzo (1997) on the basis of data from tip-of-the-tongue experiments and aphasics (but see Levelt et al., 1999, p.66, for a response). There has also been debate over whether the conceptual representations which are input to the production process should be specified in terms of sets of primitive features or in terms of lexical concepts which bear a one-to-one relationship to lemmas. Levelt et al. (1999) favor the non- decompositional approach on both theoretical and empirical grounds. They argue that "lexical Page 20 Psycholinguistics 20 concepts form the terminal vocabulary of the speaker's message construction" (p. 8). This implies that a good deal of language-specific conceptual processing needs to be done to package the intended message in such a way as it can be fed to the production process; what Slobin (1996) referred to as "thinking for speaking". At the level of form retrieval, there is convincing evidence that the phonological form of a word is not simply retrieved as a whole unit, but rather that it is constructed, or "spelled out", by inserting sub-syllabic units into syllabic frames (Levelt, 1989; Levelt et al., 1999). Speech error data have traditionally provided the strongest evidence for this assumption. When sounds exchange between two words . . .
--3000,2,750,2450,59797