Why GB Theory Alone Does Not Explain CS
The problem with these proposals based on GB theory is that they are operating at a level which is too 'purely syntactic', or too close to the surface. All approach permissible arrangements of items using a level of analysis where relatively low-order phrase structure is involved. I will argue that CS constraints come into play at a much more abstract level, which is presyntactic in that it involves such issues as form class membership and the specific subcategorization constraints this membership entails, not just 'subcategorization' in some general sense. Clues about possible constraints come from looking at models of how language is accessed and retrieved before it takes its final surface form. These models of language production will be discussed in the next chapter.
If the constraints question were put in 'language production' jargon, it would be: what mechanisms support the retrieval, directed by lemmas in the speaker's mental lexicon, of some morphemes from the ML, but some from the EL during normal, continuous speech? It has been shown that the earliest theories concerning constraints on CS were based on examination of the juxtaposition of ML and EL elements or the susceptibility of specific form classes to CS. Later, others tried to reach the answer by placing CS under constraints similar to those which GB syntactic theory posits for monolingual syntactic structures.
A review of the major proposals shows that they do not answer the question, as posed, very successfully. The proposal which comes the closest to succeeding ( Joshi, 1985) starts from a quite different perspective, by sketching a hypothesis from the standpoint of the production process. This line is also pursued by my associate, Shoji Azuma, and myself in formulating a CS model; Azuma's model, however, takes a somewhat different tack from the MLF model. My version of such a model will be presented in Chapters 4 and 5 as the MLF model.