Topic: Syn-Sem Interface
ON THE NOTION OF PHASE
In the following I shall show that there is not one single notion of phase in Chomsky (2001) and that each of the different formulations have diverse consequences.
The set of analyses and conclusions delineated in my paper (forthcoming) and Chomsky?s (1999) proposals converge to the notion that the phases of a derivation are propositional. Indeed, as Chomsky himself acknowledges, verbal phrases have full argument structure and CPs force indicators. However the same notion of phase in Chomsky (2001) is derived throughout other four different paths.
The first path to derive the notion of phase, which Chomsky has chosen, departs from the inclusiveness condition. He maintains his 1995 idea that the input of a derivation is an array of items taken from the lexicon (LA), but divides it into sub-arrays. Each phase is determined by a sub-array LA{ i } of LA, placed in the active memory. When the computation exhausts LA{ i }, forming the syntactic object K, L returns to LA, either extending K to K' or forming an independent structure M to be assimilated later to K or to some extension of K.
These premises seem to pervade Chomsky?s thoughts, but in discussing what categories demark the phases, i.e., how to label them, he presents a second view, which furthers the first vie. So, he argues that a sub-array LA
MT(1) Phases are CP and vP, and a sub-array contains exactly one C or v.
The third alternative to circumscribe phases consists of finding and summoning facts that maintain and correlate PF and LF integrity more generally, what Chomsky claims to be independent support:
MT(2) CP and vP are reconstruction sites, and have a degree of phonetic independence.
The fourth way consists of adopting the idea that the main functional categories, which may have an EPP feature and so function as targets for movement, are phases, but there are divided accordingly to their strength:
MT(3) CP and vP are strong phases, all the others are weak.
Accordingly to the thesis embraced, the spell out operation will apply simply at a phase or at a strong phase level. The options might all seem very stipulative, put this way, but they all play a crucial role in Chomsky?s consistent argumentation in favour of the idea that only the phonological component proceeds in parallel, and not the syntactic and semantic ones. This is expressed in the minimalist thesis bellow:
MT(4) a. There is no overt-covert distinction with two independent cycles; rather, a single narrow-syntactic cycle;
b. The phonological cycle is not a third independent cycle, but proceeds essentially in parallel.
We have already explained the distinction between delete and erase in Section 1, which underlies the reasoning above: by making the deleted features disappear, convergence at LF is allowed. In order to reduce computational burden and allow the phonological component to slight earlier stages of the derivation, another stipulation, the phase impenetrability condition, is added:
(PIC) Only H and its edge are accessible to operations outside a strong phase HP.
The edge being either specs or elements adjoined to HP. Under PIC, operations apply at the accessible elements of HP within the smallest strong ZP phase containing HP and not beyond, and the phonological component spells out elements that move no further. But this means that spell out interprets H and its edge as parts of ZP in a structure like:
(1) [ZP Z... [HP _ | [H YP]]]
Which leads to Ev1, already mentioned.
Let us summarise the dilemmas resulting from these views:
D1 a. A phase is the product of derivational procedures that divide lexical arrays, obeying the inclusiveness condition.
Versus
b. A phase is a special functional selector category.
D2 a. The motivation to postulate the existence of phases is semantic: they are characterised as propositional and as reconstruction sites.
Versus
b. The motivation to postulate the existence of phases is phonological: they are characterised by the cyclic application of spell-out.
D3 a. Functional categories are divided into phases and non-phases.
Versus
b. All functional categories are phases. Phases are divided into weak and strong types.
Neither alternative in the third dilemma is actually good, for phase should be maintained as a derivational notion and not treated as categorial one, by which reason (D1b) must also be discarded. Perhaps (D2) can never be solved, but it permits us to sustain the view that the notion of phase rests upon the application of an operation.
Indeed it has already been argued for the elimination of PIC in the literature. For instance, Stjepanovic' and Takahashi (2003) claim that the effects of PIC can be derived from independently necessary computational mechanisms, such as multiple spell-out and pied-piping. But any arguments in this direction also provide the means to derive the notion of phase from the application of spell-out. If, on the other hand, spell-out is not the only correspondence operation that maps the Syntactic Structure onto a representational level, i.e., if spell-out has a sister reading operation that interacts with the semantic level, then the two operations may be ordered, the application of second operation culminating in the formation of a phase. The only argument for assuming that the second correspondence operation is reading is the empirical observation that the Phonetic Structure seems always to be delayed in relation to the Conceptual Structure.
So, we have the point that PIC is not necessary, for it can be a mere effect of the cyclic application of different operations, as well as a consequence of a version of Godel?s diagonal lemma. Nevertheless, (D2b) also reveals the weak point of the narrow syntax hypothesis: it does not explain why the phonological cycle should essentially proceed in parallel, as stated in MT(7), but the semantic one should not.
(D1a) is the truly principled alternative among all others.
Posted by Tony Marmo
at 00:01 BST
Updated: Friday, 22 April 2005 01:44 BST
