Deriving the Edge: What’s in a Phase?

Chomsky (MI, DbP, BEA, OP) has claimed that cyclic transfer to the interfaces is a property of any well-designed language system, in conformance with the Strong Minimalist Thesis (SMT), i.e. the assumption of optimal, efficient design for the satisfaction of interface conditions. The phase, as the unit of cyclic transfer, optimizes the efficiency of the mapping between syntax and the external systems in at least two ways: it reduces computational burden (memory load) via the periodic ‘forgetting’ of derivational information, yielding strong cyclicity, and eliminates redundant internal levels and compositional cycles in favour of single-cycle generation with periodic transfer to the interfaces (Chomsky 2005:4, 8), in effect replacing multiple levels/cycles (such as LF and the covert cycle) with multiple spell-out. If we accept these conceptual arguments, then some notion of phase is necessary, i.e. a principled property of the language faculty FL (a ‘third-factor effect’, in Chomsky’s terms). Nevertheless, there remains the question of formal implementation: that is, what form should phases take? Adhering to Chomsky’s conception of phases, two proposals have been made. In their original formulation (MI), phases are lexical subarrays (underpinned by an empirical argument from Merge-over-Move effects – see next section), i.e. chunks of numeration. More recently, however, Chomsky (OP, AUGB) has ceased invoking subarrays, focussing instead on the central role of the phase head itself as the locus of uninterpretable, inflectional features (uF) and the ‘engine’ of the derivation. In deciding between these two alternatives, the latter approach would seem to conform better to SMT, yielding a simpler, less stipulative notion of phases. Consider the Phase Impenetrability Condition (PIC). PIC is a corollary of Chomsky’s notion of “meaningful cyclic computation” (BEA:5) in that transfer of the entire phasal category (CP/vP) at the phase level would preclude continuation of the derivation beyond the immediate phase: displacement out of phases would be barred, the edge feature EF on the phase head would be inaccessible for further selection (compromising iterability of merge), and crash would ensue at every phase level due to unvalued agreement features uF. To avoid massive undergeneration, some notion of edge is thus implied, by which the syntactic life (accessibility) is extended of uF within the phase and of EF on the phase head (label). In short, some notion of edge (and thus PIC) comes for free on the ‘forgetting’ conception of phases, with no departure from SMT. Despite this, the edge under the subarrays view of phases was one of many properties directly encoded in the definition of PIC itself, which thus has an arbitrary, stipulative flavour in the MI and DbP formulations (see, e.g., Abels 2003, Boeckx & Grohmann 2005 for relevant criticisms). In addition to the specific provision of a particular form of the edge (i.e. the accesibility of the head and specifiers of the