The notion of PHASE, introduced by Chomsky (2000, 2001), has a peculiar status in current linguistic theory: on the one hand, it is widely employed in a diverse range of studies in syntax, semantics, and (morpho)phonology; on the other hand, it is notoriously ill-understood and rarely defined explicitly. Ángel Gallego is to be applauded for his attempt to illuminate the murky state of affairs with this volume, whose goal, declared in the subtitle ('Developing the framework'), ought to be embraced by all researchers in theoretical linguistics. Given the number and length of the contributions included, they can only be discussed here in all brevity.
In a foreword to the volume, NOAM CHOMSKY traces the idea of derivational cyclicity back to early work in phonology. In this domain, as well as in current syntax, phases enforce STRICT CYCLICITY: 'for certain elements X constructed in the course of the derivation, further computation should not modify X' (4). Following his earlier work, Chomsky implements this intuition in terms of the phase-impenetrability condition (PIC), which dictates that upon completion of a phase P, the complement of the head of P is TRANSFERRED to the interfaces. Chomsky reiterates his assumption that tensed clauses (CPs) and transitive verb phrases (v*Ps) are phases in this sense, but not smaller objects (contra Müller 2011 and others). Chomsky takes cases of long-distance agreement to show that the PIC must not be too strong: transferred structure, while immune to MANIPULATION, remains VISIBLE to Agree. Transfer-induced opacity in this sense is largely redundant with Chomsky's (2007) own no-tampering condition, raising the question of what motivates successive-cyclic movement (Internal Merge 'out of' a phase should still be possible, after all). Equivalently, we might say that the PIC holds for the mapping components but not for narrow syntax (assuming that Agree operates within the latter, but see Bobaljik 2008). Notably, all other contributions adopt a stronger notion of phase impenetrability.
Gallego's extensive introduction to the volume provides a thorough overview of the development of phase theory, itself a direct descendant of the bounding nodes and barriers of earlier frameworks. Gallego discusses various empirical and conceptual arguments for phases that have been adduced, noting that they fail to provide a coherent picture. Allusions to 'computational complexity', though constantly repeated, have never been made precise. Defining phase heads in terms of uninterpretable φ-features (cf. Chomsky 2008) is at variance with proposals assuming phases other than CP and v*P, such as PP. At the same time, Gallego points out, attempts to define [End Page 357] phases extensionally, via their 'interface correlates', have never been fleshed out. And while Gallego confidently asserts that 'arguments to regard CP and v*P as cyclic objects are robust' (35), even this simple typology has been contested (den Dikken 2009).
The bulk of the subsequent contributions address syntactic consequences of phase theory. SAMUEL D. EPSTEIN, HISATSUGU KITAHARA, and T. DANIEL SEELY argue in 'Exploring phase-based implications regarding clausal architecture' that the derivational precedence of theta-role assignment over Case assignment can be derived from basic principles. In particular, NPs cannot be merged first in (nonthematic) Case positions and then raise to theta positions. In this case, the initial phase in which Case is assigned will crash: since the structure must be transferred immediately upon Case assignment (valuation), the first phase will necessarily violate the theta criterion (taken to be a subcase of the full interpretation principle). This violation, the authors argue, cannot be salvaged by operations at later cycles.
In 'Phase cycles in service of projection-free syntax', HIROKI NARITA argues that a label-free syntax with phases is conceptually and empirically superior to a theory employing conventions of 'projection', such as X-bar theory (see also Chomsky 2013). Narita proposes that Merge follows the H-α Schema, according to which for any Merge(X,Y), X must be an atomic lexical item. This requirement is met just in case...