On Feature Inheritance: An Argument from the Phase Impenetrability Condition

This squib is organized as follows: the first part (section 1) identifies a potential conceptual flaw in the feature inheritance model of phase theory proposed in Chomsky 2005; the second part (sections 2-4) suggests a revision that removes this flaw.

1 Feature Inheritance, (Non)phase Heads, and the Strong Minimalist Thesis

Chomsky (2005; henceforth OP)¹ proposes a reinterpretation of the relation between the functional heads C and T: the Agree (φ-) and Tense features associated with the inflectional system are not an inherent property of T; instead, they belong to the phase head C. Traditional subject agreement and EPP (Extended Projection Principle) effects associated with T (A-movement of the formal subject to Spec,T, expletives, etc.) then arise via a mechanism of feature inheritance, whereby uninterpretable features are passed down from the phase head to its complement. It follows that T lacks uninterpretable features unless it is selected by C. That is, T is no longer a probe in its own right; it cannot initiate operations directly or independently of C.

Clearly, in this way, feature inheritance captures the long-standing observation that raising/ECM (exceptional Case marking)-infinitival T, which lacks C, also lacks φ-features (failing to value Case on DP) and independent tense (see MI:102, 105, BEA:13, OP:9). However, where the previous system had to stipulate this connection by means of a selectional restriction (C selects φ-complete T; V selects φ-defective T), the feature inheritance model offers an arguably more explanatory account of T's featural dependence on C: the features are simply C's, not T's. This, in turn, allows a uniform characterization of phase heads (C, v*) as the locus of uninterpretable features, as is desirable on computational grounds (see section 2).

Nevertheless, there is a major hurdle to be overcome before feature inheritance can be accepted as a principled property of the language system. Whereas T now clearly needs C in order to function at all (e.g., in the Agree system), it is not at all obvious why C needs T in this feature inheritance model. In short, why should feature inheritance (and indeed T itself) exist at all? In order to conform to the Strong Minimalist Thesis (SMT), feature inheritance must be forced by considerations of good design. Unless it provides the optimal means of satisfying an interface condition or facilitates computation in some other way, feature inheritance is a redundant, extra device, a departure from optimal design that must be attributed to the unexplained residue of Universal Grammar. The challenge, then, is to find a principled reason why the Agree feature cannot simply remain on C.²

To this end, Chomsky (OP:9-10) offers an argument from the conceptual-intentional (C-I) interface: feature inheritance follows from the C-I-imposed requirement that the A/Ā distinction be structurally established. Spreading of Agree to T enables a structural distinction to be made between the A-position created by movement for C's Agree feature (Spec,T) and the Ā-position created by movement for C's edge feature EF (Spec,C, the phase edge). Further, once motivated in this way for the C-T relation, feature inheritance may reasonably be expected to hold of phase heads in general, on grounds of optimal design (OP:14). Bearing out this prediction, Chomsky notes that spread of Agree features from the v* phase head to its complement V, with A-movement of the object to Spec,V analogously with subject raising to Spec,T, immediately yields the famous "raising to object" paradigm that obtains with ECM, in which the embedded subject can bind into and take scope over matrix adverbials, indicating raising to a position in the matrix vP (see Postal 1974, Lasnik and Saito 1991, among many others, for examples, discussion, and alternative analyses). This movement has previously seemed unmotivated from the economy perspective, but now falls straightforwardly into place.

Although these arguments are suggestive, unfortunately neither has the force of necessity. The C-I-imposed A/Ā requirement would seem to imply only that we need two different types of features: the Agree...