Theory of Pied-Piping
In recent theorizing (Chomsky 2000, 2001, 2004), the movement operation is decomposed into three parts, consisting of Agree, pied-piping, and Merge. Pied-piping is the process that creates a copy of some cate gory that contains the goal of the Agree operation. The copy is then merged to become a specifier of the projection whose head contains the probe and the EPP feature. In this chapter, I would like to propose that pied-piping also makes use of a formal feature, which I call the pied-piper feature.
The pied-piper feature resolves the technical problem that accompanies the decomposition of Move into three suboperations. Pied-piping must copy some phrase that contains the goal for Agree. But the Agree operation renders the goal inactive, making it inaccessible for further computation in narrow syntax including pied-piping. Pied-piping cannot take place before Agree, however, because the goal has to be located by Agree. The pied-piper feature saves this situation. It marks the category to be copied by pied-piping.
The pied-piper feature also helps reduce the workload of the PF computation. Suppose that the pied-piper is hooked onto the phonological features of the category to be copied. This is plausible if Chomsky (2001) is right in claiming that pied-piping requires phonological content—in other words, that the category to be copied must have phonological con tent. (See also Takahashi 1997, 2000.) Suppose further that the pied-piper feature is retained in the copy, but not in the original, as in (1), where YP is raised. F(PP) is the shorthand for the pied-piper feature.
It follows that the PF materials are retained only in the newly created copy. In other words, the pied-piper feature is the device that determines pronunciation of a chain before Spell-Out without forcing the computational system to handle phonological features directly in narrow syntax. The PF computation is not called upon to make the decision for pronunciation anymore, even though we assume that deletion of phonological content of the original takes place during the PF computation.1 The idea here is that pronunciation of a chain and the need for phonological con tent in pied-piping are tightly connected.
Let us consider what happens if the decision for pronunciation is not made in narrow syntax in the model of cyclic Spell-Out proposed by Chomsky (2000, 2001, 2004). Suppose HP is a strong phase. The complement of H is sent to the phonological computation after Spell-Out. Examination of the complement of H alone does not tell us whether there is a copy remaining in the next higher phase. Take the case where H is an interrogative C that agrees with YP, as in (2).
In that case, the original YP within the complement of C no longer has an active uninterpretable feature at the point of Spell-Out. Thus, it is indistinguishable for the PF computation from a phrase ZP that is to be pronounced within the complement of H after being rendered inactive by agreement with a head below H, as in (3), where H is v in (3a) and C in (3b).
To make sure that YP within the complement of C will not be pronounced in (2), it would be necessary to look at the material to be sent to PF at the next strong phase (namely, the copy in Spec,HP = CP) as well. Thus, cyclic Spell-Out itself does not alleviate the pronunciation problem. It is the pied-piper feature that allows the decision for pronunciation to be made in a very local fashion, as part of pied-piping. In this sense, the use of the pied-piper feature is a desirable design specification for the human language faculty.
