If NP Ellipsis is really syntactically controlled, we would expect to be able to find evidence of internal structure in the ellipsis site. There are at least two ways this could be done: finding anaphors which are bound from within the ellipsis site, and finding WH-words or other moved elements which must have a trace within the ellipsis site.
In addition, we might expect the pragmatically controlled instances of quantifiers and deictics to pattern with other pragmatically controlled ellipsis processes; there should be no evidence that these ellipsis sites have internal structure.
Grinder & Postal (1971) argued that VP Ellipsis must involve a syntactic deletion rule, on the basis of sentences containing ellipsis and a pronoun which (apparently) had no antecedent on the surface. The pronoun found its antecedent inside the VP Ellipsis site; this process is known as the ``missing antecedent'' phenomenon.
To test whether a DP can find its antecedent within an elided NP, we need a very particular type of structure. We must have an elided DP and a suitable antecedent; the elided DP must itself contain a DP which can serve as the ``missing antecedent'' for the pronoun; and so the antecedent for ellipsis must also contain a DP. But we must be careful to construct the antecedent for ellipsis so that it contains a DP which is not a good antecedent for the pronoun. We will need a structure like the following:
NP is the elided NP, and NP is its antecedent. A pronoun takes DP, within the ellipsis site, as its antecedent. Crucially, DP must not be a valid antecedent for DP. I will now construct a test example. First, a DP is established containing a DP which is not an available antecedent for a following pronoun.
Next, a second DP with an elided NP, DP, is added. The ellipsis site presumably contains a DP which may be a valid antecedent. The pronoun can now find DP as its antecedent.
These data are consistent with the hypothesis that the ellipsis site in NP Ellipsis has internal structure.
If quantifiers and deictics license pragmatically controlled ellipsis, then their ellipsis sites should have no internal structure, and thus we will not expect them to be able to host antecedents for pronouns. This does in fact seem to be the case.
If the ellipsis site has internal structure, then it should be able to host a trace of WH-Movement as well. For these the structure we want is the following:
||. . .||
||. . .||
We require the DP with a trace in its ellipsis site to be possessed; and in almost all cases the possessor blocks WH movement out of the DP. In the cases where extraction is merely questionable, ellipsis then seems ungrammatical.
So we may not be able to test for internal structure of the ellipsis site in this way. Extraction from a quantified DP is similarly ungrammatical, with or without ellipsis.
So we can conclude nothing about the internal structure of the ellipsis sites from cases of extraction from quantified and possessed DPs with missing NPs.
This data raises a final interesting point. Sluicing is well known to repair island violations (Merchant 2001, among others) while VP Ellipsis does not. This difference between two similar ellipsis processes has generally been approached with the burden of explanation on what aspect of VP Ellipsis explains its failure to repair island violations. This data shows that NP Ellipsis patterns with VP Ellipsis, potentially shifting the burden of explanation onto the task of finding what aspect of Sluicing allows island violation repair instead.
For these two types of reduced DPs, I propose the following phrase structures. The structure I want to propose for the bare quantifiers consists only of a quantifier which has no NP complement in the syntax.
I have proposed that bare quantifiers are instances of Null-Complement Anaphora, meaning that the quantifier's semantic argument is recoverable from the context.
It is important not to confuse this analysis with another possible analysis, that of considering the quantifiers optionally transitive Ds (in the same way that intransitive eat has no syntactic complement and no semantic argument). H&S point out that there are two pitfalls in this analysis. First, if some were optionally transitive like eat, then it would be infinitely ambiguous between some, some disgusting chocolates, some Belgian chocolates, some Belgian waffles, and so on. This is clearly undesirable. Second, Null-Complement Anaphora obeys the Backwards Anaphora Constraint, like other anaphora:
So this structure, although it may look like a quantifier's intransitive use, is actually a quantifier with no syntactic complement, but with a semantic argument, the restrictor for the quantification, provided by the context. Better analogs of this in the verbal domain are the verbs approve, volunteer, refuse, succeed, and so on.
For NP Ellipsis, I will propose that the DP consists of the possessor DP in the spec of D, and some genitive D head. I will remain agnostic as to whether this genitive D head is realized by 's or whether it is null and requires 's on its spec. What is important is that the NP was present at some point, for example to provide an antecedent for ``missing antecedent'' anaphora, and then was elided under identity with some other NP.
In line with the DP hypothesis, this structure leaves open the possibility of an eventual unification of the structures involved in VP Ellipsis, Sluicing, and NP Ellipsis. Lobeck (1995), among others, has observed that each of these structures involves a overt specifier agreeing with a functional head, and has hypothesized that this configuration licenses ellipsis. If this configuration did license ellipsis, it would explain why bare quantifiers do not pattern with NP Ellipsis, and why NP Ellipsis demonstrates the same patterns as VP Ellipsis. I will return to this hypothesis when I discuss Lobeck's account of NP Ellipsis.
Finally, if (possibly cross-linguistic) evidence exists for a third separate functional projection in DP, between D and N, which hosts quantification or number, then the quantifiers likely sit in this projection and not in D, making the split more clear; the overt D 's licenses ellipsis, and the non-overt Ds which appear with quantifiers do not, just as VP Ellipsis is only permitted in the presence of an overt auxiliary.
There is evidence for internal structure in the site of NP ellipsis in precisely those cases of NP ellipsis involving a stranded possessor, which seem to be syntactically controlled. Surprisingly, there is not evidence for internal structure in those cases of NP ellipsis involving a stranded or lone quantifier or deictic, which seem to be pragmatically controlled. The phrase structure I propose for the two different processes is compatible with the observed facts and points to a future unification of three syntactically controlled ellipsis processes.