The most basic question to consider about NP Ellipsis is the issue of whether the antecedent for the ellipsis can be any sufficiently salient entity in the context, or whether it must be a linguistic antecedent. Hankamer & Sag (1976) argue that do it can be pragmatically controlled, but VP Ellipsis, including VP Ellipsis stranding do, must be syntactically controlled.
Example (a) shows that VP Ellipsis cannot have even a very salient event in the context as its antecedent, if that event has not been explicitly (linguistically) mentioned. Example (b) shows that this is not the case for the sentential it. Pronouns in general can find antecedents in the context. We can see that the presence of a linguistic antecedent is crucial for VP Ellipsis here, by looking at perfectly grammatical examples like this:
We can now look for similar situations within the noun phrase.( In the following examples, I will notate what may or may not be an ellipsis site with _, and enclose the entire DP with brackets. However, in cases where there appears to be pragmatically controlled anaphora there may not be an ellipsis site after all.)
First, notice that bare quantifiers are amenable to pragmatic control:
The ellipsis is interpreted in (a) as eggs, in (b) as cookies and in (c) as obsolete computers.
However, in exactly the same contexts, a bare possessor is ungrammatical, indicating that bare possessors are not amenable to pragmatic control:
These examples are infelicitous exactly if the words eggs, cookies and obsolete computers are not present anywhere in the preceding discourse. They become completely grammatical, however, if the necessary linguistic antecedent - an NP - is provided.
So DPs consisting of bare possessors are ellipsis of the kind that requires syntactic control. There's no obvious reason why possessed DPs should require syntactically controlled anaphora, while DPs headed by quantifiers should allow pragmatic control.
Many speakers report that it is very easy for them to imagine contexts where examples like those in (18) are perfectly grammatical. There are also cases of VP ellipsis which seem to be easily controlled by pragmatic antecedents:
In these situations, the antecedent is somehow super-salient, and is a valid antecedent for what is normally syntactically controlled ellipsis. We can control for this effect in a number of ways. First, we can be careful to consider the examples in (9) as if the antecedent were a bit distant, not the center of the discourse.
Or we can restrict ourselves to purely abstract entities with no clear, obvious correspondent in the context. Purely abstract nouns seem to be unable to have pragmatic antecedents, even when they are the focus of the discourse at hand.
Notice that the same pattern holds for an abstract verb, like think, describing an event which may happen and be totally non-evident from context:
The distinction between pragmatic/syntactic control then becomes a distinction which holds only for real-world objects. But the distinction still holds between syntactically and pragmatically controlled ellipsis. Real-world objects can always be antecedents for pragmatically controlled ellipsis. But syntactically controlled ellipsis still requires a linguistic antecedent, except in these strange cases of super-salience.
So with caveats about a kind of super-salience which crosses the boundary between syntactic and pragmatic anaphora, we can establish a distinction between pragmatically and syntactically controlled ellipsis of the noun phrase within the DP. Stranded possessors appear to license syntactically controlled ellipsis, and bare quantifiers appear to license pragmatically controlled anaphora.