Tomas is having problems posting things as comments on this blog, so I paste it here. In any case, what he writes is too important:
I believe your question about the possibility of pointing at nonexistents points in the right direction. I think that is what asking if there can be direct reference to nonexistents, the core question of the Millnong project, really amounts to.
At first sight, it might seem clear that there can be no such a thing as direct reference to nonexistents. This sounds natural, since there can be no causal relation between utterances of names and nonexistent objects and it happens that the causal chain theories of reference, despite its problems, are the main option to Millianism. But this reasoning is not quite right. Causal chain theories are theories of reference transmission only, that is, they are accounts of how reference is passed on from one utterance or speaker to another. The baptism of an object itself need not be explained as a causal event. Surely, in order to baptize an object one first needs to pick it out. This can be done by physically by pointing at it, given that the object is in our field of vision. But one can also pick an object out by a definite description. This is basically Graham Priest's account of why Meinongianism is not inconsistent with direct reference (Towards Non-Being, pp. 141). (Priest goes even further and insists that one can also pick an object out by mentally pointing at it. According to him, this should be a corollary of taking intentionality seriously. In its turn, this should not mean embracing internalism, since nonexistent objects are not private.) So there surely can be real causal reference transmission from one real person to another where the referent is a nonexistent object, provided we explain how a nonexistent object can be baptized in the first place. That is, provided we explain how is it possible to 'point at' nonexistents. As your question suggested, I think this is the main challenge of the Millnong project.
Naturally, an answer to the question of how one can point at nonexistent objects will heavily rely on our account of what are nonexistent objects and, ultimately, on our very notion of objecthood. I have been supposing we already have a common and solid account of nonexistent objects, which we do not, but we probably wouldn't be able to delay its details once we got into the details of pointing at nonexistents. So I suspect we'll be able to account for reference to nonexistents as soon as we account for nonexistent objects themselves. This seems to be too easy or uninteresting though, since we quickly reach a dead end. For Meinongian objects, existent and nonexistent, are normally simply defined after the primitive notion of (nuclear) property in Meinongian theories of only one kind of predication. Remember Parsons's account: for any set of nuclear properties, there is an object that has exactly the nuclear properties in that set.
Nevertheless, I sense you might be unhappy with this quick start for other reason then the obvious problem of defining nonexistent objects. As I understand the problem you pose, it seems one can not genuinely refer (with a purely referential expression) to an object which was baptized by means of a definite description. It seems in this case the object referred to would be hostage to a description and hence not enjoy the modally open horizon of life. My own guess is that an object baptized by means of a definite description enjoys the same modal life as any other object. Only we might not be sure, depending on the description, as to which object our description happened to pick out in our world - whereas pointing at an object we can perceive might seem somehow to assure us we picked out something that enjoys full modal life, as if we had picked it out by its numerical identity. It seems to me that this opposition is problematic. For instance, I doubt that the act of pointing at something is not itself a descriptive act. Perhaps Millnongism needn't rely on a Donnellan-inspired theory of ambiguous definite descriptions, though I do think one should take seriously some version of it.
Yes, the starting point is not transmission of reference, but rather specification of what is being denoted. We need to find an equivalent for ostension in nonexistent contexts. The alternatives, roughly speaking, are: a) find a way to make sense of Donnellan-like referential use of definite descriptions of what doesn't exist or b) find another way to refer to nonexistent objects. Both seem tough as existence itself plays a role in the determination of what is being denoted.
I was reading an interview with Darwich where he talks about what he does when he uses mythological characters in his poems - or rather, mythological names, as he puts it. He says names bring up a bundle with them, a bundle of associations. He can then make Helena of Troya sell bread in Paris, which is not the same as talking about any other nonexistent bread monger. Names have this role, they can be seen as tags even when tagging the nonexistent (the Mill in Millnong). But I guess we will quickly moving towards issues in possible world semantics as Kripke's original view made room only for actual objects. David Lewis' account has room for counterparts. But this seems a bit too descriptivist for me. Is there any other way out?