Objects of Fiction
Here comes a positive theory of fictional characters. Disclaimer: Only read when you are very bored. I've started thinking and reading about this topic just a weak ago, so probably the following 1) doesn't make much sense, 2) fails for all kinds of well-known reasons, and 3) is not original at all. The main thesis certainly isn't original: it is simply that fictional characters are possibilia. Anyway, I begin with an account of truth in fiction, which largely derives from what Lewis says in "Truth in Fiction".
Suppose back in 1921 some clever sociologist found out that it is very likely that Adolf Hitler will become an evil dictator. What would have happened if that sociologist, let's call him Fred, had then killed Hitler to avoid this threat? If caught, Fred would presumably have been sentenced to death. Would this have been a just sentence?
Such counterfactual reasoning is not fiction: There is no (literally) false claim in the above paragraph. On the other hand, such counterfactual reasoning closely resembles fiction. In particular, they both lead to well-known peculiarities: In a discussion started by the above paragraph, one might say things like "What Fred did was morally wrong because he wasn't sure at the time of the killing, in 1921, that Hitler would once become an evil dictator", yet we might also say that Fred didn't really kill Hitler in 1921, for as a matter of fact, Hitler wasn't killed in 1921. As a matter of fact, Fred doesn't even exist. We would also consider many questions about Fred to be somewhat inadequante and indeterminate, e.g. questions about his blood type.
These similarities are a good reason to treat fictional discourse in some way like counterfactual discourse. Fiction, too, is an invitation to suppose certain things. And like counterfactual reasoning, thinking about fiction involves thinking about what would be the case if the supposition were true. So very roughly, in fictional discourse a statement S is true iff S were true given that certain suppositions were also true.
It is a bit difficult to spell out the "certain suppositions" in more detail. They are not just the statements explicitly made in the fictional text: Often the fiction is told from the perspective of a fictional character who gets certain (fictional) facts wrong. And even if there is no fictional narrator, the actual author himself can make false statements in his story about what happens in this very story. (This is where we get imaginative resistance on the part of the reader.) Moreover, there are problems with inconsistent fictions and with background beliefs in the author's community, and with "carry-over" from other fictions (see Lewis, "Truth in Fiction"). All these difficulties don't look insurmountable though, and I'm rather confident that there already are good proposals to deal with them.
So let's assume we can spell out the "certain suppositions". Then a fictional statement S (that is, a statement about what is the case in a particular fiction) made in an appropriate context C is true iff S is true in all the possible worlds in which the certain suppositions associated with C are true, or -- depending on how the suppositions are spelled out -- in those of these worlds that are closest to actuality.
I don't think we need a story prefix like "in the fiction" for this account. Merely saying that a fictional statement S is to be understood as "in the relevant fiction, S", without explaining how this latter sentence is to be understood hardly qualifies as an analysis. But if we're going to analyse away the story prefix anyway, there is no real need to introduce it in the first place. Of course the story prefix also occurs in ordinary fictional discourse, so we need an account of how it works. It seems that in fact it largely determines the "certain suppositions" that must be taken as antecedents for the implicit counterfactual conditional whose consequent is the embedded sentence. In other words, it largely determines the possible worlds at which the embedded sentence is to be interpreted. But there is no need to claim that every fictional statement works in exactly the same way as the corresponding story-prefixed statement.
This, then, is the general idea: When we talk about what happens in a fiction, we talk about what is true not in the actual world but in the worlds selected by the relevant context. "Fred killed Hitler in 1921", uttered in an appropriate context C1, doesn't contradict "Hitler wasn't killed in 1921", uttered in an appropriate context C2, because sentences like these are only true relative to certain worlds, and the worlds selected by C1 are not the same as the worlds selected by C2 (the latter contain only the actual world, I guess).
Now over to fictional entities. What is Sherlock Holmes? That is, what does "Sherlock Holmes" refer to? Well, it depends on the context of utterance. In strictly non-fictional discourse, having all our quantifiers restricted to actuality, it doesn't refer at all. But such a context is unusual, because merely by using the name "Sherlock Holmes" you strongly suggest that you intend to quantify over more than what there actually is. In such a more usual context, "Sherlock Holmes" refers roughly to the realiser of the Sherlock Holmes role in the worlds selected by the relevant context. This is rough firstly because it is not entirely clear what we have to take as the Sherlock Holmes role: Is it the role attributed to Sherlock Holmes in the Holmes stories? Not quite. Again, we have to take into account unreliable narrators, background beliefs, carry-over from other fictions, etc. Presumably (though I'm not sure about it) we also have to take into account the intentions of the story-teller so that even if the Holmes stories surprisingly turn out to be true, "Sherlock Holmes" doesn't denote a real person. Second, there are several different realisers of the Holmes role in the different worlds selected by the context. So we should rather say that "Sherlock Holmes" is referentially indeterminate between all these candidates.
Kripke complains about this indeterminacy, but to me it seems that it is exactly what we need to account for the indeterminacy of many statements about fictional characters: Saying that "Sherlock Holmes" is indeterminate between some detectives with blood type A and some with blood type B explains why Sherlock Holmes' blood type is intuitively indeterminate. "Sherlock Holmes lives at 221B Baker Street", on the other hand, is determinately true because all the referents between which "Sherlock Holmes" is indeterminate live at (a counterpart of) 221B Baker Street.
And "Sherlock Holmes doesn't really live at 221B Baker Street" is true because "really", like other features of context, invokes a restriction to actuality: In the actual world, Sherlock Holmes (who only exists at some other possible worlds) doesn't live at 221B Baker Street. I can now even explain why "Sherlock Holmes doesn't really live at 221B Baker Street" sounds slightly odd to me: Usage of "Sherlock Holmes" normally opens the quantifier domains to include possibilia from Sherlock Holmes worlds, whereas usage of "really" restricts them to the actual world. This is why "Actually, no person called 'Sherlock Holmes' ever lived at 221B Baker Street" sounds much better to me: Here there is no clash between implicit operators on quantifier domains.
Identifying fictional characters with possibilia and doing away with the story-prefix analysis also allows for not-too-complicated interpretations of "mixed" statements that are neither clearly fictional (so that we could prefix the story operator) nor "serious" (all quantifiers restricted to actuality): The hero of the Sherlock Holmes stories -- over there at his (indeterminate) world -- consumes drugs that are -- actually, here at our world -- illegal. It's just an ordinary quantification into modal context, as in "every drug that is actually illegal could have been legal".
Unlike van Inwagen and Thomasson, I can respect the intution that Sherlock Holmes is not an abstract, invisible, ghostly thing, unlocated in space and never doing anything. At the same time I can respect the intuition that Sherlock Holmes is not located at any actual place in space, and that he hasn't actually solved any murder cases.
Nor do I have problems with the vague boundary between stories about real characters and about fictional characters: When you gradually alter the Oliver Cromwell biography into a Sherlock Holmes story, you're first determinately telling a story about Oliver Cromwell, then it gets indeterminate whether you're refering to him, and finally you're definitely not refering to him any more. Clearly whether or not you still write about Cromwell depends to a large extent on your intentions. It also depends on the context in which we talk about your work: Even if you've intended to write about Cromwell and make only very small mistakes about him -- even if you make no mistakes at all --, we might interpret your story as being about a fictional person, so that we can say that your Cromwell, unlike the real Cromwell, did this and that, or that your Cromwell did this and that whereas it is uncertain whether the real Cromwell did. I don't think we need to decide what your story is "really" about.
I think I can also handle easily most of statements from literary critism used by advocates of the ghostly theory. Most of these only show that we are committed to fictional characters, not that they are ghostly: "There are characters in some nineteenth-century novels who are presented with a greater wealth of physical detail than is any character in any eighteenth-century novel." -- This sounds exactly like "There are characters in some nineteenth-century biographies who are presented with a greater wealth of physical detail than is any character in any eighteenth-century biography". Certainly this latter statement doesn't commit us to the view that all eighteenth- and nineteenth-century biographies were biographies of immaterial ghosts.
Only statements about the alleged ontological dependence of fictional characters on acts of story-telling seem to create problems: I can hardly acknowledge that Conan Doyle really and literally created Sherlock Holmes, nor that Sherlock Holmes literally exists now but had not existed in 1800. But I don't share the intuitions that these statements must be literal truths. I would of course agree that Conan Doyle in some sense created or invented Sherlock Holmes. But did he bring it about that Sherlock Holmes really exists? This is far less clear, since like most people I also believe that a) Sherlock Holmes is human, but b) Conan Doyle never created a human being, and anyway c) Sherlock Holmes does not really exist. So whatever we mean when we agree that Conan Doyle "created" Sherlock Holmes, it seems doubtful that we mean that he literally brought it about that Sherlock Holmes exists.