The null hypothesis for fiction and logical indiscipline

The literature on the semantics of fiction is long-standing and voluminous. The null hypothesis, however, is rarely seriously entertained. Such a hypothesis simply denies that the fiction/non-fiction distinction is a semantic one, and so just like other statements, fictive ones of all kinds might be true or false depending on how the world is, and their truth conditions involve no ontological exotica or bespoke semantic machinery for their specification. As far as language goes, we might say, there just is no fiction. The present paper attempts nothing as ambitious as a full articulation and defence of this position; still less a refutation of the extant alternatives that are the focus of contemporary discussion. Much of the work in this regard, however, has been done in various ways by Ludlow (2006), Azzouni (2010), Friend (2012), Crane (2013), and Collins (2021a). Instead, my aim is to raise and dispel what might seem conclusive evidence against the null hypothesis. If nothing else, then, I'd like the null hypothesis to be rendered as a genuine null hypothesis from which we need a reason to depart.

The problem to be addressed is one of logical indiscipline, that is, if fiction is treated as on all fours with non-fiction, then inconsistency quickly follows in the shape of patently false conclusions appearing to follow from accepted premises, and contradictions being formulated by the conjunction of truths. A kind of sceptical solution will be presented: the relevant inferences break down outside of fiction, so the null hypothesis is not to blame; a naïve view of the extent of natural language's discipline is the problem. That said, we still need to differentiate fiction from non-fiction, if we are not to be confused, but this is not a semantic achievement. Before all of that, the following section will set out some preliminary semantic assumptions and the section after will present a general clarification and partial defence of the null hypothesis.

Firstly, I shall assume throughout that an adequate semantic theory for a language will assign truth conditions to each sentence of the language in a compositional manner; perforce, the semantics for each lexical item and phrase will specify how the expression contributes to the truth conditions of its potential host sentences. I shall leave in abeyance how such an assignment might best be realised, what general ontology the assignment assumes, and what might guide the compositional assignment (whether, say, the semantics interprets an independently specified syntactic structure or some intermediate translation such as a formal logical language). In short, my considerations do not depend either upon an endorsement of any contentious semantic view that might be deemed independently implausible or a rejection of any semantic bells and whistles that might accompany truth-conditional semantics; my claim is simply that, per the null hypothesis, an account of fiction does not require anything not on general semantic duty beyond fictional discourse.

An immediate problem, it might be thought, is that a truth-conditional theory specifies the semantics of a nominal in terms of its referent, for that is what contributes to the truth conditions of the nominal's potential host sentences. Yet what referents do fictional expressions have? Per the null hypothesis, they cannot be bespoke ontological exotica or abstract artefacts, but then what? This quandary is not really a problem for semantic theory, qua a chapter of linguistics, for the theory is happy to specify the semantic contribution of a nominal via a meta-linguistic term such that, roughly, Donald Trump refers to Donald Trump and Donald Duck refers to Donald Duck. Just what it is to be Donald Trump or Donald Duck that contributes to the truth conditions of ‘F(DD)’ is left entirely open (cf., Azzouni, 2017; Collins, 2017a; Glanzberg, 2014; Pietroski, 2018).1

Second, I shall remain neutral on what ontology there might be. Let one's ontological jungle be as dense as one likes. Again, the null hypothesis is only claiming that whatever makes a fictive statement true or false is just the stuff that makes a non-fictional statement true or false. Fiction is not special, at least as far as language is concerned, although it might be special as a genre, as Friend (2012) suggests.

Third, I acknowledge, of course, a fiction/non-fiction distinction; the null hypothesis only says that it is not registered in the semantics of natural language. Fiction is illocutionary in the sense that it involves the telling of a story or some other make-believe, and includes the linguistic expressions from such story telling put to other uses (see immediately below). Non-fiction is liable to reality independent of the story telling. This distinction will become clearer in the next section. It bears emphasis now, however, that the null hypothesis is not beholden to great precision on this matter. If the distinction between fiction and non-fiction is ultimately vague or indeterminate, then so much the better for a hypothesis that assumes that language does not register the distinction. If the distinction is ultimately precise, it is much less plausible that natural language will register it. The point here has been noted before. The immersive character of fiction relies upon its fictionality not being linguistically registered, at least not in any straightforward way.2

Fourthly, while fiction is myriad, I shall assume that it comes in three basic types.

In-fiction: this category covers what might be in a text (broadly construed) or directly implied by a text. For example:

Out-fiction: This category covers the use of in-fictional expressions outside of a text in relation to the broader world. For example:

The distinction between these two cases can be thought of as a distinction between the person (or thing) in the fictional world and the character or role that the fiction or story grounds in the sense of being created by an author. The two notions are obviously related, but distinct in the respect that what is true of a character, such as being popular or influential or making lots of money for an author, is often times not true of the person (/thing) in the fiction. James Bond is not popular in the fiction—the guy is a spy! Further, while the two notions are asymmetrically related—you cannot have a character without some originary story—the relation is somewhat nebulous in various ways. First, a speaker might be competent with Oedipus or James Bond, say, without having any knowledge whatsoever of the originary texts (such might well be the norm). Second, what might be truly said of a character is not necessarily dependent on an originary text. For example, most people are familiar with Bond via the films rather than the books. Only the obtuse would insist that Bond really smokes, say, because he does so in the books. Third, there might not be any identifiable originary text. This holds for Greek drama, say, where the origin of many of the stories lies in a long-lost oral tradition, likewise for Robin Hood and the knights of the round table. In short, for many cases, the actual relation between in-fiction and out-fiction is opaque. All the distinction depends upon, though, is the difference between predications that are part of some story and those that bear upon extra-story matters.3 What is clear, though, is that it will not do to say that an in-fiction is made true by whatever is in a canonical text, for no such text need be relevant, metaphysically, epistemically or semantically.

Copredication: This category covers cases where an expression has a simultaneous in-out reading:

The crucial point about these cases is that the single nominal (Agamemnon, Media, Oedipus) supports in- and out-predications, or simultaneously refers, if you like, to both the person in the text (who killed family members, say) and the trans-textual character who relates to extra-textual things, such as Aristotle, a film, or Freud. In this respect, they are like other cases of copredication where a single polysemous nominal supports predicates that appear to select for different kinds of entity.4 For example:

Such cases can be multiplied indefinitely, for both nominal and verb. There is no settled approach to the phenomenon, or even what exactly counts as copredicative.5 What is clear is that copredication is not ambiguity, for homophonous terms do not admit conjunction reduction or anaphoric binding.6 Equally, copredication appears not to be a species of context sensitivity, that is, the multiplicity of senses activated by the two or more predicates in the construction is not licensed by variation in context.7 For my purposes, I make no stand on the general phenomenon or how it applies to the fiction case specifically. I only assume that the kind of constructions are well-formed, that is, they can have perfectly coherent truth conditions (their syntactic soundness is beyond question). I should add, however, that copredication cases are not easily assimilable to standard treatments of fiction precisely because they meld different ways in which fiction is used, and so prima facie preclude approaches designed exclusively to deal with in or out cases, respectively, and also dual treatments that assume fictional terms to be somehow ambiguous between, say, character as abstract artefact and pretence (Kripke, 2013). If a traditional account is to work on copredication, something additional is required (Collins, 2021a). My argument will not rely upon any doubts in the direction of filling the lacuna, though.

This is all I assume. Recall, my purpose is not to refute extant accounts of fiction or fully defend the null hypothesis; the aim is only to show that the hypothesis can be cleanly stated and that it evades what might appear to be a damning consequence.

Let the semantic null hypothesis about fiction be the following:

The theoretical import of SNH is that if one has an adequate semantic theory for non-fictional or literal language, then, eo ipso, the theory will apply to fiction, too, in exactly the same manner with exactly the same results, that is, sentences will be are made variously true or false depending on how the world is and what the meaning of the sentences specify about the world. This is not a deflationary or eliminative claim, however, as if fiction were a fiction, as it were; rather, the implication is that the basis and interest of fiction lies beyond what a semantic theory should specify. In a phrase: fiction has no linguistic signature.

Three points of clarification are in order.10

Firstly, SNH is the null hypothesis simply because it recognises no semantically significant linguistic distinction between fiction and non-fiction. The claim can be understood both linguistically, as in clause (ii), and metaphysically, as in clause (iii). At the level of language, the hypothesis says that fiction does not call for any special meaning or speaker attitude or uniform truth value, as if all fiction is false or indeterminate or a mode of pretence, say. Perforce, fiction does not call for any special semantico-syntactic structure, such as covert operators or some other kind of fictive index. Semantically speaking, fiction is just like any other discourse. At the level of the world, the hypothesis says that fiction does not call for any special or additional ontology of ficta, abstract objects, or artefacts of a kind one would otherwise have no mind to endorse.11 To be sure, the truth and falsity of statements about Sherlock Holmes, say, depend upon certain human creations (texts, films, etc.) in ways that the truth and falsity of (some) statements about Donald Trump do not. The crucial point here is that, according to SNH, the relevant texts are not what a speaker refers to in using Holmes; nor even do the texts enter into a specification of the semantic properties of the name in any way whatsoever. Still less are speakers referring to an additional entity somehow conjured from the texts. The texts make true or false what a speaker of (the relevant) fiction says, but the literal content of what is said does not advert to any such texts.

So, SNH does not claim that there is no difference between fiction and non-fiction, as if James Bond is on all fours with Ian Fleming. The claim is only that the distinction is not a linguistic one to which the semantics of a language is sensitive. I want to assume that this semantic claim is defensible in the absence of a full-dress theory of fiction as a distinctive mode of human cognition and activity. Still, as just intimated, I take all fiction to involve some cognitive creation. So, at a first pass, the difference between statements involving Holmes and ones involving Trump is going to feature somewhere down the line an appeal to authorship in the first case, but not the second (more on this anon). Put ontologically, this thought might appear to endorse ficta as cognitive artefacts. The point, however, is that such putative artefacts are not the semantic values of fictional terms.

Secondly, by ‘semantic’, as it occurs in SNH, I mean a property of a linguistic expression (word, phrase, sentence) that contributes to what can be literally said by way of the expression, where a saying has essential truth conditions. There are broader extra-linguistic conceptions of the semantic that would render my hypothesis trivially false. For example, I take it that it is a feature of many concepts that they express fictional notions, and so, at the level of concepts, SNH is surely false. If, however, we restrict the semantic to the linguistic, SNH is a substantive hypothesis. We may note this by recognising that the truth of SNH is not decidable without serious consideration of the relevant linguistic phenomena. SNH would be false, for instance, if there were a fictive or pretence mood, along with the declarative, imperative, optative, and interrogative moods, or if fictional terms carried a certain feature that marked them as such, or had a special index of some sort. There is not the least evidence for any of that, and I do not know anyone who has even entertained such notions. Still, there are other ways in which SNH might be false that are less committal than the syntactico-semantic ideas just mooted. It could be that all fictions are false, precisely because some constituent expression of the fictional statement lacks a semantic value. That would count as a linguistic property, albeit not a syntactically registered one. It could also be, instead, that fictional expressions have bespoke referents, such as abstract artefacts of a certain kind, or abstract objects of a more thoroughgoing kind. It might be that fictions always involve a covert operator, which would virtually be a fictive mood. Equally, SNH will be true, if no such proposal holds that distinguishes fiction from non-fiction in terms of a uniform semantic property. It bears emphasis that SNH is perfectly consistent with fiction being a wholly conceptual notion, or a notion that is registered in discourse and how we take statements to be made true or not. All that SNH insists upon is that the distinction is not marked in the language itself.

Thirdly, I intend SNH to apply to all types of fictional discourse: in-fiction, out-fiction, and copredication. Suppose we take Media married Jason. Per SNH, we can say that it is true, but here we have not necessarily invoked any covert intensionality, ontological exotica, or merely made as if to assert the claim. What makes the claim true are some ancient texts by Euripides and Apollodorous and myths that predate them. Yet what makes the sentence true is not part of its semantics in any sense. One could be perfectly competent with the sentence but not know of Euripides, say, or even know that the sentence expresses a fiction (it could turn out not to be). All that is demanded is that whatever makes the sentence true can be accurately specified as Media marrying Jason, but a competent speaker need know no more than that. Likewise for out-fiction. Take, Freud made Oedipus into a complex. Again, what makes this true are various facts about texts of Sophocles and Freud and the dissemination of ideas throughout the 20th century. Oedipus as another entity in reality is not among such facts, but a competent speaker need not know that. I dare say that many probably think there was such a king of Thebes (why not?). They would be ignorant of history, but not semantically incompetent. All they need to know is what whatever makes the sentence true is specifiable in terms of Oedipus, which of course it is, for how else would one make sense of the relevant texts. Finally, copredication comes out as expected. Take: Oeadipus married his mother and remains a by-word today for a tragic figure. This is made true by a complex set of facts pertaining to Sophocles' texts and their cultural transmission over the proceeding couple of millennia. All the competent speaker need know is that such facts can be specified as involving Oedipus marrying his mother and so on, regardless of whether it is fact or fiction.

In essence, then, SNH claims that the world makes true or false fictional statements, and since semantic competence is insensitive to how statements are made true, the relation may or may not involve a text or story telling more generally with semantic competence and semantic value remaining invariant.

Much more could be said on each of these points. All I hope to have shown is that SNH can be stated in a way that dispels its first blush absurdity. In the next section, I shall entertain what appears to be a decisive problem for the view.

If SNH is at least to be seriously entertained, it cannot be rejected on the grounds that it denies a fiction/non-fiction distinction, for it is perfectly happy with the distinction; its only claim is that it is not one to which natural language semantics is sensitive. Nor can the position be rejected because it ushers in a really true and truth-in-fiction distinction. All kinds of fictions, according to SNH, are really true, if true. It is just that some of them are made true by what is in a text or story and others are made true in a text-free way, but being cognisant of which condition holds for which sentence is not constitutive of semantic competence. What does present a problem for SNH is where fictional truth clashes, if you will, with non-fictional truth. Such a clash would appear to be semantic in character rather than merely a difference in what makes the respective claims true (text or non-text), for we do not want to credit the world itself as somehow inconsistent, at least not merely on the basis of such clashes.

The problem is that, even if we are minded to treat fiction as true simpliciter, we better be able to corral it from other truths lest we utter falsehoods, such as would occur if we included Holmes in a biographer of Gladstone. The underlying problem here does not so much concern the very idea of true fictions, but more that if a semantic distinction is not drawn between the fiction and non-fiction, then a logical indiscipline follows, which suggests, in contradiction of SNH, that the distinction must be semantic after all.

Let a language have logical discipline if properties of consistency and entailment hold of it as a matter of the semantic properties of (at least some) lexical items and principles of composition. Since such properties are aspects of the semantics of the language, a competent speaker will appreciate how certain elementary logical principles are realised in their language. Of course, performance limits the exercise of competence, in both logic and language. No speaker may properly acknowledge all of the infinite consequences of some one of their statements, even though such consequences are genuine ones given what the statement in fact means. The point is not that speakers must be logically consistent and complete in their endorsements (logically omniscient), but only that their competence gives a certain logical shape to their performance, such that bare contradictions and simple entailments are acknowledged, and much more complex logical properties can be robustly appreciated as following from their competence.12 Also, the bare idea of logical discipline does not entail or even suggest a particular logical framework in which the various disciplinary notions are to be proprietorially encoded. There might be no such framework independent of the contingent facts of the syntax-semantics interface of natural language.

Suppose, then, that we seek to impose logical discipline upon fictional discourse, as we must, it seems, if SNH holds, for that hypothesis exactly says that fiction has no special semantic status and so whatever discipline holds generally should hold for fiction in particular. Logical discipline poses a challenge to every account of fiction, although some positions appear better suited to meet the challenge than others.13 A position that claims all fiction to be false has an easy time of it, albeit at the expense of denying what appear to be simple truths (both in-fiction and out-fiction). The real problem arises for those positions that consider fiction to be variously true and false, such as SNH. Since I wish to defend SNH, hereafter I shall just consider how that position might handle logical discipline.

Felix Lighter is Bond's CIA friend. We also know that Ian Fleming created Lighter and that George Bush was director of the CIA (1976–77). Thus, the sentences in (5) appear to be true:

Given that, we can readily concoct some valid arguments with false conclusions:

One may obviously proceed in a like manner for all cases of fiction we reckon to be true.

The general recipe here is to take some fictional truths (in-fiction and/or out-fiction) and put them together in a premise set with non-fictional statements, and then simply draw the relevant conclusions based upon the shared predications, such as being a member of the CIA or, say, living in London, which renders Holmes and the Queen residents of London. The basic moral is that if we treat in-fiction and out-fiction as true, then nothing appears untoward, so long as we keep to describing the fictional world or the created character; if we infer to a statement concerning the fiction in the non-fictional world or in relation to non-fictional elements, then we appear to land with absurdity.

The first thing to note is that, in fact, copredication appears to give rise to the same problem. (8a) appears to be true, and (8b) appears to be entailed by it, but (8b) strikes us as obviously false, if not a pun:

An initial diagnosis for the problem here is that while we readily consider in-fiction true, it also appears to be isolated, or a world unto itself. The CIA of the Bond stories is just not the CIA Bush headed, just as the London of Holmes is just not the London in which the Queen resides. If so, we cannot generalise over both worlds, as it were, or infer from one to another. The problem is how to square such restrictions with logical discipline, which, precisely because it is logical, does not essentially recognise such divisions.

Given the diagnosis, there are, it seems, two basic approaches to this problem one may take. A Meinongian (for want of a better term) or anyone who thinks of ficta as abstract might respond to the problem by saying, ‘Exactly! Ficta exist independently of actual or concrete things, so there is a fictitious London, a fictitious CIA, and so on, besides the concrete ones. Don't mix the two up, for they are different things!’. Someone who thinks of ficta as abstract artefacts might make much the same reply, save for thinking of there being an artefactual CIA and a non-artefactual one, and so and so forth. Regardless of how ultimately coherent such approaches might be, an advocate of SNH must eschew them, for fiction does not, ex hypothesi, introduce any novel ontology or semantic apparatus. In this light, I shall suggest on behalf of SNH a response to the logical indiscipline problem.

We assumed above that logical discipline is a fair condition on non-fictional discourse, and for many constructions in many contexts discipline surely does hold sway for speakers. It does not follow, however, that discipline is a general constraint specific to non-fictional discourse, for there are perfectly fine, literal constructions that lead to absurdity in much the way fiction does once discipline is imposed. These are cases of copredication as already intimated.

Consider, then the following arguments:

The premise sets here can be perfectly true, but their conclusions are at best puns.14 There is, however, nothing metaphorical or otherwise non-literal about the premise sets. Logical discipline is contravened because the relevant nominals are polysemous and so cannot be construed as picking out a respective univocal object across all predicates. Logical principles that target the nominal and its predicates fail, because they abstract away from the very polysemy that allows the premise set to be true. Again, it bears emphasis that polysemy is not ambiguity, precisely because polysemous nominals do admit conjunction reduction and anaphora, in the way ambiguous nominals do not.

As mentioned previously, there is little-to-no consensus on how to treat polysemy and copredication. My present point does not presuppose any tendentious position on the linguistic phenomena. The relevant fact is that logical discipline fails to hold outside of the fictional realm; so, that a position on fiction faces the problem of logical indiscipline does not all by itself sink the position. Of course, one might think it does, because copredication is somehow illicit. I take it, though, that an obtuse denial of the linguistic phenomena does not count as an argument. The premise sets of (8) and (9) can be true, and the oddity of the conclusions is to be explained in light of that fact. Still, even if logical indiscipline occurs in and out of fiction, an explanation of how SNH squares with such apparent incoherence would be welcome.

Recall that the implementation of SNH I suggested above makes a distinction between how the world might be such that a statement can be true and what a speaker knows such that she is competent with the relevant sentences. Logical indiscipline, as we are supposing to occur in fiction and literal copredication, pertains to the latter semantic conditions on what speakers know. That is to say, semantic competence is sometimes out of sync with how we suppose the world to be in order to make our statements true. It does not follow, however, that speakers therefore consider the world to be illogical or contradictory in any way, shape, or form.

For example, argument (8) concludes with Felix Lighter had Bush as a boss, and argument (9) concludes with Bush was the boss of a Fleming creation. What has gone wrong here, according to SNH, is not that one of the fictional premises is in fact false, but simply that we cannot see how the world could make true the conclusions of the arguments, even though we can see how the same world would make true the premises. Thus, various facts about Ian Fleming, the content of his books, and the governance of the CIA in the 1970s make true the various premises, but the same facts cannot make true the conclusions. The world is coherent alright (in relevant respects), but it is not at the behest of semantics. The same holds for non-fictional copredication. The world can be such that a statement about a book as a concrete particular and a statement about it as a body of information (a story, say) can both be true. It does not follow that the world contains a thing that is both content and concrete.

A lot more could be said on this topic. My modest ambition has only been to show that logical indiscipline does not sink SNH. Indeed, it strikes me as a positive virtue of the account that it recognises that logical indiscipline occurs both in and out of fiction, and that it can be handled in exactly the same way without extra ontology or rendering the world incoherent.

The response to logical indiscipline offered is a sceptical solution insofar as the problematic indiscipline is endorsed; it turns out not to be a peculiar flaw of fiction, but a far more general phenomenon resting upon the ubiquity of polysemy and copredication, and otherwise sourced to semantic univocity being misaligned with the different ways in which tokens of the one sentence type can be made true. Still, as is often the way in philosophy, there is the part where you take it away, and the part where you give it back. What a good sceptical solution in this realm should give back is some sense that fiction can be carrolled so as to save us from confusion. This may be achieved just as it is with copredication.

We are not minded, for example to think that the world contains geographical areas with psychological properties, but we happily use New York or London as subjects of geographical and psychological predicates with no linguistic or semantic condition being flouted. We can, in such cases, readily understand how the world may make true what we say without hypothesising peculiar objects. So it is in the case of fiction. If we know that an expression is used fictively, then we can corral it, understanding that the host statement is made true by some text/story. If we do not know that an expression is fictive, then we shall inevitably fall into confusion. Such confusion, however, is not semantic incompetence, but mere ignorance of the world. Where, therefore, we hit patent falsehoods such as Bush being Felix Lighter's boss, this is simply the world not being such as to make it true that Bush was Lighter's boss, but there is no semantic interdiction against such thoughts. Imagine someone quite naïve, such as a child. She might, given the appropriate information, reason to the conclusion that Lighter had Bush as a boss. In doing so, she has made no semantic error. The only error she does make is to copredicate in her conclusion which results in a falsehood. In this sense, semantic composition is not a sure guide to how the world might be, but we can avoid confusion by knowing what is fiction and what is not. Merely being competent with a language does not suffice for that.

The general reasoning here is supported by the same considerations applying outside of copredication cases.

Consider fictions that contain what we would think of as general non-fictional falsehoods pertaining to our world, not characters or things specific to a fiction. Imagine a sci-fi novel where the universe is a simulation or, as the ancient Greeks thought, there are only four elements.15 We clearly want to avoid outright contradictions, but how might we do so? The same formula applies as with copredication. If we are talking about the relevant fiction, then, indeed, it is true that the universe is a simulation, and it is made true by the story, otherwise not. We simply need to keep track of what will make true or false what we say. We are not guaranteed of success merely be being semantically competent. Suppose someone takes the sci-fi novel to be a philosophical tome (the novel is a touch experimental) and is suitably credulous. They have not made any kind of semantic error when they declare, ‘Everything is just run on an alien supercomputer! The planets, the tables and chairs, and you and me’. In particular, they are not speaking of the denizens of a novel, as we would be if we were talking of Middle Earth, say. That is why they have said something false. Equally, when speaking of the novel, one could say the exact same thing, with the exact same semantic properties, and be saying something true. What makes it true is not the planets et al., but the story. Just so, if it turns out that the universe is in fact a simulation, then both statements are true, albeit for different reasons.

The same considerations cross over to particular denizens of fiction and reality, such as London where the Queen lives and London where Holmes lives. The nominal London is not ambiguous or even relevantly polysemous (although it is in other respects). If someone is confused and takes Conan Doyle to be a true-crime writer, then they might end up saying false things about London, such as that Baker St. contains a 221B address (it does not), but they will also say lots of true things from their reading, such as that Baker St. is in the West End. The clued-up person may also speak of 221B Baker St. but now be saying something true, for it is made true, not by the addresses in a London street, but by the story. Again, no semantic error or incompetence need be in play; the fault of the confused speaker lies in an ignorance of the world, and so what might or night not make one's statements true. Only by knowing something about what makes what true can one so much as understand fiction, and so corral truth accordingly between fiction and non-fiction, but knowing about the world is not something semantics caters for.

My ambition has not been to mount a compelling case for SNH, but only to defend its bare coherence, and as what should be expected in the absence of further semantic inquiry. To end on a positive note, we should reflect why the position is not treated as a genuine null hypothesis in most of the philosophical debates. The reason, no doubt, is that there is a fiction/non-fiction distinction and if semantics is in the business of relating language to the world, then it should be incorporated, since, surely, non-fiction relates to the world and non-fiction does not. We might think differently. There is, to be sure, a fiction/non-fiction distinction, but it is not a semantic one. This thought does not scupper truth-conditional semantics or truth as a world-involving notion, for we can see how fictions can be made variously true or false by the one world that does not contain ficta, but does contain stories and books, and it is not the semantic responsibility of a speaker-hearer to know what is fiction and what is not, or how a particular truth is made true. If we are to move from the null hypothesis, evidence is required.16