7
Presupposition
The discussion of definites in the last chapter leads us to the even thornier issue of presupposition. One of the interesting things that a definite NP does is to presuppose the existence of its referent. Our friend Frege (of ‘sense/reference’ fame) made the following observation:
If anything is asserted there is always an obvious presupposition that the simple or compound proper names used have a reference. If one therefore asserts ‘Kepler died in misery’, there is a presupposition that the name ‘Kepler’ designates something.
(Frege 1892, cited in Levinson 1983: 169)
This makes sense; it would be odd to assert something of Kepler if the word ‘Kepler’ had no referent, so the use of the name Kepler presupposes the existence of this referent.
So consider the following statement, the analysis of which has been debated for more than a century:
(1) 
This sentence seems to assume that there is presently a King of France and assert that he is bald. That is, it presupposes the existence of a King of France. It’s as though the King’s existence weren’t up for debate. For instance, suppose someone responded to (1) by saying (2):
(2) 
This person would be assumed to be taking issue with the assertion that the King is bald, not with the fact that he exists. In order to take issue with the fact that he exists, you have to do something more extensive than simply say “no” or “that’s not true” or in some other way express disagreement; you have to actually make clear that it’s the presupposition that you’re disagreeing with, by saying something like (3):
(3) 
As it happens (and as you presumably know), there actually isn’t a King of France at the moment; France currently isn’t a monarchy. So the presupposition in (1) is false, yet I can’t express that falsity by replying to (1) simply with (2).
Relatedly, Strawson (1950) noted that you can’t get rid of the presupposition simply by negating the original statement, as in (4):
(4) 
This statement presupposes the existence of the King of France just as much as (1) does; now we’re merely saying that the King isn’t bald, not that he doesn’t exist.
What we’ve seen is that what a sentence presupposes behaves quite differently under negation from what it asserts, and this is in fact a standard test for presupposition. Presuppositions in general survive negation; that is, if the King of France is bald, the King of France exists, and if the King of France is not bald, the King of France exists. They also survive questioning: The question Is the King of France bald? also presupposes the existence of the king.
But as we saw in the last chapter, Russell (1905) sees things differently. In Russell’s view, the meaning of (1), with its definite article the, consists of three parallel propositions:
(5) 
Russell’s analysis of the definite thus has two aspects: It characterizes the meaning of the definite article as entailing both existence and uniqueness. And under this analysis, the falsity of any of the three propositions in (5) renders the entire statement false. Since in the current real world there is no King of France, this means that in this world, the statement The present King of France is bald is false.
The problem, however, is that these three pieces of meaning don’t, intuitively, have the same status at all: The point of the utterance seems to be to convey (5c), not (5a) or (5b), both of which are presupposed (and therefore survive negation and questioning)—and Russell’s account fails to capture this difference.
Now notice a second difficulty: If the nonexistence of the King of France means that any statement featuring The King of France as its subject is false, it follows that both of the following statements are false:
(6) 
That is, (6b) should be false if it has a Russellian meaning like (7):
(7) 
And worse, since the function of negation is to reverse the truth-value of the corresponding non-negative statement, if (6b) is false (as Russell claims) then (6a) must be true. But of course (6a), by virtue of featuring a nonexistent King of France as its subject, is false. Therefore, Russell’s view of definiteness as entailing the existence of its referent would appear to entail (6a) being simultaneously both true and false.
But under a fairer representation of Russell’s account, the negation of The King of France is bald negates not just the verb phrase but the whole clause, as in It is not the case that the King of France is bald. Because of a quirk of English grammar, (6b) is the usual way of negating (6a), but as we’ve seen, (6b) is usually understood as negating only the verb phrase, as in (7). But for Russell, the negation of The King of France is bald is a negation of the entire clause, so it has the following meaning:
(8) 
In the current world, (8) is clearly true. So this second difficulty with Russell’s account, although it’s the one that has gotten the most attention, is only apparent.
But the first problem is real. It seems clear that someone saying The King of France is bald does not ‘weight’ the three subpropositions in (5) equally; they are primarily asserting his baldness, not (for example) that there isn’t a second King of France lurking somewhere. And it still seems counterintuitive that the absence of a King of France, or the existence of a second King of France (under some weird new monarchy scheme), should render The King of France is bald every bit as false as if there’s an actual King (and only one) but he’s got a full head of hair.
Frege took a different view. He believed that if the presupposition is false, the sentence containing it has no truth-value whatsoever. So in his view, (6b) is neither true nor false—which tidily rescues us from the problems with Russell’s view. But it’s not quite as easy as that, because a standard bivalent—that is, two-valued—system of logic doesn’t allow for a proposition to have the status ‘neither true nor false’. So if we’re going to allow a false presupposition to render a sentence truth-valueless, we’ll need to abandon the standard bivalent truth-conditional logic.
Strawson (1950) agreed with Frege that a statement with a false presupposition has no truth-value, and in that sense his account, like Russell’s and Frege’s, is semantic. But he makes an additional interesting point, which is that the truth of the presupposition depends on when the sentence is uttered. France hasn’t always had its current political structure; for much of its history, it did indeed have a king. So 500 years ago, if I had said The King of France is bald, the presupposition (‘there is a King of France’) would have been true, and the truth of the utterance would have depended entirely on whether he was bald. In our terms, this means that the truth of the statement is context-dependent, which in turn means we’ve landed in pragmatic territory.
As further evidence in favor of a pragmatic account of presupposition, note that presuppositions may sometimes be canceled, as in (9a), or suspended, as in (9b):
(9) 
Correspondingly, to answer a question like (10a) with a flat no is to accept the presupposition in (10b), but to answer it as in (10c) is to reject the presupposition:
(10) 
Cancelability is a hallmark of pragmatic meaning, as we saw in the tests for implicature. Thus, the fact that presuppositions may (sometimes) be canceled aligns them with conversational implicatures, which, as we saw in Chapter 3, are cancelable by definition. An aspect of meaning that can be canceled or suspended would seem to be by definition context-dependent and therefore pragmatic.
In contrast, entailments cannot be canceled. For example, (11a) entails (11b), but any attempt to cancel (11b), as in (11c), renders (11a) necessarily false:
(11) 
Note that in a Frege/Strawson system, constancy under negation is still a defining feature of presupposition: If a statement is true, its presupposition is true; and if it’s false, its presupposition is still true. But if the presupposition is false, then the statement itself is neither true nor false.
Since presuppositions can in some circumstances be canceled or suspended, like (pragmatic) implicatures but unlike (semantic) entailments, and since their truth depends on the context rather than purely on the truth of the statement that contains them, there is good reason to look to pragmatics for a more satisfactory account of presupposition. A pragmatic approach to presupposition disentangles us from the question of the effect of the presupposition on the truth of the statement by asking not whether the larger statement is true, but instead whether it’s appropriate. We will consider pragmatic analyses of presupposition in detail, after taking a moment to examine expressions that serve as triggers for presupposition.
Presupposition triggers
So far we’ve focused on proper names and other definite noun phrases as giving rise to presuppositions, but as we saw in (10), there are other expressions and constructions that do so as well; these are called presupposition triggers. The range of such triggers includes (but is not limited to) those listed in (12):
(12) 
There are others, but you get the idea. And you can check for yourself that in each case, negating the main verb preserves the presupposition (e.g., I don’t regret that I broke the vase still presupposes that I broke the vase; What Steve wanted wasn’t to win the race still presupposes that Steve wanted something). For more discussion of the structure of cleft sentences, see Chapter 8.
These presuppositions have real-world consequences. Consider, for example, the Second Amendment to the US Constitution, which states:
(13) 
Arguments about gun control in the US frequently make reference to this amendment, but there is a great deal of disagreement about how to interpret it. And one aspect of the controversy involves presupposition. The initial clause expresses a presupposition, i.e., that a well-regulated militia is necessary to the security of a free state. As always, you can check this by negating the main verb, or in this case—since the verb is already negated—removing the negation:
(14) 
The main clause here becomes a rather odd amendment, but more importantly for our purpose, the presupposition expressed in the first clause is unaffected.
So let’s suppose that a well-regulated militia is not necessary to the security of a free state, on the grounds that there are plenty of secure, free states that don’t currently have such a militia (like, for example, the US). That means that the presupposition is false. What, in turn, does that mean for the main clause, i.e., the assertion that the right of the people to keep and bear arms shall not be infringed? This goes back to our earlier question: If the presupposition is false, what does that do to the status of the statement that presupposes it? (See Kaplan’s (2012) argument that the amendment “guarantee[s] the right of the people to keep and bear arms exactly to the extent that a secure free state depends for its security on a well-regulated militia. That is, not at all.”) As you see in this example, a lot can hinge on the way we interpret presuppositions.
Pragmatic views of presupposition
We have looked at a variety of problems that arise with a semantic view of presupposition. By ‘a semantic view of presupposition’ I mean a definition of presupposition that’s based on truth-values and how they affect other truth-values: what happens to the truth of a statement if it presupposes something false, what happens to the truth of the presupposition if the presupposing statement is false, and so on. An alternative is to see what happens if we assume that the relationship between the statement and its presupposition isn’t one of truth-value effects (as is the case for, say, the relationship between a statement and its entailments), but rather appropriateness in context. That is, an alternative to a semantic view of presupposition is a pragmatic view.
The question then becomes not whether The King of France is bald is true if There exists a King of France is false, but rather when it is and isn’t appropriate to predicate something of the King of France. Put simply, a pragmatic view would maintain that the fundamental problem is that it’s silly to be saying anything at all about the hair (or lack thereof) of a nonexistent entity. Unfortunately, things are, as always, not quite that straightforward; pragmatic approaches bring their own problems.
Let’s begin by going back to the (loose) definition we started the chapter with. I said there that the sentence The King of France is bald seems to assume that the king exists and assert that he’s bald. That is, the existence of such a person is taken for granted. In Stalnaker’s (1974, 1978) terms, it’s treated as part of the common ground—the background assumptions that the speaker believes they and the hearer share (or consider uncontroversial, as we’ll see shortly). This approach works reasonably well for the examples in (12): To say The steak thawed is pointless unless both participants are willing to take it as common ground that the steak had previously been frozen; if one person objects to this proposition, the statement becomes inappropriate and there’s a problem to resolve. To take another example, for me to say I regret that I broke the vase is just bizarre if you and I both know perfectly well that someone else broke the vase; I regret that I broke the vase requires for its felicity that we both know—or are willing to take as a given—that I broke the vase.
Since this view depends on our mutual beliefs about the presupposition rather than its truth, it’s pragmatic—and by virtue of being pragmatic, it has the handy feature of also accounting for the fact that the presupposition can be canceled or suspended, as we have seen, as well as the fact that context can affect its truth, and that presuppositions are not all of equal strength, as seen in (15):
(15) 
In (15a) we see our old friend the presupposed King of France, and we recall the sense of unease that we might feel (or at least that many people feel) at labeling this sentence false in the absence of a King of France. But no such sense of unease arises with (15b): In view of the nonexistence of a King of France, most people will happily state that (15b) is false. And correspondingly, most people would hesitate to say (in the present world) The King of France is not bald, but they’d be perfectly happy to assert that Jane did not have lunch with the King of France. And again, if someone were to assert (15a), the average person would have trouble responding with a simple That’s not true and leaving it at that, whereas that same person would have little trouble giving the same response to (15b). And finally, if someone were to query either of those two propositions—Is the King of France bald? and Did Jane have lunch with the King of France?, most people would feel more comfortable responding ‘no’ to the latter than the former, because such a response to the former would seem to accept the presupposition in a way that it doesn’t in the case of the latter. In short, the presupposition in (15a) is stronger than that in (15b). (See Strawson 1964 for discussion.)
One could go through a similar set of exercises with (15c) and (15d), neither of which presupposes the existence of a King of France as strongly as does (15a). In (15e), on the other hand, the factive verb realizes causes the presupposition in the lower clause to ‘percolate up’ to the higher clause, and again the sentence seems bizarre rather than simply false. Linguists call this the projection problem—the question of whether a presupposition in an embedded clause is preserved—or ‘projects’ up to—the larger clause that contains it. Some expressions, like factive verbs, are considered to be holes that allow this projection; others, including wish and believe, are considered to be plugs that do not allow it (see Karttunen (1973), who fills out the list with filters, which sometimes do and sometimes do not allow it), though some researchers have argued against the existence of plugs (e.g., Levinson 1983), and even Karttunen acknowledges that plugs tend to ‘leak’.
A final example of variability in presupposition can be seen in temporal expressions like before:
(16) 
In general, to say that A happened before B is to presuppose that B happened: Both (16a) and (16b) presuppose that I ate my dessert. But now consider (17):
(17) 
Here, there is no presupposition that Felix finished his dessert; quite the contrary, in fact. In sum, we see that there are many contexts in which presuppositions don’t hold, can be canceled or suspended, or vary in strength. All of these can be accounted for by thinking of presuppositions as pragmatic rather than semantic. And thinking of them as constituting the common ground for an utterance accounts for our sense that if the presupposition doesn’t hold, the sentence isn’t simply false; instead, a more complicated failure has occurred. If the presupposition is the common ground, it makes sense that its falsity would render the entire utterance bizarre; why are we treating the King of France as common ground if we both know perfectly well that there’s no such person?
Accommodation
Unfortunately, we still haven’t resolved all of the issues surrounding presupposition. It’s all well and good to say that it’s a pragmatic phenomenon involving what we take as common ground (or uncontroversial, nonasserted, mutually known, given, etc., depending on the researcher), but there are plenty of examples in which we treat information as presupposed that isn’t actually common ground at all. One interesting example is given in (13), from the Second Amendment (repeated here):
(18) 
When this amendment was added to the Constitution, the presupposition may indeed have been common ground; that is, it may have been obvious to everyone (or nearly everyone) that a well-regulated militia was necessary to the security of a free state. But times have changed, and now most people probably agree that such a militia isn’t actually necessary to the security of a free state (given that, for example, the US itself lacks one). So what’s common ground at one moment may not constitute common ground at another. But there are other cases in which, even at the time of utterance, the presupposition is not shared between the interlocutors:
(19) 
These are fine even if the addressee has no previous knowledge that the speaker has a dog, a brother, or a garage, respectively. Instead of actually being common ground, the information in question is being treated as common ground, and by treating it in this way, the speaker causes it to in fact become common ground. And this isn’t just a matter of definiteness; we see it with our other presupposition triggers as well:
(20) 
In (20a), we see the factive verb regret introducing a clause that isn’t actually expected to constitute common ground for the speaker and the addressees; instead, it’s being used to inform the addressees of the fact that the performance has been canceled. In (20b), the cleft structure is being used to inform the reader of a fact of which they are assumed to be unaware, which is that Ignacy Krasicki published the first modern Polish novel. And in (20c), the change-of-state verb stop triggers the presupposition that the speaker used to drive to Munich, though the addressees (readers of tripadvisor.com) aren’t expected to have previously known this. Thus, just as with the definites in (19), the supposed ‘presupposition triggers’ in (20) are used with information that isn’t previously shared by the interlocutors—and in (20a)–(20b), the whole purpose of the utterance is to present these facts as new information. Example (20b) is an instance of what Prince (1978) calls an ‘informative-presupposition’ it-cleft, which is distinct in function from other clefts and serves to present the supposed presupposition as new information. Example (20a) is similar, in that the complement of the factive verb not only is not being taken for granted, but is in fact the point of the utterance. So we can distinguish a class of utterance with the form of a presupposition but an informative function.
Perhaps even more interesting are the other cases—the definites in (19) and the change-of-state verb in (20c), in which the presupposed material isn’t the point of the utterance. The goal of the utterance in (19a) isn’t to inform the hearer that the speaker has a dog, and similarly for the other examples—but it’s not information that’s assumed to be common ground, either. Rather, these utterances make use of what Lewis (1979) calls accommodation: By treating a piece of information (like ‘I have a dog’) as though it were common ground, the speaker cues the hearer to likewise treat it as though it were common ground—and by doing so, the speaker succeeds in actually causing it to become common ground. As Lewis puts it, it’s like ‘changing the score’ on the scoreboard of the conversation: A moment ago, the proposition ‘speaker has a dog’ wasn’t in the shared discourse model, but now it is. By treating the proposition as though it were shared, the speaker causes it to in fact be shared. From the hearer’s point of view, the reasoning is roughly: If the speaker is treating this dog as though we both know that it exists, I’ll give them the benefit of the doubt and assume that it does in fact exist; I’ll accommodate it. And now it is shared knowledge.
We can, in fact, use accommodation to gently—or insidiously, depending on the context—affect our hearers’ beliefs. Loftus and Zanni (1975) ran a clever study in which they showed subjects a video of a car crash, and afterward asked them about what they had seen. Half of the subjects were asked, “Did you see a broken headlight?” and the other half were asked, “Did you see the broken headlight?” Despite the fact that there was actually no broken headlight in the video, the subjects who were asked the version with a definite noun phrase—that is, the version that presupposed the existence of a broken headlight—were far more likely to answer “yes”; that is, the presupposition seems to have influenced their belief (or at least their response): They now had a broken headlight in their model of the accident.
You won’t be at all surprised, however, to learn that accommodation also has its problems. First off, there’s the fact that not everything can be accommodated. Second, there’s the related problem of how to know what can and what can’t be accommodated. And finally, there’s the problem of how we can formulate a testable account of presupposition if every violation of our account is simply labeled an ‘accommodation’ rather than a counterexample. We’ll take these problems in order.
First, let’s examine some cases in which accommodation doesn’t work:
(21) 
If someone has invited you to a party, (21a) is an odd way to refuse—not because parties are better than pies (they’re not), but because in the absence of prior shared knowledge of the pie in question, the hearer can’t easily accommodate it. Likewise, if the melon is not previously known, (21b) is bizarre; and as a way of informing someone that you can’t make it to their party, (21c) is also odd.
To be fair, the fact that not everything can be accommodated shouldn’t quite be considered a problem; if it were the case that absolutely anything could be accommodated, that would mean that absolutely anything could be presupposed, which in turn would leave presupposition without a meaning or function. But it does bring us to the second problem: Why the difference? Why is it okay to presuppose a previously unknown dog (19a) but not a previously unknown pie (21a)? Why is it okay to use a possessive for a previously unknown brother (19b) but not a previously unknown melon (21b)? And why is it okay to use an informative-presupposition it-cleft to convey information about a Polish novelist (20b) but not to convey your inability to attend a party (21c)? The question is how we know what can and can’t be accommodated.
Some researchers have reframed the notion of presupposition as uncontroversial or nonasserted content (see, e.g., Abbott 2000, 2008), which helps with, for example, previously unknown brothers; it’s uncontroversial that someone might have a brother, and that’s not what the speaker’s main assertion is in (19b). But that still doesn’t help us with the equally uncontroversial melon in (21b): People have brothers and people have melons, but the one sounds fine and the other sounds just plain silly. Is it something about the categories in question? After all, you could replace my brother in (19b) with virtually any other family member (and change the later pronoun accordingly) and get a felicitously accommodated presupposition: my mother, my uncle, my daughter, my cousin…And replacing melon with another food gets similarly odd results: his avocado, his carrots…But not all food is quite that bad: John cut himself yesterday slicing his steak seems fine, and evokes a scenario in which John was eating steak for dinner when the event occurred. And his bread seems much better than his avocado or his onion, despite the fact that it’s not especially more controversial for someone to be slicing an onion than to be slicing bread. You can multiply the examples for yourself in other semantic categories. And the syntactic context matters, too; virtually any food can appear in an utterance like (22):
(22) 
Here, it seems that in the context of a request and mention of the kitchen, the hearer is likely to accommodate almost anything one might reasonably find in a kitchen. (I feel as though clock and dictionary work here, too.)
As you’ve probably noticed, the question of when accommodation is possible overlaps with the question we examined in the last chapter of when definiteness is felicitous. So, recall the following examples from that chapter:
(23) 
Given that definiteness is a trigger for presupposition, it’s very hard to tease apart whether the problem here is a question of definiteness or of presupposition. That is to say, is the infelicity in (23b) due to the fact that the constraint on definiteness (whether it’s framed in terms of familiarity, uniqueness, or something else) hasn’t been satisfied? Or is it due to the fact that a cab is being presupposed that cannot be accommodated in the context? Or to phrase it in terms of luckless John and his slicing ability:
(24) 
Again, is the problem in (24b) that the constraint on definiteness has not been met, or that the melon cannot be accommodated in context? Or is there no real difference between those two questions; that is, are the constraints on accommodation the same as those on definiteness? But notice the similarity in the issues surrounding definiteness and presupposition: In both cases, it’s difficult to arrive at a unified account of the phenomenon, in both cases, both semantic and pragmatic accounts have been proposed, both involve tricky issues of mutual knowledge, and both are somehow ‘rescued’ by accommodation. The jury is still out on all of these questions.
And that, at last, brings us to the third problem: how to formulate a noncircular account of presupposition. The problem is this: Any account of presupposition will include (or constitute) an account of what is presupposed in a discourse. So suppose we say that presupposed material is material that’s in the common ground. If we can presuppose material that is not in the common ground and count on our hearer to accommodate it, that means there’s no constraint on what can be presupposed. If, instead, we come up with a constraint on what can and can’t be accommodated, how can we distinguish ‘normal’ presupposed material from that which does not satisfy the constraints on presupposition but instead is simply being accommodated? How can we come up with a constraint on presupposition that distinguishes between presupposed and nonpresupposed material if there’s an additional category of ‘accommodated’ material that can be treated identically? An account of presupposition that says ‘you can presuppose material if and only if it’s in the common ground—and if it’s not, you can still presuppose it and call it accommodation’ essentially proposes a constraint and then states that the constraint needn’t be met, which means that it’s no constraint at all. And if anything can be accommodated, anything can be definite, and that’s clearly not the case, as seen in (25):
(25) 
So we’ve got an interesting problem: Not everything can be expressed with a definite, but when it can, it triggers a presupposition. Not everything can be presupposed, but when it cannot, we can accommodate it. But not everything can be accommodated, so we need to explain when it can and when it cannot—meaning there is still work to do. Meanwhile, we will encounter some of these issues again in the next chapter, where presupposition will be among the factors for determining the felicity of different noncanonical word orders within a sentence.