I will begin with some clarifications of my view. Firstly, by claiming that the Principle of Substitutivity does not hold, I do not mean its blanket failure. On the contrary, I believe that it only fails when some specific conditions are met. Secondly, I do not maintain that non-doxastic attitude operators prevent substitution of the embedded clauses that have the same semantic contents. Yet again, I believe the opposite—namely, that the semantic contents of the embedded clauses must manifest a difference (although this difference does not amount to the difference in truth conditions) in the cases where they produce substitutivity failure. Furthermore, I am aware that the recognized phenomenon illustrated by (12) and (13) is not completely unfamiliar as it has been discussed in the investigation of presupposition projection (e.g., see Heim 1992: 194–195). However, I believe that the particular explanation offered in the discussion concerning presuppositions is not satisfactory for the reasons presented in a moment.
Are presuppositions the key?
I will start with the following hypothesis. Looking at the earlier examples, one may say that the difference in the truth conditions of the ascriptions such as (12.a) and (12.b) lies, in fact, in a difference in presuppositions of the embedded clauses. Indeed, it is easy to illustrate the problem with using a textbook example of a pair of sentences with different presuppositions, e.g.,
a. Brad stopped smoking.
b. Brad smoked and he stopped smoking.
Observe now that, for instance, wondering whether (14.a) is the case is not the same as wondering whether (14.b) is the case. Generally speaking, non-doxastic attitude ascriptions with (14.a) and (14.b) as the relative clauses have different truth conditions. It seems that much the same explanation would hold for the earlier examples. It is standardly assumed that (11.a) presupposes that Sarah is late while (11.b) does not trigger such a presupposition. Likewise, we may try to explain the difference between (10.a) and (10.b). As some theorists have argued, adverbs of manner are presupposition-triggers in the sense that a sentence of the form “S Ves ADly” presupposes that S Ves (see: Abbott 2000). In sum, the hypothesis is that the truth-conditional asymmetry between (12.a) and (12.b), or between (13.a) and (13.b), is a result of presuppositional differences of their embedded clauses.
The above hypothesis can actually provide a precise explanation why the Substitutivity Principle fails in the considered cases. Namely, we are able to explain the truth-conditional asymmetry between these ascriptions by an appeal to some well-established facts about “presupposition projection”. In brief, if p presupposes q, then a sentence of the form “S wants that p” (or with a similar non-doxastic attitude verb) presupposes that S believes that q (see: Karttunen 1974; Heim 1992). For example, the statement that Joan wants Brad to stop smoking presupposes that Joan actually believes that Brad smokes. This observation is enough to justify the difference between (2) and (3) as embedded in the context of (8) (repeated below):
Hans wonders whether there is a ghost in his attic.
Namely, the second statement (8.b) is infelicitous because it triggers a presupposition that Hans believes that there is a ghost, which contradicts the first part of the utterance, i.e., that Hans wonders whether there is a ghost. Altogether, the difference between (2) and (3)—and ascriptions (12)-(13) as well—is explained by a theory of presupposition projection.
Once we grant that the presuppositional analysis is the key to the truth-conditional asymmetry between pairs such as (2) and (3), it will be bad news for the quantificational theory of descriptions. This is because, in a standard version, this theory does not claim that descriptions have existential presuppositions. In particular, for a diehard Russellianist, subject-predicate and existential-there sentences do not differ in presuppositions. Hence, if (i) the apparent substitutivity failure is strictly related to the difference in presuppositions of the embedded clauses, (ii) from the viewpoint of Russellianism, the presuppositions of the embedded clauses in (2) and (3) are identical, then how can Russellianism explain the truth conditional asymmetry between these ascriptions?
Nonetheless, I believe that the presented objection to Russellianism is not effective since the above considerations have been on a wrong track from the very beginning. Namely, I do not think that the problem with the Principle of Substitutivity can be explained away in terms of presuppositional differences of the embedded clauses in attitude ascriptions. Let me start with an observation provided by Schoubye (2013: 514–15).Footnote 6 The author considers a theory which combines the quantificational analysis with the presuppositional one—the view which claims both that subject-predicate sentences have Russellian truth conditions and that a use of “the F” imposes the requirement that it is common ground that there is a unique F. Given such a view, we are ready to explain the contrast in (8). However, the problem has not been solved at all. The point is that if there were only a presuppositional difference between (2) and (3) (i.e., they both have the same content but the former presupposes that Hans believes in the existence of a unique ghost, while the latter does not), they both should have the same truth value under the supposition that this presupposition is satisfied—or so Schoubye claims. Yet, even in such a case, the truth-conditional difference emerges. Let us compare (15.a) and (15.b) below:
Hans believes that there is a unique ghost and
Obviously, (15.a) and (15.b) may have a different truth value, that is, Hans may want the ghost to be quiet without wanting the existence of a ghost.
Shoubye is definitely correct in saying that the above attitude ascriptions have different truth conditions. Yet, the question arises whether the presuppositional analysis cannot in fact capture this difference. As a matter of fact, Heim’s prominent analysis (1992) cannot distinguish between the meanings of (15.a) and (15.b).Footnote 7 Yet, if we appeal to some other theories of presuppositions which deal with attitude ascriptions, we will be able to show that these sentences are not equivalent, since the presupposition that there exists a ghost—triggered by the embedded clause in (15.b)—will be analyzed as one that escapes the scope of the attitude verb (see Geurts 1998 or Maier 2015 for concrete formal analysesFootnote 8). However, if we modify the example a little bit, we will get a sentence that will have a problematic reading according to these theories, despite the fact that the mere presuppositional analysis will be correct. This will show that the problem with the Substitutivity Principle is not related to an inadequate analysis of the presupposition behavior. Consider the following two ascriptions:
Hans believes that there exists a ghost and
The first version (15.c) is reported by native speakers to be “true” in the story about Hans, while the second (15.d) is definitely not.Footnote 9 The problem is that many formal analyses of attitude ascriptions based on possible-worlds semantics seem to predict that (15.c) entails (15.d). Take Maier (2015) for example. In his account—set in the framework of DRT—the mental representations of (15.c), (15.d) would have the following formsFootnote 10, Footnote 11:
Given the semantic system presented by Maier (2015, sec. 3.2), the subDRS in the scope of “Att” in DRS-15d (Fig. 1) will capture the mental state of Hans if the corresponding subDRS in DRS-15c (Fig. 2) does so.Footnote 12
The above observation signalizes that prominent theories of presuppositions cannot capture the difference between (15.c) and (15.d) in the sense that the latter is validated by the former within the formal semantic frameworks adopted by those theories. Most importantly, we can see that the problem with the Substitutivity Principle is a phenomenon quite independent from the issue of how presuppositions behave in attitude ascriptions. The problem that (15.c) counter-intuitively validates (15.d) seems to be generated by the fact that in (15.c), the information that there exists a ghost in the state of affairs desired by Hans is smuggled in the modified nominal “a quiet ghost”. Hence, although the presupposition of the existence of a ghost disappears from the scope of desire in (15.c), the asserted content of the embedded clause entails that the individual is a (quiet) ghost, which, in turn, entails that there exists a ghost (and it is quiet). In other words, although we eliminate the problematic component entailing the existence of a ghost from the scope of “wants”—thanks to our presuppositional analysis—it “comes back” in the asserted content as a consequence of using an adjective combined with a noun. This shows that no matter how well we treat presuppositions, the problem with the Substitutivity Principle may show its teeth anyway.
Another kind of evidence against the presuppositional explanation of the problem with the Substitutivity Principle is that there are cases where the principle fails but there are no apparent presuppositional differences between the embedded clauses of the ascriptions. The following pairs include typical instances of entailments:
a. Fred lied in court. ⊧
b. Fred was in court at least once.
a. The president has been assassinated. ⊧
b. The president is dead.
Observe now that these (b)-entailments do not project out of non-monotonic environments, which is good evidence that they are not presuppositions of the (a)-sentences. Consider some examples with negations and conditionals:
a. Fred did not lie in court. (He lied in a private conservation and has never been in court.)
b. If Fred has ever lied in court, I do not want to hire him.
a. The president has not been assassinated by the kidnappers. (Luckily, he is still alive.)
b. If the president has been assassinated by the kidnappers, I will be very sad. (I hope that he is still alive.)
As we may observe, neither (18.a) nor (18.b) presupposes that Fred was in court and a similar remark can be made about (19.a), (19.b)—none of these presupposes that the president is dead. Treating these examples as representative, we may conclude that the entailments (16.b), (17.b) do not generally survive embeddings and thus are not presuppositions of (16.a), (17.a). Consequently, the following pairs of sentences which are plainly equivalent do not differ in terms of their presuppositionsFootnote 13:
a. Fred lied in court ⇔
b. Fred was in court at least once and Fred lied in court.
a. The president has been assassinated ⇔
b. The president is dead and he has been assassinated.
Observe now that it is not difficult to come up with examples of non-doxastic attitude ascriptions containing these sentences as the embedded clauses, which have significantly different truth conditions. Consider:
a. Anna regrets that Fred lied in court.
b. Anna regrets that Fred was in court at least once and that Fred lied in court.
a. I wonder whether the president has been assassinated.
b. I wonder whether the president is dead and he has been assassinated.
Ascription (22.a) says that Anna regrets Fred’s lying in court and not his being in court, while (22.b) clearly commits Anna to regret the latter. Ascription (23.a) can be true when I know that the president is dead and wonder whether his death was natural or he has been killed; on the other hand, (23.b) plainly assumes that it is a part of my wondering whether the president is dead.
To conclude, we see that the problem with the Principle of Substitutivity is not restricted to the cases with presuppositional differences, and this shows that any explanation of the problem which appeals to the theory of presupposition projection does not get to the heart of the puzzle.
Two features of non-doxastic attitudes
In this subsection, I want to say a few words about why the Principle of Substitutivity fails and in particular, to specify what circumstances actually entail its failure. Before I move on to my explanation, some further clarifications are in order. I do not aim by any means to present in this subsection a full-blown theory of non-doxastic attitude ascriptions, a theory which, among others, is able to fully predict what modifications of the embedded clauses in such ascriptions preserve the truth conditions and what modifications do not. Developing such an account is beyond the scope of this paper. What I only aim to show is that a there is a certain kind of a modification which alternates the truth conditions and that the examples with descriptions (i.e., examples (2) and (3)) actually involve this kind of modification.
In my view, the Principle of Substitutivity fails in the cases where non-doxastic attitude ascriptions exhibits two features (in presenting these features, I use “want” as a paradigmatic example of a non-doxastic attitude verb). The first one has been actually recognized by Kaplan, namely, non-doxastic attitude verbs do not support entailments:
There are cases in which S wants that p, p entails q, but S does not want that q.
As we have already seen, there are good examples which illustrate the above claim. The second feature of the non-doxastic attitude verbs is that they usually support at least some kinds of entailments, in particular, the following one:
If S wants that p and q, then S wants p and S wants q.
This assumption seems to be very plausible in light of the meaning of “and”. When I say that I want p and q to be the case, I intuitively mean that I want both conjuncts to be the case, i.e., I want p and I want q. That is, non-doxastic attitude verbs support conjunction elimination.
Some theorists have suggested that “wants” does not actually support conjunction elimination. Namely, there are cases where someone wants both that p and q, but does not want p at the same time. Consider the following statements:
Nicolas wants to fly on the Concorde and pay nothing for the trip,
John wants to swim in the shark-infested waters and survive.Footnote 14
For example, assume that Nicolas does not want to spend all his savings on buying a ticket for a Concorde flight but he would like to fly on the Concorde if someone else paid for his flight. In such a case, (24) is intuitively true, while (26) seems to be false:
Nicolas wants to fly on the Concorde.
A similar story can be made about (25). Nonetheless, I am not entirely convinced that (24) is a real counterexample to Feature 2. The problem is that the intuition saying that (26) is false in the described circumstances is not fully appealing. Observe that falsity of (26) entails that Nicolas does not want to fly on the Concorde, but this claim sounds contradictorily in the context of (27):
# Nicolas wants to fly on the Concorde and pay nothing for the trip. And he does not want to fly on the Concorde.
It is infelicitous to say that Nicolas does not want to do p right after saying that he wants to do p under such-and-such conditions. However, if both claims are compatible and true, how can we then explain the infelicity? Naturally, it would not be infelicitous to add that Nicolas does not want fly on the Concorde under any condition—but this is not what is claimed here.
Another way to highlight the conflict between the negation of (26) and (24) is to consider a situation of responding “Yes/No” to a straight question whether Nicolas wants to fly on the Concorde. Consider the following dialogue between Mary and Nicolas:
Observe that a denial of the content of the question sounds somewhat infelicitous when the respondent at the same time claims that he wants to fly and not pay for it. Again, this suggests that (26) cannot be false (or its negation cannot be true) when (24) is true. A quite similar doubt can be rephrased with respect to (25).
Altogether, it is questionable whether (24)/(25) is a good counterexample to the claim that non-doxastic attitude support conjunction elimination (Feature 2). The reason why it appears to be so is, in my view, a result of confusing two ways of understanding desire ascriptions. Firstly, the ascription “I want p” may simply express that p is, for some reasons, attractive to me. For example, flying on the Concorde may be attractive because of the feeling of speed. Let me call this kind of attitudes “straight desires”. Secondly, by saying “I want p”, I may convey a thought equivalent to “I want p all things considered”, “all in all, I want p”. In this case, the ascription should be interpreted as an expression of the subject’s willingness to take up a certain activity, which is based on careful considerations including an estimation of all benefits and possibilities of risks of taking up that activity. For example, I may want to work in my own country all things considered, because I have come to the conclusion that it is more important to me to be close to my family than to get a better salary abroad. (Still, working abroad may be attractive to me.) Let me call this second kind of wanting “a rational desire”. It is important to emphasize that wanting p all things considered does not entail wanting p no matter what will eventually happen. For example, I may want to fly on the Concorde all things considered, because I have taken into account the fact that plane accidents are extremely rare; however, if I were informed that the plane would be shot down, I would certainly not want to fly on the Concorde in such a case. Wanting p all things considered implies that—so far as I know—obtaining p does not exclude fulfilling my other desires which are more important to me than p itself (like saving my life and etcetera).
The ascriptions of the two indicated kinds of desires are not equivalent in the sense that S may have a straight desire while not having a rational desire of doing p. This will be the case when p is attractive to S, but she has different desires which stay in conflict with p, that is, cannot be jointly fulfilled with p, and which are stronger or more important to her; thus, in such a case, S does not have a rational desire of p, that is, she does not want p all things considered.
When it comes to examples (24) and (25), I would say that Nicolas does not want to fly on the Concorde all things considered, but he does want to fly in the sense that a ride on the Concorde is attractive to him. Similarly, John does not want to swim with sharks all things considered (because he does not want to die—which is a stronger desire than, e.g., making an impression on his friends). But he may want to swim with sharks in the straight sense, that is, in order to make an impression on his friends. So the intuition that (26) is false is an effect of taking it as an ascription of the rational desire. However, (26) is true when interpreted as an ascription of the straight desire. Altogether, “wants” seems to support conjunction elimination so far as we understand it as a feature of straight desires ascriptions.
Finally, let me observe—quite independently from the above considerations—that a rejection of Feature 2 would be problematic in itself. In most cases, ordinary speakers of language seem to infer the conclusion that S has attitude Α towards p (or towards q), once they learn that “S Αes p and q”. If we reject Feature 2, how are we going to explain that the inferences like the following ones are intuitively valid?
a. Fiona wonders whether to go the cinema and visit Joan beforehand.
b. So Fiona wonders whether to go to the cinema.
a. Johnny wants to have a dog and to go for a walk with it every day.
b. So Johnny wants to have a dog.
It seems that the conclusions (28.b), (29.b) simply follow from the premises (28.a), (29.a) respectively. In passing, let me note that Shoubye (2013, ft. 18) acknowledges that it is reasonable to infer the conclusion that Hans wants there to be a unique ghost in his attic from (3) (repeated below)
Hans wants there to be a unique ghost in his attic and for it to be quiet.
What licenses this conclusion is exactly Feature 2.
Once we agree that non-doxastic attitude verbs support conjunction elimination but do not generally support entailments, the Principle of Substitutivity is no longer valid. We can demonstrate it briefly below.
The Argument against Principle of Substitutivity
For the sake of indirect proof, assume that the Principle of Substitutivity is actually valid. From Feature 1, we know that there is S, p, q such that (i) S wants that p, (ii) p entails q, (iii) S does not want q. Fix such S, p, q. Consider the statement “q and p”. Observe that since p entails q, the statement “q and p” is equivalent to p. From the Principle and (i), we derive “S wants that q and p”. From Feature 2, we can conclude then that S wants that q; contradiction with (iii).
The argument shows that the Principle of Substitutivity fails in light of the fact that non-doxastic attitudes have Features 1 and 2. More importantly, it shows how a counterexample to the Principle can be made. Take such p and q that exemplify Feature 1; observe that the conjunction “q and p” is actually equivalent to p; assuming Feature 2, you already have got a counterexample to the Principle of Substitutivity: p, “q and p”.
In my view, the example involving definite descriptions actually belongs to the indicated class of counterexamples to the Principle of Substitutivity (assuming perhaps slight adjustments). Firstly, let us observe that the following entailment holds:
a. The ghost in the attic is quiet. ⊧
b. There is a unique ghost in the attic.
This seems to be correct, though the way how we exactly characterize this relation depends on the treatment of definite descriptions. On the quantificational account, (30.b) is a logical consequence of (30.a); on the presuppositional account, in turn, it is claimed that a felicitous utterance of the first one presupposes the second one. However, it is important to observe that even on this account, we may say that the truth of (30.a) “guarantees” the truth of (30.b) in the sense that a felicitous utterance of the former—the one which expresses the truth or falsehood—requires the latter to be satisfied (in other words, provided that (30.a) expresses the truth in a context, it suffices to conclude that (30.b) must also be true). Secondly, we assume that in the context of the story about the ghost, Hans wants (30.a) but not (30.b) to be the case (Feature 1). Thirdly, we observe that (30.a) is equivalent to a sentence being a conjunction of it and (30.b) (written below as (31)):Footnote 15
There is a unique ghost in the attic and the ghost in the attic is quiet.
However, Hans does not want (31) to be the case. It is because this ascription implies—in line with Feature 2—that Hans wants there to be a unique ghost. Altogether, the example with definite descriptions is constructed in such a way that it falls into exactly this class of counterexamples to the Principle of Substitutivity, which is indicated in the above argument. (Let me note here that even if one insists that not all non-doxastic attitude ascriptions support conjunction elimination, the example with the ghost does not seem to be the case where this principle actually fails.)
Altogether, the consideration of this section delivers the conclusion that the failure of the Principle of Substitutivity is a result of the fact that non-doxastic attitude ascriptions support only some kinds of entailments, particularly, conjunction elimination. As I emphasized at the beginning, by saying that this principle fails, I do not mean that every substitution of the embedded clause in a non-doxastic attitude ascription with an equivalent clause must produce a difference in the truth conditions. Obviously, there can be modifications of the embedded clause which preserve the truth conditions. For example, consider the following pair:
a. I want Sarah to be exactly 5 min late.
b. I want Sarah to be late and her delay to be exactly 5 min.
It is hard to imagine a situation in which (32.a) would have a different truth-value than (32.b). One may wonder what I can say about these cases. As I have said, I have not promised a full account of non-doxastic attitude ascription. Nonetheless, I believe that a certain explanation can be offered, based on what was said above.
What I have established is that when S Αes that p, p entails q, and S does not Α that q, then we have a counterexample to the Principle of Substitutivity (p, “q and p”). Certainly, it would not be a counterexample if S Αes that p and Αes that q, contrary to the earlier assumption. However, observe that this situation actually obtains in the above case (32). Namely, if I want Sarah to be exactly 5 min late, then I want her to be late (at least, such a general desire can be ascribed to me). Hence, no confusion arises if someone says that I want both, i.e., I want Sarah to be late and her delay to be precisely 5 min—which is exactly what (32.b) says. In general, my hypothesis is that a modification of the embedded clause alternates the truth conditions of an ascription if and only if—roughly speaking—it “splits” the clause into separate “pieces” (which are entailed by the clause and which, taken together, entail the clause itself) and the subject does not hold the attitude in question towards some of these “pieces”. In the above case, we do not encounter such a phenomenon.
Finally, I want to indicate a certain corollary of the above analysis of non-doxastic attitude ascriptions. As it is usually assumed, attitude ascriptions are generally statements of relations between the subjects and “propositional contents”. That is to say, “Α” used in “S Αes that p” expresses a relation between S and the propositional content of p. However, in light of the above observations, our choice of an account of propositional contents is somehow limited. Namely, we cannot adopt any account which implies that analytically equivalent sentences actually express the same proposition. Provided that such sentences express the same propositions, a hopeless puzzle arises of how it is possible that they yield different truth conditions while being embedded under propositional operators. To sum up, an important upshot of my analysis of non-doxastic attitude ascriptions is that, given the propositional-accounts, only the fine-grained conceptions of propositions are correct (e.g., Larson and Ludlow 1993; King 2013).
Another corollary of my analysis is that the semantics of attitude ascriptions must be more complicated than one expects as they do not allow for even such “innocent” substitutions as the ones illustrated by examples (22.a, b), (23.a, b). We need a characterization which will block the inferences from (a)-sentences to (b)-sentences in these pairs of ascriptions—one which is surely more sophisticated than Hintikka’s or even Stalnaker’s semantic frameworks. It is worth saying at this point that many philosophers or linguists have recently signalized various problems concerning some specific features of attitude ascriptions and proposed elaborated semantical analyses which focus on addressing these particular problems, and they step away from the standard accounts (e.g., see Anand and Hacquard 2013 on epistemic modals embedded in attitude ascriptions, or Ninan 2012 on de re uses of the ascriptions of counterfactual attitudes).