Sometimes a Great Notion: A Critical Notice of Mark Crimmins’
Talk about Beliefss (Cambridge, MA: MIT Press, 1992. pp. xi + 214).
Anyone weary of endless philosophical debate on belief reports will find welcome relief in this book. Talking not just about belief talk but about belief itself, it offers much that is new, interesting, and subtle. The central thesis, though interestingly and subtly developed, is not exactly new. It is a version of the “hidden indexical theory” (HIT) of belief reports invented some time ago by Stephen Schiffer (1977). But Crimmins has given HIT new punch. I will take up, of the many interesting and often subtle claims sprinkled throughout the book, only those most relevant to the overall theory. Otherwise, suffice it to say that the only way to appreciate this engaging and rewarding book is to read it. The worries to be registered here are not meant to discourage that.
Crimmins focuses exclusively on singular belief reports, which take the form ‘S believes that b is F’, where ‘b’ is a referential term. Assume that belief is a relation to a proposition and that an individual or a property cannot be thought of simpliciter but only in some way or another, and consider the following example,
(1) Penguin believes that Batman is brawny.
where we will pretend, lest we get ensnarled in special problems about fiction, that these characters are real people, residents of that well-known metropolis Gotham City. Ostensibly, what (1) does is to specify a relation between Penguin and the proposition that Batman is brawny. Popular wisdom these days has it that this is a singular, “Russellian” proposition, literally containing Batman and the property of being brawny. This conflicts, but for good reason, with the traditional Fregean view of propositions as containing “modes of presentation” (ways of thinking of individuals or properties). The trouble with the Fregean view is that it violates “semantic innocence”—it has a sentence like ‘Batman is brawny’ expressing something different when embedded in a belief context than when standing alone. But semantic innocence has a problem of its own.
This is the well-known substitution problem that arises when referential terms occur in belief contexts. According to what Crimmins calls the “naive” view, belief is simply a relation to a Russellian proposition and belief reports simply report instances of that relation. That’s what a belief report like (1) appears to do, describing Penguin as believing the singular proposition that Batman is brawny. Yet (2) seems to be false,
(2) Penguin believes that Bruce Wayne is brawny.
even though Bruce Wayne is Batman. Given semantic innocence, the ‘that’-clauses in (1) and (2) express the very same proposition, which contains Batman = Bruce Wayne, and how he is referred to or thought of does not enter into that proposition. So how can (1) and (2) themselves, identical except for the occurrence of different names for the same individual, differ in truth value?
On the naive view, they don’t: (2) is true, just like (1). Not only that, (1) is not contradicted by (3) nor is (2) by (4).
(3) Penguin believes that Batman is scrawny
(4) Penguin believes that Bruce Wayne is scrawny.
They are all true, however odd that may seem in the case of (2) and (3). Indeed, on the naive view Penguin believes not only that Bruce Wayne is not Batman but also that Bruce Wayne is Batman, although he doesn’t realize this. That is counterintuitive. The intuitive trouble with the naive view, despite its semantic innocence, is that it allows that one can unwittingly believe blatantly contradictory propositions and coherently believe a proposition while being mistaken about which proposition it is.
Nathan Salmon (1986) has ingeniously defended the naive view. Yes, belief is a two-place relation between believers and propositions, but equally it is the “existential generalization” of a three-place relation ”BEL,” whose third place is filled by a “guise” (a way of believing). The naive view’s seemingly awkward implications about belief are really acceptable consequences of the fact that one can coherently BEL a proposition under one guise, disbelieve it under another, and withhold belief from it under a third. However, belief reports are silent as to the guise under which the ascribed belief is held. Even so, our intuitions are implicitly sensitive to guises, to specific ways of thinking of Batman/Bruce Wayne, so that we accept (1) and (4) but reject (2) and (3). We read something into them that is merely “pragmatically imparted.” For while they specify the proposition believed, it is only by way of pragmatic implicature that they indicate (or at least constrain) the way in which it is believed. The latter doesn’t enter into truth conditions —it is not part of the semantic content of a belief report.
Crimmins finds the naive view unnacceptable, even with its refinements. For belief reports don’t just suggest information about how an agent is said to believe a proposition, this information is part of what they claim, part of what is “literally said.” But rather than just appeal to intuition, he rebuts various arguments for the naive view. He shows that the naive view is not an inevitable consequence of broad semantic principles, such as direct reference and compositionality, and that Grice’s criterion of cancelability lends little support to the naive view. He then proceeds to develop a more sophisticated view that is as semantically innocent as the naive view but is more intuitive.
Crimmins holds that belief reports do not simply express relations between believers and things believed but tacitly refer to ways of believing. The sentences used to make belief reports are like sentences containing indexicals, not true nor false simpliciter but only with respect to context-relative references, in this case to ways of believing. As a result, even though a pair of belief reports like (1) and (2) (or (3) and (4)) have the same agent believing the same thing (at the same time), (2) can be false even though (1) is true.
How can this be so unless they differ in some way? To be sure, they contain different names, ‘Batman’ and ‘Bruce Wayne’, but this cannot be the relevant difference, not if both names innocently make the same semantic contribution. In Crimmins’ view, the relevant difference does not meet the eye (or ear): the proposition expressed by a belief report contains an “unarticulated constituent,” namely, the way the agent is being said to believe the specified proposition. The occurrence of ‘believes’ determines that some way of believing is being tacitly referred to, but which way of believing this is, like any provision of an unarticulated constituent, is determined pragmatically (roughly, by virtue of relevance and salience). Believing is not a two-place but a three-place relation (Salmon’s BEL, in fact), and belief reports express instances of this relation, but not explicitly. Belief reports are like utterances of other sentences in which one term of the relevant relation is not expressed, such as,
(5) It is raining. [where?]
(6) Dumbo is small. [relative to what?]
(7) Daffy is finished. [with what?]
(8) Dewey is ready. [for what?]
In each case a tacit reference to (or, I should say, at least a quantification over) some additional argument or parameter is required in order for the utterance to express a (complete) proposition, to be true or false. But with belief reports there is no syntactic basis for positing a component of the sentence to provide that reference. Unlike the empty categories of Government-Binding Theory, for example, not only can you not see or hear this alleged component, it isn’t even there. Because “there is no expression in ‘It’s raining’ to serve as the indexical” (p. 17) referring to the relevant place, Crimmins describes what is missing as an “unarticulated constituent” of the expressed proposition.
Part of what distinguishes Crimmins’ view from the hidden-indexical view introduced by Schiffer (1977), as well as the intricate version developed by Mark Richard (1990), is that for him a way of believing is composed of particular representations. That is, belief reports literally refer to particular items in the minds or heads of believers. The unarticulated constituents in belief reports are particular representations because, according to Crimmins, what is reported are “instances of believing,” and these are individuated by what is believed together with how it is believed. One believes Russellian propositions by means of mental representations, which are concrete, “cognitive particulars,” not abstracta like belief-state types, belief properties, or Fregean modes of presentation (pp. 35-53). A way of believing is made up of “the concrete particular representations employed by the agent to represent objects, properties and relations that the proposition is about. The ‘how’ information conveyed by a belief report, then, must constrain the representations involved in the alleged belief” (p. 100). Without committing himself to any specific theory of the nature or format of mental representations, Crimmins calls representations of individuals “notions” and representations of properties and relations “ideas” (he doesn’t say how connectives and quantifiers fit into the picture). He isn’t sure “what determines a notion to be of a thing,” but he does insist that acquaintance with a thing is necessary for believing a singular proposition about it, i.e., for this thing to be a constituent of the proposition; otherwise, what is believed is merely a “general proposition,” in which case “the thing [can] bear on its truth [only] in a more indirect, contingent way” (p. 81). Whatever acquaintance is exactly, ”purely general knowledge … does not suffice for it” (p. 82). Nevertheless, Crimmins takes “a liberal view of what it takes to have a notion of a particular individual” (p. 86); “One can form notions of what one knows to exist” (p. 89).
Crimmins applies this liberal view to notions of notions and of ideas. It allows him to claim that “When we utter a belief sentence, we are talking about an agent’s ideas and notions, and these notions and ideas become unarticulated constituents of what we say. … What we claim is that the agent believes a certain proposition in a way such that certain ideas and notions are responsible for representing certain constituents of the proposition” (p. 152). We need not be reporting that the agent uses those notions and ideas to represent the proposition as a whole, however, for the belief need not be explicit. Indeed, belief reports often concern merely tacit beliefs and tend to be insensitive to the difference between the two kinds of belief. After making some perceptive points about previous accounts of tacit belief, Crimmins proposes a “virtual-belief” account, which he defends against various natural objections (pp. 65-73). His main purpose here, considering that beliefs need not be and generally are not explicit, is to prepare the way for explaining precisely how his account of belief reporting avoids imputing to the agent an actual representation of the proposition believed while maintaining that belief reports implicitly refer to certain of the agent’s notions and ideas of the constituents of that proposition.
If notions and ideas are tacitly referred to, how does a belief reporter manage to communicate which ones he is talking about in a given case? The intended way must be relevant and salient in the context of utterance. Crimmins does not go into details, but the idea is that the speaker, in Gricean fashion, intends his audience to rely on the presumption that what he saying is relevant and truthful and on that basis identifiable under the circumstances. In the case of tacitly referred to notions and ideas, most commonly the audience presumes that the speaker is talking about “normal” notions (and ideas), ones of a sort commonly associated with, e.g., the names ‘Batman’ and ‘Bruce Wayne’. But sometimes it will be clear that the relevant notions and ideas are, whether or not normal, the ones that characterize the agent’s perspective in perception or in action. Third, in the case of self-attributions the expressions the speaker uses in a ‘that’-clause may be presumed to provide the notions and ideas attached to his very use of those expressions. Finally, the expressions used in what are properly called de dicto belief reports provide notions and ideas that essentially involve those very expressions.
Crimmins’ discussion of de dicto reports (pp. 141-145) is of special interest because of how he draws the familiar analogy between them and Quine’s example, ‘Giorgione was so-called because of his size’. Crimmins argues that the name plays only its usual semantic role of referring to Giorgione and that the fact that another name of him cannot be substituted for ‘Giorgione’ salve veritate can be explained pragmatically. What explains “opacity” is that the reference of the demonstrative ‘so’ in ‘so-called’ is context-relative. The name ‘Giorgione’, rather than ‘Mr. Big’, say, serves as its reference simply because it is salient in the context; and occurring earlier in the sentence is just one way of being salient. In this way, what ‘Giorgione’ contributes is not its reference (or any other semantic property) but its presence. And so with the occurrence of proper names and other expressions in de dicto belief reports. As Crimmins argues (pp. 165-168), when mention of them is necessary to refer (tacitly) to a believer’s ways of thinking of the constituents of the proposition believed, the expressions don’t play any semantic role beyond their usual referential role. They make an additional contribution simply by their presence, but that contribution is merely pragmatic.
Now, let us ask, how are we to decide between Crimmins’ version of the hidden-indexical theory and Salmon’s naive view? (Let’s go along with their shared assumption that singular belief reports express relations to singular propositions.) Their essential difference is on whether ways of believing enter into the truth conditions of belief reports or are merely conveyed pragmatically. In my opinion, there are problems on both sides. Salmon has not explained how it could be that the word ‘believes’ should, unbeknownst to its users, express a two-place relation that is the merely the “existential generalization” of a relation for which there is no word. That would be a strange linguistic fact if it were a fact. Nor has he shown that because ways of thinking are “pragmatically imparted,” they do not enter into propositions expressed by belief reports. As Crimmins explains, this is not required by semantic innocence, and besides, innocence should not be counterintuitive. But has Crimmins shown that belief reports possess truth values only relative to tacitly referred to ways of believing?
First of all, to sustain HIT Crimmins needs to show that a way of believing is an argument of the relation of believing. This certainly doesn’t follow from the mere fact that to believe something one must believe it in some way or another. One must walk at some pace or another, but that doesn’t make walking a relation between walkers and paces. Perhaps a way of believing is like pace with respect to walking (or volume respect to talking). Only if ways of believing are constituents of the propositions expressed by belief reports can it be supposed that belief reports must refer to them. Only then would failing to refer, even tacitly, to a way of believing the thing believed entail that the belief report did not express a complete proposition. A belief report would be like an utterance of (7) or (8), for example,
(7) Daffy is finished.
(8) Dewey is ready.
which are neither true nor false except relative to some unspecified argument or parameter. That is, the speaker must mean something more specific, e.g. that Daffy has finished eating or that Dewey is ready to sing. Now in these cases the tacit reference is required by the lexical entries for ‘finished’ and ‘ready’, both of which take complements, but only optionally. In the case of belief reports, however, it is not clear that there is any lexical basis in the word ‘believe’ for the alleged semantic requirement that there be a tacit reference to a way of believing. Crimmins hasn’t really ruled out the alternative, that belief reports do not require tacit reference to unarticulated constituents. Perhaps they are more like (9) and (10),
(9) Everyone is going to get drunk.
(10) Scrooge has had a bath.
utterances of which are likely to involve, but do not require, tacit reference to unarticulated constituents, in these cases to a restricted domain of quantification. This is evident if we compare them to the syntactically parallel sentences (11) and (12),
(11) Everyone is going to die.
(12) Scrooge has had a heart attack.
whose typical utterances do not involve a tacit reference to a restricted domain of quantification. Strictly speaking, (9) is true iff everyone in creation will get drunk and (10) is true iff Scrooge has had a bath at some previous time, but of course these are not what users of those sentences are likely to mean: taken literally, (9) is too improbable and (10) is too uninformative to be what the speaker means, and what the speaker does mean is surely more specific. It could be made explicit only with the insertion of additional words, like ‘at this party’ or ‘today’. What the speaker thus means is a pragmatic matter, going beyond the semantic content of the words actually used. However, it is not a matter of Gricean implicature, which involves meaning both what is said and something else as well, but what I call “impliciture” (in Bach forthcoming). In impliciture the speaker does not mean two things but only one, namely, some restricted version of what he is saying.
By employing the notion of impliciture rather than implicature, a proponent of the naive view does not have to say that tacitly referring to a way of believing is needed only for the truth-conditionally irrelevant purposes of making a belief report germane and informative and of keeping it from being misleading. He could accept Crimmins’ contention that the way of believing enters into the proposition expressed by a belief report but deny that this is the bare-bones proposition literally expressed by the sentence used to make the belief report. It is relevant to the truth condition all right, but only to the truth condition of an enriched version of what is said, not to the truth condition of what is said strictly speaking.
Even if Crimmins is correct in claiming that believing is a three-place relation with a way of believing as one of its terms, it doesn't follow that belief reports must tacitly refer to particular ways of believing in order to have truth values. Implicit existential quantification over them could suffice. This quantification would require not merely the existence of some set of notions and ideas of the constituents of the believed proposition but would impose a restriction on them, requiring each to be of a certain tacitly specified type (this is roughly how Schiffer formulates HIT). That would be enough to explain how, for example, unpuzzled Pierre could, without logical error, believe that London is pretty and believe that London is not pretty. This version of HIT avoids a serious flaw in Crimmins’ particularist version, on which belief reports are about “instances of believing,” as partly individuated by particular representations. He supposes that because belief reports are made true by instances of believing, they are about instances of believing. They are no more about instances of believing than a “running report,” such as ‘Bill ran to work today’, is about an instance of running. This report is true just in case there was that day an instance of running to work by Bill, but the report is not about that instance of running. The report would have been true even if Bill had run to work at a different time, in a different way, and by a different route. Its content is indifferent to which of all possible instances of running (to work that day by Bill) makes it true, and it is therefore not about any particular one. Besides, its content would be the same even if it were false and there were no instance of running for it to be about. Similarly, belief reports are not about instances of believing, even though instances of believing make them true. Thus there is no need for Crimmins to insist that belief reports typically refer to particular representations (notions and ideas).
Moreover, he has not really justified his assumption that when we make belief reports we are in a position to refer to notions and ideas, i.e., to “specify” rather than merely “describe” them (p. 156). Specifying them requires having notions of them. But on Crimmins’ liberal view of singular thought, one can form a notion of something if one is acquainted with it (in a loose sense, not Russell’s), thereby knowing it to exist (p. 89). Not only is this essentially the epistemic conception of singular thought that many find dubious, his formulation of it is either circular or regressive. If one can form a notion of what one knows to exist, one must have some other notion of it. Otherwise, the best one can do is think of it under a description, as the unique thing of a certain sort, but that, by Crimmins’ own lights, is not to specify it but merely to describe it. If his version of HIT is to have the distinctive character he claims for it, he needs to develop a clearly non-descriptional account of singular thought and show that it applies to thought of, and reference to, others’ notions and ideas. Instead, he simply speaks vaguely of how they are “provided in” or “supplied by” the context.
There are some other trouble spots in Crimmins’ account of belief reports. For one thing, it falsely implies that if the ‘that’-clause in a belief report expresses a singular proposition, then the belief being ascribed must be a belief in a singular proposition. If you say, “Van believes that Ortcutt hides in unlikely places” but it is evident that Van doesn’t realize that Ortcutt is the shortest spy, you could be ascribing to Van not a singular belief about Ortcutt but merely a descriptive belief about whoever is the shortest spy. Secondly, Crimmins’ treatment of the occurrence of empty names in belief contexts seems unable to handle the case in which the speaker (mistakenly) thinks the name refers. For example, suppose that on Christmas Eve one child says to another, “Billy believes that Santa Claus will be coming tonight.” Crimmins denies that the ‘that’-clause expresses a proposition for Billy to believe. But the speaker thinks it does, and his belief report may well be true. Crimmins seems committed to denying that the belief report even expresses a proposition. Finally, he appears to assume that the sentence embedded in the ‘that’-clause of a belief report is ipso facto a “content sentence” (p. 146), i.e., a sentence that specifies the content of the ascribed belief. Yet this need not be so. For example, seeing that something has caught Bill’s attention, you say, “Bill believes that something is happening.” You could be reporting that Bill has a certain very unspecific existential belief, but in fact, you are reporting that Bill has a belief whose specific content you are not giving (you could have added, “I know what it is but I won’t tell you”).
In sum, there are two main difficulties with Crimmins’ version of the hidden-indexical theory. First, its claim that a viable alternative to the naive view must invoke tacit reference to particular representations rather than to ways of believing (types of modes of presentation) relies on the dubious assumption that because instances of believing make belief reports true, belief reports must refer to instances of believing. Second, it does not explain how people reporting beliefs can have, much less communicate, singular thoughts of other’s notions and ideas. Unfortunately, having concentrated on the central thesis of Talk about Beliefs, I haven’t left space to extol its many virtues. The most distinctive one is its rigorous account of the structural relations between ways of believing and propositions believed, of how notions and ideas are “responsible” for constituents of propositions. As for its other virtues, look and see for yourself.
So Crimmins exaggerates when he bills his theory as “the new game in town” (p. 204). He does cite its immediate predecessor Crimmins and Perry (1989), as well as Richard (1990), but there is nary a mention of Schiffer (1977) or any place else where Schiffer has discussed HIT (he does not endorse it). This oversight is ironic, in light of the glowing (though conditional) praise from Schiffer quoted on the sleeve of Crimmins’ book.
He never explains why he limits himself so, thereby risking loss of generality, but presumably he is worried primarily about occurrences of singular terms in belief reports.
On both views a proposition is something truth-valuable that is abstract but objective. It seems to be a rather terminological question whether the honorific ‘proposition’ should apply to the ones that contain objects and properties or the ones that contain modes of presentation (or even the hybrid ones that have cropped up in the literature). There has been a similar, largely terminological dispute over the word ‘content’.
An essential part of Frege’s theory, needed to preserve compositionality of reference, was that expressions embedded in belief contexts do not have their “customary” references.
Schiffer (1987) forcefully argues that the naive view, the “‘Fido’-Fido theory of belief,” cannot account for this fact, but Salmon (1989) thinks he has an answer. Crimmins’ discussion of the naive view overlooks their exchange.
The hidden indexical theory discussed by Schiffer (1977, 1987, 1992) is, despite its name, not committed to hidden indexical expressions but only to tacit reference. I don’t mean to suggest anything different when calling Crimmins’ account a version of HIT.
Unarticulated constituency is a special case of the semantic undetermination discussed, e.g., by Sperber and Wilson (1986) and by Bach (1987 and forthcoming), who distinguish it from indexicality.
Richard makes the puzzling claim that the verb ‘believe’ is an indexical, expressing different belief relations in different contexts.
This is not Crimmins’ position on denials of belief reports and on belief reports about groups of agents (pp. 181-193), cases where conditions on representations are “raised to constituency.”
According to Crimmins, even though a belief report does not impute to the agent a representation of the proposition believed, the notions and ideas that are imputed jointly constitute the “thought map” that charts the way of believing that would be used by the agent were he to represent the proposition in question (pp. 61, 124-130). Unfortunately, Crimmins does not explain why, if an imputed belief can be tacit, it should be required that in order to represent the constituents of a proposition tacitly believed, the agent would, or could, use notions and ideas that he currently possesses. In particular, Crimmins seems to be assuming that the representation the agent would employ must be decomposable into existing notions and ideas. This Humean assumption is subject to serious doubt in light of the evidence to be found in Fodor, Garrett, Walker, and Parkes (1980). Also. Crimmins never rules out representational redundancy, i.e., the possibility that an agent could have two functionally equivalent notions/ideas of the same thing/property. And he never considers the possibility of ambiguity or content change in a single notion or idea.
In his critique of HIT, Schiffer (1992) argues that, inter alia, it is far-fetched to suppose that speakers really have communicative intentions to refer to modes of presentation.
Crimmins nicely explains (pp. 92-98) how a normal idea of a property (or kind) is often misleadingly described as the concept of that property.
The same form of argument may be used to show that there is no semantic difference between the anaphoric and the deictic use of , e.g., ‘his’ in ‘John lost his wallet’, hence no reason to regard the pronoun as ambiguous or the sentence as having two readings (Bach 1987, p. 222-225).
It seems quite implausible to suppose that they have determinate—but minimal—truth conditions, expressing the “minimal” propositions that John is finished with something and that Martha is ready for something.
Interestingly, ‘complete’ requires a complement: ‘John has completed’, though semantically equivalent to (7), is ungrammatical.
This structural parallelism strongly suggests there is no syntactic slot to be filled by a specification of a domain of quantification. There is no linguistic reason to incorporate that into the structure of sentences whose typical utterance includes more than what is said. For further discussion of these issues see Bach (1987, pp. 69-82, and forthcoming), and for criticism of my approach see Récanati (1989).
This is also evident from the fact that a phrase of the form ‘S’s believing at t that p in way w‘ is a definite description, applicable to an instance of believing, whereas a sentence of the form ‘S believes at t that p in way w’ expresses a putative fact.
Unfortunately, Crimmins makes no contact with the considerable recent work on singular thought. I for one have argued (Bach 1987, pp. 15-16) that knowing that something exists is not a necessary condition for thinking of it; also, the epistemic conception incorrectly implies that singular thought is parasitic on general thought.
Another case in which the ‘that’-clause is not a content clause, which Stephen Neale (p.c.) has spotted, involves quantification and bound pronouns, as in ‘Most competitors believe that they will win’.
Bach, K. 1987: Thought and Reference. Oxford: Oxford University Press.
Bach, K. forthcoming: Semantic Slack: What is Said and More.
Crimmins, M. and J. Perry 1989: The Prince and the Phone Booth: Reporting Puzzling Beliefs.
Journal of Philosophy, 86, 685-711.
Fodor, J., M. Garrett, E. Walker, and C. Parkes 1980: Against Definition. Cognition, 8, 263-367.
Recanati, F. 1989: The Pragmatics of What is Said. Mind & Language, 4, 295-328.
Richard, M. 1990: Propositional Attitudes. Cambridge, Eng.: Cambridge University Press.
Salmon, N. 1986: Frege’s Puzzle. Cambridge, Mass.: MIT Press.
Salmon, N. 1989: Illogical Belief. Philosophical Perspectives, 3, 243-285.
Schiffer, S. 1977: Naming and Knowing. Midwest Studies in Philosophy, 2, 28-41.
Schiffer, S. 1987: The ‘Fido’-Fido Theory of Belief. Philosophical Perspectives, 1, 455-480.
Schiffer, S. 1992: Belief Ascription. Journal of Philosophy 89, 000-000.
Sperber, D. and D. Wilson 1986: Relevance. Cambridge, Mass.: Harvard University Press.