Belief Ascriptions, Prototypes and Ambiguity
A belief ascription such as "Oedipus believes that his mother is the queen of Thebes" can be understood in two ways, one in which it seems true, and another in which it seems false. It can seem true because the woman who was, in fact, Oedipus' mother was believed by him to be the queen of Thebes. It can seem false because Oedipus himself would have sincerely denied that Jocasta could be correctly characterized as "Oedipus's mother." Belief ascriptions thus seem to admit of two interpretations, and this has suggested to many that belief predicates such as "________ believes that his mother is the queen of Thebes" are ambiguous between a de dicto and a de re reading.  However, the impression of ambiguity is a function of the narrow ranges of examples that philosophers focus on. When we consider our ascriptional practices as a whole, the suggestion that belief predicates are ambiguous is neither plausible nor needed to explain the de dicto/de re distinction. The following will argue that understanding paradigmatic de dicto and de re ascriptions in terms of disavowals from a more basic sort of ascription is preferable to positing an ambiguity in which each of the two sorts of ascription are conceptually primitive.
The difference between de dicto and de re ascriptions seems clear when we look at pairs of ascriptions such as the following.
(I) Oedipus believes that his mother is the queen of Thebes.
(II) Oedipus believes that his mother is the queen of Corinth.
Both (I) and (II) are ascriptions that we could find ourselve making, the first because Oedipus believes that Jocasta is queen of Thebes, and the second because he believes that Merope is queen of Corinth. Of course, while we would endorse the characterization of Jocasta as "Oedipus's mother," Oedipus would not, and while Oedipus would endorse the characterization of Merope as "Oedipus's mother," we would not.  Ascriptions such as (I) which attempt to characterize Jocasta in terms that we (but not necessarily Oedipus) would accept, are characterized as de re. Ascriptions such as (II), which attempt to characterize Merope in terms that Oedipus (but not necessarily we) would accept, are characterized as de dicto.  De re ascriptions thus commit the ascriber herself to the characterization of the objects in the belief clause, but not necessarily to the ascribee's endorsement of those characterizations. De dicto ascriptions, on the other hand, commit the ascriber to the ascribee's endorsement of the characterizations of the objects in the belief clause, but not necessarily to their actual correctness. De dicto belief ascriptions thus reflect the ascribee's point of view, while de re ascriptions reflect what the ascriber takes this point of view to be a point of view on. De re ascriptions allow substitution of coreferential terms, de dicto ascriptions need not.
Still, this difference between de dicto and de re ascriptions should not be understood as reflecting an ambiguity in our belief predicates. To see why, we should first notice that many, if not most, ascriptions fail to be characterized exclusively as either de dicto or de re. Since the belief ascriptions discussed in the philosophical literature (those involving Oedipus, Superman, Pierre etc.) often involve characterizations that are accepted by just one of the acriber or the ascribee, it can be easy to lose sight of the fact that most of our belief ascriptions involve characterizations of the topic that are accepted by both the ascriber and the ascribee.  For instance, suppose I make the ascription:
(III) Oedipus believes that his daughter is good to him.
The characterization of Antigone as Oedipus's daughter is one that both I and Oedipus would accept. Since ascriptions such as (III), like most attitude ascriptions, involve characterizations of the topic that are accepted by both the ascriber and the ascribee, they apparently satisfy both the criteria for being de re and de dicto, and are thus not comfortably classified as just one or the other. 
Why most actual ascriptions seem neither purely de dicto nor purely de re should be clear when we consider how the possibility of acceptance by the ascriber and the ascribee divide characterizations into four classes.
A1. Characterizations acceptable to both the ascriber and the ascribee.
A2. Characterizations acceptable to the ascriber but not the ascribee.
A3. Characterizations acceptable to the ascribee but not the ascriber.
A4. Characterizations acceptable to neither the ascriber nor the ascribee.
Ascriptions intended to be in the first class seem to be neither purely de re nor purely de dicto. Ascriptions intended to be in the second class could be characterized as 'pure' or 'paradigmatic' de re, while those intended for the third class could be characterized as 'pure' or 'paradigmatic' de dicto. Belief ascriptions involving characterizations in the fourth class are comparatively rare, and they may seem to be neither de re nor de dicto. 
Ascriptions involving characterizations of type A2 tend to be flagged with the "believes of" locution, while those involving characterizations of type A3 are often flagged with the "so-called" locution.  Cases of type A4 can be represented with a combination of both. Accordingly, a regimented English can represent Peter's belief that Frege is German in the following four ways which reflect our own and Peter's attitude towards Frege's purported status as the father of modern logic.
B1. Peter believes that the father of modern logic is German.
B2. Peter believes of the father of modern logic that he is German.
B3. Peter believes that the so-called 'father of modern logic' is German. 
B4. Peter believes of the so-called 'father of modern logic' that he is German. 
In the regimented English, B1-B4 are used for cases A1-A4 respectively, while in an unregimented English, B1 may be used for cases of type A1, A2 and A3.  While most of our ordinary ascriptions would be like B1, cases of misidentification may call for ascriptions like B2 and B3. For instance, if Peter buys a rat thinking that it is a hamster, we may characterize his beliefs with ascriptions such as: "Peter believes of his rat that it is a hamster, and he thinks that his mother will be delighted when he brings this so called 'hamster' home with him."
Sentences like B1-B4 are good ways to get across that one is dealing with cases like A1-A4, though it will be suggested here that they commit their users only to the following. 
D1. Commitment to characterization both attributed and undertaken
D2. Commitment to characterization undertaken but not attributed
D3. Commitment to characterization attributed but not undertaken
D4. Commitment to characterization neither attributed nor undertaken.
On such an account, B2 can be truly uttered in cases of type A1 and A2, and B3 can be truly uttered in cases of type A1 and A3. However, B2 and B3 typically implicate that one is dealing with cases of sort A2 and A3 because we typically use the simpler form B1 unless it is inappropriate. For instance, while B2 is true in both cases of type A1 and A2, it would be uneconomical and uncooperative to use it if one knows that one is dealing with a case of type A1. Consequently, in using the longer and less committal B2, one implicates that the ascribee does not (or may not) endorse the characterization in question, and thus that one is dealing with a case of type A2 (or that one can't tell whether one is dealing with a case of type A1 or A2). 
Unless we specify otherwise,  it is assumed that both the ascriber and the ascribee accept the characterization used in the ascription. This is why (if one is unfamiliar with Oedipus's story) statements like (I) and (II) can be misleading: the default assumption will be that both the ascriber and the ascribee would endorse the characterizations used.  These facts about our default assumptions suggest that 'pure' de dicto and de re ascriptions should be understood as special cases in which we explicitly disavow some of the commitments associated with standard ascriptions. In particular, with de dicto ascriptions we disavow any commitment to the characterizations used, while with de re ascriptions we disavow any commitment to the ascribee's accepting the characterizations used. These disavowals are what we explicitly flag with the "believes of" locution in the case of de re ascriptions and the "so-called" locution in the case of de dicto ascriptions. 
Such disavowals are understood best if belief predicates, rather than being treated as ambiguous, are viewed as having both 'prototypical' and 'extended' uses.  The healthiest and 'prototypical' belief ascriptions involve characterizations accepted by both parties, though attributions can still be made with characterizations accepted by just one (and, if sufficiently flagged, neither) of the parties. According to such an account, the 'core' ascriptions characterize the subject matter in ways acceptable to both the ascriber and the ascribee, and the other types of ascription are reached by disavowing commitments involved in the core. Such a proposal is, of course, very different from the suggestion that belief predicates are ambiguous between independent de dicto and de re forms, each of which ascribes a single commitment. Someone who wanted to keep de dicto and de re ascriptions independent of the proposed core ascriptions must argue that, rather than characterizing ascriptions of types D2 and D3 as disavowals of the commitments involved in ascriptions of type D1, we should take ascriptions of type D1 as a combination of the commitments involved in ascriptions of types D2 and D3. 
However, there are a number of reasons to think that we should treat cases of type D1 as more primitive than those of types D2 or D3. 
The first is that it produces a regimentation that is more recognizably English than those that take cases of type D2 and/or D3 as primitive. If we are trying to explicate belief ascription in English, rather than creating an alternate language for describing the attitudes, some fidelity to our actual talk of the attitudes is important. In light of this, it is important to note that our everyday ascriptions often treat cases of types D2 and D3 as "marked."  Such marking tends to go with the derived rather than the primitive cases, so the fact that pure de dicto and de re ascriptions are marked gives one prima facie evidence that they are not primitive. Also, having one primitive rather than two fits better with our having one word for belief, which we occasionally modify, rather than two words for belief, which we occasionally combine.
A second reason in favor of the account suggested above is that it brings the semantics of belief predicates comparatively close to our actual intuitions about their truth and assertability conditions. For instance, the prototype account predicts that "Peter believes that the father modern logic was German" would be maximally acceptable if both Peter and the ascriber thought of Frege as the father of modern logic, moderately acceptable if just one of them did, and barely acceptable at all if neither of them did. This corresponds, I think, to our intuitions about the acceptability of such ascriptions.  If belief predicates were ambiguous, there shouldn't be such gradations. If Peter didn't think of Frege as the father of modern logic, we should assume that the ascription was either de re and straightforwardly true, or de dicto and straightforwardly false. How an ascription could be acceptable, but still have a marginal feel is something that the ambiguity account cannot explain. The prototype theory explains the marginality and predicts that it won't be 'fully' acceptable until the "believes of" hedge is added.
Another reason for understanding D2 and D3 in terms of D1, rather than vice versa, is that the proposed regimentation has more expressive power. A serious problem for the ambiguity account is that if one treats ascriptions of type D2 and D3 as primitive,  then one makes it difficult to account for the possibility of ascriptions of type D4. If one starts with primitive ascription-types geared towards attributing commitments to the ascriber and the ascribee, one will not be able to use these primitives to get a case where a commitment to the characterization is attributed to neither party.  That is to say, by treating de dicto and de re ascriptions as primitive, one rules out the possibility that a commitment could be attributed to a speaker using characterizations that neither the ascriber nor the ascribee endorsed. On the other hand, if one takes unmarked cases of type D1 as primitive (and cases of types D2 and D3 as associated with hedges that strip ascriptions of their usual commitments), then cases of type D4 could be understood in terms of the application of both mechanisms. Indeed, it would actually be a problem for this type of account if cases of type D4 were not possible.
Fortunately for the view defended here, cases of types D4 are possible. For instance, suppose that neither Peter nor I believe that Frege is the father of modern logic, but Paul does. If Paul were to ask me "Does Peter believes that Frege, the father of modern logic, is German?" I could reply to him "Peter believes of the so-called father of modern logic that he is German." The commitment to Frege's being German is attributed, but the ascription involves no commitment to either me or Peter endorsing the characterization of Frege as the father of modern logic. Such ascriptions may not be very popular, but they can be useful when the ascriber wants to make clear that neither he nor the ascribee accepts a characterization endorsed by the ascriber's audience (or some other salient group).
Finally, the suggestion that belief ascriptions are prototype driven rather than ambiguous seems supported by the behavior of the sorts of expression that we use to 'mark' ascriptions like B2 and B3. For instance, the suggestion that sentences like B3 involve an explicit disavowal of a commitment involved in B1 (rather than a taking on of a single commitment that helps make up B1) is suggested by other uses of the "so called" locution. If D2 and D3 were primitive, the "so called" locution would involve attributing to the speaker a commitment to the characterization of the topic. If D1 is primitive, the "so called" locution involves denying that one is committed to the characterization oneself. Both accounts would seem to accommodate our intuitions about sentences like B3, but other uses of "so called" do not sit well with the hypothesis that it involves the attribution rather than the disavowal of a commitment. This can be seen in the special case of self-ascriptions. If I claim that "I believe that the so-called father of modern logic is German," I am precisely not attributing a commitment to the characterization to the ascribee (myself). Rather, I am disavowing any commitment to the characterization.  Furthermore, if the "believes of" and "so called" locutions involved the actual undertaking and attributing of primitive commitments to the characterization, then an utterance of "Peter believes of the so-called father of modern logic that he is German", should involve the speaker both undertaking and attributing to Peter a commitment to the characterization in question. The fact that it seems to do precisely the opposite strongly suggests that the hedges strip commitments from more basic ascriptions rather than adding commitments of their own.
In conclusion, belief predicates give the impression of ambiguity because our application of them can extend from their prototypical use in a number of systematic and predictable ways.  Such a prototype-driven analysis undermines, but explains the appeal of, a number of standard approaches to the semantics of belief predicates in English. Since Frege, analyses of belief ascriptions have typically treated their truth-conditions as corresponding to something like D1, and ascriptions that failed to capture the ascribee's 'mode of presentation' were simply treated as false.  Such 'Fregean' accounts are adequate for prototypical ascriptions (and would be correct for unflagged belief ascriptions in the more regimented English suggested above), but they do not seem to adequately capture the truth conditions of all ascriptions made in everyday English. As defenders of 'Millian' semantic theories have pointed out, we can make attitude ascriptions that seem in line not with D1, but rather with the more modest D2.  However, such theorist's often go on to argue that a general analysis of belief should be along the lines of D2. The account outlined above suggests that the procrustean feel that such theories have stems from their basing a general account of belief on legitimate, but non-paradigmatic, ascriptions. Just because ascriptions can be (marginally) acceptable without being faithful to the ascribee's mode of presentation, it does not follow that such modes of presentation play no role in a semantics for belief. Thinking otherwise comes from accepting a semantic framework radically at odds with the assumptions behind prototype-driven analyses of language such as the one presented here. Both the Millian and the Fregean accounts presuppose that the semantics of a belief predicate will provide one with a set of necessary and sufficient conditions that are, or are not, satisfied in any given context. A mode of presentation either is, or is not, part of such a set, and it is thus semantically necessary (as in Fregean theories) or semantically irrelevant (as in Millian theories). The account above suggests that less prototypical ascriptions (such as (I) and (II)) while not, strictly speaking, false, still stray from the conditions that come out of the semantics of belief. A semantics for belief is hard to give precisely because belief predicates are more flexible than standard accounts of them have supposed.
1 See Stich 1986 for a critical discussion of the pervasiveness this assumption. While there has been widespread agreement that belief predicates are ambiguous, there has been some disagreement about whether the ambiguity should be taken to correspond to two actual types of belief or merely to two types of ascription. (For a discussion of the distinction in terms of types of belief, see Kaplan 1968, Burge 1977. For a discussion of the distinction in terms of types of ascription, see Searle 1983, Dennett 1982, Brandom 1994.) Many who understand the distinction as between two types of ascriptions deny that there are actually two types of belief involved, but those who understand the distinction in terms of two types of belief must admit that there are still two types of ascriptions (namely, those that ascribe de dicto beliefs and those that ascribe de re beliefs). Consequently, this paper will focus on the shared claim that there are two types of ascription.
2 Actually, things aren't quite this simple, since there is a perfectly good sense of "mother" in which it is true that Merope is Oedipus's mother. However, for the purposes of this paper, "mother" will be used as short for "biological mother."
3 What follows will draw heavily upon Brandom's recent and detailed characterization of the difference between de dicto and de re ascriptions (see Brandom 1994, especially ch. 8). It will, however, ultimately reject his claim that "'Belief' is ambiguous in scorekeeping terms, referring sometimes to doxastic commitment and sometimes to acknowledgment of such a commitment." (Brandom 1994, 260.)
4 Proper names help provide a type of common currency that makes such ascriptions easier.
5 For instance, Brandom describes the distinction as follows:
[Where the ascriber employs] only commitments the ascriptional target would acknowledge, the content specification is de dicto. Where the ascriber has employed substitutional commitments he himself, but perhaps not the target, endorses, the content specification is de re. (Brandom 1994, 507.)
On such an account, ascriptions that involve characterizations accepted by both parties fit the criteria for being both de dicto and de re.
6 It will be later suggested that they are, in some sense, a combination of the two. If all ascriptions were intended to be of types A2 or A3, then the claim that belief predicates were ambiguous would be uncontroversial. There would be just two types of ascription, each of which involved an independent commitment. However, the fact that there are four, rather than two, classes of ascription makes the ambiguity thesis harder to justify.
7 In print, Sscare quotesS play a similar role.
8 Or, "Peter believes that the Sfather of modern logicS is German"
9 Or, "Peter believes of the Sfather of modern logicS that he is German."
10 Indeed, instances of such uses can be seen in (III), (II) and (I) respectively.
11 One might think that one should map the commitments involved in B1-B4 directly onto A1-A4. Namely,
C1. Commitment to the characterization both attributed and undertaken
C2. Commitment to characterization undertaken but rejection of characterization attributed
C3. Commitment to characterization attributed but rejection of characterization undertaken
C4. Rejection of characterization both attributed and undertaken.
However, such an analysis treats the speaker as committed to more than she intuitively seems to be committed to. For instance, the claim "Peter believes of the father of modern logic that he is German" might suggest that Peter doesn't think of Frege as "the father of modern logic," but it doesn't actually say it. The sentence would still be true if Peter did think of Frege as "the father of modern logic." Indeed, this is why sentences like B2 can be used when we simply don't know whether Peter believes that Frege was the father of modern logic.
12 For the relation of such cooperative maxims to implicature, see Grice 1989.
13 Or the context makes it obvious, as in "Ullyses S. Grant believed that the site of the 1996 summer Olympics had to be captured."
14 Statements such as (I) and (II) would not have such potential to be misleading if belief predicates were straightforwardly ambiguous, since there would be no such default assumption. For instance, for paradigmatically ambiguous words such as "bank," even if one of the two meanings has a certain 'default' status in various contexts ("I keep my money in the bank" etc.) it certainly isn't the case that the default is ever to pick out both of the putative meanings. We don't naturally assume that the speaker has in mind a financial institution on the side of a river. However, this seems to be precisely what happens in the case of belief ascriptions we assume that both the ascriber and the ascribee accept the characterization.
15 Such 'hedges' work to make the 'marginally' acceptable ascriptions fully so, in the way that 'roughly flat' will make can be uncontrovertially said of something that is marginally flat. (For a discussion of hedges, see Lakoff 1972)
16 For a discussion of such prototype driven accounts of categorization, see Rosch 1975, Lakoff 1987.
17 This is the line Brandom suggests when he talks about how the cases in which ordinary that-clauses both characterize the speaker's conception and pick out the relevant object can be regimented as, say "Alfred believes of the man in the corner as the man in the corner that he is a spy." (Brandom 1994, 546.)
18 I will not go into it here, but I suspect that the language learning data would support a model where cases of type D1 are primitive, and that mastery of ascriptions for cases of type D2, and D3 comes later.
19 "Believes-of" and "so-called" locutions respectively. Cases of D4 would thus be doubly marked.
20 I am assuming here, for simplicity's sake, that the ascriber's beliefs about how the topic should be characterized are true.
21 Thus treating the 'so called' and 'believes of' hedges as if they were in the business of adding primitive commitments.
22 Brandom, for instance, characterizes the de dicto/de re distinction as one between "two different styles in which the content of the commitment ascribed can be specified." (Brandom 1994, 503, italics mine.) However, in cases like D4 there is no way for his account to get a grip on how the commitment involved is specified.
23 Much the same could be said for the use of "so called" outside of belief contexts, as in "I've seen the so-called 'greatest show on Earth', and it wasn't that great."
24 Ways that are explicitly flagged with the hedges "so called" and "believes of".
25 See Frege 1892. A typical example of such a 'false' ascription would be "Lois Lane believes that Superman works for the Daily Planet."
26 See, for instance, Salmon 1986. B1 itself is a decent example of such a case. Typically such ascriptions will be more acceptable if the ascribee is ignorant or ambivalent about the ascription used, they can seem less acceptable if the ascribee actually rejects the characterization.
27 Any more that 'flat' just commits you to the minimal gradation you can 'get away with' asserting.
28 For a general discussion of how categories may frequently be ill suited for the sort of analysis in terms of necessary and sufficient conditions, see Lakoff 1987.
Brandom, R. 1994. Making it Explicit, Cambridge: Harvard University Press.
Burge, T. 1977. "Belief De Re", The Journal of Philosophy. vol. 74.
Dennett, D. 1982. "Beyond Belief", in Dennett, D. The Intentional Stance. Cambridge: MIT. 1987.
Frege, G. 1892. "On Sense and Meaning", in Frege, G. Collected Papers, Oxford: Blackwell, 1984.
Grice, P. 1989. Studies in the Way of Words, Cambridge: Harvard University Press.
Kaplan, D. 1968. "Quantifying in", in Davidson, D. and Hintikka, J. Words and Objections. New York: Humanities Press. 1968.
Lakoff, G, 1972. "Hedges: A Study in Meaning Criteria and the Logic of Fuzzy Concepts" Journal of Philosophical Logic 2 (1973) 458-508.
Lakoff, G. 1987. Women, Fire and Dangerous Things, Chicago: The University of Chicago Press.
Rosch, E. 1975. "Family resemblances: studies in the internal structure of categories", Cognitive Psychology, 7.
Salmon, N. 1986. Frege's Puzzle, Cambridge: MIT.
Searle, J. 1983. Intentionality, New York: CUP.
Stich, S. 1986. "Are Belief Predicates Systematically Ambiguous?" in Bogdan, R. (ed.). Belief: Form, content and function, New York: OUP. 1986.