Frege's Puzzle and Belief Ascriptions

(in Künne, W., Newen, A. & Auduschus (eds.) Direct Reference, Indexicality, and Propositional Attitudes)

Pierre Jacob

This paper is about belief ascriptions. Belief ascriptions are utterances whereby one person tries to convey to another person the content of the belief of yet a third person. Like other utterances, they are subject to pragmatic principles of interpretation. I am interested in belief ascriptions in which the embedded "that"-clause is a simple predication containing a predicate and a singular term, as when I utter sentences such as "Claire believes that Cicero was tall" or "Claire believes that you are tall". I will consider cases in which it may be assumed that the predicate stands for (or expresses) a monadic property and the singular term (whether a proper name, an indexical or a demonstrative pronoun) possesses a referent.

Presumably, the propositional content or the truth condition of the belief ascription depends on the proposition expressed by the embedded "that"-clause. My topic will be to understand the propositional contribution of the embedded "that"-clause to the truth condition or the propositional content of belief ascriptions. What does the simple predication contribute to the proposition expressed by the belief ascription of which it is a syntactic constituent? What is the relation between the proposition expressed by the utterance of a simple predication and the contribution of the simple predication when it is embedded under "believes that" to the truth condition of the belief ascription? Is the former identical to the latter? There is a plausible response to this question which, following Crimmins and Perry (1989), I will call the thesis of 'Semantic Innocence'. My goal in this paper is to question Semantic Innocence.

1. What Is Frege's Puzzle?

Consider pairs of sentences (1a, b) and (2a, b) in which respectively "Cicero" has been replaced by the coreferential name "Tully" and the indexical pronoun "I" has been replaced by the coreferential pronoun "you":

(1a) Cicero was tall.

(1b) Tully was tall.

(2a) I am sick.

(2b) You are sick.

In recent philosophy of language, there has been an important controversy over whether an utterance of (1a) can be said to express the same proposition as an utterance of (1b) and whether an utterance of (2a), in a given context, can be said to express the same proposition as an utterance of (2b). If they do, then they express what Kaplan (1978, 1989) has famously called a Russellian singular proposition and Schiffer (1992) has called a Kaplan proposition. Such propositions can be represented as ordered pairs whose first coordinate is the individual who happens to be the common reference, respectively, of the names "Cicero" and "Tully" in (1) or of "I" and "you" in (2) and whose second coordinate is the property expressed by the predicate. As Kaplan (1978, 385) has put it, the concrete individual is "right there, trapped in [the] proposition." If the two utterances do not express the same proposition, then the proposition expressed by an utterance of (1a) must differ from the proposition expressed by an utterance of (1b) in that the former contains a reference to a mode of presentation of Cicero associated with the name "Cicero" (a "Cicero" mode of presentation of Cicero), whereas the latter contains a reference to a mode of presentation of Cicero associated with the name "Tully" (a "Tully" mode of presentation of Cicero). Similarly, an utterance of (2a) and an utterance of (2b) would express different propositions in virtue of the fact that the same individual is presented differently by a use of "I" in (2a) and a use of "you" in (2b).

I will refer to the former view as the theory of direct reference and to the latter as the Fregean view. The former identifies the proposition expressed by the utterance of a simple predicative sentence with its truth condition - something like the state of affairs which must obtain for the utterance to be true. The latter identifies the proposition expressed with something like the object (the content) of the belief expressed by the speaker's utterance of the sentence.

Let us consider belief ascriptions such as (3a) and (3b) of which (1a) and (1b) respectively are syntactic constituents:

(3a) Claire believes that Cicero was tall.

(3b) Claire believes that Tully was tall.

It is, I think, generally assumed both that Frege, in "On Sense and Reference," has provided a serious challenge to the theory of direct reference and that he set up the agenda for an appropriate theory of belief ascriptions, by advancing an argument which can be reconstructed as follows and which, following current practice, I will call "Frege's Puzzle."

(I) If two singular terms "Cicero" and "Tully" share the same reference, then, given that sentences (1a) and (1b) differ only in that in (1b) "Cicero" has been replaced by coreferential "Tully", the proposition expressed by an utterance of (1a) is the same as the proposition expressed by an utterance of (1b).

(II) If an utterance of (1a) and an utterance of (1b) express the same proposition, then so do utterances of (3a) and (3b).

(III) However, utterances of (3a) and (3b) may differ in truth value. A fortiori, they may express different propositions.

Premise (I) is a statement of the theory of direct reference. I shall call it DR. Premise (II) says that what a "that"-clause contributes to the proposition expressed by (or the truth condition of) a belief ascription just is the proposition expressed by an utterance of the sentence following "that". I shall call it 'Semantic Compositionality.' Premise (III) I will call 'Common Sense' for obvious reasons.

2. Two Views of Belief Ascriptions Criticized: the Opacity View and the Millian View

There are two important views of belief ascriptions which can be seen as responses to Frege's challenge, and which I would like to criticize. One view, which derives from Frege himself, I shall call the Opacity view. The other view I shall call the Millian view (by reference to Mill's own view that the only propositional contribution of a name is its referent). What I call the Millian view is what Schiffer (1987, 1992) has nicknamed "the 'Fido'-Fido theory of Belief" and what Recanati (1993, chap. 17) has labeled "the Implicature theory". These two views disagree as sharply about the propositional structure (or truth condition) of belief ascriptions as they disagree about which of the premises in Frege's Puzzle is the wrong one. The Opacity view puts the burden on premiss (I), the Millian view on premise (III). However, both assume the truth of premiss (II), Semantic Compositionality. The view I would like to defend is a version of what Schiffer (1992) calls the "hidden-indexical" view. I would like to argue that it can be seen as rejecting the truth of premise (II).

2.1. The Opacity View

In Fregean terminology, whereas a singular term contained in a simple predication has both a sense (or Sinn) and an ordinary (or direct) reference (or Bedeutung), a singular term in a "that"-clause does not have its ordinary or direct reference. It has an oblique reference: it refers to its ordinary Sinn, i.e., to the mode of presentation of its ordinary reference.[1] On this view, which I call Opacity, a singular term in a "that"-clause is thus deprived of its ordinary reference. This asymmetry between the semantic value of a singular term in and out of a "that-"clause is crucial to the Fregean Opacity view. What a singular term in a "that"-clause contributes to the proposition expressed by (or to the truth conditions of) the belief ascription of which it is a syntactic constituent is a particular mode of presentation of its ordinary reference. Suppose I utter (3a). In order to know which proposition I have expressed and whether my utterance of (3a) expresses a truth, one must first determine which particular mode of presentation of Cicero I have referred to.

What seems to drive this view is a parallel between belief contexts and a restricted set of quotation contexts, such as (4):

(4) "Cicero" has six letters.

In (4) the property expressed by the predicate "has six letters" is ascribed, not to Cicero (the man), but to the name "Cicero". Even though "Cicero" and "Tully" are coreferential, we cannot replace salva veritate "Cicero" by "Tully" in (4) since it is not true that "Tully" has six letters. If "Cicero" in (3a) behaves as it does in (4), then we can understand why it cannot be substituted by "Tully" in (3a) (when it cannot). But does it so behave in (3a) as it does in (4)?

This view has been effectively criticized. I will merely summarize two kinds of criticisms. First, although there is a genuine insight in the parallel between belief contexts and quotation contexts, the quotation contexts referred to above are far too restricted. In the quotation contexts envisaged - direct quotations -, what the quotation refers to is assumed to be the sentence type which has been inserted between quotation marks. Belief contexts, however, are much more like indirect quotations than they are like direct quotations. As I will argue at the end of the paper, the determination of the reference of indirect quotations is both more complex and more context-dependent than the reference of direct quotations. The comparison between belief contexts and direct quotation leads the Opacity view to the implausible claim that, by uttering (3a), a speaker has no more referred to Cicero (the man) than he would have, had he uttered (4). Whether or not "Cicero" is replaceable salva veritate by "Tully" in (3a), it is hardly plausible to deny that, in uttering (3a), the speaker has referred to Cicero: he has said that a person (Claire) has a belief about Cicero.

Second, on the view under discussion, the proposition expressed by my utterance of (3a) refers to a particular mode of presentation of Cicero. In other words, the proposition expressed by my utterance of (3a) depends on the sense of sentence (1a). But, it seems, sentence (1a) will have different senses according to who utters it. And of course (1a) can be uttered by different persons. Suppose I know things about Cicero which Claire does not. Whose mode of presentation of Cicero should prevail? Whose utterance of (1a) should determine the mode of presentation of Cicero referred to by my utterance of (3a)? Mine or Claire's? The belief ascriber's or the believer's? On both horns of the dilemma, it seems, the believer and the ascriber must share exactly the same particular mode of presentation of the referent of the singular term in the "that"-clause for the ascriber to ascribe truly a belief to the believer. And this is not very plausible.

2.2. The Millian View

I now turn to the Millian view. Whereas on the Opacity view, a belief ascription refers to a particular mode of presentation of the referent of the singular term in the "that"-clause, on what I call the Millian view, no reference whatsoever is made to any mode of presentation of the referent of the singular term in the "that"-clause. Schiffer's nickname, "The 'Fido'-Fido Theory of Belief", emphasizes one component of the Millian view: The fact that what is said in the utterance of a belief ascription, according to this view, contains a singular proposition - a Kaplan proposition - and no mode of presentation. Recanati's label, "the Implicature theory", emphasizes a complementary feature of the view: the fact that, on this view, reference to a mode of presentation is never part of the truth condition of the belief ascription, but is conveyed indirectly via conversational implicatures. Nathan Salmon (1986, 1989) has defended such a view. As I said earlier, he is committed to denying the third premise of Frege's Puzzle; on his view, if an utterance of (3a) expresses a truth, then so must an utterance of (3b).

On Salmon's (1986, 1989) view, the proposition expressed by (or the truth condition of) either an utterance of (3a) or an utterance of (3b) is a general proposition involving an existential quantification over modes of presentation of Cicero:

($m) BEL (Claire, p, m)

where "BEL" is a three place predicate, "p" is the Russellian singular proposition <Cicero, being tall> and "m" is a variable ranging over modes of presentation of Cicero. On this view, reference to a particular mode of presentation is always external to the truth conditions of the utterance of a belief ascription; it is external to the proposition explicitly expressed, or to what is said, by the utterance of a belief ascription. Prima facie, this view may fit the truth conditions of de re, not de dicto belief ascriptions, i.e., ascriptions in which the "that"-clause contains a singular term which the believer would not use to express his or her belief. The Millian strategy is to claim that we do not commonly appreciate the difference between what constitutes the genuine propositional content (or truth conditions) of an utterance and the additional information pragmatically conveyed by means of Gricean conversational implicatures carried by an utterance. On the Millian view, what is said by the utterance of a belief ascription is a general proposition involving an existential quantification over modes of presentation. And reference to a particular mode of presentation of the referent of the singular term in the "that"-clause is always conveyed indirectly by means of Gricean conversational implicatures.

Not only, in the words of Crimmins and Perry (1989, 249-50), does the Millian deploy a "strategy of denying the accuracy of our strong intuitions about truth and falsity." This strategy also, I think, faces a nasty dilemma, as soon as we consider an utterance whose content we would intuitively consider as the negation of the content of an utterance of (3b), such as (5):

(5) Claire does not believe that Tully was tall.

Let us suppose that (3a) expresses a truth and let us suppose that Claire gives all conceivable evidence in favor of what intuitively we would consider the falsity of (3b) and the truth of (5). She refrains from assenting to (1b) when asked; she utters the negation of (1b): "Tully was not tall"; and so on. The intuitive view is that (5) is true: it is the denial of (3b), which is false.

On Salmon's view, in such circumstances, an utterance of (3b) would presumably be true and misleading. It would be true, since it expresses the same true general proposition as does an utterance of (3a): the proposition that there exists a mode of presentation under which Claire does believe (or recognize) the singular proposition about Cicero that he was tall. It would be misleading since it would convey indirectly, by means of a Gricean conversational implicature, the false information that Claire believes the singular proposition under a mode of presentation, namely the "Tully" mode, under which she does not recognize the singular proposition in question. But now, it is incumbent upon the Millian to account for our intuition that, in the circumstances, an utterance of (5) expresses a truth. Even though the Millian holds that our intuition is the result of some confusion, he is bound to explain how our intuition arises.

Given his views, the Millian has a choice: Either he may assume that what is negated by an utterance of (5) is what is said by an utterance of (3b), that is, the proposition explicitly expressed by an utterance of (3b) or the truth condition of an utterance of (3b). Or alternatively, he may assume that what is negated by an utterance of (5) is the overall content which is communicated by an utterance of (3b), where the overall content communicated by an utterance of (3b) includes both what is said by, the proposition explicitly expressed by, or the truth condition of, an utterance of (3b), together with the conversational implicatures of the utterance. Neither horn seems very satisfactory.

If he assumes both that an utterance of (3a) is true and that an utterance of (5) denies what is said by an utterance of (3b), then he must conclude that what is said by an utterance of (5) is false, not true. Now, in response to a criticism made by Schiffer (1987), Salmon (1989, 250-53) has explicitly rejected the view that what happens in this case is similar to the interpretive process involved in a hearer's understanding of an ironical or metaphorical utterance (or speech-act). A speaker's utterance of the sentence "John is a lion" where "John" refers to a human male may be a metaphorical speech-act. A speaker's utterance of "His lecture was brilliant" where the context makes it clear that the speaker thought that the lecture referred to was dull is an ironical speech-act. Arguably, in such cases, the hearer starts with the assumption that the speaker said something literally false, and then infers a related true proposition. But applied to the case under consideration, the utterance of (5), in Salmon's (1989, 251-2) own terms, "this alleged explanation is incoherent; it purports to explain ordinary speakers' belief that a given sentence is true by means of their belief that it is false. Clearly, no attempt to explain the widespread view that (5) is literally true can proceed from the initial hypothesis that ordinary speakers typically believe that (5) is literally false".

Clearly then, Salmon denies that, in the case of an utterance of (5), the hearer would first uncover a false proposition from which he would then infer a related true proposition. So, it seems, the option of assuming both that an utterance of (3b) expresses a truth and that (5) denies what an utterance of (3b) says explicitly does not sit comfortably with Salmon's overall position.

Consider now the second option. He may assume that what is denied by an utterance of (5) is not the proposition explicitly expressed by an utterance of (3b). Suppose he assumes that what an utterance of (5) denies is the overall content communicated by an utterance of (3b): i.e., the combination of the truth condition of (3b) together with the conversational implicatures carried by an utterance of (3b). This seems consistent with the acknowledgment that an utterance of (5) may be true. But the problem is now to explain how the insertion of a negation, yielding (5) from (3b), has converted something which was merely communicated by an utterance of (3b), not part of the truth conditions of an utterance of (3b), into a constituent of the truth conditions of an utterance of (5). Why should the introduction of negation turn into a truth conditional component something which was not truth conditional prior to the introduction of negation? Unless the Millian can provide the beginning of an answer to this puzzle, the latter horn of the dilemma does not seem more hospitable to the Millian view than the former horn. What this brief examination of the Millian view reveals, I believe, is that it is wrong to assume that reference to a mode of presentation of the referent of a singular term in a "that"-clause is always external to what is said by (or to the truth conditions of) a belief ascription.

3. A Third View: The Hidden-Indexical View

Let us take stock. The Opacity view claims that the proposition expressed by a belief ascription contains a reference to a particular mode of the reference of a singular term in the "that"-clause. On the Millian view, reference to a particular mode of presentation of the singular term is always external to the explicit content of the belief ascription. On the view I would like to defend, the truth condition of a belief ascription may well - but it need not - include a reference to a type of mode of presentation of the referent of the singular term in the "that"-clause. It need not for in cases of de re belief ascriptions, as I mentioned previously, the speaker need not try to capture the point of view of the believer. In this case, something like Salmon's Millian view of the truth conditions of a belief ascription may well be correct: the speaker need not commit himself to more than to the claim that there exists a mode of presentation of the referent of the singular term in the "that"-clause under which the believer believes the singular proposition. The view I would like to defend is a version of the view Schiffer (1992) calls the "hidden-indexical" theory of belief ascriptions (henceforth HIT), which has also been advocated by Crimmins and Perry (1989). In particular, it makes use of Perry's (1986) notion of an unarticulated constituent.

Let me first quote what I take to be Schiffer's latest statement of HIT:

As applied to the paradigmatic example

[A] Ralph believes that Fido is a dog.

the hidden indexical theory says that the logical form of an utterance of this sentence may be represented as

[B] ($m) (F*m and B(Ralph, <Fido, doghood>, m))

where F* is an implicitly referred to and contextually determined type of mode of presentation. By a type of mode of presentation I mean merely a property of modes of presentation... The reference to a type of mode of presentation is implicit in that, although the sentence requires the speaker to be referring to a type of mode of presentation whenever the sentence is uttered, there is no word in [A] which refers to that type (whence the 'hidden' in 'hidden-indexical theory'). The reference to the type of mode of presentation is "contextually determined" in that different types may be referred to on different occasions of utterance (whence the 'indexical' in 'hidden-indexical theory')... [B] tells us that [A]'s 'that'-clause... is a referential singular term whose referent is the singular proposition <Fido, dogwood>, and it tells us that the references of 'Fido' and 'dog' in that 'that'-clause are Fido and dogwood, respectively. It tells us that an utterance of [A] is true just in case [B], where F* is the mode-of-presentation type referred to in the utterance. And it tells us that this contextually determined reference to a type of mode of presentation is by a "hidden indexical" in that there is no actual indexical in [A] which carries this reference (Schiffer 1992, 503-4).

Something like HIT was already argued for in Schiffer (1977). In the course of arguing against the '''Fido"-Fido theory of Belief" Schiffer has made the point that, as he put it, "the complete content" of a belief ascribed must be what he calls (Schiffer 1978, 182) a "quasi-singular proposition", not a singular proposition. What Schiffer calls a quasi-singular proposition is something like a complex ordered pair whose second constituent is the property expressed by the predicate and whose first constituent is itself an ordered pair whose first coordinate is the referent of the singular term and whose second coordinate is the mode of presentation of the referent, such as <<Cicero, m>, being tall>. I take HIT, unlike the Millian view, to assume that reference to a type of mode of presentation (or to a property of modes of presentation) of the referent of the singular term in the "that"-clause is part of the truth condition of the belief ascription, not external to it. But, unlike what I called the Opacity view, on the HIT view, as Schiffer makes clear, reference is made to a type of mode of presentation, not to a particular mode of presentation of the referent of the singular term in the "that"-clause. Furthermore, reference to the type of mode of presentation is context sensitive. Possibly, no word in the "that"-clause triggers or prompts the reference in question.

Consider the fact that, on the HIT view, there may be no word in the "that"-clause to trigger reference to the type of mode of presentation of the referent of the singular term. And consider the Millian appeal to the generation of Gricean conversational implicatures (in which, according to the Millian, reference is made to a particular mode of presentation). Then, arguably, there is a (tenuous) connection between HIT and the Millian view: They both assume that reference to the type F* of mode of presentation of the referent of the singular term in the "that"-clause depends on Gricean cooperative maxims. I take it this is what Crimmins and Perry (1989, 250) have in mind when, after rejecting the 'Fido'-Fido Theory of Belief plus Implicature theory, they write: "We shall present an account that does not ignore pragmatic features, but assign them a more honorable role. They do not create an illusion, but help to identify the reality the report is about." In other words, I take Crimmins and Perry to advocate the view that pragmatic features (such as Gricean maxims) can be involved in determining what is explicitly said in the utterance of a belief ascription. The feature I mean to emphasize here is that, on the HIT view, reference to a type of mode of presentation can depend on pragmatic features (such as Grice's cooperative maxims) and still be part of the truth conditions (or explicit content) of a belief ascription.

There are three things I would like to do in the remainder of this paper: First, I want to briefly examine two of Schiffer's own objections to HIT; second, I want to offer a response to Frege's Puzzle which differs both from Frege's own and from the Millian's in questioning Semantic Innocence; third, I want to examine the kind of contextual features involved in determining the reference to what Schiffer calls a type of mode of presentation of the referent of the singular term in a "that"-clause.

4. Schiffer's Objections to the Hidden-Indexical Theory

Although he claims that HIT is the best theory of belief reports available, Schiffer (1992, 1994) thinks that it faces insuperable difficulties, two of which are what he calls the "logical form problem" and the "meaning-intention problem". Though obviously related, the two objections are distinct from each other.[2]

The problem of logical form is partly syntactic and partly semantic. As he himself presents it, HIT assumes that "believe" is a three-place predicate (with one argument place for modes of presentation). Is it really the case that "believe" is a three-place predicate with one argument place for the believer, one for a propositional content, and one for a mode of presentation? Schiffer considers the uncontroversial case of the three-place predicate "give". Although the ternary structure of "give" may be disguised in "Mary gave her house", it is easily revealed in "Mary gave her house to her husband" where "her husband" is an uncontroversial argument of "give". And just as we can say "Mary gave her house to her husband", can't we reveal the ternary structure of "believe" in "Ralph believes that Fido is a dog" by expanding it into "Ralph believes that Fido is a dog in way w (or from one point of view w, or under mode of presentation m)"?

Well, according to Schiffer (1992, 518-19), there are two difficulties with the analogy between "give" and "believe." The first and (in my view) less convincing reason why the above expansion of the "believe" sentence does not reveal the genuine ternary nature of "believe" is that "in way w" or "under mode of presentation m" is "no ordinary language specification but technical jargon". Granted, the expressions 'mode of presentation' or 'way of thinking' are technical expressions. But why, one may ask, should logical form be specifiable in ordinary language? I am not sure how much weight ought to be given to ordinary language specifications in such matters as the logical form of belief ascriptions. Should Davidson's hypothesis that action verbs take an argument for events be responsive to ordinary language specifications?

Schiffer's second (and stronger) reason for doubting that the above expansion of the "believe" sentence reveals the genuine ternary nature of "believe" is that, although "kiss" is obviously a binary verb, we can meaningfully utter the sentence "Ralph kissed her in the most exciting way" in which "in the most exciting way" is an adverbial adjunct, not a genuine argument of "kiss". So we need to find evidence which will allow us to decide whether "believe" is more like the ternary "give" or more like the binary "kiss". But, according to Schiffer (1992, ibid.), the decisive evidence for the argument status of "her husband" in "Mary gave her house to her husband" "is revealed by the fact that we can answer 'her husband' in response to the question 'To whom did you wonder whether Mary gave the house?'". Now, on Schiffer's view, what reveals that "believe" is really like binary "kiss", and not like ternary "give", is that one cannot answer "way w" or "mode of presentation m" in answer to the question: "In what way /under what mode of presentation did you wonder whether Ralph believes that Fido is a dog?".

Now, it seems to me, in considering possible responses to the question "In what way/under what mode of presentation did you wonder whether Ralph believes that Fido is a dog?", we should distinguish the issue of whether the sentence is grammatical, acceptable, or meaningful from the issue of whether it belongs to ordinary language or to some technical jargon. Only the former issue, not the latter, it seems to me, is relevant to whether "believe" is ternary or binary.

The second related problem is the meaning-intention problem. As made clear in both Schiffer's (1992) and Crimmins and Perry's (1989) versions, one model for the hidden-indexical theory of belief ascriptions is Perry's notion of an unarticulated reference to a place in "It's raining". As Schiffer (1992, 504) has remarked, "in uttering 'It's raining,' the speaker must be referring to some place at which it is raining (typically, this is the speaker's location, but it need not be: a speaker in Chicago may reply 'It's raining' when asked about the weather in New York)". The important point about unarticulated constituency is that typically a hearer of an utterance of "It's raining" does not take the speaker to be saying that it is raining in some place or other. The hearer does not expect the speaker to have expressed a general proposition involving existential quantification over raining places. The latter proposition would be a kind of minimal proposition capable of being truth-evaluable. Typically, the audience of an utterance of "It's raining" does not take the speaker to have expressed such a minimal proposition; she takes the speaker to be referring to some specific place.

Now, a speaker, who utters "It's raining" with no word referring to the place where it is raining, could presumably, if asked, specify which place he has in mind. However, what Schiffer calls the meaning-intention problem is that the speaker who utters a belief ascription will presumably be unable to specify (or articulate) the unarticulated property F* of the mode of presentation m under which the belief-relation is said (according to HIT) to hold between the believer and the proposition believed.

I want to consider the alleged asymmetry between the unarticulated reference to a place in "It's raining" and the hypothetical reference to a property of modes of presentation in belief ascriptions, according to HIT. In the former case, the speaker can indeed supply a missing word such as "here" or "in Paris". However, uncontroversial cases of unarticulated constituency are not restricted to hidden reference to a place in "It's raining" in which the speaker can easily supply a missing word for the raining place. Other cases of unarticulated constituency involve statements of time which are relative to a time zone and reports of velocity or simultaneity between events which can only be true relative to a frame of reference. Now, it is far from clear that speakers can always specify in ordinary language, e.g., the frame of reference relative to which an assertion of simultaneity between two events or a report of velocity ought to be understood. Even though, strictly speaking, the explicit content (or the truth conditions) of such assertions (of time, velocity, or simultaneity between events) includes the specification of a time zone or a frame of reference, it is by no means clear that speakers are conscious of times zones, let alone of frames of reference. Nor is it clear that such specifications belong to ordinary language as opposed to technical jargon.

As Perry (1993, 221) has written:

The general phenomenon is using an n-place predicate or concept to deal with an n + 1-ary relation. Suppose I judge perceptually that two events happen simultaneously, and I am right. The fact that makes me right is that those two events were simultaneous relative to my certain frame of reference... The frame of reference in question is not determined by a representation in my thought, but by the broader situation in which my judgment takes place. A theorist who is analyzing the way an agent handles information and uses it to guide action may have to pay attention to factors the agent's cognitive system can safely ignore. The theorist's interest may be precisely how these factors can be ignored - how architectural or external constraints make internal representations unnecessary. It is the speed of light that allows us to get by with a 2-place concept of simultaneity. It is the shortness of our arms compared to the width of time-zones that allows us to ignore the latter when we read our watches.

Seen in the light of the general phenomenon of "using a n-place predicate to deal with an n + 1-ary relation", the requirement that the speaker of a belief ascription ought to be able to specify (or articulate) the unarticulated property F* of the mode of presentation m ought to lose some of its grip. The notion of a mode of presentation is a theoretical notion of which a speaker may have no conscious ordinary concept.

Second, as already noticed above, in the latter case, the objection cannot be just that, in the case of a belief ascription, the speaker could not find some words or other standing for the property of modes of presentation. For in fact, the speaker who uttered a belief ascription could use words referring to a property of modes of presentation such as "in one way" or "from one point of view" or "under one mode of presentation", as opposed to "in some other way" or "from another point of view" or "under another mode of presentation" (as in 'under the "Cicero" mode, not under the "Tully" mode'). Rather, it would seem, the objection must be that the reference of such words is vague, indeterminate or underdetermined.

Why, I wonder, could not the advocate of HIT bite the bullet and, in Schiffer's (1994, 9) own terms, "allow for the belief ascriber to be making an indeterminate reference to a type of mode of presentation"? Isn't the reference of "here" similarly vague or indeterminate? Consider Schiffer's (1994, ibid.) example:

Suppose you call Ernie Lepore in New Brunswick and ask him where Jerry Fodor is. "He's here," Ernie replies. To what does the utterance of "here" refer? To New Brunswick? To Rutgers University? To Douglas Campus? To Davison Hall? To Ernie's office?... Almost certainly, Ernie's utterance of "here" doesn't refer to some definite region of space. The word is being used to make a vague or indeterminate reference.

Now, if, as Schiffer concedes, the truth conditions of Ernie's utterance of "He's here" include a reference to some indeterminate region of space, why couldn't the advocate of HIT claim that the truth conditions of a belief ascription contain a reference to an indeterminate type of mode of presentation of the referent of the singular term in the "that"-clause? Granted, this move exposes the advocate of HIT to the question: What does it mean to claim that such an indeterminate type of mode of presentation of the referent of the singular term in the "that"-clause is part of the truth conditions of a belief ascription? I will come back to this question in Section 7 of this paper.

5. Richard's Puzzle and the Second Premise of Frege's Puzzle

Both by way of offering a further motivation for HIT, and for the purpose of probing the second premise in Frege's Puzzle, I will examine a puzzle put forward by Richard (1983, 1990):

Consider A - a man stipulated to be intelligent, rational, a competent speaker of English, etc. - who both sees a woman across the street, in a phone booth, and is speaking to a woman through a phone. He does not realize that the woman to whom he is speaking - B to give her a name - is the woman he sees. He perceives her to be in some danger - a runaway steamroller, say, is bearing down upon her phone booth. A waves at the woman; he says nothing into the phone (Richard 1983, 184).

In this setting, we do have the intuition that, unlike A's utterance of (7), A's utterance of (6) would express a truth:

(6) I believe that she [he points across the street] is in danger.

(7) I believe that you [he speaks into the phone] are in danger.[3]

Suppose, however, with Richard, that upon seeing A's waving, B, speaking to A, were to utter (8) on the phone:

(8) The man watching me believes that I am in danger.

She would thereby express a truth, i.e., a true belief she may have. Now, trusting B's statement, A might echo B's utterance by uttering (9):

(9) The man watching you believes that you are in danger.

A would thereby express a truth. And now, Richard claims, from the truth expressed by A's utterance of (9) together with his utterance of (10), A could deduce the truth of (7):

(10) I am the man watching you.

So now we have what seems like an argument for the joint truth of (6) and (7), when our intuition is that (6) is true, but (7) is not. In Crimmins and Perry's (1989: 260) words, this may well be "the ultimate doxastic puzzle."

There are two things I would like to notice about Richard's derivation of the truth of (7). First, I would grant that A's utterance of (9) expresses a true proposition. But what proposition does it express? If A trusts B (if he has no reason not to trust her), he may be confident that his utterance of (9) expresses a truth. But which truth does he thereby express? Which truth does he think he thereby expresses? A's utterance of (9) is reminiscent of a famous example of Kaplan (1989, 508-10). By uttering "I am here now", a speaker may know a priori - he may know with certainty - that his utterance expresses a true proposition. But considering that he may neither know where he is nor what time it is (not to speak of his own identity), he may not know which true proposition he just expressed. By uttering (9), in the story, A is merely echoing B's utterance of (8). He seems to be expressing his higher-order belief that some unidentified man has a belief about B that she is in danger.

Second, Richard's derivation, it seems to me, begs the question in the following sense. Remember that the derivation is based on A's reasoning, i.e., on beliefs accessible to A. Now, were A in a position to utter (10), he would thereby express a true statement. But he hardly seems to be in such a position. And if he is not, then there is a missing step in the derivation. In other words, if he can form a belief he would express by uttering (10), then ipso facto, it would seem, he would be in a position to form a belief directly expressible by his utterance of (7). So either he is in a position to form the belief expressible by uttering (10), or he is not. If he is, then the derivation is superfluous. If he is not, then the derivation does not go through.

In any case, the view I would like to take is that the truth condition of A's utterance of (7) differs from the truth condition of A's utterance of (9). And furthermore, the difference in truth conditions is due to a difference in the content contributed by the embedded "that"-clause, i.e., "you are in danger". This is where the puzzle does, I think, teach us something important which can be accommodated by HIT. As Crimmins and Perry (1989, 275) say, "Richard's case is especially interesting because it shows how a contextual shift can be brought about by a change in wording outside the embedded sentence in a belief report." The sentence type "you are in danger" is a syntactic constituent of both sentences (7) and (9). We can assume that the nonlinguistic contextual features responsible for fixing the reference of the pronoun "you" in both cases remain constant: A's two uses of "you" refer to B. What I want to argue is that, although the embedded "that"-clause is unchanged and although the nonlinguistic contextual features involved in fixing the reference of "you" are invariant, still, the "that"-clause does not contribute the same content to the truth conditions of utterances of (7) and (9).

The relevant difference between (7) and (9) resides in the way A refers to himself in the matrix or main clause: either as the reference of "I" in (7) or as the reference of "the man watching you" in (9). Following Crimmins and Perry, I would argue that this difference in wording preceding the "that"-clause can affect the content expressed by the "that"-clause. Crimmins and Perry argue that A's use of (7) makes the claim that he has a belief involving an acoustic mode presentation of B as the person being talked to through the phone to the effect that this person is in danger. A lacks such a belief. A's use of (9) makes the claim that he has a belief involving a visual mode of presentation of B as the person he is currently seeing in a phone-booth to the effect that this person is in danger. A has such a belief. The difference in the way of referring to A induces a difference in the way of thinking of B, the reference of "you" in (7) and (9). I would claim that, for an utterance of (9) to be true, the two occurrences of "you" have to be thought by A (the speaker) to be coreferential. For A to realize that the two occurrences of "you" are coreferential, he must think of the reference of the second occurrence of "you", not only as an individual he is currently addressing on the phone, but also as an individual who is salient in the visual field of someone or other. No such constraint is present in (7).

In Richard's puzzle, words outside the "that"-clause induce a change in the content expressed by the "that"-clause. A similar phenomenon has been noted by Nunberg (1991, 16) and Recanati (1993, 309-10; 352): "John is giving me my fiftieth chess lesson; I have just made a dubious move." Now consider two sentences which John might utter in this context:

[N1] You often get into trouble with that move.

[N2] According to Horowitz, you often get into trouble with that move (Recanati 1993, 352).

Nunberg and Recanati argue that, by uttering [N1], in a given context, the speaker has expressed a singular proposition about his hearer. By uttering [N2] in the very same nonlinguistic context, given that plausibly Horowitz has never had the speaker's hearer in mind, the speaker would express a general proposition in which the propositional contribution of the pronoun "you" is not the hearer but "the concept 'whoever makes the move which has just been made'" (Recanati 1993, 352). The point, as in Richard's puzzle, is that two tokens of the same sentence type are uttered in the same nonlinguistic context. However, as in Richard's puzzle, in Nunberg's example, the linguistic contexts in which two tokens of the same sentence-type are embedded are not the same. The change in linguistic context prompts a change in the proposition contributed by the two tokens of the same sentence type.

What reflection on Richard's puzzle and Nunberg's example suggests is a response to Frege's Puzzle, which differs from both Frege's and the Millian's. The former rejected DR (premise I); the latter rejected premise III, which I called Common Sense. In Richard's puzzle, we seem to have a pure instance of the antecedent of premiss (II), the one I called 'Semantic Compositionality', and a violation of its consequent. We are to imagine two distinct utterances of one and the same sentence type "you are in danger" in the very same nonlinguistic context of utterance, where "you" would twice be used to refer to the same individual, B. Presumably, if two tokens of the same sentence-type "you are in danger" were uttered by A in the same (nonlinguistic) context, referring twice to B, the two utterances ought to express the same proposition. But now, when two tokens of the same sentence-type are embedded within two main clauses with two distinct and coreferential singular terms in subject position, they do not express the same content. Or so I claim. Similarly, in Nunberg's example, by changing the linguistic embedding of two tokens of the same sentence type and keeping the nonlinguistic context invariant, we change the proposition contributed by the two tokens of the same sentence type. And this is surprising.

6. Semantic Innocence and Unarticulated Constituency

Consider (3a) again. The name "Claire" in subject position in the matrix (or main) clause of sentence (3a) is, it seems, substituable salva veritate by the coreferential description "my oldest daughter". If Claire is my oldest daughter, and if (3a) is true, then so is "My oldest daughter believes that Cicero was tall". So by virtue of premise II of Frege's Puzzle (Semantic Compositionality), given that "I" in (7) and "the man watching you" in (9) are coreferential, we should expect that, if two utterances of "you are in danger" express one and the same proposition, then if an utterance of (10) expresses a truth, then so should an utterance of (7). I take Richard's puzzle and Nunberg's example to be evidence against premise II of Frege's Puzzle (Semantic Compositionality).

What, I think, makes premise II of Frege's Puzzle (Semantic Compositionality) plausible is a view which seems, on the face of it, irresistible, but which I would like to examine critically. It is a view which Crimmins and Perry (1989, 250), following Davidson (1969) and Barwise and Perry (1981), call 'Semantic Innocence'. In their terms, it is the view that "the utterances of the embedded sentences in belief reports express just the propositions they would if not embedded." This is a view which I found very plausible too. Of course, Semantic Innocence directly contradicts the Fregean slogan that, in a belief context, a singular term contained in the "that"-clause does not have the same reference as in a simple predication. Semantic Innocence - which is well accommodated by the Millian view - has been explicitly endorsed by, e.g., Fodor (1989, 169):

Consider the expression 'believes that E' where it is used to attribute to some agent the state of believing that E... How does the "E" part work? ...I think this works in the following simple and aesthetically satisfying fashion. The proposition that is the object of the belief-state that is attributed by using the formula 'believes E' is the very same proposition that is expressed by using the unembedded formula E. So, for example, the expression 'believes that it's raining' is used to attribute a belief-relation to the proposition that it's raining; and this is the very same proposition that the unembedded formula 'it's raining' is used to express.

This is as good a statement of Semantic Innocence in Crimmins and Perry's sense, I think, as one can find. Semantic Innocence is the thesis that I would like now to question. It is interesting that Fodor should choose one of the very examples used by Perry (1986) to illustrate his notion of an unarticulated constituent, namely, the utterance of the sentence "It's raining". As Perry's younger son's utterance of the sentence "It is raining" one Saturday morning in Palo Alto famously shows, in Perry's (1986, 206) words, "Palo Alto is a constituent of the content of [his] son's remark, which no component of his statement designated; it is an unarticulated constituent."

The point of unarticulated constituency is that two utterances of "It is raining" may contain two distinct unarticulated references to two different places. But if one accepts the notion of unarticulated contituency, then, it seems, Semantic Innocence (at least in Fodor's version) must go. According to Fodor's statement of Semantic Innocence, "the expression 'believes that it's raining' is used to attribute a belief-relation to the proposition that it's raining; and this is the very same proposition that the unembedded formula 'it's raining' is used to express." What is the proposition expressed by an utterance of "it's raining"? Is there a single such proposition? According to unarticulated constituency, it cannot be the proposition that it is merely raining. Nor is it, I have claimed, the proposition that it is raining somewhere or other. No, it is the proposition that it is raining in some contextually determined definite place.

What kinds of contextual mechanisms are involved? In stating the HIT view, Schiffer has emphasized two features of the way the reference to a type of mode of presentation of the referent of the singular term in the "that"-clause is determined: it is indexical and it is implicit or hidden. Consider the difference between an utterance of "It is raining" and "It is raining here". By uttering the latter, a speaker would be taken to refer to some specific place too. What is the difference? Let me again quote Crimmins and Perry (1989, 265):

The phenomenon of underarticulated constituency is similar to that of indexicality in the reliance on context. But the two phenomena should not be conflated. If we say "It is raining here", an expression in our statement identifies the place. The place is articulated in a context-sensitive way. In the case of indexicals, expression and context share in the job of identifying the constituent, according to the conventional meaning or character of the indexical. In a case of underarticulation, there is no expression to determine the constituent in this way.

Now, it seems to me, as Kaplan's work on demonstratives shows, when we think of the role of context in determining the reference of an indexical word, we can distinguish two kinds of contextual features. Consider the case of a "pure" indexical like "I". It seems that Kaplan's notion of a character suffices to determine the reference (or the contribution of a use of "I" to the singular proposition expressed). It seems that, for a person to know who the reference of a given token of "I" is, it suffices to know who uttered the token in question. I shall call extensional such contextual features. Consider now a token of a demonstrative like "he". Arguably, in order to help the hearer determine the reference of a token of "he", the speaker may supply a "demonstration" of the referent of the demonstrative. But of course, the speaker may so point demonstratively to the referent of his token of "he" only if the referent is perceptually accessible to both speaker and hearer. When the referent is not so perceptually accessible, the hearer's task may in addition involve the interpretation of the speaker's intentions and/or perhaps knowledge or previous elements of the content of the speaker's thinking and/or the topic of his or her speech. I shall call intensional those further aspects of the context relevant to determining the reference of a demonstrative. In other words, what a person may know in virtue of his or her semantic competence may not suffice to determine the reference of a demonstrative like "he". What a token of "he" refers to may depend on the interpretation of the speaker's intentions. It may not be completely determined by a rule of grammar.

As suggested by Crimmins and Perry, a token of "here" is usually meant to refer to the place of utterance. Purely extensional features of the context of utterance, however, may not be sufficient to determine the reference of "here" or "now". Intensional aspects of the context may be relevant too to determine the scale. As already made clear by Schiffer's example of the indeterminate reference of a token of "here", I could use "here" to refer to a chair, a room, a house, a street, a neighborhood, a city, a country, the Earth, the solar system, etc. Similarly "now" could be used to refer to intervals of time of varying length: to a second, an hour, a year, to the time interval corresponding to a generation, to a century, or to a longer time interval. The point I would therefore like to make is that both Schiffer's and Crimmins and Perry's use of the term "indexical" should not blind us to the fact that the contextual features relevant to determining the contribution of a "that"-clause to the truth conditions of a belief ascription - i.e., the fixing of the reference to a type of mode of presentation of the singular term in the "that"-clause - may well involve what I call intensional, and not only extensional, contextual features. It might be less misleading, in this respect, to link the reference to the type of mode of presentation of the referent of the singular term in the "that"-clause to the reference of demonstratives rather than to the reference of "pure" indexicals.

On the version of HIT which I would favor then, the rejection of Semantic Innocence does not force one to accept the Fregean view that a singular term in a "that"-clause does not have its ordinary reference. However, it does entail that the fact that the singular term in a "that"-clause is embedded under "believe" may trigger a process of reference to a property of the mode of presentation of the referent of the singular term. Given that the nonlinguistic context is fixed, no such pragmatic (interpretive) process need be triggered in the utterance of a simple predication containing the singular term in question. The pragmatic process at work in belief ascription is a process of enrichment due to the linguistic context.

7. A Pragmatic View of the Truth Conditions of Belief Ascriptions

I said above that there is, on my view, one point in common between HIT and the Millian view. It is time to say what it is: Both views find it relevant to turn to something like Grice's cooperative maxims of conversation in order to determine the type of mode of presentation of the referent of the singular term in the "that"-clause. However, unlike the Millian view, HIT assumes that the reference to the type of mode of presentation in question is not external to what is said by (or to the truth conditions of) an utterance of a belief ascription: it is part of its truth conditions.

For lack of space, I will illustrate all too briefly what I have in mind by ascribing to HIT the thesis that Grice's maxims may be used to determine, not just the implicatures of an utterance, but its truth conditions too. First, I notice that, in defense of the 'Fido'-Fido theory of belief plus Implicature, Salmon (1989: 252) has argued for an analogy between the interpretation of belief ascriptions and the interpretation of conjoined utterances:

Surely there can be... a mechanism that, when employed, sometimes has the unintended and unnoticed consequence that speakers mistake what is conveyed ("implicated") for the literal content. Consider, for example, the conjunction 'Jane became pregnant and she got married' , which normally carries the implicature that Jane became pregnant before getting married. Utterers of this sentence, in order to employ it with its customary implicature, need not be aware that the sentence is literally true even if Jane became pregnant only after getting married. Some utterers may well become misled by the sentence's customary implicature into believing that the sentence literally means precisely what it normally conveys - so that, if they believe that Mary became pregnant only after getting married, they would reject the true but misleading conjunction as literally false.

(For a similar analogy, see Richard 1990, 120).

Salmon's view of the content of utterances of sentences conjoined by "and" is faithful to Grice's own view. But this view has been recently challenged in the framework of Sperber and Wilson's (1986) Relevance-based approach. In particular, an alternative view has been put forward in an insightful paper by Carston (1988). It should be striking how Carston's Relevance-based approach fits in with Perry's notion of unarticulated constituency. I will succinctly consider a number of examples provided by Carston in which a process of enrichment seems at work very much like Perry's view that the context supplies unarticulated constituents:

(11) a. He gave her his keys and she opened the door.

b. John has been insulted and he is going to resign.

c. He ran to the edge of the cliff and jumped.

d. I went to the exhibition and ran into John.

e. She took the gun, walked into the garden, and killed her mother.

f. I had a holiday in Austria and did some cross-country skiing.

The question raised by all these examples from the point of view assumed by Salmon's quotation is: What is said, as opposed to what is implicated, by any of the utterances in (11)? In (11b), is the fact that the event described in the second conjunct is a consequence of the event described in the first part of the explicit content of an utterance of (11b)? In (11c), is the fact that the person who is said to have run to the edge of the cliff and to have jumped did jump from the edge of the cliff part of the explicit content of an utterance of (11c)? In (11d), is the fact that I ran into John at the exhibition to which I went part of the explicit content of an utterance of (11d)? And so on. Notice that, in the relevant cases, words corresponding to some of the content are missing so that the hearer must infer the relevant content from contextual clues - where the context involves linguistic elements from previous utterances.

Let us concentrate on (11a). In order to understand what the speaker said by her utterance of (11a), not only must the hearer determine the reference of the third-person pronouns and other referential expressions; he must also assign the proper coreference and anaphoric relations. He must even mentally provide a missing prepositional phrase, not overtly present in the sentence: He must understand that she opened the door with the key he gave her. As Carston (1988, 161-2) has argued persuasively, once the hearer has gone that far in the interpretation of the explicit content of an utterance of (11a), why not include an explicit representation of the temporal ordering between the events mentioned in the two conjoined sentences? It is therefore plausible that the proposition explicitly expressed by a utterance of (11a) should include, in addition to the coreference and/or anaphoric relations, the temporal ordering as in the following representation:

He1 gave her2 [his1 key]3 at t and she2 opened the door at t + n with [his1 key]3

The contrast between Salmon's view of the contents of utterances of sentences conjoined by "and" and Carston's is that the former takes to be external to what is said what the latter takes to be an integral part of what is said. However, on Carston's view, the gap between the semantic structure of the sentences uttered and the proposition explicitly expressed by the utterance is filled in by the kind of Gricean pragmatic mechanisms which Salmon appeals to in order to determine the implicatures of the utterance: i.e., a process of enrichment. My suggestion is that the contrast between the Millian view and the hidden-indexical view of the reference to the mode of presentation of the referent of the singular term in the "that"-clause reflects the contrast between the two views of the contents of utterances of sentences conjoined by "and".

In all of Carston's examples, we have a process of enrichment reminiscent of the process which, I claimed (in Section 6), is triggered by a belief context, and which gives rise to a reference to a type of mode of presentation (of the referent of the singular term in the "that"-clause). Now, I want to go back to the analogy suggested by Schiffer (1994) between the indeterminate reference of a token of "here" and the indeterminate reference made by a belief ascriber to a type of mode of presentation (of the referent of a singular term in the "that"-clause). This analogy prompted (toward the end of Section 4) the question: How should one understand such a reference to an indeterminate property of modes of presentation?

Consider (12), an example due to Katz (1977, 19-20), and reanalyzed by Sperber and Wilson (1986, 192-93) in the context of an argument for their thesis that not every thought can be expressed by (or "encoded by a sense of") some sentence of any natural language:

(12) Thank God, he is gone.

In Katz's (1977, 20) scenario, an utterance of (12) would be a shorthand for "The man who just asked the stupid question about the relation between the mental and the physical has, thank God, left the room". Of course, here we are not dealing with a type of mode of presentation of a singular term in a "that"-clause. We are rather dealing with a type of mode of presentation which is a constituent of the thought expressed by a speaker who uttered a simple predication containing the demonstrative "he".

First, Sperber and Wilson argue that unless the various references to times and places are fixed, no full proposition has been explicitly expressed by Katz's candidate for the full disambiguation of (12). Second, they argue (ibid., 193) that, for communication between speaker and hearer to succeed, it is not necessary that speaker and hearer think of the referent of "he" under the very same mode of presentation:

... two people may be able to think of the same man that he has gone, without being able to think exactly the same thought, because they might not individuate him in exactly the same way. Similarly, by saying "He has gone" I may induce in you a thought which is similar to mine in that it predicates the same thing (that he is gone) of the same individual, but which differs from mine in the way you fix the reference of "He". It seems to us neither paradoxical nor counterintuitive to say that there are thoughts that we cannot exactly share, and that communication can be successful without resulting in an exact duplication of thoughts in communicator and audience.

To say, as Sperber and Wilson do, that no single mode of presentation of the referent of "He" in an utterance of (12) need to be entertained by both speaker and hearer for the latter to understand what the former has said, is to say that the thought (or proposition) expressed by an utterance of (12) contains a mode of presentation of the referent of "he" and that this mode of presentation remains indeterminate. Following their lead, I would like to think of the property of modes of presentation which is, according to HIT, referred to in a belief ascription as similarly vague or indeterminate in the following sense. The speaker does not take full responsibility for the property which the mode of presentation (of the singular term in the "that"-clause) must have (or the type under which it must fall) for the hearer to grasp what the speaker has said. The hearer has some latitude for filling in the indeterminacy and make her own contribution to the determination of the property in question.

Concluding Remarks on the Parallel between Quotations and Belief Contexts

As I said above when considering the Opacity view (Section 2.1), there is an interesting parallel to be made between belief contexts and quotation contexts appropriately enlarged to include indirect as well as direct quotations. The parallel carries over the references respectively of quotations and "that"-clauses. Before ending this paper, I will take a brief look at the role of pragmatic factors involved in determining the reference of quotations.

Following Sperber and Wilson (1986, 227-30) and Jacob (1987), consider cases in which a bilingual speaker B might report to an English-speaking audience C what a speaker A originally said in French. He might directly quote the original French utterance; a different token of the initial French sentence type would occur within quotation marks. Arguably, the direct quotation refers to the sentence type a token of which is within the quotation marks. Notice, however, that if C did not understand A's original utterance, then C is not likely to understand B's direct quotation any better. So for C's benefit, B might try as literal a translation as possible, in which case his report would be an indirect quotation preceded by "said that". Although B's translation of A's French utterance attempts to preserve both the proposition originally expressed in French and the semantic structure of the French sentence uttered, it requires interpretation on the part of B. Consider now a case in which A made a long speech in French and B's report is an English summary of A's original utterance. Arguably, in the case of literal translation, the part of B's utterance following "said that" refers to the content or proposition originally expressed by A's utterance. But in the case of a summary, what the part of B's utterance following "said that" refers to cannot be strictly speaking the content originally expressed by A's utterance. A might have uttered many sentences expressing many distinct propositions. B's summary will consist of a shorter set of sentences expressing fewer propositions. If B summarized A's utterance, then the content of the "that"-clause in B's utterance will be something which suitably resembles the content originally expressed by A's utterance. The former will not be identical to the latter.

What translation as a method of reporting utterances shows is that one and the same "that"-clause can be used to refer to different sentence types belonging to different languages (to which the reported utterances belong). For example, a single English "that"-clause might be used to refer to sentence types belonging to many different languages. What the possibility of summary shows is that even if the language of the report is the same as the language of the reported utterance, then one and the same "that"-clause could be used to refer to different sets of original utterances. What the examination of "that"-clauses in indirect quotations reveals is that the determination of their reference is contextual. The analogy between the reference of "that"-clauses in indirect quotations and the reference of "that"-clauses in belief contexts suggests that the same kind of pragmatic (enrichment) processes are at work in the determination of the latter. Since the contribution of the "that"-clause to the truth conditions of a belief ascription depends on the type of mode of presentation of the referent of the singular term in the "that"-clause, and since the latter may depend on pragmatic processes of enrichment, it follows that the contribution of the "that"-clause to the truth condition of the belief ascription too depends on such pragmatic processes of enrichment.

References

Barwise, J. and J. Perry. 1981. Semantic Innocence and Uncompromising Situations. In P.A. French, T.E. Uehling, Jr., H.K. Wettstein, eds. 1981.The Foundations of Analytic Philosophy. Midwest Studies in Philosophy, vol. VI, 387-404. Minneapolis: University of Minnesota Press.

Carston, R. 1988. Implicature, Explicature, and Truth-Theoretic Semantics. In R. Kempson, ed. 1988. Mental Representations, the Interface between Language and Reality. Cambridge: Cambridge University Press.

Crimmins, M. 1992. Talk about Beliefs. Cambridge, Mass.: MIT Press.

Crimmins, M. and J. Perry. 1989. The Prince and the Phone-Booth: Reporting Puzzling Beliefs. In J. Perry. 1993. The Essential Indexical and Other Essays. Oxford: Oxford University Press.

Davidson, D. 1969. On Saying that. In D. Davidson, 1984. Inquiries Into Truth and Interpretation. Oxford: Clarendon Press.

Fodor, J.A. 1989. Substitution Arguments and the Individuation of Beliefs. In J.A. Fodor, 1990. A Theory of Content and Other Essays . Cambridge, Mass.: MIT Press.

Frege, G. 1892. On Sense and Reference. In P. Geach and M. Black eds. 1970. Translations from the Philosophical Writings of Gottlob Frege, Oxford: Blackwell.

Jacob, P. 1987. Thoughts and Belief Ascription. Mind and Language, 2, n°4: 301-25.

Kaplan, D. 1978. Dthat. In P.A. French, T.E. Uehling, Jr. and H.K. Wettstein, eds. 1979. Contemporary Perspectives in the Philosophy of Language. Minneapolis: Minnesota University Press.

Kaplan, D. 1989. Demonstratives. In J. Almog, J. Perry and H.K. Wettstein, eds. 1989.Themes from Kaplan, Oxford: Oxford University Press.

Katz, J.J. 1977. Propositional Structure and Illocutionary Force: A Study of the Contribution of Sentence Meaning to Speech Acts. New York: Crowell.

Perry, J. 1986. Thought Without Representation. In J. Perry (1993) The Essential Indexical and Other Essays. Oxford: Oxford University Press.

Recanati, F. 1993. Direct Reference, From Language to Thought. Oxford: Blackwell.

Recanati (forthcoming) Quasi-singular Propositions: the Semantics of Belief Reports.

Richard, M. 1983. Direct Reference and Ascriptions of Belief. In N. Salmon and S. Soames, eds. 1988. Propositions and Attitudes. Oxford: Oxford University Press.

Richard, M. 1990. Propositional Attitudes. Cambridge: Cambridge University Press.

Salmon, N. 1986. Frege's Puzzle. Cambridge, Mass.: MIT Press.

Salmon, N. 1989. Illogical Belief. In Philosophical Perspectives, 3: 243-85.

Schiffer, S. 1977. Naming and Knowing. In P. A. French, T.E. Uehling, Jr. and H.K. Wettstein, eds. 1979. Contemporary Perspectives in the Philosophy of Language. Minneapolis: the University of Minnesota Press.

Schiffer, S. 1978. The Basis of Reference. Erkenntnis, 13: 171-206.

Schiffer, S. 1987. The 'Fido'-Fido Theory of Belief. Philosophical Perspectives. 1: 455-80.

Schiffer, S. 1992. Belief Ascriptions, The Journal of Philosophy, 89, n°10: 499-521.

Schiffer, S. 1994. Descriptions, Indexicals, and Belief Reports: Some Dilemmas (But Not the Ones You Expect). Chapter 10 in this volume.

Sperber, D. and D. Wilson. 1986. Relevance, Communication and Cognition. Cambridge, Mass.: Harvard University Press.

---

[1] In this paper, I only consider the mode of presentation of the referent of the singular term (in or out of a "that"-clause). Both the advocate of the Millian view and the advocate of the "hidden-indexical theory" are prone to talk of the mode of presentation of the singular proposition (i.e., the ordered pair comprising an object and a property). Now, such a mode of presentation will include the mode of presentation of the object and the mode of presentation of the property. However, for purposes of simplification, I simply ignore the mode of presentation of the property.

[2] In this paper, I examine the force of Schiffer's objections to the version of the hidden-indexical theory of belief ascriptions based on the assumption that "believe" is a three-place relation with one argument place for modes of presentation. While thinking about this issue, I have seen Recanati (forthcoming) in which he argues in favor of a different ternary logical form of belief ascriptions. In his version, modes of presentation are constituents of quasi-singular propositions. Quasi-singular propositions are arguments of the belief relation. However, modes of presentation as such are not. Whether Recanati's version should count as another version of the hidden-indexical theory is an open question.

[3] Arguably, utterances of (6) and (7) may be said to have two readings. On one reading, they have the same truth conditions, i.e., one and the same individual is believed to be in danger. On this reading, the difference in modes of presentation of the person who is believed to be in danger is irrelevant. On the other reading, the mode of presentation of the person referred to in the "that"-clause is relevant to the truth conditions of the utterances of (6) and (7). I consider only the latter reading.