Hostname: page-component-cd9895bd7-7cvxr Total loading time: 0 Render date: 2024-12-26T18:07:29.449Z Has data issue: false hasContentIssue false
  • English
  • Français

Beyond Singular Propositions?

Published online by Cambridge University Press:  01 January 2020

Scott Soames
Scott Soames*
Affiliation:
Princeton University, Princeton, NJ08544-1006, USA
Get access Rights & Permissions [Opens in a new window]

Extract

Propositional attitudes, like believing and asserting, are relations between agents and propositions. Agents are individuals who do the believing and asserting; propositions are things that are believed and asserted. Propositional attitude ascriptions are sentences that ascribe propositional attitudes to agents. For example, a propositional attitude ascription α believes, or asserts, that S is true iff the referent of a bears the relation of believing, or asserting, to the proposition expressed by s. The questions I will address have to do with the precise nature of propositions, and the attitudes, like belief, that we bear to them.

I will assume both that propositions are the semantic contents of sentences, and that the proposition expressed by a sentence is a structured complex made up of the semantic contents of the parts of the sentences that express it.

Type
Research Article
Information
Canadian Journal of Philosophy , Volume 25 , Issue 4 , December 1995 , pp. 515 - 549
Copyright
Copyright © The Authors 1995

Access options

Get access to the full version of this content by using one of the access options below. (Log in options will check for institutional or personal access. Content may require purchase if you do not have access.)

References

1 The ideas presented in the first three sections of this paper were developed in 1992, and first presented at the Instituto De Investigaciones Filosóficas, Autonomous University of Mexico, Mexico City, on 27 January 1993. Later versions of the paper were given at the Department of Linguistics, The Graduate Center of the City University of New York, on 14 October 1993, and at the Pacific Division Meeting of the American Philosophical Association on 26 March 1994. The final two sections (IV and V) of the paper were written in response to Mark Richard's reply at the Pacific Division Meeting to the material in sections I - III. The material in these new sections was presented informally in discussion at that meeting, but was not written up until later. I would like to thank Ali Akhtar Kazmi for useful discussions of this new material.

2 Boldface italics are used to indicate corner quotes; italics alone are used in place of single quotes.

3 See Salmon, Nathan Frege's Puzzle (Cambridge, MA: The MIT Press 1986)Google Scholar; and Soames, Scott ‘Substitutivity,’ in On Being and Saying: Essays for Richard Cartwright, Thomson, Judith Jarvis ed. (Cambridge, MA: The MIT Press 1987) 99–132Google Scholar.

4 In (4) and (5) I am treating the verb phrases ‘is visible only in the morning’ and ‘is visible only in the evening’ as having simple, unstructured semantic contents, rather than having contents that are themselves structured complexes of the semantic contents of the semantically significant constituents of the verb phrases. This is a simplification designed to increase readability. It does not affect the argument.

5 More precisely, it is sufficient to grasp the proposition by means of any sentence or representation that expresses it, and to accept the proposition when grasped in that way. In most cases this just amounts to accepting and understanding a sentence or representation that expresses the proposition. However, delicate issues arise in special cases - for example, in cases involving Kaplan's ‘dthat’ operator, as well as in cases involving so-called descriptive names (names introduced by stipulations that their referents are to be the denotations of certain specified descriptions). In such cases a description may be used to specify the content of a directly referential singular term, even though the speaker does not know which object is denoted by the description, and hence does not know which object is the content of the term. Such a speaker may accept the sentence If there is a unique D, then dthat(D) is D in the sense of knowing that it expresses a truth, without knowing which truth it expresses. Does the speaker understand the sentence? In a sense he does, since he understands all the words, as well as how they are combined in the sentence. However, in a sense he doesn't, since he does not know which proposition the sentence expresses; he does not grasp the proposition by means of that sentence. It is not important for the issues I raise in the text to decide precisely how to characterize these special cases. I will say, however, that in the sort of case just imagined the speaker's acceptance of the sentence containing dthat, together with his understanding of the words in the sentence and their mode of combination, is not, in my view, sufficient for him to believe a singular proposition concerning the referent of the description D.

6 Further arguments specifically directed against attempts to analyze names or indexicals as rigidified descriptions (using the actuality operator) can be given by appealing to certain facts about propositional attitudes. Limitations of space do not allow them to be presented here.

7 I set aside here delicate questions about the precise nature of the vehicle mediating the belief relation. For further discussion see Salmon, Frege's Puzzle, chapter 8.

8 For further discussion see chapter 8 of Salmon's Frege’ s Puzzle, especially section 8.4, 114-18; in addition see sections VII and VIII, 117-22, of my ‘Substitutivity.’

9 Richard, Mark Propositional Attitudes (Cambridge: Cambridge University Press 1990)CrossRefGoogle Scholar

10 This is a simplification of Richard's official position. According to the official position, the belief predicate expresses a three-place relation between individuals, annotated propositions, and correlation functions. (More on these below.) In simple cases, belief ascriptions are evaluated by existentially generalizing on the argument place standing for correlation functions. In such cases, Ralph believes that S is true in a context C iff there is some correlation function satisfying the restrictions involving Ralph in C which maps the annotated proposition expressed by S in C onto some annotated proposition accepted by Ralph (in the world of C). If all cases were like this we could simply take the belief predicate to express a two-place relation between individuals and annotated propositions that is the existential generalization of a three place relation between individuals, annotated propositions, and correlation functions.

However, Richard wishes to extend his theory to cover discourses containing multiple belief ascriptions involving what is understood to be the same agent. For example, in evaluating a discourse, Ralph believes that n is F. He also believes that n is G, Richard wants to have the freedom to assign it either, or both, of the following readings: (i) there is a correlation function f satisfying constraints involving Ralph in the context which maps the annotated proposition expressed by n is F in the context onto an annotated proposition accepted by Ralph and there is a correlation function g satisfying constraints involving Ralph in the context which maps the annotated proposition expressed by n is G in the context onto an annotated proposition accepted by Ralph; (ii) there is a correlation function f satisfying constraints involving Ralph in the context which maps the annotated proposition expressed by n is F in the context onto an annotated proposition accepted by Ralph and which also maps the annotated proposition expressed by n is G in the context onto an annotated proposition accepted by Ralph. The freedom to generate both readings is achieved by analyzing the belief predicate as expressing a three place relation between individuals, annotated propositions, and correlation functions, and allowing different scope possibilities in discourses for the existential quantifier binding the third argument place (including the possibility that a single such quantifier may bind the final argument place in several different occurrences of attitude predicates in the discourse).

Since the issues concerning Richard's theory that I will be raising do not directly involve discourses like these, I will adopt the simplifying assumption that, for Richard, the belief predicate expresses a relation between individuals and annotated propositions.

11 (20) is also a simplification. For Richard, the things we accept need not be sentences of any public language, but rather can be internal, mental representations. This is important for his treatment of the beliefs of creatures who do not use any public language, and for his handling of the monolingual version of Kripke's puzzle involving Peter and Paderewski. However, it is not important for the issues that I raise in this paper. Thus, in the interest of simplicity, we may continue to think of belief as involving the acceptance of sentences. It must be remembered, however, that this is an expository convenience.

12 Strictly speaking, this claim must be qualified slightly to cover a feature of Richard's approach that arises in the special case of belief ascriptions the complement clauses of which contain multiple occurrences of the same term. Regarding such cases, a strictly Russellian approach has both (i) and (ii) as consequences; whereas Richard's approach has (i), but not (ii), as a consequence, when the contexts in question are restricted to those in which there are no constraints on correlation functions. (i) Let t be any term, t’ and t* be any pair of distinct terms, and let C be a context (and A an assignment of values to variables) such that the Russellian content of t with respect to C (and A) is the same as the Russellian content of both t’ and t* with respect to C (and A). H Ralph believes that … t … t … is true with respect to C (and A), then Ralph believes that … t’ … t* … must also be true with respect to C (and A). For example, if Ralph believes that Hesperus is Hesperus is true, then Ralph believes that Hesperus is Phosphorus must also be true. (ii) Let t be any term, t’ and t* be any pair of distinct terms, and let C be a context (and A be an assignment of values to variables) such that the Russellian content of t with respect to C (and A) is the same as the Russellian content of both t’ and t* with respect to C (and A). H Ralph believes that … t’ … t* … is true with respect to C (and A), then Ralph believes that … t … t… must also be true with respect to C (and A). For example, if Ralph believes that Hesperus is not Phosphorus is true, then Ralph believes that Hesperus is not Hesperus must also be true.

The disparity between (i) and (ii) on Richard's approach results from the fact that correlation functions take expression types as arguments. Hence multiple occurrences of a single term t have to be provided with identical counterparts. This means that Ralph believes that … t … t… will be true only if Ralph accepts some corresponding sentence … t# … t# … which contains multiple occurrences of some term with the same Russellian content as t. By contrast, Ralph's acceptance of such a sentence is sufficient, but not necessary, for the truth of Ralph believes that … t’ … t* … , in a context in which there are no constraints on correlation functions (since in such a context the distinct terms t’ and t* can be mapped onto any term or terms with the right Russellian content.)

Like other complications, this special feature of Richard's approach is independent of the issues raised in this paper, and so may be ignored.

13 There are now several ‘linguistic’ accounts of belief ascriptions in the literature which analyze such ascriptions as reporting relations to objects that include the expressions of the complement clauses of such ascriptions. Typically, however, such accounts have nothing to say about the belief relation, and so end up making no predictions whatsoever about the logico-semantic relationships between any pair of belief ascriptions whose complement clauses differ in any way.

14 Subject to the qualification in note 12.

15 See Propositional Attitudes, 135-6.

16 A different technical modification that would accomplish the same result would be to allow the general intentions of speaker-hearers in the context to determine world-indexed restrictions on whoever is the most advanced Babylonian astronomer in an arbitrary world. On this conception, restrictions are quadruples rather than triples, or perhaps functions from worlds to original Richardian triples. The general intention of speaker-hearers to use the names Hesperus and Phosphorus to represent their Babylonian translations when ascribing attitudes to the most advanced Babylonian astronomer, whoever he may have been, can then be seen as determining restrictions, (i, ii), which are functions that assign to an arbitrary world a triple whose first constituent is the individual who was the most advanced Babylonian astronomer in that world, and whose second and third constituents are as before.

(i). the actual world (world of the context) → Hammurabi, <'Hesperus,’ Venus>, ﹛< the Babylonian translation of ‘Hesperus,’ Venus> ﹜> world w → <Sam, <'Hesperus,’ Venus>, ﹛ <the Babylonian translation of 'Hesperus,’ Venus> ﹜>

.

.

.

(ii) the actual world → <Hammurabi, <'Phosphorus,’ Venus>, ﹛< the Babylonian translation of ‘Phosphorus,’ Venus> ﹜> world w → <Sam <'Phosphorus,’ Venus>, ﹛ < the Babylonian translation of 'Phosphorus,’ Venus> ﹜ >

.

.

.

With this expanded conception of contextually determined restrictions, examples like (32) can be given a unified treatment that avoids the earlier difficulty.

This conception of world-indexed restrictions was the one I originally presented at the APA symposium in March of 1994. The conception of generalized restrictions given in the text (with properties as the initial constituents of restrictions) was formulated by Richard in his response to the difficulty with (32). For present purposes there are no significant empirical differences between these two technical proposals, and I am happy to accept Richard's formulation as simple, natural, and easy to grasp.

17 This was Richard's proposal in his response to the version of my paper that was presented at the 1994 Pacific Division Meeting of the APA.

18 Richard emphasized this point in his response to my paper at the APA session.

19 For example, consider the following discourse. I have heard it said that the wealthiest members of most professions are among the tallest members of those professions. Now that I think about it there seems to be something to that. The wealthiest F is tall, the wealthiest G is tall, and the wealthiest H is tall. Here, what the speaker means is that the wealthiest F is tall for an F, the wealthiest G is tall for a G, and so on.

20 Although in his AP A response Richard emphasized the (alleged) semantic parallel between ‘tall’ and ‘believe,’ he apparently did not notice this crucial aspect of that parallel. In connection with this parallel, I will introduce a technical distinction between two conceptions of context. The first conception is that of a context of utterance for an English sentence S. These are circumstances in which an agent uses (utters, inscribes, etc.) the sentence S, with specific intentions about how the occurrences of the words in S are to be understood. The second conception is essentially that of Kaplan, David ‘Demonstratives,’ and ‘Afterthoughts,’ in Almog, J. Perry, J. and Wettstein, H. eds., Themes from Kaplan (Oxford: Oxford University Press 1989).Google Scholar This conception of context is the one needed for a (nontrivial) logic of indexicals. It is one in which sentences are evaluated in contexts, independently of whether the agent in the context is using them (or any other expression.) On this conception, contexts assign expressions (not utterances or even individual occurrences of expressions) contents.

For example, in a given context of utterance an agent may utter the sentences The wealthiest F is tall and The wealthiest G is tall with the intention that the two occurrences of tall be provided with different reference classes. In representing this example in a form suitable for evaluation in a system of logic for indexicals, the first sentence would be represented by one ‘logical form'- Tall (the wealthiest F, r1)while the second sentence would be represented by Tall (the wealthiest G, r2). Contexts provided by the logic (formal semantics) would specify referents for the indexical terms r1 and r2 - thereby securing the reference classes for occurrences of the predicate. If believe parallels tall, then a similar analysis should hold for it.

21 There are different technical approaches that could be taken to deal with the problems noted above in sections III and IV, but none that I know of seems very natural or promising. One thought would be to retain the fundamentals of Richard's original system while treating mutual, or potentially conflicting, restrictions on the same agent in a special way. For example, one might let conflicting restrictions cancel out, leaving no restrictions on the agent in the context. A different possibility would be always to merge multiple restrictions on the same individual in a given context, so that different intentions that turned out to correspond to the same individual i and term t would not result in different (original) Richardian restrictions

(ia) < i, < t, o >, S>

(ib) < i, < t, o >, S*>

but rather would result in a single restriction

(ic) < i, < t, o >, S u s* >

in which the set of translation targets onto which t can be mapped is simply the union of those specified by (ia) and (ib ). Under either of these suggestions conflicts between restrictions would be eliminated, thereby allowing some ascriptions of attitudes to the agent in the context to be true.

Despite this virtue, these proposed modifications are substantially flawed. One complication to be noted right away is that both modifications ignore the modal aspect of the difficulty with Richard's (original) restrictions. As I have pointed out, whether speakers’ intentions give rise to multiple restrictions on the same individual (which might then be subject to cancellation or merging) is not something to be settled once and for all in the world of the context. Restrictions that place compatible constraints on ascriptions to different individuals in the world of the context may place incompatible constraints on ascriptions to the same individual in other worlds. Thus, if cancellation or merging is to take place, it must do so on a world by world basis. Although this is certainly not incoherent, it seems ad hoc and cumbersome.

More important, both potential modifications give rise to counterintuitive truth conditions of just the sort that Richard's system is designed to avoid. The strategy of canceling conflicting restrictions on a given agent (at a given world) gives rise to essentially Russellian truth conditions for all ascriptions involving that agent (at that world). This cannot be a happy prospect for a critic of the Russellian account. The consequences of merging multiple restrictions on a single agent (at a given world) are little better. This can be seen by considering the limiting case of this strategy, in which one of the two or more contextual restrictions applying to a given agent, and a particular expression, allows us to map that expression onto anything with the same Russellian content. This will have the effect of assigning Russellian truth conditions to certain ascriptions in the context that the speaker intends to be taken in a non-Russellian way.

The effect of this is to create within the modified system a Russell-like problem for every apparently non-Russellian example that motivated Richard in the first place. To see this, consider a context C, and an ascription,

(ii) α believes that … t … ,

where the speaker of C resists substitutions for t, and intends to convey the information that a accepts the very sentence ....t .... the speaker used to report the belief. These are cases in which the assignment of Russellian truth conditions are found to be intuitively problematic, and for which the Russellian must resort to a pragmatic treatment. Richard's central goal is to avoid this.

The difficulty, under the modification we are considering, is that the Russellian problem avoided in the context C will break out again in a more complicated context C’ in which not only the ascription (ii), but also the ascription

(iii) β believes that … t …

is under consideration. Let us suppose, first, that ? and β denote the same individual (at a given world), second, that the speaker intends (ii) to be understood in just the non-Russellian way it is understood in C, while (iii) is to be understood in a Russellian manner, and, third, that restrictions concerning the same individual are to be merged (at the given world). With these suppositions in place, the problem for the Russellian posed by example (ii) in context C will arise in the modified version of Richard's system when (ii) is placed in context C.’ In other words, accepting the proposal to merge restrictions involving the same individual relocates, without solving, the original problems of Russellianism.

In addition, the proposal to merge restrictions generates a class of counterintuitive predictions that fall on a continuum between the liberal semantic substitutivity sanctioned by Russellianism and the strict restrictions on such substitutivity originally motivating Richard. (In these cases, the union of translation targets does not do justice to any of the speaker's multiple intentions.) For these reasons, the proposal to merge multiple restrictions involving the same individual does not, in my view, provide a satisfactory solution to the problems raised in the text.

22 The terms α and β in (42) may either be names, indexicals, or singular definite descriptions. In light of the difficulty just noted with (40), the best illustrations of the point I wish to make about (42) are those in which at least one of the terms is a description.

23 The reason this is objectionable is not that predicates never function in this way. Certainly, it is arguable that in the case of ‘tall’ the argument form (i) really is not truth preserving in all contexts of utterance.

(i) α is tall

α = β

Therefore, β is tall

For, if different occurrences of ‘tall’ really can receive different contextually specified reference classes in the same context of utterance, then it should, in principle, be possible (even if it is uncommon in practice) for there to be contexts of utterance in which in which (i) has the force of (ii), where the standards of being tall for a G are more stringent than those for being tall for an F.

(ii) α is tall (for an F)

α = β

Therefore, β is tall (for a G)

In such contexts of utterance the premises may be true and the conclusion false. (Note, in the abstract sense of a context required by a logic for demonstratives there will never be a context in which the logical forms- Tall (α., r1) and α = β are true in the context while the conclusion Tall (β, r1) is false. However these logical forms do not represent (ii). See note 20.)

In light of this one might wonder what is so bad about a semantic account of the belief predicate that characterizes (42) as failing to preserve truth in some contexts of utterance. The answer is just that we don't ordinarily think of the belief predicate in this way.