Propositional Attitudes: Issues in Semantics
PROPOSITIONAL ATTITUDES: ISSUES IN SEMANTICS
Propositional attitudes like knowledge, belief, and assertion play an important foundational role for semantic theory, the goal of which is to specify the meanings of sentences and their semantic contents relative to contexts of utterance. Meanings are plausibly regarded as functions from such contexts to semantic contents, which in turn are closely related to the assertions made, and the beliefs expressed, by utterances. For example, the semantic content of I live in New Jersey in a context C with x as agent and t as time is standardly taken to be the proposition that x lives in New Jersey at t. To understand the meaning of this sentence is, to a first approximation, to know that a competent speaker x who sincerely and assertively utters it in C asserts, and expresses a belief in, this proposition. Roughly put, if p is the semantic content of S in C, then an assertive utterance of S in C is an assertion of p, and is standardly taken as indicating the speaker's belief in p. Whether the semantic content of a sentence is always among the propositions asserted by an utterance of the sentence, and whether, in those cases in which it is, the assertion of any other proposition by the utterance is always parasitic on the assertion of the semantic content, are matters of detail. Though important, they do not affect the foundational point. A semantic theory for a language is part of a larger theory that interprets the assertions and beliefs of its speakers. This, more than any other fact, allows one to subject semantic theories to empirical test. Competent speakers of a language are relatively good at identifying the propositions asserted and beliefs expressed by utterances. To the extent to which assignments of semantic content issued by a semantic theory lead to verifiably correct characterizations of speakers' assertions and beliefs, the semantic theory is confirmed; to the extent to which these assignments lead to verifiably incorrect characterizations, it is disconfirmed.
This point is closely related to the use of attitude ascriptions
(1a) N asserted that S
(1b) N believed that S
to test different semantic analyses of S. It is convenient to express this in terms of the relational nature of the attitudes. Consider assertion. In each case of assertion there is someone, the agent, who does the asserting, and something, the object of assertion, that is asserted. The term proposition is used to designate things that are objects of assertion (and other propositional attitudes) and bearers of truth value. Assertion is a mediated relation holding between agents and propositions. An agent asserts a proposition p by doing something or employing some content-bearing representation associated with p. The most familiar cases are those in which the agent asserts a proposition by assertively uttering a sentence.
Ascriptions like those in (2) report the assertions of agents:
(2a) Edward asserted the proposition that Martha denied.
(2b) Edward asserted the proposition that the Earth is round.
(2c) Edward asserted that the Earth is round.
That in (2a) asserted is flanked by two noun phrases suggests that it is a two-place predicate and that a sentence ⌜NP assert NP⌝ is true if and only if the first (subject) noun phrase designates an agent who bears the assertion relation to the entity designated by the second (direct object) noun phrase. This analysis also applies to (2b), which is true if and only if Edward asserted the proposition designated by the proposition that the Earth is round. On the assumption that this proposition is also designated by that the Earth is round, this analysis can be extended to (2c), which is equivalent to (2b). Similar remarks hold for other propositional attitude verbs, including believe, deny, refute, and prove.
With this in mind, one can return to the ascriptions in (1). If, as many theorists believe, (i) ⌜that S⌝ in (1) designates the semantic content of S (in the context), (ii) these ascriptions report relations between agents and those contents, and (iii) sometimes substitution of sentences with necessarily equivalent semantic contents fails to preserve the truth values of such ascriptions, then semantic contents must be more fine-grained than the sets of possible world-states in which they are true. On these assumptions substitution in such ascriptions can be used to discriminate different but intensionally equivalent semantic analyses of S.
Propositions, Possible World-States, and Truth Supporting Circumstances
This has significance for possible world semantics. In this framework a semantic theory is a formal specification of truth with respect to a possible context of utterance and circumstance of evaluation. The semantic content of S in C is the set of possible circumstances E such that S is true with respect to C and E. Circumstances of evaluation are traditionally identified with possible world-states—thought of as maximally complete properties that the world genuinely could have had. As a result, the semantic contents of all necessarily equivalent sentences are taken to be identical. This, plus the standard treatment of attitude ascriptions as reporting relations between agents and the semantic contents of their complement clauses, leads to the counterintuitive prediction that substitution of necessarily equivalent sentences in such ascriptions never changes truth value. If one adds the apparently obvious fact that (3a) entails (3b),
(3a) A asserts/believes that P&Q
(3b) A asserts/believes that P&A asserts/believes that Q
one gets the further counterintuitive results (i) that anyone who asserts or believes a proposition p asserts or believes all necessary consequences of p, and (ii) that no one ever asserts or believes anything necessarily false, since to do so would involve simultaneously asserting or believing every proposition.
In 1983 Jon Barwise and John Perry attempted to evade these results by constructing a semantic theory in which metaphysically possible world-states were replaced by abstract situations—thought of as properties that need be neither maximally complete, nor genuinely capable of being instantiated by any parts of the world. This strategy was shown to be unsuccessful by Scott Soames (1987), where it was demonstrated that variants of the problems posed by attitude ascriptions for standard possible worlds semantics can be re-created for any choice of truth-supporting circumstances used in formal characterizations of truth with respect to a context and a circumstance. Robert C. Stalnaker explored a different approach in 1984. After providing a naturalistic argument that semantic contents must be sets of metaphysically possible world-states, he suggested that counterexamples could be avoided by (i) allowing for exceptional cases in which attitude ascriptions report relations to propositions other than those expressed by their complement clauses, and (ii) resisting the claim that the agent believes the conjunction of p and q in many cases in which the agent believes both conjuncts. These suggestions are rebutted in later work by Mark Richard (1990) and Jeffrey Speaks (forthcoming).
Structured Propositions, Millianism, and Descriptivism
The problems posed by attitude ascriptions for possible worlds semantics have led many theorists to characterize the semantic content of a sentence S as a structured complex the constituents of which are the semantic contents of the semantically significant constituents of S. In essence this was also the classical position of Gottlob Frege (1892/1948) and Bertrand Russell (1905, 1910). A variant of this position, growing out of the possible worlds framework, was championed by Rudolf Carnap (1947). For Carnap, two formulas are intensionally isomorphic if and only if they are constructed in syntactically the same way from constituents with the same intensions (functions from world-states to extensions). In effect, semantic contents of syntactically simple expressions are identified with intensions, while semantic contents of syntactically complex expressions are structured complexes the constituents of which are the semantic contents of their grammatically significant parts. This view was criticized by Alonzo Church (1954), who argued that semantically complex, but syntactically simple, expressions require a stronger notion of synonymy than sameness of intension. Church's modification of Carnap—which relies on rules of sense to induce a notion of synonymous isomorphism—is a variant of the classical Fregean position.
In the late 1980s the assignment of structured semantic contents to sentences was given a neo-Russellian twist by David Kaplan (1986, 1989), Nathan Salmon (1986), and Soames (1987). On the Russellian picture structured propositions are recursively assigned to formulas, relative to contexts and assignments of values to variables. The semantic content of a variable v relative to an assignment f is just f(v), and the semantic content of a closed (directly referential) term relative to a context C is its referent relative to C. Semantic contents of n-place predicates are n-place properties and relations. The contents of truth-functional operators may be taken to be truth functions, while the semantic content of a formula ⌜λx [Fx]⌝ is identified with a propositional function g that assigns to any object o the structured proposition expressed by ⌜Fx⌝ relative to an assignment of o to "x." ⌜∃x [Fx]⌝ expresses the structured proposition in which the property of assigning a true proposition to at least one object is predicated of g. In this framework the attitude ascriptions (1a and 1b) express structured semantic contents in which the relation of asserting or believing is predicated of a pair consisting of an agent and the structured proposition semantically expressed by S. The semantic theory is completed by specifying the intensions determined by structured semantic contents, including the truth conditions of structured propositions in all possible world-states.
The signature commitment of this approach is to the possibility of asserting and believing singular propositions—which include as constituents the very objects they are about. On this approach to believe de re of an object that it is F is to believe the singular proposition about that object, which says that it is F. Sentences like (4), involving quantifying-in, are quintessential examples of de re belief ascriptions.
(4) There is a planet x such that when the ancients saw x in the morning they believed that x was visible only in the morning and when they saw x in the evening they believed that x was visible only in the evening.
If, as Kaplan (1989) contends, the semantic contents of sentences containing indexicals are also singular propositions, then belief ascriptions containing indexicals in their complement clauses are also de re and hence share the basic semantic properties of ascriptions like (4). Salmon (1986) and Soames (2002) take this a step further, arguing for the Millian view that the semantic content of an ordinary proper name is simply its referent. One potentially problematic consequence of this view is that since Ruth Barcan and Ruth Marcus are coreferential, (5a) is characterized as semantically expressing the same proposition as (5b) and hence as having the same semantically determined truth value, even though it seems evident to many that it is possible to believe that Ruth Barcan was a modal logician without believing the Ruth Marcus was:
(5a) John believes that Ruth Barcan was a modal logician.
(5b) John believes that Ruth Marcus was a modal logician.
Different Millians respond to this problem in different ways. Salmon (1986) and David Braun (2002) argue that the intuitions that (5a) and (5b) can differ in truth value are mistaken because speakers tend to confuse the identical beliefs ascribed to John by these ascriptions with the different manners of holding these beliefs associated with their different sentential complements. Soames (2002, 2005a) argues that even though the semantic contents of these sentences are the same, assertive utterances of them may indeed result in assertions of propositions with different truth values. In "A Puzzle about Belief" (1979) Saul Kripke takes a different tack. While neither advocating nor denying the Millian view, he argues that substitutivity problems of the sort illustrated here are independent of Millianism and indicate a breakdown of the basic principles underlying our belief-reporting practices.
By contrast, descriptivists, following in Frege's footsteps, have wanted to assign different semantic contents to the two names and hence to the complement clauses in (5a and 5b). The problem has been to find a way of doing this that does not run afoul of Kripke's refutation of descriptivism in Naming and Necessity (1972). One of Kripke's arguments holds that since names are rigid designators, their semantic contents cannot be given by any nonrigid descriptions. This argument is not easily avoided by rigidifying candidate descriptions. As shown by Soames (2002), an analysis that takes the semantic content of Aristotle to be given by ⌜the actual F⌝ will, all other things being equal, identify the semantic content of Aristotle was a philosopher with the singular proposition (about the actual world-state @) that the unique individual who "was F" in @ was also a philosopher. Assuming that the analysis also includes the standard relational treatment of belief ascriptions, one then gets the result that for any possible agent a and world-state w, ⌜x believes that Aristotle was a philosopher⌝ will be true of a with respect to w only if in w a believes that the unique individual who "was F" in @ [not w] was also a philosopher. Since this is obviously incorrect, names can neither be nonrigid descriptions, nor descriptions rigidified using the actuality operator.
What about descriptions rigidified using Kaplan's dthat operator? Even if, contra Kripke, a correct reference-fixing description ⌜the x: Dx⌝ could be found for each name, the semantic content of ⌜dthat [the x: Dx]⌝ would simply be its referent, in which case the descriptivist would be saddled with precisely the Millian predictions about attitude ascriptions that the theory was designed to avoid. One possible response, suggested by David Chalmers (2002), is, in effect, to take a belief ascription ⌜a believes that S⌝ to report that the belief relation holds between the agent and pair consisting of the semantic content of S (in the context) and the meaning (function from contexts to such contents) of S. However, now a different problem arises. To avoid Kripke's nonmodal arguments against familiar candidates for reference-fixing descriptions, post-Kripkean descriptivists have had to resort to egocentric, metalinguistic descriptions of the sort the individual I have heard of under the name "n." Although this move assigns different objects of belief to the complement clauses of (5a) and (5b), it does not solve the problem. The point, after all, is not simply to assign different belief objects in these cases, but to explain the different information one gathers about John from utterances of (5a) and (5b). As Soames (2005b) argues in Reference and Description: The Case against Two-Dimensionalism (2005), it is hard to see how these egocentric, metalinguistic descriptions could, realistically, contribute to this.
Davidson's Linguistic View
A different approach to problems involving substitutivity is to take ascriptions ⌜x says/asserts/believes that S⌝ as reporting relations either to S itself, or to a complex in which S is paired with its semantic content. Either way, since substitution of one expression for another in S always produces a new complement S′, attitude ascriptions that differ in this way always report relations to different objects, whether or not the semantic contents of S and S′ are the same. This encourages the thought that such ascriptions can always differ in truth value.
An early and influential version of this approach was developed by Donald Davidson (1968–1969), who argued that (6a) should be understood on the model of (6b), in which that is treated as a demonstrative, utterances of which refer to utterances of the independent sentence that follows it:
(6a) Galileo said that the Earth moves.
(6b) Galileo said that. The Earth moves.
On this analysis what is said by an assertive utterance u of (6a) is that one of Galileo's utterances stands in, as Davidson puts it, the samesaying relation to the subutterance u* of the Earth moves. Although this analysis promised a simple way of capturing the logic of attitude ascriptions, it foundered on certain recalcitrant facts, including the fact that some ascriptions, like Every mother said that her son was lovable, cannot be broken up into separate and independent sentences in the manner of (6b) and the fact that the assertion made by an utterance of (6a) could have been true even if the subutterance u* had never existed, indicating that the Davidsonian truth conditions are incorrect.
Later Linguistic and Representationalist Views
Beginning in the 1990s improvements of Davidson's idea, including, most notably, that of Richard Larson and Peter Ludlow (1993), avoid these difficulties by dispensing with utterances and by treating attitude ascriptions as reporting relations between agents and the interpreted logical forms of their sentential complements. These are abstract, syntax-encoding structures that contain both the expressions occurring in sentences and their referential contents. Abstracting, one has here a version of the structured propositions approach in which linguistic expressions are included in the propositions sentences express. Although this version has potential virtues, it shares a crucial problem with Davidson's original analysis. Just as Davidson's silence about the intension of the samesaying relation prevented his theory from making any predictions about when (if ever) substitution of coreferential names or indexicals in a says that ascription changes truth value, so Larson and Ludlow's silence about the intension of the belief relation, alleged to hold between agents and interpreted logical forms, prevents their theory from making any predications about similar substitution in belief ascriptions (see Soames 2002). Since some such substitution clearly does preserve truth value, the problem is a daunting one.
Arguably, the most sophisticated approach of this general type is Richard's (1990), which combines context-sensitivity with linguistically augmented, structured Russellian propositions. For Richard, a belief ascription ⌜x believes that S⌝, used in a context C, is true of an agent a if a accepts some sentence S′ with the same Russellian content in a's context as S has in C, while being similar enough to S to satisfy the belief-reporting standards in C. As indicated by Soames (2002), the evaluation of this view crucially depends on identifying similarity standards present in contexts and assessing their impact. Although there are certain evident problems here, opinions of their import vary. Finally, a different sort of context-sensitive view, advocated by Mark Crimmins and John Perry (1989), takes belief ascriptions to report that an agent believes a structured, Russellian proposition by virtue of having ideas of a certain sort—where these are mental particulars in the mind of the agent that are either implicitly demonstrated, or implicitly characterized as being of a certain type, by the one uttering the ascription. This view is usefully criticized by Jennifer Saul (1993).
Extension: Intensional "Transitive" Verbs
Example (2a), in which assert occurs as an ordinary transitive verb operating on the extensions of its noun-phrase arguments, shows that not all attitude ascriptions contain sentential clauses. The examples in (7) show that there are also verbs, the grammatical objects of which are not overtly clausal, which are intensional in nature:
(7a) John wants a perpetual motion machine.
(7b) John is looking for the fountain of youth.
(7c) John imagined a room full of unicorns.
(7d) John worships many gods.
The relationship between these examples and ordinary propositional attitude ascriptions is a matter of ongoing investigation. How is it that (7a to 7d) can be true even though there are apparently no real entities described by their postverbal arguments? Are some or all these sentences covertly clausal? For example, are (7a) and (7b) to be assimilated to (8a) and (8b)?
(8a) John wants it to be the case that he has a perpetual motion machine.
(8b) John is trying to bring it about that he finds the fountain of youth.
These and related questions have been discussed by philosophical logicians and linguistic semanticists including Richard Montague (1974), Graeme Forbes (2000), Richard (1998), and Marcel den Dikken, Larson, and Ludlow (1997).
See also Intensional Transitive Verbs.
Barwise, John, and John Perry. Situations and Attitudes. Cambridge, MA: MIT Press, 1983.
Braun, David. "Cognitive Significance, Attitude Ascriptions, and Ways of Believing Propositions." Philosophical Studies 108 (2002): 65–81.
Carnap, Rudolf. Meaning and Necessity. Chicago: University of Chicago Press, 1947.
Church, Alonzo. "Intensional Isomorphism and Identity of Belief." Philosophical Studies 5 (1954): 65–73.
Crimmins, Mark, and John Perry. "The Prince and the Phone Booth: Reporting Puzzling Beliefs." Journal of Philosophy 86 (1989): 685–711.
Davidson, Donald. "On Saying That." Synthese 19 (1968–1969): 130–146.
Dikken, Marcel den, Richard Larson, and Peter Ludlow. "Intensional 'Transitive' Verbs and Concealed Complement Clauses." In Readings in the Philosophy of Language, edited by Peter Ludlow, 1041–1053. Cambridge, MA: MIT Press, 1997.
Forbes, Graeme. "Objectual Attitudes." Linguistics and Philosophy 23 (2000): 141–183.
Frege, Gottlob. "On Sense and Reference." Zeitschrift fur Philosophie und philosophische Kritik 100 (1892): 25–50; translated by Max Black. Philosophical Review 57 (1948): 207–230.
Kripke, Saul. Naming and Necessity. Cambridge, MA: Harvard University Press, 1980; originally published in Semantics of Natural Language, edited by Donald Davidson and Gilbert Harman. Dordrecht, Netherlands: Reidel, 1972.
Kripke, Saul. "A Puzzle about Belief." In Meaning and Use: Papers Presented at the Second Jerusalem Philosophical Encounter, April 1976, edited by Avishai Margalit. Dordrecht, Netherlands: Reidel, 1979.
Larson, Richard, and Peter Ludlow. "Interpreted Logical Forms." Synthese 95 (1993): 305–356.
Richard, Mark. "Commitment." Philosophical Perspectives 12 (1998): 255–281.
Richard, Mark. Propositional Attitudes: An Essay on Thoughts and How We Ascribe Them. Cambridge, U.K.: Cambridge University Press, 1990.
Russell, Bertrand. "Knowledge by Acquaintance and Knowledge by Description." Proceedings of the Aristotelian Society 11 (1910): 108–128.
Russell, Bertrand. "On Denoting." Mind 14 (1905): 479–493.
Salmon, Nathan. Frege's Puzzle. Cambridge, MA: MIT Press, 1986.
Saul, Jennifer. "Still an Attitude Problem." Linguistics and Philosophy 16 (1993): 423–435.
Soames, Scott. Beyond Rigidity: The Unfinished Semantic Agenda of Naming and Necessity. New York: Oxford University Press, 2002.
Soames, Scott. "Direct Reference, Propositional Attitudes, and Semantic Content." Philosophical Topics 15 (1) (1987): 47–87.
Soames, Scott. "Naming and Asserting." In Semantics vs. Pragmatics, edited by Zoltán Gendler Szabó. New York: Oxford University Press, 2005a.
Soames, Scott. Reference and Description: The Case against Two-Dimensionalism. Princeton, NJ: Princeton University Press, 2005b.
Speaks, Jeffrey. "The Priority of Belief: Stalnaker on Intentionality." "Is Mental Content Prior to Linguistic Meaning?" Nous. Forthcoming.
Stalnaker, Robert C. Inquiry. Cambridge, MA: MIT Press, 1984.
Scott Soames (2005)