Even for sentential guises, identity and distinctness are not guaranteed to be transparent to speakers: someone may be confused as to whether “Paderewski,” the name of the politician, is the same name as “Paderewski,” the name of the pianist (Kripke 1979). A single speaker at a single time may associate different mental files with the same word of a natural language, or the same mental file with different words of the language. Speakers may also be confused as to whether they are calling on two mental files or one. What needs to be found is not the mythical level of description at which perfect transparency to the subject is guaranteed but rather a perspicuous level of description at which the relevant cognitive phenomena are individuated in a way that is neither so coarse-grained that the most relevant distinctions cannot be drawn nor so fine-grained that they are drowned out by a crowd of irrelevant ones. Since philosophical debates involve many interacting individuals, sentential guises usually provide an appropriate level of description.
We also need an epistemology for logical truths in the first place. To that, the notion of Frege-analyticity contributes nothing. In particular, that a sentence is Frege-analytic does not imply that mere linguistic competence provides any insight into its truth, or constitutes more than the minimal starting-point for inquiry it does for ordinary synthetic truths.
How many philosophical truths are Frege-analytic? As a simple example, take the true sentence “Persons are not events” (if you think that persons are events, take “Persons are events” instead). It is not itself a logical truth, on any standard conception of logic. In particular, “person” and “event” seem not to be logical constants, and the logical form “Ps are not Es” has false instances such as “Parisians are not Europeans.” What logical truth could “Persons are not events” be synonymous with? “Persons who are not events are not events” is a logical truth, but not synonymous with the original. Granted, “persons” and “persons who are not events” have the same intension (function from circumstances of evaluation to extension) in every context of utterance.20 Still, they are not literally synonymous, for whatever the semantic structure of “persons,” it is finite, and therefore a proper part of the semantic structure of “persons who are not events”; thus the two expressions differ in semantic structure. One can try to construct non-circular analyses of “person” and “event” or both whose substitution into the sentence would yield a logical truth: “To be a person is to be a QRS.” However, “person” and “QRS” are unlikely to be literally synonymous. Almost certainly, someone will produce a purported counterexample to the analysis: “Such-and-such would be a person but not a QRS” or “So-and-so would be a QRS but not a person.” Direct reference theorists will tend to expect just such counterexamples to the claim that the apparently simple term “person” and the complex description “QRS” have the same intension; direct reference theories partly originate from Kripke and Putnam’s counterexamples to a host of similar descriptivist claims. Opponents of direct reference may be less pessimistic about the prospects for a complex description with the same intension as “person.” However, on their finer-grained views of meaning, on which synonymy is as transparent as possible to competent speakers, a purported counterexample need not be correct to defeat the claim of synonymy: what counts is that its proponent is neither linguistically incompetent nor fundamentally irrational. Contemporary proponents of a descriptivist view of meaning as a rival to direct reference theory usually envisage a loose semantic connection with a cluster of descriptions rather than strict synonymy with a single description. Whichever side of the debate one takes, there are good grounds for skepticism about the supposed synonymy of “person” and “QRS.” The best bet is that “Persons are not events” is not Frege-analytic. The point does not depend on peculiarities of the example; it could be made just as well for most other philosophical claims.21 In contemporary philosophy, few who propose complex analyses claim synonymy for them.22
One might react by loosening the relation of synonymy to some equivalence relation that would have a better chance of holding between the analysandum and the analysans in philosophically significant analyses. Call the looser equivalence relation “metaphysical equivalence.” A wider class of philosophical truths might be transformable into logical truths by the substitution of metaphysically equivalent terms. Call the truths in the wider class “quasi-Frege-analytic.” The poor track record of philosophical analysis does not suggest that the class of quasi-Frege-analytic truths will be very much wider than the class of Frege-analytic truths.23 In any case, the looser metaphysical equivalence is, the more problematic it will be to extend an epistemology for logical truths to an epistemology for quasi-Frege-analytic truths. The aim of the loosening is to permit some distance between the meaning of the analysandum and the meaning of the analysans; that will tend to make even the coextensiveness of the analysandum and analysans less cognitively accessible. There will be a corresponding tendency to make the material equivalence of the original quasi-Frege-analytic truth to the logical truth less cognitively accessible too.
For instance, one might define “metaphysical equivalence” as sameness of intension in every context. The question is then how the sameness of intension in every context of the substituted terms could enable one to advance from knowing or justifiably believing the logical truth to knowing or justifiably believing the merely quasi-Frege-analytic truth. No guarantee has been provided that we can know or justifiably believe the universally quantified biconditional of the substituted terms. By hypothesis, that biconditional will in fact express a necessary truth in every context; the problem merely shifts to how such truths can be known, just as in the case of modal-analyticity. If that problem were already solved, there would be little to gain from appealing to quasi-Frege-analyticity in order to explain how core philosophical truths can be known.
Even if many philosophical truths are quasi-Frege-analytic, it does not follow that we can gain cognitive access to them simply on the basis of our logical and linguistic competence.
Yet another proposal is to consider as (metaphysically) analytic just the logical consequences of true (or good) semantic theories. It is presumably in the spirit of this proposal to interpret semantic theories not as stating straightforwardly contingent, a posteriori facts about how people use words but as somehow articulating the essential structure of semantically individuated languages; in this sense, the word “green” could not have meant anything but green in English. Even so, the definition does nothing to trace any special cognitive access that speakers have to semantic facts about their own language to any special metaphysical status enjoyed by those facts. It also counts every logical truth as analytic, since a logical truth is a logical consequence of anything, without illuminating any special cognitive access we may have to logical truths. Of course, if someone knows the relevant semantic truths about their own language and is logically proficient, then they are also in a position to know the analytic truths as so defined. But, on this definition, we do nothing to explain how the semantics and logic are known in the first place by saying that they are analytic. As in previous cases, the account of analyticity merely shifts the burden from explaining knowledge of analytic truths to explaining knowledge of some base class of necessary or logical or semantic or other truths. Once the analyticity card has been played to effect this shift of the explanatory burden, it cannot be played again to explain knowledge of the base truths, by saying that they are analytic, for they count as analytic simply because they belong to the relevant base class, and the question remains how we know truths in the base class.
5
Unless one is a skeptic about meaning or modality, one can define several notions of analyticity in semantic