The A priori Isn’t All That It Is Cracked Up To Be,
But It Is Something
David Henderson and Terry Horgan
Alvin Goldman’s contributions to contemporary epistemology are impressive—few epistemologists have provided others so many occasions for reflecting on the fundamental character of their discipline and its concepts. His work has informed the way epistemological questions have changed (and remained consistent) over the last two decades. We (the authors of this paper) can perhaps best suggest our indebtedness by noting that there is probably no paper on epistemology that either of us individually or jointly have produced that does not in its notes and references bear clear testimony to the influence of Professor Goldman’s arguments. The present paper is no exception (and this would be a particularly inapt place to break with our tradition of indebtedness). Professor Goldman has produced a series of discussions that we find particularly important for coming to terms with the venerable idea that there may be truths that can be known a priori (Goldman 1992a, 1992b, 1999). We do not altogether follow his lead, while he draws on the idea that a priori justification has something to do with innateness or processess, we prefer to accentuate the idea that a priori justification turns on a conceptually grounded truths and access via acquired conceptual competence (at least in many significant philosophical cases). Still, in developing our understanding we have been aided by much that Professor Goldman says regarding concepts, conceptual competence, and related psychological processes. The influences should become progressively clear, particularly in the later sections of this paper.
What would it take for there to be a priori knowledge or justification? We can begin by reflecting on a widely agreed on answer to this question—one that purports to identify something that would at least be adequate for a priori justification. The answer will then serve as one anchor for the present investigation, a bit of shared ground on which empiricists and rationalists can, and typically do, agree. Of course the empiricist and rationalist traditions fall out over the extent of what can be known a priori. Many empiricists allow that there are conceptually grounded truths, and that epistemic agents can come to have a distinctive kind of justification for believing such truths by virtue of drawing on their conceptual competence. Empiricists who recognize a place for a priori justification at all insist that conceptual-competence based reflective justification constitutes all the a priori justification there is. Rationalists agree that such conceptual-competence based justification provides a priori justification, but go on to insist that there are yet further a priori justification to be had. They insist that the truths that can be known by way of reflection are not limited to truths that are simply conceptually grounded. These parties thus dispute whether there are cases of a particular kind of a priori truth; they dispute whether there are synthetic ones. Still, they commonly are in agreement that there are a priori truths. More to the point, they agree that the following would make for a priori knowledge and justification: a claim is a priori if (a) it is conceptually grounded and (b) individual epistemic agents can be justified in holding it by virtue of their reflectively drawing on their conceptual competence. This may not be the only way in which a claim may qualify as a priori; it may not represent a necessary condition for a priori justification. But, it at least expresses a sufficient condition for a claim being a priori.
An account that provides for conceptual-competence-based reflective justification has good claim to providing an account of a priori justification, or at least of an important class of such justifications. Accordingly we can claim to provide an account of a priori justification, as we argue that there are claims that can be so justified—so there are a priori truths. However, we also argue for a particular understanding of the reflective justification, or at least of that reflective understanding that commonly features in philosophy. On our understanding, a priori justification itself must have an empirical dimension. We mark this somewhat paradoxical result by writing of “low-grade” a priori justification—as opposed to the “high-grade” a priori justification of epistemic lore.
One can think of the matter this way. Central to the concept of a priori justification is the idea of an epistemic justification by reflection—justification that can be obtained “from the agent’s armchair,” without “going out” and collecting empirical evidence regarding what the actual world is like. Supposedly such reflective justification is exhibited in familiar philosophical and mathematical cases and is distinct in kind from that empirical reasoning found in the sciences. (Although it is typically acknowledged that the relevant form of reflective justification does play a limited role in scientific contexts. Still, it is less prominent there.) Empiricists hold that reflective justification turns on agents drawing on their competent understanding of their own “ideas,” or “meanings,” or “concepts.” That is, empiricists hold that the reflective understanding that makes for a priori justification turns on competence with some manner of semantic entity, or some entity with a semantics. Rationalists hold that that is part of the correct story—but insist that there are other kinds of justificatory reflection—kinds that give justification for truths that are not conceptually grounded. Without prejudice to this rationalist claim, we hope to pick up on the empiricist suggestion and give an account that vindicates reflection drawing on conceptual competence as a distinct justificatory route—though not as distinct as has commonly been supposed.
So, we take as a starting point the idea that conceptually grounded reflection provides a priori justification:
(CGR) a claim is a priori if (a) it is conceptually grounded and (b) individual epistemic agents can be justified in holding it by virtue of their reflectively drawing on their conceptual competence
The picture is roughly this. All truths are true in part by virtue of the concepts there featured and their mode of composition. This dependency holds no less for claims like, (1) “There are brown dogs,” than it does for those like, (2) “Water is H2O,” or those like (3) “Water is stuff with the same microstructure as the stuff in our prominent samples (filling lakes and rivers, …),” and (4) “Capacitors store electric charge.” But there is this notable difference. Given the semantics of the concepts and their mode of composition (and thus given whatever it is about the world that made or makes for the semantics of those concepts) the latter three claims could not help but be true—they are necessary (true in all possible worlds). We say that such truths are conceptually grounded—and this much is a matter of their semantics, and not of epistemology. Thus, CGR says that a claim is a priori if it is so conceptually grounded and (without “leaving the armchair) agents can relate to it in a way that is appropriately sensitive to this fact.
The three conceptually grounded truths above are of very different epistemic characters. The differences turn on the extent to which the elements of the conceptual semantics (on which the truth of the claims depends) may be “internally accessible” to one by virtue of acquiring a competence with the relevant concepts. It is now a commonplace to note that “meanings are not in the head.” Commonly, the semantics for a concept is not determined wholly by internal states of one who counts as having acquired the concept. To use the common example, our concept of water refers to stuff with the same microstructure as the stuff in a set of historical and contemporary samples. Its reference is fixed by the historical interactions of a community (or set of related communities) with a stuff in the world. Nonetheless, the (externalistic) semantics for water guarantees that water is H2O. Because of this semantical guarantee, (2) counts as a conceptually grounded truth. But it is clearly not an a priori truth. One is generally required to learn certain things in acquiring a concept. The truths that may be counted among the internally accessible components of a concept’s semantics are those that have both of these features: they are conceptually grounded necessary truths, and they also are truths that one has to learn in order to acquire the concept. One need not have come to appreciate all that is conceptually grounded, but by virtue of having acquired the concept, one must have come to appreciate or be sensitive to some of what is conceptually grounded. One does not count as having acquired the concept of a capacitor if one does not appreciate that capacitors store electric charge. Thus, (4) is not just a conceptually grounded truth, it is also one that is grounded in internally accessible elements of semantics. As a result, one who has acquired the featured concepts should be able to reflectively appreciate that (4) is true, where this reflective appreciation draws on one’s conceptual competence with those concepts. Unlike (2), (4) is grounded in internally accessible elements of the semantics of its featured concepts, and thus is a priori.
Claim (3), “Water is stuff with the same microstructure as the stuff in our prominent samples (filling lakes and rivers, …),” is a delicate matter. It seems that a closely related claim is a priori: (3*) “Water is stuff of the same natural kind as the stuff in our prominent samples.” But, one may certainly doubt that it is a priori that being of the same natural kind (of substances) turns on microstructure (in any substantive sense of “microstructrue.” After all, the discovery that natural kinds for substances do turn on microstructural similarities and differences—as opposed to brute qualitative kinds—is a point that seems to have been empirically supported by the work of physicists such as Lavoisier, Dalton, and Maxwell. Also, it would seem that the prominent samples need not have been drawn from lakes and rivers; as folk on an arid world in which naturally occurring water is now subterraneous could yet share with us the concept of water. Perhaps it would be best to say that, while (3*) like (4) may well be conceptually grounded and a priori, (3) is strictly speaking like (2): viz., conceptually grounded in a way that involves an externalistic dimension, and hence is not a priori.
This then adds up to the suggestion for a route to a priori justification. When one has acquired a competence with the concepts featured in a claim, and when the claim itself expresses a truth that is grounded entirely in internally accessible elements of the semantics of the featured concepts, then a rather distinct route to justification would seem to be available: drawing on conceptual competence, the agent can come to appreciate that the claim must be true. Rather than consulting experiential evidence regarding features of the actual world, one comes to see that the semantics of the relevant concepts adds up to ensure that the world is as claimed—no matter what possible world is the actual world. Put simply (as we will see, perhaps somewhat deceptively simply, but not falsely) the individual agent “need only” access what are the internal elements of the semantics of the concepts that he or she has acquired in order to appreciate that the claim is true. Insofar as one need only draw on one’s conceptual competence, and does not have to look beyond that to determine what world is the actual world, this justificatory route is indeed distinctive.
Since the epistemic justification just entertained would not depend on experiential evidence about what the actual world is like, this form of justification is commonly held to be uncontaminated by the empirical. As we will soon see, while reflecting a kernel of truth, this claim to empirical uncontamination is often something of an overstatement. Here we will focus on an important special case of reflective investigation, for which a modified understanding is in order: viz., reflective inquiry, as practiced in philosophical contexts, into the kinds of concepts that figure importantly in philosophy. What we now hope to bring into focus is the character of philosophical reflection insofar as it conforms to the general understanding of a priori reflection advanced above—insofar as it involves reflection that draws on conceptual competence, and thus on the internally accessible elements of the semantics of concepts, and thereby leads one to appreciate that certain claims are necessarily true. We hold that such reflective justification of claims has rightly been understood as an epistemic breed apart—although just how apart is in need of discussion here—and that the long philosophical tradition that sees it as a priori justification is correct. However, we believe that this reflection also possesses an empirical dimension—and is not “high-grade” a priori in the sense envisioned in much philosophical lore. Accordingly, we term it “low-grade” a priori.
Philosophical reflection about philosophically important concepts, then, will serve as our principal example of low-grade a priori reflective inquiry. On the other hand, considerations to be adduced are potentially generalizable to various other putative cases of the a priori. We will leave this generalizability issue largely open here.
2. The Low-grade A priori
2.1: The easy model, an easy mistake
So, the mode of philosophical inquiry often called “conceptual analysis” has served as one prominent example of a priori reflection. Such reflection (or a large and important portion of it) can be understood as drawing on conceptual competence, and thus on internally accessible elements of the semantics of concepts featuring in the claims at issue. It is tempting to move freely from this understanding of reflection (which we believe to be essentially correct) to the idea that philosophically interesting results emerge “fully formed,” as direct deliverances of conceptual competence. We will call this the “easy model” of reflection. On the easy model, after “turning ideas over in one’s mind,” one “just sees” on the basis of one’s competence with the featured concepts that certain general claims are necessarily true. On this view, the philosophical claims somehow just “float into consciousness” with a certain kind of obviousness—as conceptually guaranteed. (The model is suggested by BonJour’s formulations in which general a priori results are “grasped” in a “direct” and “immediate” fashion.)
We term this account the “easy model” for two reasons. First, it makes a priori reflection a fairly straightforward, one-step, affair. Turn over some ideas in your mind and, provided you start with an adequate grasp of the concepts, the general results putatively stand out as immediately and obviously true. Insofar as one is then spared the toil of careful inferential moves of a sort we will soon discuss, the model makes reflection out to be rather easier than what it actually is.
Second, it is easy to simply fall into holding this model without careful examination: given that reflection draws on one’s conceptual competence with the internally accessible elements of conceptual semantics, should not what comes in this way be simply or directly accessible? No. We will argue that general claims can be grounded in elements of conceptual semantics, elements that are in an important respect internally accessible, without being direct deliverances of conceptual competence. This point is crucial; it and the supporting understanding of conceptual competence will be central to our account of low-grade a priori justification.
2.2: Three important observations, and the need for a two-stage model
If one is to have a sober understanding of what can be expected from reflection that draws upon conceptual competence, and if one is to have a considered view of just how reflection itself can deliver justified beliefs, one must not succumb to unexamined presumptions regarding its workings. We think that the following three conjectures are jointly more than reasonable.
First, a prominent way of drawing upon one’s conceptual competence, particularly in evidence in much well-received philosophical work, involves generation of judgments about relatively specific and concrete scenarios. For example, one begins:
[Suppose that] somewhere in the galaxy there is a planet that we shall call Twin Earth. Twin Earth is very much like Earth; in fact, people on Twin Earth even speak English. In fact, apart from the differences we shall specify in our science-fiction examples, the reader may suppose that Twin Earth is exactly like Earth. …
One of the peculiarities of Twin Earth is that the liquid called ‘water’ is not H2O but a different liquid whose chemical formula is very long and complicated. I shall abbreviate this chemical formula simply as XYZ. I shall suppose that XYZ is indistinguishable from water at normal temperatures and pressures. … (Putnam 1975a, 223).
In response to such a scenario, conceptual competence spawns in a direct way fittingly specific and concrete judgments:
On Twin Earth the word “water” means XYZ (Putnam 1975a, 223)
Our concept water refers to H2O; it does not refer to the XYZ on Twin Earth.
Water is H2O.
Perhaps it should not be surprising that human conceptual competence is particularly suited to the generation of responses to such concrete specific scenarios. After all, in everyday contexts, where conceptual competence serves us in a largely unnoticed way, it would seem to function largely in the direct and automatic generation of applications of concepts.
Second, conceptual competence is much less steady and reliable when called upon to directly generate conceptually grounded general truths. Perhaps conceptual competence can generate general truths with reasonable success when they turn on certain concepts of relatively straightforward sorts that do not commonly come to feature in philosophically interesting claims. For example, having acquired the concepts of capacitor and electrical current, perhaps the general claim that capacitors store electrical current comes fully formed when one pauses to wonder. But, we think that the sorry track record of philosophers grasping for conceptually grounded general truths (regarding freedom and determinism, or regarding causation, epistemic justification, or the character of mental states, for example) suggests that at least greater caution is needed.  Particularly with respect to certain concepts that have commonly been significant for philosophers, one’s conceptual competence may be much better at directly delivering presumably veridical judgments regarding specific applications of our concepts than it is at directly generating veridical judgments regarding generalities.
Third, by drawing upon what conceptual competence does provide, one can manage to justifiably believe certain conceptually grounded generalities that are not themselves the direct deliverances of that competence. Consider the justly famous result of Putnam’s (and Kripke’s 1972) reflections:
In all possible worlds, water is stuff with the same micro-structure as the stuff in the salient samples.
This conceptually grounded general truth is justified in terms of more particular judgments about the referent of ‘water’ in various concrete scenarios. We believe that such general results are rightly taken to be a priori. The central question becomes just how one can get to such truths.
The above three observations strongly indicate that, at least as it has application to philosophical cases, an adequate model of a priori reflection drawing on conceptual competence will need to recognize at least two stages: one in which reflection on specific concrete scenarios generates correspondingly particular judgments (the direct deliverances of conceptual competence, mentioned in the first observation), and another in which one reflectively draws on these particular judgments to support a judgment whose content is abstract and general (the need for this is evident in the second and third observations above).
2.3: An instructive parallel: investigations of syntax in theoretical linguistics
Lest traditional assumptions regarding the character of philosophical reflection have an undue impact on one’s thinking about the two stages, it may be helpful to consider a parallel two-stage epistemic process: the methodology typically employed by linguists in constructing and evaluating theories of natural-language syntax. The empirical data for syntactic theory includes certain judgments of competent language-users—in particular, judgments concerning the grammaticality or ungrammaticality of various sentence-like strings, and concerning grammatical ambiguity or nonambiguity of various sentences. Notably, such judgments are relevant simultaneously to psychological theories of human language processing, and also to linguistic theories about the syntax of language itself. Native speakers, after all, can be expected to have judgment dispositions about these matters that reflect a solid mastery of their own language (or their own regional dialect, at any rate). So, when native speakers are intersubjectively consistent and also uniformly confident about such syntactic judgments, then normally these judgments reflect the natives' syntactic competence, their mastery of the syntactic norms or syntactic structures underlying their language. And this psychological hypothesis, in turn, has a direct implication for linguistic theory--viz., that under an adequate theory of syntax for the natives' language (or dialect), those syntactic judgments will turn out mostly correct.
Although intersubjective consistency in grammaticality judgments is important, much of the crucial data for syntactic theorizing about one’s own language is available from the armchair, in the form of introspectively accessible, confidently held, first-person judgments about grammaticality. The linguist’s own grammaticality judgments already constitute a very rich, and evidentially very significant, body of empirical data vis-à-vis syntactic theory—because the default presumption in the first person case too is that such grammaticality judgments are normally the products of one’s own linguistic competence, and hence are mostly correct.
A two-stage structure is evident in linguistic inquiry into natural language syntax. When the linguist proposes certain abstract general syntactic principles, claiming that they are the rules of grammar for a given language, these proposals are not themselves the direct deliverances of the linguist’s cognitive mechanisms of grammatical competence. Rather, the proposed rules of syntax are theoretical hypotheses about the language. Thus, first, grammatical competence generates grammaticality judgments about specific sentence-like word-strings. Although these judgments presumably are generated in a manner consistent with the general rules of syntax—so that the competent speaker possesses an implicit mastery of those rules, whatever they are—grammatical competence does not generate explicit beliefs whose contents are the syntactic rules themselves. Then, second, the linguist’s own grammaticality judgments, with their default status of presumptive correctness, are empirical data vis-à-vis these proposed rules. The evidential connection between the data and the theoretical hypotheses—here, as elsewhere in science—is inference to the best explanation.
As is virtually always the case with inference to the best explanation, the evidential considerations involved, in proffering the proposed rules of syntax as the putatively best explanation of the data, operate via wide reflective equilibrium. Considerations of fit vis-à-vis wider theory, and with various kinds of facts less directly connected to the matter at hand, are potentially relevant. For example, some kinds of possible rules might fit less well than others with extant psychological understandings concerning human language-processing, or with extant theory in linguistics.
As this last point makes clear, the relevant wide-reflective-equilibrium methodology (WRE methodology) certainly need not be confined to armchair-obtainable data. Often a linguist can arrive at a reasonably well confirmed theoretical hypothesis from the armchair, but (i) the linguist’s background theoretical knowledge (e.g., knowledge of linguistic theory) may well play an important role in the overall evidential support the hypothesis possesses, and (ii) this evidential status certainly can be further strengthened epistemically, and/or weakened, by potential evidence obtainable beyond the armchair. Relevant considerations can include, for example, the grammaticality intuitions of other people. They can also include theories and results in cognitive science concerning natural-language processing, since ultimately an adequate formulation of the rules of syntax would have to mesh with a detailed cognitive-scientific account of how those rules get implicitly accommodated by human language-processing mechanisms.
Another important dimension of the WRE methodology employed in the nondeductive inference from the linguist’s introspective data to the proffered rules of syntax is the availability of plausible supplementary hypotheses for “explaining away” any recalcitrant data—in particular, recalcitrant grammaticality judgments, or judgment-tendencies, that do not conform with the proposed syntactic rules. Common examples are so-called center-embedded sentences, such as this one:
The man the cat the dog chased scratched sued for damages.
Such word-strings are difficult for humans to process, and are often judged ungrammatical. But, to the extent that cognitive science can provide independent reasons why such center-embedded strings are hard to process whether or not they are grammatical, ungrammaticality-judgments vis-à-vis such cases should not count heavily against certain proposed syntatic rules that count them as grammatical. Moreover, a recalcitrant grammaticality-judgment will be even less of a problem if it can be reversed or mitigated with the aid of a suitable paraphrase, such as this one: The man sued for damages who was scratched by the cat that the dog chased.
Thus, in producing their accounts of the grammar of a language, theoretical linguists begin with judgments about grammaticality that are defeasibly presumed to be the relatively direct results of the grammatical competence of informants. To make the matter vivid, a linguist might even produce a model of his or her own idiolect—and in so doing, the linguist would begin with his or her own judgments, from the armchair. This first stage provides the data for subsequent inferential moves. While the set of such grammaticality judgments at least contains a healthy proportion that are direct products of the linguist’s own grammatical competence, the full set of such judgments enters into subsequent (second-stage) linguistic theorizing in a somewhat different epistemic modality. The judgments become data for a theory that must provide an explanation of their status and their source. This theory will necessarily have dual dimensions—one regarding the rules of the grammar for the language, and the other involving psychological hypotheses regarding the extent and form of the informant’s (linguist’s) own grammatical competence. Accordingly, this second stage clearly involves an empirical abductive inference, rather than itself being the direct product of the linguist’s grammatical competence.
These points are worth emphasis. Consider what is needed to account for the data, the particular judgments produced in response to specific sentence-like word-strings. These judgments will conform to the grammar of the language to the (to be determined) extent that the rules of grammar are successfully tracked by the individual’s (the linguist’s) grammatical competence. A full psychological account of that competence would include an account of the grammar; to speak of such a competence is to speak of a capacity to conform to the rules. If no theoretical-linguistics account of the grammar of the language is forthcoming, then there is no place or purchase for an a psychological account of the individual’s competence—which it itself a capacity of conforming to that grammar (within reasonable limits). But, any actual competence is subject to interferences, distortions, and systematic errors. Various performance errors are to be expected of one who is competent. Actual competence is imperfect. Accordingly, any satisfactory accounting for sets of judgments will need to sort out both the grammar of the language and the character of the individual’s competence. Given the judgments the individual makes, any account of the grammar will have implication for which judgments are competent judgments and which are performance errors. One’s account of grammar thus makes demands on one’s understanding of competence and of the ways that human competence falls short of ideal competence. Thus, if certain intuitive grammaticality judgments conflict with a proposed account of grammar, and if these judgments cannot be plausibly explained away as performance errors by appeal to credible psychological hypotheses about the limits of human grammatical competence, then the grammatical account is itself problematized. Any theory of grammar ultimately calls for an accommodating psychological account of competence. Overall then, the linguist’s proposed grammar will be informed by background and broadly empirical understandings of the finite and fallible character of people’s actual (as opposed to ideal) grammatical competence; the account will then at least implicitly be committed to a rough understanding of this competence and its salient limitations; and this understanding will be rightly affected by, and answerable to, broadly empirical hypotheses about various human cognitive capacities and limitations. The results of such theorizing are clearly not the direct product of the linguist’s grammatical competence. Nonetheless, syntactic theorizing can be effectively pursued by reflection in the armchair, drawing on the individual’s own linguistic competence. But, as just witnessed, there are also strong grounds for thinking that the justification of these results is not devoid of an empirical element. As we have seen, the investigation cannot plausibly be undertaken without one eye on a set of psychological issues having to do with the investigator/informant’s competence. Psychological hypotheses about one’s conceptual competence and its limitations and linguistic theorizing about the rules of grammar must be pursued together—as parallel tracks—as mutually reinforcing strands.
2.4 The two-stage model of philosophical reflection
The observations of section 2.2 strongly suggest that the reflective investigation appropriate to many philosophically interesting and important issues is best understood as having a two-stage structure that parallels the linguist’s reflective investigations of grammar. As in the case of linguistic investigation, philosophical reflection draws on the scattered concrete direct deliverances of a cognitive competence—conceptual competence, the analogue of grammatical competence. It seeks to arrive at conceptually grounded necessary truths (CGNT’s), the analogues of the rules of grammar the linguist seeks to discover. Further, the relevant CGNT’s are not reflectively generated in a noninferential or direct way out of one’s conceptual competence; so the ensuing reflection will need to employ these deliverances as data for abductive inference. As we will soon see, this reflective procedure has both psychological and conceptual faces, like investigations of grammar, and it thus properly understood as having an empirical dimension.
Philosophical reflection has an empirical dimension importantly analogous to the empirical reasoning that the armchair linguist employs. There are certain armchair-accessible (empirical) facts, including in particular facts about one’s own intuitive judgments about how terms (e.g., ‘water’) and the concepts they express apply to various concrete scenarios, actual and hypothetical (e.g., Twin Earth). These judgments are defeasibly presumed to emanate fairly straightforwardly from one’s own conceptual/semantic competence—perhaps operating directly and non-empirically (when the resulting judgments are high-grade a priori), perhaps operating in combination with fairly non-tendentious, well justified, empirical beliefs (e.g., the belief that the stuff we call ‘water’ or earth is composed of H20 molecules, and the belief that H20 is a chemical natural kind). Given this presumption about the competence-based etiology of the concrete judgments—itself a broadly empirical assumption, notice—the occurrence of these various judgments constitutes empirical evidence in support of the philosophical hypothesis in question, a putative CGNT.
The abductive reasoning in support of the hypothesis will ultimately rest in part upon an available side-explanation concerning recalcitrant aspects of the data—for instance, why certain mistaken judgment-tendencies are present that go contrary to the philosophical hypothesis in question; and this explanation itself will often be an empirical (and partly psychological) hypothesis, whose credibility or lack thereof affects the overall wide-reflective-equilibrium credibility of the principal inference to the reflective claim that such-and-such is a CGNT. When discussing the case of theoretical grammar, we noticed that accounting for the data demanded a coordinated account of grammar and competence—where the latter included an understanding of various performance errors. Similarly, the reflective abductive move from the data (from the particular judgments) to an account of certain elements of the semantics for the relevant concepts, and to certain general CGNT’s, turns on mutually supporting understandings of both conceptual competence and certain corresponding performance errors. Thus, not only do one’s judgments qualify as data for an account of conceptual semantics or CGNTs by virtue of a background of broadly empirical understandings of one’s own conceptual competence, but also the eventual evidential status of particular subsets of that data, particular judgments, must be affected by understandings of one’s own conceptual competence and tendencies to performance error.
Ideology is the term we recommend for the kind of broadly empirical inquiry into the workings of concepts here described (cf. Horgan 1993, Graham and Horgan 1994, Henderson and Horgan 2000a, and in press (b)). Ideology is a multi-discipline enterprise, whose philosophical dimension is continuous with relevant work in disciplines like cognitive science, linguistics, and sociolinguistics. Insofar as ideology relies primarily on armchair-accessible data, it can be effectively practiced by philosophers. Philosophical thought expermiments, like Putnam’s Twin Earth case, really are experiments: they generate empirical data for ideological theorizing, much as a linguist’s intuitive grammaticality-judgments constitute data for syntactic theorizing.
Although the linguistics analogy is helpful and suggestive, the kinds of armchair-obtainable data that are pertinent to philosophical ideological inquiry appear to be fairly diverse (more so than in the linguistics case), with some kinds being more directly analogous to grammaticality judgments than others. The types of data that can figure in philosophical ideological reflection include the following:
1. Intuitive judgments about what is correct to say concerning various concrete scenarios, actual or hypothetical.
2. Facts about conflicting judgments or judgment-tendencies, concerning the correct use of certain concepts in various actual or hypothetical scenarios.
3. Facts about standardly employed warrant-criteria for the use of various concepts.
4. Facts about the key sociolinguistic purposes served by various terms and concepts.
5. General background knowledge, including untendentious scientific knowledge.
Facts of all these kinds can go into the hopper of wide reflective equilibrium whereby ideological claims are defended in philosophy. One makes a case for a certain ideological hypothesis—for instance, the contention that the meaning of natural-kind terms depends on the language-users’ environment—by arguing that it does a better job, all things considered, of accommodating the relevant data than do any competing ideological hypotheses. Such reasoning is broadly empirical: inference to the best explanation, in which empirical data of all the kinds 1-5 are potentially relevant. Although it can be conducted from the armchair and it is aimed at discovering CGNT’s, it is not high-grade a priori.
Again, a compelling reason for thinking of the results of such investigations as a priori is that one there draws on deliverances of one’s conceptual competence—deliverances that are at least presumptively a priori—in a reflective process that can be pursued “in the armchair,” to establish certain conceptually grounded necessary truths. Notably, the truth of such claims can be appreciated without looking to acquire data about whether the world is as there represented.
On the other hand, there are multiple reasons for recognizing that such investigation—and in particular, the second stage of ideological reflection from the armchair—is not itself devoid of empirical dimensions. For now, we wish to focus on one pivotal general point. (We will return to develop several related and reinforcing considerations in section 4.) As explained, ideological reflection takes its departure from those judgments concerning concrete, specific scenarios that are the most characteristic direct deliverances of our conceptual competence. Insofar as responses to scenarios are the direct output of one’s conceptual competence, they may be thought of as high-grade a priori. Of course, not all such judgments are really the results of conceptual competence.
Variations on this theme will occupy us at length in section 4.) So, such judgments enter as input into subsequent reflection as only presumptively and defeasibly a priori. These become data for a kind of abductive generalizing process. What is the character of the abductive inference?
There are those who think that fundamental processes and methods of nondeductive inference must themselves have a kind of a priori status—somehow we can just see a priori that such a method is appropriately truth-conducive. BonJour (1998) holds such a view. We suspect that something like this understanding of epistemic processes, or at least of those at play in reflection, could play a central role in obscuring the empirical dimension of such reflection. We have serious doubts regarding such rationalist-foundationalist understandings of epistemic processes and methods. But for present purposes we will concede the point for argument’s sake, in order to evaluate what really would follow regarding the results of philosophical reflection. In effect, we will argue that the second-stage of armchair-based ideological theorizing involves an important empirical dimension anyway—even granting both (i) that the concrete judgments at the first stage are presumptively high-grade a priori, and also (ii) that the non-demonstrative principles and methods that are applied to these concrete input-judgments in the second stage are themselves high-grade a priori.
So, reflection begins with particular or concrete judgments that are, at least commonly, the deliverances of our conceptual competence applied to concrete scenarios. Insofar as these judgments are the direct deliverances of conceptual competence, they may be thought of as high-grade a priori. Further, for the sake of argument we are now supposing that the reflective process that takes us beyond these judgments is itself a priori indicated—that somehow one can just see a priori that such a method is appropriately truth-conducive. It is tempting to view the results of such a process operating on such input as themselves high-grade a priori truths. After all, by hypotheses the process is an a priori indicated way of moving from high-grade a priori starting places. This would seem to be sufficient to make the results themselves high-grade a priori. However, on closer inspection, the temptation just enunciated reflects something of an oversimplification, and the temptation should be resisted.
We can begin to appreciate the problems with this overly a priorist understanding when we notice that the careful and cautious formulation with which the previous paragraph began spoke of the concrete judgments being “at least commonly the deliverances of conceptual competence.” It was noted that, when such judgments really are deliverances of conceptual competence, then they reasonably may be taken to be high-grade a priori. But, this does not support the blanket claim that the second-stage processes start with “a priori input.” At best, they take their start from input that are commonly a priori. Accordingly, these judgments must be taken up in further reflection as largely, and thus presumptively, a priori.
It is an important empirical point, one that must be marked well, that not all the judgments produced in response to concrete scenarios are really the products of conceptual competence. There are performance errors of various kinds to be found in the first stage. Thus, it is commonly recognized that, as one settles on a satisfactory general philosophical result, some of the “data points” generated in the first stage will ultimately need to be explained as erroneous. The data set for subsequent reflection is thus impure—and accordingly must be taken up as only presumptively, only defeasibly, a priori. This point about the input to the second stage of reflection crucially affects the epistemic character of that reflection—it bestows on it an ineliminable empirical dimension.
As presented in our initial characterizations of both the case of theoretical linguistics and the case of philosophical reflection, the move to generalizations turns on a kind of inference to the best explanation. Such inference cannot be understood simply as an a priori move to a generalization that accounts for a pure set of high-grade a priori judgments by subsuming them under an embracing generalization. The inference cannot be understood as a kind of pure a priori codification of a pure set of high-grade a priori “data points.” For, since the inputs are not pure, the best explanation for the set of them will explain some of them as performance errors of various kinds. The empirical dimension to the investigation should then be obvious. Philosophical reflection must be sensitive to issues regarding the character and content of human conceptual competence, issues that are essentially at the level of psychology; reflection must be sensitive to these matters while pursuing issues regarding the content of that conceptual competence. The best explanation of judgments pertaining to various concrete scenarios (e.g., Twin Earth scenarios, brain-in-vat scenarios, “Gettier cases,” etc. ) will ultimately involve empirical hypotheses about the workings and character of human semantic/conceptual competence and at the same time uncover certain conceptually grounded necessary truths.
3. A Tempting Alternative
Christopher Peacocke (1993, 1998) has advanced a view of a priori knowledge that is in many respects like ours—but is also significantly at variance with our account. Like us, Peacocke is interested in accounting for a priori justification as a matter of drawing on conceptual competence—on what he terms the “possession” or “mastery” of the concepts featured in the relevant claims. His account of the character of concept possession leads him to a two-stage model of the a priori justification one can have for general truths: various particular truths may be derived directly from one’s conceptual competence, and the subsequent justification of general truths inferentially builds upon these particulars. In these respects, we are in agreement. However, Peacocke’s over-idealized take on the epistemic situation of one who has acquired the relevant concepts leads him to understand the second stage (the generalizing stage) as devoid of empirical elements and thus as high-grade a priori.
On Peacocke’s well known account, concepts are identified and individuated in terms of what it takes for an arbitrary individual to possess them: “Concept F is that unique concept C to possess which a thinker must meet [certain] conditions” (1992, p. 6). Such possession conditions for any given concept can be understood in terms of transitions, and thus simple claims, that one must “find primitively compelling.” The reference of concepts, and their semantic value generally, and thus the semantics of the claims featuring those concepts are, on Peacocke’s account, fixed by a charitable treatment of the possession conditions: we are to “assign semantic values to concepts in such a way as to ensure that the belief-forming procedures mentioned in the possession conditions lead to the formulation of true beliefs” (1993, p. 190). So, the claims that fall out of the primitive moves associated with the possession conditions for concepts will be true (1993, pp. 180-1, 189-90). On this understanding of concepts, possession conditions constitute concepts and demand a determinate semantics that makes the conforming judgments true and the conforming transitions truth-preserving. Accordingly, those judgments and transitions spawned by one’s possession of the relevant concepts provide an epistemic route to the truth. Such are truths delivered directly by one’s conceptual competence, on Peacocke’s account.
Without adopting Peacocke’s full understanding of concepts, one can at least recognize a virtue of his account: he keeps clearly in view a distinction between what are the direct deliverances of conceptual competence and the wider set of conceptually grounded truths that can be known drawing on conceptual competence. He appreciates that what falls directly out of conceptual competence are particular responses to concretely presented scenarios. In effect, these judgments are applications of the relevant concepts to particular cases—actual or hypothetical. He writes of taking one’s capacity for the application of concepts “off-line” to apply it to a range of cases: “As in any other simulation exercise, he then exercises a capacity off-line. This capacity is the very same, understanding-based capacity he would be exercising in a real case” (1998, p. 45).
One may take this as as a fair description of what we have in mind as the first-stage reflective generation of (high-grade) a priori truths—and as characterizing a set of judgments that are relatively direct deliverances of conceptual competence. Moving beyond these basics will turn on a range of inferential moves. Consider the straight logical consequences of the set of particular concrete truths that can be derived from conceptual competence. One may reasonably think of these as being as much high-grade a priori as are those direct verdicts of conceptual competence. However, these would also not be interesting general truths. (The conjunction: ‘The XYX in the lakes and rivers on TE-1 is not water and the XY2Z in the lakes and rivers on TE-2 is not water’ is not much closer to a philosophical generality than are its conjuncts.) Interesting a priori philosophical results require more daring inferential moves.
Peacocke’s treatment of such inferential moves is somewhat underdeveloped. He clearly recognizes that the project turns on “finding general axioms from which truths already known to be a priori truths follow” (1993, p. 196). But, one surely would not want to say flatly that all claims from which those basic a priori truths follow are themselves a priori justifiable; this would be far too permissive and needs to be restricted. There are too many and too various a class of general truths from which the antecedently available particular a priori judgments would follow. So, what is needed is not simply an inference to some (or just any) subsuming generalization, but rather an inference to one that accounts for those concrete judgments in the best fashion. What one needs is not just inference to any or some “explanation” (or subsuming generalization), but inference to the best explanation. The point is recognized in Peacocke’s guarded formulation:
The method obviously has affinities with Piercian abduction. It is not, though, required by recognition of the fourth method that the general axioms discovered by it be regarded as explanatory in exactly the same sense as theoretical general hypotheses of the empirical sciences (1993, p. 196).
Peacocke’s formulation seems to be motivated by a perceived need to forestall any rush to a view such as that we are urging in this paper—one in which philosophical generalizations are arrived by way of an investigation that has an ineliminable empirical dimension. However, one reading Peacocke (1993) will find very little in the way of a fleshed out positive characterization of what makes for a better and worse explanation.
The near absence of positive accounting of the character of the needed inference to the best explanation is perhaps excusable after a fashion. It is natural enough to think that there are clear cases to point to—Peacocke commonly points to mathematical examples (such as Peano’s axioms) where generalizing inference is in play. But, we believe that Peacocke’s sketchy characterization facilitates a convenient over-idealization in his general treatment of the a priori. The over-idealization commences long before Peacocke mentions any a priori abductive process. It is already in play in his characterization of concept possession and the epistemic situation of the concept-possessing agent.
For Peacocke, one who has come to possess a concept (or set of concepts) has acquired certain dispositions: dispositions to find certain transitions “primitively compelling.” We agree with this much. Concept acquisition turns on internalizing certain information and coming to have associated dispositions. What remains to be determined is the exact character of the epistemic situation that results: just how an agent is thereby situated for purposes of generating general and philosophically interesting truths. To determine this one must know more. Peacocke recognizes that what are “primitively compelling” by virtue of having acquired concepts are relatively particular truths. But, this is only the beginning of the story. It is crucial to also recognize that the set of claims an agent finds primitively compelling may yet not be limited (indeed is not substantially limited) to truths that are conceptually grounded. While one who has acquired the relevant concepts may thereby come to find certain transitions and claims primitively compelling, this only ensures that certain conceptually grounded truths will be among those that such an individual finds primitively compelling. Furthermore, those judgments that are rooted in conceptual competence (those that turn on concept possession) and those that are not so based need not come so tagged at their inception. That is, our judgments in response to various scenarios may result either from conceptual competence, or from central bits of learned non-conceputally-grounded information, or from various performance errors—and, as the individual responds to scenarios, the various judgments themselves may simply arise without the corresponding labels (“conceptually grounded” or “not conceptually grounded”). If “primitively compelling” judgments are such an undifferentiated lot, then this will importantly condition the character of the epistemic task of generalizing to conceptually grounded philosophical truths. The “primitively compelling” judgments will confront the agent as a varied and untagged array.
For now, it is worth noting how the situation is portrayed by Peacocke. He begins with an agent who has mastered the relevant concepts—who by definition will find some class of particular judgments primitively compelling. This is fine, as far as it goes. But this presumptive focus moves off-stage certain elements of the epistemic action (without denying, or even mentioning, them). To begin with, the reflective agent proceeds in a way that presupposes, at least implicitly, that he or she has indeed “mastered” the concept(s) under ideological investigation. The point is reflected in BonJour (1998) who envisions rationalist successes as turning on an adequate grasp on the relevant concepts. But, whether or not the agent does possess that grasp is not ultimately internally accessible. At best, it is the sort of external matter that an agent can (imperfectly) gauge. Thus, BonJour concludes that, in proceeding, the agent is at least presuming an answer to an externalist issue—one that the agent can have some (nondispository) evidence for taking as settled one way of another. This point should be taken: whether a given “primitively compelling” judgment of the agent is indeed produced as Peacocke supposes is itself an empirical question that the agent probably has some purchase on, and to which the agent should be sensitive. Peacocke provides little notice of the epistemic tasks and the needed sensitivity.
Focusing on the conceptually competent agent, Peacocke explains that such an agent can generate a set of conceptually grounded judgments in response to concrete scenarios. The project is then understood as one of moving beyond those conceptually grounded particulars to conceptually grounded generalities. But, understanding the project in just such terms obscures questions of great epistemic moment that the agent must confront—as it supposes an idealized conceptual competence delivering a delimited set of conceptually grounded truths that can then unproblematically be treated as conceptually-given data points constraining some high-grade a priori inference to a subsuming generalization (when such is finally hit upon).
We do not deny that part of the epistemic action is finding a generalization that subsumes the direct deliverances of conceptual competence—as these are typically spawned in judgments responding to concrete scenarios. But, we insist that there are families of questions that must be sorted out and answered at the same time. As we have indicated, the result is an epistemic project with empirical as well as “high-grade” a priori constraints. First, as just noted, there are questions regarding the agent’s own conceptual competence—whether the agent is competent with the relevant concepts. Second, there are questions regarding the concrete judgments. Which of these are the deliverances of conceptual competence? This question is crucial, and it is forced on the agent by the mixed and undifferentiated character of the agent’s set of concrete judgments. It is just this issue that is most obscured in Peacocke’s picture of an ideally competent agent with a clearly delimited set of conceptually grounded judgments awaiting an adequate subsuming generalization. Third, there are questions regarding the character and structure of conceptual competence. For example, can conceptual competence best be understood as working at a surface and at a deeper level? If so, are judgments reflecting the deeper-level workings sometimes obscured by more automatic surface-level workings? Presumably, the deeper-level workings should be more constraining on one who seeks conceptually grounded philosophical truths.
The real epistemic situation, the issues it forces on the reflective agent, and epistemic hurdles to be cleared, can be appreciated once we leave the over-idealized picture behind.
4. Real Conceptual Competence
Granted, there are things one must learn in acquiring a given concept. In order to have acquired the concept (in some important and recognizable sense that is more demanding than what is required for merely “using” the concept in a blindly deferential fashion), there are claims and transitions that one must find “primitively compelling”. As just noted, saying this much is at best the preface (or perhaps the first chapter) to an account of the epistemic situation of one with conceptual competence. To light the terrain that must be crossed as a conceptual competent in order to arrive at the promised land of a priori justification for philosophically interesting generalities, we can begin with some very general observations.
Consider the typical situation in which one learns concepts. This is not a situation in which one’s informants are interested solely, or even primarily, in one’s acquiring the concept. Typically, the occasion for instruction arises when one is not capable of understanding some substantive point that the teacher wishes to convey. Perhaps one’s instructor is trying to explain why a given monitor is better than another. The instructor mentions the pixel-density of the two, and gets a blank look in return. One apparently needs to acquire the concept of pixel in the course of this instruction. Instruction is then forthcoming both in certain concepts and in particular information about the monitors in question. And the two lessons may be given in a highly integrated and largely undifferentiated course of instruction. Similarly, we suspect, early childhood lessons about morally right behavior are not well understood as lessons solely about what happens to be morally right or solely about the concept morally right. The child (or colleague) is instructed in both within an integrated and largely undifferentiated lesson. Claims in that lesson are not marked out as being conceptual instruction or “substantive” doctrine—and many claims in the lesson may function both ways. Thus, the bottom line: what one learns while acquiring a concept is not what one learns in or by virtue of acquiring that concept. From the start, then, one acquires a mix of conceptually-based and non-conceptually-based information and transitions.
Again (by way of emphasis on what has been said already), in acquiring this mix of information, the pieces do not come tagged, categorized, or otherwise reliably marked as either “conceptually based” or “other.” And this largely undifferentiated mix can then spawn largely undifferentiated judgments in response to scenarios. This is to say that the judgments the conceptually competent person produces in response to scenarios can reflect central bits of non-conceptually-based lore as well as conceptual competence. The judgments one finds primitively compelling may reflect this mix.
These observations force on the reflective agent seeking to capitalize on his or her conceptual competence a question mentioned in the closing paragraphs of the last section: the question of which of the compelling judgments really are the direct deliverances of one’s conceptual competence. While the judgments reflect conceptual competence, they may and typically do reflect more (rather than emanating directly from conceptual competence alone). Other information, other processes, cast their shadows across this then imperfect reflection of one’s conceptual competence. Thus, one is not entitled to any simple categorical presumption regarding the judgments produced in response to the sort of scenario that is common philosophical seed.
One who would identify general philosophical truths drawing on concrete judgments in response to scenarios, will need to come to terms with shadows. The shadows may result from several sources. In the final pages of this paper, we discuss four very general and rather common sources. To say that one must come to terms with shadows in the set of presumptively a priori particular judgments is to say that one must ultimately account for which judgments are conceptually grounded and which are from other sources. As we will soon see, it will also be important to say just in what respects a judgment is conceptually based.
The first kind of shadow is something we have been lately emphasizing: typically concepts are acquired in a context that is marked by mixed learning agendas. As a result, what is learned is ultimately a undifferentiated mixed bag of conceptually grounded and non-conceptually-grounded elements. The concrete judgments that a conceptually competent person finds primitively compelling will likewise be a poorly differentiated mixed bag. That is, central pieces of (non-conceptually grounded) information will cast their shadows in the judgments that must be accounted for. Coming to terms with these shadows is a matter of ultimately determining that some particular judgments have nonconceptual bases. Since we have already emphasized the importance of this situation for the partly empirical character of the epistemic task facing the reflective investigator, there is little that needs to be added here. We simply lead off our list of general shadow sources by mentioning this highly general and very pervasive source of shadows: broadly empirical information (that is commonly learned early on, commonly while also acquiring one’s conceptual competence and) that remains largely undifferentiated from more conceptually grounded elements in one’s understanding.
It is important to realize that there may be aspects of the semantics of some concepts that invite a kind of hasty conceptual generalization that is ultimately distorting. Of particular importance here—a second major source of shadows affecting ideological reflection—are contextual aspects of the semantics of various concepts. Whether certain concepts apply in a case or scenario may be sensitive to a conversational (or other background) context that affects what counts “for present purposes” as “one of those.” Lewis (1979) has been instrumental in calling to our attention the importance of conversational contexts in the application of concepts. If indeed there are such contextually-grounded context effects for certain philosophically important concepts—if it is part of the semantics of a given concept that its vagueness be managed in application by context-sensitive standards—then one reflecting on the concept will need to take care to have a rich enough diet of contexts featured in one’s data set. Failure to do so will be pitfall to be avoided, in reflective inquiry.
This point about hasty generalization can be usefully illustrated by recent contextualist treatments of the concept of knowing (e.g., Lewis 1996, Cohen 1987, 1988, Goldman 1992a, DeRose 1995, Lewis 1996). Crudely, it seems that what is required to know that P turns (within limits) on the context in which the claim to know (or not know) that P is considered. Contextually salient claims set more or less demanding standards. Again crudely, to know that P, one’s belief regarding P (or not-P) would need to “track” the truth or falsity of P across a more or less wide set of possible worlds. One’s belief that P would need to be sensitive, in this sense: were P not to obtain, one would not then still believe P. Across how wide a set of possible worlds would one’s belief regarding P need to track the facts in order for one’s belief that P to count as a piece of knowledge? How demanding are the standards for sensitivity that must be satisfied for a given belief to count as knowledge? One plausible answer is that this is determined by what are the nearest possible worlds in which various contextually salient claims are false. Contextually salient claims then set a contextual standard for how sensitive one’s beliefs must be in order to be knowledge. These suggestions are elaborated and defended in DeRose (1995). And, although we believe that something along these lines is correct, for present purposes we need only consider the implications of such conceptual sensitivity for reflection.
Skeptical philosophers commonly attempt to keep certain skeptical scenarios prominent in one’s reflections. When a knowledge claim is in the offing (say, that I know that my gas tank is close to empty), they ready their thought experiments (making vivid the work of demons and brain scientists). In so doing, they seek constantly to condition the contexts in which one reflects on cases. While this may lead to consideration of a fairly wide range of possible worlds, it also impoverishes the reflective diet of judgments in contexts. As a result, one may readily overlook the contextual elements of the semantics of the concept knowing, and may then be led to suppose that knowing requires a single contextually invariant (and very high) degree of sensitivity of belief. The suggestion: flatfooted skepticism and flatfooted denials of skepticism are both conceptually mistaken, and both arise from reflecting on judgments drawn in overly narrow sets of contexts. (Perhaps this is mistaken. But, if so, one can only determine that it is mistaken by a consideration of judgments made in a sufficiently varied set of contexts.)
The lesson is that when reflecting on a concept with a contextual element to its semantics, if the particular judgments in one’s data set are all generated in or with respect to contexts within some narrow range, then this will itself obscure aspects of the conceptual semantics. The “biased” data, while yet the product of one’s conceptual competence (albeit employed in a narrow range of contexts), will yet cast a distorting shadow over one’s reflective inquiry, obscuring from view other elements of the conceptual semantics. The resulting reflections will yield a flawed appreciation of the status of certain claims—which will not be conceptually grounded, although one is led to them by a (flawed) reflection drawing upon one’s conceptual competence. If not handled appropriately, then contextual elements of the semantics of concepts can cast their shadows over the reflective “data” and the resulting abductive generalizations.
A third general source of shadows in the judgments that constitute data for ideological reflection is context effects that are not themselves anchored in the semantics of the relevant concepts. One can construe ‘context’ more inclusively and broadly. For example, Alvin Goldman writes of “various of [an individual’s] other beliefs, background information, and so forth, which I shall lump together under the term ‘context’” (1992a, p. ). Goldman argues that there are general psychological context effects, and that philosophers engaging in concept analysis will need to be alive to their marks or results. Sometimes such psychological contexts may distort our judgments, by “priming” a kind of differential sensitivity to certain elements of the presented concrete situation. Such priming may skew the individual’s judgmental responses. One who wishes to produce an understanding of the semantics of contexts must ultimately “control for” such distortions in evidential judgments by accounting for their nonconceptual roots.
It is undeniable that beliefs that the subject has, and which have been made salient by conversational or contemplative context, can affect the subject’s judgment in ways that distort that individual’s application of concepts. Recent harrowing experiences, or recently recalling or even reading about such experiences, could “prime” an individual to categorize presented people, animals, or objects as objectively threatening, for example. Attorneys in courts of law seek to induce context effects in jurors. It is commonly noted that defense attorneys seek to manipulate the context in which jurors approach a case so as to make proof beyond a reasonable doubt seem equivalent to proof beyond any doubt. They seek to frame convictions in terms of risks, and reflect on the bare possibility of mistaken conviction. Prosecutors may work to produce a mirroring set of conceptual mistakes—inducing the jurors to conflate proof beyond a reasonable doubt with a showing that something is merely probable, given the balance of evidence. While the standards for proof beyond a reasonable doubt are subject to conceptually appropriate and mandated variation, these do not encompass the standards for proof beyond any doubt or for being probable, given the balance of evidence. (At least they do not encompass the standards for these other concepts in the context of real law courts.) So, the competing strategies would seem to boil down to inducing errors by setting up various context effects. Even without the help of lawyers, we are prone to some context effects that are not conceptually appropriate, and these can cast shadows over the set of concrete judgments that serve as one’s basis for reflective generalization. The best explanation for those judgments will occasionally explain away some of them as mistakes induced by context.
Finally is a fourth kind a shadow affecting the intuitive concrete judgments that serve as data-points for ideological reflection: what is learned in acquiring a concept may itself be somewhat varied in its conceptual depth. Put differently, one’s conceptual competence may itself encode and reflect more and less deep elements of the semantics of the relevant concepts—both deep and surface elements of conceptual semantics. To get a handle on what is intended here, consider the plausible account of conceptual dynamics suggested by Goldman’s “two-stage reliablism.” Goldman argues that a flatfooted treatment of concrete judgments can yield an incomplete understanding of the relevant conceptual dynamics, and that this can foster ready generalizations that obscure or distort deeper elements of the relevant conceptual workings (Goldman 1992b, 1999). He suggests that the best account of our particular judgments regarding justification is that those judgments most directly reflect internalized lists of approved processes and methods. Ultimately, however, these list-structures are themselves influenced by a deeper evaluative basis: understandings of real-world reliability. The lists are a part of one’s conceptual competence, and do have some (defeasible) conceptual and epistemic standing—and they strongly and most directly affect the judgments one makes about concrete cases. The underlying evaluative basis is, however, a deeper and more enduring element of the semantics—although it commonly affects concrete judgments only indirectly. We think that there is much to this suggestion, although we maintain that reliablism is only one component of the underlying conceptual dynamics (Henderson and Horgan, in press (a)).
If indeed there are both deep and surface elements to the semantics of some concepts, then this gives rise to certain pitfalls in generalizing from a set of concrete judgments. Too simple an accommodation of the particular judgments may fix upon relatively surface elements, thereby obscuring the underlying or deeper elements of conceptual semantics. To use Goldman’s plausible caution: it may fix upon elements of a surface list structure while obscuring the deep evaluative basis for the list itself. Common responses to scenarios may reflect what might be termed a “surface-competence.” One must sort out what turns on surface-competence (e.g., common list applications) and what turns on deep-competence matters (e.g., responsiveness to deep evaluative basis). As Goldman rightly notes, this understanding of the workings of common conceptual competence invokes empirical psychological hypotheses.
We began with the idea that, while rationalist and empiricist philosophers may disagree concerning the extent of a priori knowledge, and concerning some purported ways of getting to a priori justification for claims, they agree on at least one point concerning what would be sufficient to make for a priori justification. When one has acquired a competence with the concepts featured in a claim, and when the claim itself expresses a truth that is grounded in internally accessible elements of the semantics of the featured concepts, then a rather distinct route to justification would seem to be available: drawing on conceptual competence, the agent can come to appreciate that the claim must be true. Rather than consulting experiential evidence regarding features of the actual world, one comes to see that the semantics of the relevant concepts suffices to ensure that the world is as claimed—no matter what possible world is the actual world. Insofar as one need only draw on one’s conceptual competence, and does not have to look beyond that to determine what world is the actual world, this justificatory route is indeed distinctive. We believe that there are cases of such justification to be had, and that it is epistemically distinctive and important enough to honor with the designation “a priori.” Accordingly, we have set out to take a closer look at this a priori justification.
On the account we have developed here, with philosophical reasoning about philosophically important concepts as a paradigm case, a priori justification for interesting generalizations typically will come by way of a two-stage process. In the first stage, one employs one’s conceptual competence to do what it does best: to generate particular judgments regarding concrete scenarios. However, such direct but particular results of conceptual competence do not have the generality that is sought in philosophy. This set of particular judgments provides the basis for a reflective generalization that is abductive in character. Our central objective here has been to urge on the reader an understanding of the epistemic character of this generalizing stage.
We have shown that the abductive inference required to get one from the set of concrete judgments made in response to scenarios to general truths has an ineliminable empirical dimension. One must take the concrete judgments as defeasiblly a priori, and one then seeks to provide the best explanation one can for those judgments. To say that they are defeasibly a priori marks several things at once. Insofar as a concrete judgment is itself the direct result of the agent’s conceptual competence, it is a priori. But, from the perspective of the inquiry that would take the agent from such concrete judgments to to a justified grasp of the general truths that are wanted, the concrete judgments are a mixed-bag—only some of which are a priori. Thus, they become data points, many of which will be accounted for as the direct results of a conceptual competence, but some of which will be accounted for as shadows from other sources. This is largely what we intend by saying that the particular judgments are “taken up for purposes of further inquiry” as “defeasibly a priori.” Any account adequate to explain the set of judgments is bound have two parts: (1) a part that explains many of the concrete judgments as being directly produced by conceptual competence, and as being instances of certain general claims that are held to be conceptually grounded necessary truths, and (2) a part that explains away certain recalcitrant concrete judgments as performance errors of various sorts. The account will thus yield certain general claims that are put forward as conceptually grounded necessary truths. But such general claims will be the outcome of an abductive inference that relies partly on hypotheses of a unmistakably psychological, unmistakably empirical, character.
Christopher Peacocke has recognized the two-stage structure of philosophical reasoning to a priori truths. However, we argued that he starts with an over-idealized understanding of the epistemic situation of the reflective agent, and thus produces an account that misses the empirical elements in philosophical reflection. Peacocke’s over-idealizations and mistakes arise because his account ignores the essentially mixed-bag character of the judgments with which the reflective agent begins. In arriving at an adequate understanding of one’s concepts, and in arriving at conceptually grounded general truths, the reflective agent must deal with myriad “shadows” in the data—in the set of concrete judgments. These shadows stem in part from the unmarked contributions of conceptual and nonconceptual information in the concrete judgments. They also arise from the various conceptual sources that require special care that has not been common in philosophy. Context effects provide examples of both. Some context effects are conceptually appropriate—and yet have led to mistakes in philosophical reflection because of insensitive generalization from a too limited range of contextually spawned judgments. Other context effects are not conceptually appropriate, but will cast their shadows over the set of concrete judgments. Additionally, one’s grasp on concepts will typically reflect both deep and surface semantic elements. This is nicely captured in Goldman’s suggestion that an individual’s concept of justification commonly takes the form of a list structure (the surface element) with an underlying evaluative element (the deep element). Without attending to such possibilities, generalizations reached by ideological reflection may be hostage to surface semantic elements, and thereby hostage to passing understandings of what satisfies the deeper evaluative basis. In discussing such suggestions, we have been concerned to catalog a few of the prominent and highly general sources of shadows in the data. These shadows make for the necessity of an account that has an empirical dimension.
The upshot is a route to epistemic justification that has been pointed to repeatedly by philosophers—although they have had various descriptions of what is there going on. We think that they have been dealing with samples of an epistemic kind: justification of (putative) conceptually grounded, necessary, general truths that draws uoon conceptual competence. Philosophers’ descriptions have commonly been mistaken, by ignoring the empirical abductive character of the relevant inquiry. Accordingly, while we are happy to retain the designation “a priori” for the epistemic kind referred to, we add the qualifier “low grade” as a caution: the relevant inquiry does not have one feature that has commonly been thought to mark the a priori, viz., it is not uncontaminated with the empirical. When philosophers undertake to reflectively discover interesting, general, conceptually grounded necessary truths, the kind of justification they should expect to obtain—and typically have obtained, despite widespread misconceptions about their own reflective methodology—is low-grade a priori justification.
Brandom, R. 1994: Making it Explicit. Harvard University Press
BonJour, L. 1998: In Defense of Pure Reason. Cambridge: Cambridge University Press.
Cohen, S. 1987: “Knowledge, Context, and Social Standards,” Synthese 73: 3-26.
Cohen, S. 1988: “How to be A Fallibilist,” Philosophical Perspectives 2: 91-123.
DeRose, K. 1995: “Solving the Skeptical Problem,” Philosophical Review 104: 1-52. Reprinted in K. DeRose and T. Warfield (eds.), Skepticism: A Contemporary Reader. Oxford: Oxford University Press (1999).
DeRose, K. and Warfield, T. 1999: Skepticism: A Contemporary Reader. Oxford: Oxford University Press.
Goldman, A. 1992a: “Psychology and Philosophical Analysis,” in A. Goldman, Liaisons. M.I.T. Press, pp. 143-53.
Goldman, A. 1992b: Epistemic Folkways and Scientific Epistemology” in A. Goldman, Liaisons. M.I.T. Press, pp.155-75.
Goldman. A 1999: “A Priori Warrant and Naturalistic Epistemology,” Philosophical Perspectives 13: 1-28.
Graham, G. and Horgan, T., 1994: “Southern Fundamentalism and the End of Philosophy,” Philosophical Issues 5, 219-47.
Henderson, D. and Horgan, T, (2000a): “What Is A Priori And What Is It Good For?” Southern Journal of Philosophy: Spindel Supplement …
Henderson, D. and Horgan, T. (2000b): Iceberg Epistemology,” Philosophy and Phenomenological Research …
Henderson D. and Horgan, T., in press (a): “Practicing Safe Epistemology,” Philosophical Studies …
Henderson D. and Horgan, T. in press (b): “What Does It Take to Be a True Believer? Against the Opulent Ideology of Eliminative Materialism” in C. Erneling and D. Johnson (eds.), Mind as a Scientific Object: Between Brain and Culture. Oxford: Oxford University Press.
Horgan, T., 1979: “‘Could’, Possible Worlds, and Moral Responsibility, Southern Journal of Philosophy ” 17, 345-58.
Horgan, T., 1985: “Compatibilism and the Consequence Argument,” Philosophical Studies 47, 339-56.
Horgan, T., 1993: “The Austere Ideology of Folk Psychology, Mind & Language 8, 282-97.
Kripke, S. 1972: Naming and Necessity. Cambridge, MA: Harvard Univ. Press.
Lewis, D. 1979, “Scorekeeping in a Language Game,” Journal of Symbolic Logic 8: 339-59.
Lewis, D. 1996: “Elusive Knowledge,” Australasian Journal of Philosophy 74: 549-67. Reprinted in K. DeRose and T. Warfield (eds.), Skepticism: A Contemporary Reader. Oxford: Oxford University Press (1999).
Peacocke, C. 1992: A Study of Concepts. Cambridge, MA: M.I.T. Press.
Peacocke, C. 1993: “How Are A Priori Truths Possible?” European Journal of Philosophy 1: 175-199.
Peacocke, C. 1998: “Implicit Conceptions, Understanding and Rationality,” Philosophical Issues 9: 43-88.
Putnam, H. 1975: “The Meaning of Meaning,” in Putnam, Mind, Language, and Reality: Philosophical Papers, Vol. 2. Cambridge: Cambridge Univ. Press, 215-71.
 These semantic entities—the concepts—have been variously understood, and how they are construed affects how the notion of a conceptually grounded truth is best understood. In this paper we remain neutral about such questions, although we do assume that there are indeed conceptually grounded truths—whatever might be the right account of concepts and conceptual grounding.
 Or at least it does not depend on experiential evidence in the standard or traditional fashion. As we will argue, in drawing on reflection one must sometimes treat the most direct deliverances of one’s reflective thought-processes as evidence for an empirical understanding of our concepts. The investigation is not purely a priori in the sense of being free from empirical contamination—but neither is it devoted to determining whether the world is as advertised in the relevant claims. It has rather to do with drawing on the results of one’s reflections to determine the character of those claims.
 This theme is also developed in Henderson and Horgan 2000a.
 This paper is importantly an outgrowth of Henderson and Horgan 2000a, where we develop at somewhat greater length the case for recognizing as a priori the sort of philosophical reflection we describe there and here. When that paper was presented at the 1999 Spindel Conference in Philosophy, many participants seemed to think that the two-stage process what we describe (the two-stage process of reflectively drawing on conceptual competence) does indeed give rise to a priori justification. However, we found ourselves pressed (notably by Albert Casullo, Michael DePaul, Richard Feldman, ) to better explain why the second stage of reflection should be understood as having the empirical dimension that we envision. Making this case is then the central task we face as we here develop our views. We expect to be pressed from two directions. There are those who, like many at the Spindel Conference, would like to see the results of philosophical reflection as high-grade a priori, and resist recognizing the empirical dimension that is important to the second stage of that reflection. And there are those, perhaps Goldman (see 1999, p. xx, who are quick to appreciate the empirical dimension of second-stage reflection, but who will resist calling the results a priori, even low-grade a priori. Here are here concerned to address the concerns of the first camp in this paper. In doing this, we draw on ideas Goldman has developed (1992a).
 Proposed conceptual analyses very often end up encountering counterexamples—including analyses that initially seem intuitively obvious, like the putative analysis of knowledge as justified true belief.
 Related examples include metalinguistic parallels to the object-language truths. For example:
Natural-kind substance concepts are rigid.
The reference of a natural kind substance concept depends upon an interaction of a community with a stuff (with some underlying structure) in the world.
On our understanding, the family of conceptually grounded necessary truths typically arrived by reflection includes both object-level and meta-level claims.
 This leaves it open whether what we are here calling “implicit mastery” is a matter of the subconscious deployment of explicit representations of general syntactic rules by human language-processing mechanisms, or instead occurs without the deployment of rule-representations even at the subconscious level. Either way, rule-mastery is implicit in the sense that language processing conforms to the rules without deploying conscious explicit representations of them.
 It is not implausible to suggest that these judgments, when they are the products of properly exercised grammatical competence, are themselves (high-grade) a priori, being the product of the linguist’s implicit mastery of the concept grammatical in L (where L is the linguist’s own language). But this suggestion is not essential to our use of armchair syntactic theorizing in linguistics as a model for armchair philosophical inquiry in pursuit of conceptually grounded necessary truths.
 The present points are broadly empirical. The linguist approaches the study of any language with these points in mind; they inform how the linguist “takes up” the judgments as data for further investigation. It is worth noting, however, that nonlinguists possess some inarticulate grasp of these points. After all, the ideas of performance error and competence would seem to be bound up with the notions of correctibility and correctness that are implicit in the everyday grammatical practice ultimately being accounted for. This observation is reflected in Brandom (1994). All grammatically competent speakers would seem to have some understanding of these matters in place.
 Linguists draw on various understandings of the general form of competence and or competence-related cognitive processes when developing their accounts of grammar. The point is reflected in earlier remarks on expectations regarding center-embedded sentences, for example. But, again, linguists are not alone in having and making use of broadly empirical understandings of competence and tendencies for error. All would seem to have and employ understandings of some of the dimensions affecting the rate of performance errors (or of those contexts where they may well need correction). For example, it would not be news to anyone that one is more likely to err when judging long and complex sentence-like strings. Further, one’s judgments may be expected to include rough judgments of complexity as well as judgments of grammaticality.
 Note that the occurrence of a specific concrete judgment is an empirical fact, even if the judgment itself happens to be high-grade a priori. The empirical fact of the judgment’s occurrence, in combination with the empirical presumption that the judgment emanated fairly straightforwardly from the judge’s conceptual/semantic competence, provides empirical support for the relevant theoretical/philosophical hypothesis.
 As in the linguistics case, the operative understandings of conceptual competence and performance-error tendencies need not amount to a full-dress psychological theory about such matters. Rather, typically these understandings will take the form of plausible psychological hypotheses. Such hypotheses are themselves empirical, of course, being ultimately susceptible to vindication or to refutation by ongoing theoretical developments in cognitive science. And their status as empirical carries over to the status of the abductive reasoning in which they figure as background assumptions, since that reasoning typically relies upon such supplementary hypotheses (e.g., ones about performance-error tendencies) as a way of explaining away recalcitrant concrete judgments that do not accord with the abductive conclusion that such-and-such general claim is a CGNT.
 Often, but not always, pertinent empirical considerations are effectively “built into” the scenarios by stipulation, so that the judgments are about conditionals of the form “Given that …., the correct way to describe the case is …” Judgments of this form are perhaps most directly analogous to intuitive grammaticality judgments in the linguistics case, and also are candidates for high-grade a priori status themselves.
 For a piece of ideological refection that illustrates the working of each of these kinds of data vis-à-vis the ideological hypothesis that the concept of freedom does not presuppose the falsity of causal determinism, see the example discussed in Henderson and Horgan (2000a). See Henderson and Horgan (in press (b)) for a similarly structured piece of ideological reflection that illustrates the workings of each of these kinds of data vis-à-vis the ideological hypothesis that folk-psychological concepts like belief and desire are ideologically “austere” in such a way that the truth of folk-psychological ascriptions would not be undermined by the kinds of scenarios for mature science typically envisioned by eliminativists.
 At any rate, they may be thought of this way provided that any pertinent empirical considerations—for instance, the general background knowledge that microstructure is the basis for physical natural kinds—are effectively “built into” the scenarios by stipulation, so that the judgments are about conditionals with a form such as this: “Given that …., the correct way to describe the case is …” Cf. note 12.
 The observation in note 10 is pertinent again here: the occurrence of a concrete judgment about a specific scenario is an empirical fact, whether or not the judgment itself is (high-grade) a priori. Such concrete judgments qua occurrences are empirical phenomena, and when such a judgment fails to be a priori, this is an empirical fact about it. (We are inclined to add that even when such a judgment is indeed a priori, this too is an empirical fact about the judgment qua empirical occurrence—and that this alone is already enough to inject an empirical dimension into the second stage of ideological reflection, thus preventing it from being high-grade a priori. But we do not pursue these further claims in the present paper; the issues they raise are subtle and complex.)
 In a related vein, Peacocke believes that, an individual who satisfies the possession conditions for a set of concepts thereby has some ability to fix on referents of those concepts (or on their semantic values generally): “Possessing a concept is knowing what it is for something to be its semantic value” (1992, p. 23). He writes of this as having some knowledge of the “determination theory” for the concepts. Thus, he insists that one who possesses the relevant concepts has a priori access to “contents whose truth follows from the limiting principles of the determination theories for the concepts they contain” (1993, p. 195). He seems to think of this access as fairly direct (although perhaps not as direct as that associated with those judgments mandated by the possession conditions themselves). On Peacocke’s account, then, this knowledge of the determination theory of the relevant concepts provides a source of judgments that is supposedly akin to that deriving from the satisfaction of the possession conditions themselves. It might be thought of as a second component of what is behind the generation of particular judgments at the first stage of reflection. We are not clear on just how this second component is supposed to work epistemically, and we are thus unable to determine the extent to which we would want to embrace this aspect of Peacocke’s two-stage model.
 We doubt that all of concept identification and individuation can be managed or understood in terms of such dispositions. But these concerns need not detain us here. For the epistemology of a priori justification, what matters is that one who has acquired the concept(s) thereby has access to certain conceptually grounded truths.
 We are not suggesting that the agent need articulately or explicitly address this issue in order to proceed with reflection. Rather, we chose to write of a “sensitivity” to the matter for a reason: this seems to us to represent an epistemic task that is commonly (and commonly best) handled by cognitive processes that accommodate information that cannot be fully articulated. It would seem to present an example of the sort of epistemically crucial use of “morphological content” that we discuss in (Henderson and Horgan 2000b).
 There is seldom a test to pass. Seldom does one get certified as one who has acquired a set of concepts—in the sense of having “mastered” or come to really “possess” them. One commonly gets to employ concepts on which one has only a tenuous grasp, and one uses concepts in attributing beliefs to folk who have only such a tenuous grasp of some of the featured concepts. (Peacocke marks these contexts by distinguishing between possession conditions, and attribution conditions.) So, even one who is employing a concept may yet be acquiring the concept in the same conversation.
 While we ourselves are very sympathetic to these accounts, it will suffice for present purposes that they might well be correct. If they are, then they bring to light an important type of shadow looming over ideological reflection about the concept of knowledge.
 It is worth noting Goldman’s own understanding of philosophical analysis presents it as a two-stage process in which one elicits “intuitive” judgments and then seeks a general understanding of the relevant concepts and the operative psychology that accounts for those judgments. “The question as I see it is one of trying to give a plausible, parsimonious explanation of people’s (linguistic) intuitions” (1992, p. 146). As he explains, the reflective investigator must then manage to sort out the mixed sources of the judgments:
Presumably, the intuitions of an informant, or judge, are the product of (at least) three factors: (A) (his grasp of) the meaning of ‘know’, (B) his beliefs about the state of the putative knower in the target example, and (C) various of his other beliefs, background information, and so forth, which I shall lump together under the term ‘context’. Although the analyst is primarily trying to pinpoint the effects of (A), she needs to assess the contributions of the other factors as well. It is at this juncture that psychology becomes germane. (1992, p. 146)
Thus, Goldman would presumably join us in criticizing Peacocke for presenting an overly idealized understanding of the epistemic situation of the reflective investigator—and for underplaying the empirical contamination of the reflection. He would thus join us in insisting that “philosophical analysis” cannot be a high-grade a priori affair. However, he would part company with us about what we call “low-grade a priori justification” in connection with philosophical reflection. But we believe that the continuity with philosophical tradition makes it reasonable to believe that we are talking about just those philosophical practices that were dubbed “a priori” earlier, and that these practices have features sufficiently distinctive to warrant still employing that term (Henderson and Horgan 2000a, p. XXX).
 Plausibly, there is a contextual dimension to the semantics for the concept of threatening. Thus different degrees of presented risk may qualify something or someone as threatening in different contexts. But, not all context effects can be chalked up to such conceptually appropriate variation. Sometimes contexts effects may make one more sensitive to presented risks than is (contextually) appropriate.
 Both of the examples employed here are of non-conceptually-appropriate context effects involving concepts with contextual elements to their semantics. Perhaps it is to be expected that concepts with contextual dimensions should be subject to certain non-conceptual context effects as well. In both the cases considered here, the mistakes occur by a context-induced employment of standards that are beyond the range that is contextually appropriate.