Some Post-Kantian Variations
of the Analytic-Synthetic Distinction
by Stephen Palmquist (stevepq@hkbu.edu.hk)
This Appendix will review and assess some of the variations which philosophers since Kant have imposed upon Kant's distinction between analytic and synthetic knowledge. As we saw in IV.2, Kant uses these terms to distinguish between two mutually exclusive kinds of knowledge, based respectively on the laws of logic and on intuition. I will begin here by calling attention to the limitations of Quine's empirical version of this distinction in terms of blurred degrees. After discussing Kant's own empirical distinction between analytic and synthetic methods of argumentation, I will proceed to discuss several other attempts to root this distinction in the empirical perspective. I will then briefly examine the feasibility of defining analyticity in ways other than those Kant suggested. Next I will summarize my argument in Pq9, that what Kripke defends as the 'contingent a priori' status of naming is more properly viewed (assuming the meanings Kant gives these terms) as a way of extending analyticity so that it has an a posteriori domain as well as its commonly recognized a priori domain of logical truths. Finally, I will conclude by exploring some additional applications and implications of this new analytic a posteriori class.
Philosophers who reject Kant's Transcendental Perspective in favor of an empirical approach to philosophical problems are likely to side with Quine in denying the validity of distinguishing between analytic and synthetic kinds of knowledge and reformulating the distinction in terms of blurred degrees. Yet what Quine fails to recognize is that, by limiting his discussion to the implications of 'recalcitrant experience' [Q1:40], he forfeits the chance to argue against those who, like Kant, regard the distinction as one of kind, because for them it refers not to experience but to reflective knowledge. The type of blunder that results is evident in the following example. After complaining 'I do not know whether the statement "Everything green is extended" is analytic' [31], Quine analyzes the terms 'green' and 'extended'. His indecision is due to his convenient neglect of the term 'everything', which can be regarded from two distinct perspectives. If 'everything' is taken to include such 'things' as ideas, attitudes (e.g., envy), etc., then the sentence is synthetic in form, because we must go outside the concept 'anything describable as green' in order to see its application to 'extended things'; yet it is false. But if it is restricted to 'sensible objects', then the sentence is analytic and true, because (at least, according to Kant) every sensible object is by definition extended [Kt1:11; s.a. H4:ln]. The sentence Quine ends up analyzing would be more accurately expressed as 'Green is extended', which is indeed incoherent (and so, neither analytic nor synthetic) because 'green' refers to a thing's quality and 'extension' to its quantity. His difficulty in determining the logical status of such sentences is a direct result of his refusal to adopt a perspective outside the 'periphery' of the empirical [Q1:40].
The most Kantian way of formulating an analytic-synthetic distinction in which both terms can be used to describe empirical situations is to connect it with the ancient distinction between analytic and synthetic methods of argumentation [see note IV.13], which should be carefully distinguished from the contrast between analytic and synthetic knowledge [Kt1:395n; Kt2:276n; cf. C13:382-90]. In the latter case the distinction applies to 'transcendental logic', while in the former it applies to 'general logic' [see H4:xxii and III.2]. A synthetic method begins with a set of premises and attempts to determine the validity of some conclusion. An analytic method, conversely, begins with a conclusion and attempts to determine its validity by dissecting it into its logical constituents [see Kt10:149(149)]. The former investigates how a concept (or empirical knowledge) arises, and the latter, how we can understand it once it has arisen. Ultimately, the two methods should yield reciprocal results [cf. E5:40 and B6:35]; thus Kant says in Kt19:387: 'just as analysis does not come to an end until a part is reached which is not a whole, that is to say a SIMPLE, so likewise synthesis does not come to an end until we reach a whole which is not a part, that is to say a WORLD.' Indeed, it is primarily in this sense of method that Bradley's remark is correct when he says, rather cryptically, that 'Analysis is the synthesis of the whole which it divides, and synthesis the analysis of the whole which it constructs' [B26:471]. A prime example of this use of the analytic-synthetic distinction is when Kant says he employs a synthetic method in Kt1 and an analytic method in Kt2 [Kt2:263].
The notions of judgment and method are combined too indiscriminately in Caird's assertion that 'all judgments are synthetic in the making and analytic when made' [C1:1.269]. Had he considered more carefully the difference between judgment and method, Caird might have said something more like: 'All judgments regarded as "in the making" must be described according to a synthetic method; but when regarded as "made", they must be described according to an analytic method.' Or, if he really intended to say something about judgment rather than method, then he could have expressed Kant's view by saying something more like: 'All judgments are synthetic in the making (i.e., in immediate experience); but, upon reflection, some turn out to be analytic while others remain synthetic.' A similar conflation of judgment and method probably lies at the basis of Paulsen's unfortunate misunderstanding of Kant's terms, which leads him not only to deny the importance of Kant's analytic-synthetic distinction [see e.g., P4:135-6,144n], but also to make the rather strange claim that Kant, had he thought of questioning the possibility of synthetic a posteriori knowledge, would have realized that this is an impossible contradiction in terms [see e.g., P4:147]! Given a proper understanding of Kant's terms, the latter view is so obviously wrong that it requires no refutation.
Bennett's empirically oriented version of the analytic-synthetic distinction is rather more adequate. He argues that 'to be able to say of a given sentence that it is [analytic] we must be able to relate it ... to an individual person propounding an argument about [some] situation' [B16:185]. Seen in the context of its underlying method, there would be 'an important difference of kind between the argument in which a sentence is up for possible revision of some sort and the argument in which it cannot be up for revision at all.'[1] The former would occur in the context of a synthetic method, and the latter in the context of an analytic method. However, once sentences were regarded objectively-i.e., outside their methodological context-the distinction would revert again to one of degree. In any case, inasmuch as Kant does not use the distinction in this way, I shall not pursue its implications here.
Quinton's discussion of analyticity in Q4 represents one of the more influential tendencies among analytic philosophers. He relates the analytic-synthetic and a priori-a posteriori distinctions to the necessary-contingent distinction, defining 'the necessary as that which is true in itself, no matter what' [112] and the contingent as that which 'is true dependently on or because of something else, something outside itself' [109]. The latter distinction makes 'an exclusive and exhaustive division of the realm of truths' [109] which implies not only that 'all necessity is logical' [124], but also that 'all a priori statements are analytic' [107]. In other words, the analytic-synthetic and a priori-a posteriori distinctions both 'coincide with the distinction of the necessary from the contingent' [109]. Since Quinton's unkantian definitions of these technical terms are the same as those adopted by Kripke, and discussed below, and since I offer in Pq9:258-60 a more detailed criticism of Quinton's approach, it will suffice to point out here that his definition of necessity in Q4:112 is problematic in the extreme, since it neglects the perspectival principle that all truths are dependent upon some governing perspective.
Bird is right when, in arguing against this tendency to treat 'analytic' and 'a priori' as synonymous terms, he insists that 'analytic truths have something which synthetic [and some a priori] truths lack, namely something that mediates the bearing of ... facts on their truth' [B19:234]. But when he supposes this 'something' to be 'that analytic statements owe their truth to linguistic conventions ... [which] are adopted or rejected often as a result of the recognition of certain facts' [235], he reveals that he too is favoring an empirical version of the analytic-synthetic distinction. As a result, there are two differences between Bird's account and Kant's distinction in terms of logical kinds. First, according to the Kantian distinction, all knowledge, whether synthetic or analytic, depends on linguistic conventions. The importance of recognizing this fact is brought out by Schulze, one of Kant's closest disciples, when he explains that 'if one wishes to decide about a judgment, one must in each case know previously what should be thought under the subject as well as the predicate' [S7:175; cf. Kt22:232]. And second, the mediating factor which truly distinguishes between analytic and synthetic (and so also between analytic and a priori) is that the former alone is wholly dependent for its validity (given the linguistic conventions) on the laws of logic.
When the logical roots of the analytic-synthetic distinction are recognized [see IV.3], borderline cases, such as 'An unprotected human being cannot survive prolonged exposure to temperatures above 100°C' [see note IV.9] or 'Water boils at 100°C' or 'His mother did not die two weeks before he was born' [W15:493-4], all turn out to be synthetic; for the necessity they possess holds only because the natural laws which limit our experience make their contradiction physically impossible. The inevitability of including some such empirical factors in any definition contributes to the inadequacy of 'deducible from definition' as the sole criterion of all analytic a priori knowledge [see note IV.17]. This criterion is used by Waismann when he argues that 'a statement is analytic if it can, by means of mere definitions, be turned into a truth of logic' [W1:31]. Yet his example, 'All planets move around the sun', is actually logically synthetic: it can be 'turned into' an analytic 'truth of logic' only by taking up into the definitions of its terms various 'idiomatic (linguistic) operators' which are contingent upon empirical rather than logical verifiability. As Allison points out in A11:35n, any 'judgments about words in contrast to the intension of concepts' are 'normally not analytic', since they 'are empirical claims about linguistic usage.'
Without a doubt, the most novel twist in the many discussions of Kant's analytic-synthetic and a priori-a posteriori distinctions over the past few decades is suggested by Kripke in K12, where he defends the notions of 'necessary a posteriori' and 'contingent a priori' truth. In Pq9, however, I argue that the latter category is actually a perfect example of what (given a proper understanding of Kant's terms) should be called 'analytic a posteriori' truth. Kripke argues that any proposition which 'fixes the reference' of a name, such as 'One meter is the length of stick S at t0' [see K12:54-5], is 'a priori' since its 'truth follows from a reference-fixing "definition"'[64n] and is contingent since the name is a 'rigid designator' (i.e., since 'in every possible world it refers to the same object' [48]) and its description is non-rigid. Yet for Kant, the fact that a proposition's truth follows from a definition would make it not a priori, but analytic [see note IV.17]; and the fact that an empirical object's description 'could have been otherwise' makes it epistemologically a posteriori (as well as physically or ontologically contingent). (What sense would it make to fix the reference of 'one meter' by referring to a stick in Paris, but stipulating that no one is able to experience the stick in question, since it has an a priori status?) Thus Kripke's analysis of naming provides us with an excellent example of analytic a posteriori knowledge.
Accepting the analytic a posteriori as a legitimate epistemological category enables us to distinguish, in a way neither Kant nor Kripke succeeded in doing, between the status of 'naming' and 'defining'. To name requires us to adopt a hypothetical perspective, according to which we act 'as if' (or stipulate that) a certain object is to be rigidly designated by a certain word. That is, we subsume an object as experienced (a posteriori) under a given concept (analytically). To define, by contrast, requires us to adopt a logical perspective, according to which we devote all our attention to accumulating a set of properties uniquely describing a concept. That is, we subsume a set of general characteristics (a priori) under a given concept (analytically). Naming is in certain key respects prior to the synthetic a posteriori knowledge we gather from the empirical world, whereas defining generally comes after and on the basis of a good deal of empirical information. (We therefore honor babies by naming them, but honor the elderly or the dead by telling stories, or even writing biographies, about their lives.) Kant himself alludes to the significance of this distinction between naming and defining in K2:11.73-4(Z1: 158)e.a.: 'something ... lies only in reason itself [i.e., analytically], something that we can name (viz., freedom ...) but that we cannot grasp [i.e., define a priori].'
Once the possibility of analytic a posteriori knowledge is admitted on the basis of an analysis of naming, various other types of propositions can also be included in this category. Propositions which use a word or words in some 'counter-conventional' way, such as is often found in poetry and other creative activities, 'may be striking enough to recommend a new convention' [Q4:118] and so earn the status of analytic a posteriori. Another example taken from philosophy is Descartes' famous cogito argument, which is cogent only if it is regarded from this hypothetical perspective. To say 'I think, therefore I am' expresses a necessary truth only if it is taken as an analytic a posteriori principle: 'I know from experience (a posteriori) that I can and do think, and this implies (analytically) that I exist.' The mistake of Descartes and others who regarded such arguments as establishing the a priori certainty of the immateriality of the soul, or even the a priori certainty of one's own existence, is, as Kant himself argues in the Paralogisms [Kt1:A341-405,399-432], to confuse logical implication (i.e., analyticity) with real implication (i.e., syntheticity). Of course, interpreting the cogito as a hypothetical proposition divests it of much of its philosophical interest, inasmuch as it no longer confers absolute (a priori) certainty on its conclusion.
Kitcher follows Kripke in arguing for the contingent a priori status of statements such as 'I exist' [K7:92-3], 'I have some beliefs' and 'There are thoughts' [K6:18; K8:30]. In K7:92-3, for example, Kitcher argues: 'so long as I think about the issue and my belief is a product of my reflection upon it, ... I know a priori that I exist.' Yet we can now see that his reference to the acts of 'thinking about the issue' and forming a 'belief' imply that the statement 'I exist' is not a priori, but a posteriori, since the relevant aspect of these acts is that they are experienced. Likewise, the fact that this belief is implied as 'a product of my reflection' makes it analytic, even though any given person's actual existence is indeed ontologically contingent. As such, analytic a posteriori knowledge is knowledge of particular facts (such as the reference of a name) which is based primarily on a conceptual process (such as stipulating a reference).
What then is the difference between the 'analytic a posteriori' and Kant's special 'synthetic a priori' class of knowledge? The analytic a posteriori and synthetic a priori are similar classes of knowledge insofar as both are concerned with conditions imposed on the world by the subject (in contrast to the analytic a priori and synthetic a posteriori types, which are concerned with information that can be drawn out of, or deduced from, what we find in experience), but they differ by virtue of the fact that the former imposes particular conditions (a posteriori) with conceptual (analytic) content, whereas the latter imposes general conditions (a priori) with intuitive (synthetic) content. Thus, for example, this difference can be used to distinguish between the status of the religious rites of initiation and sacrificial offering, respectively. Interpreted as an example of the former, the Christian ritual of infant baptism represents the official naming of a baby: a conceptual (analytic) form, the name, is imposed upon the baby through a specific transforming experience (a posteriori). Interpreted as an example of the latter, the celebration of the Eucharist, by contrast, represents the necessary condition for the possibility of communion with God: an intuitive (synthetic) content, the body of Christ as symbolized by the bread and wine, becomes the agent for conveying God's imposition of a universal (a priori) 'yes' upon humanity-an acceptance which is repeatedly celebrated because it does not depend on any particular experience of the accepted individual.
[1] B16:188. Garver adopts a similar position when he says: 'whether a judgment is analytic or not depends upon the perspective or intention of the person making the judgment... [Therefore] a proposition might be analytic for one person and synthetic for another person' [G3:253; s.a. G4:409-14].
