1. The Basic Symptoms of Vagueness
Paradigmatic vague predicates such as “is bald,” “is a heap,” or “is tall” exhibit anomalous properties.Footnote 1 One striking property of these predicates is that, in contrast to precise expressions, they do not draw sharp boundaries across their range of signification. When Fx counts as vague, then there are objects to which F is unambiguously applicable and there are objects to which not-F is unambiguously applicable, but in the vast majority of cases the extension and anti-extension of F appear to be not exhaustive. This means that there are borderline cases of application where it seems indeterminate or unsettled whether a particular object belongs to the extension or the anti-extension of F.
The property of having borderline cases can be distinguished from the property of being tolerant. Vague natural-language predicates are tolerant in the following sense. As part of its lexical semantics, each vague F is associated with some dimension of comparison. The application of the predicate “is bald” depends, presumably, on the number and distribution of hairs on one’s scalp. The predicate “is a heap” is effected by the spatial position and number of grains. What is remarkable is that sufficiently small changes in the relevant dimensions of comparison do not modify the applicability conditions of these predicates. If “is bald” is true of a man with n hairs, then (in the same context) a man with n+1 hairs may also properly be classified as bald. And in the same way, if “is a heap” may truly be applied to a collection of n grains of sand, then (in the same context) a collection of n–1 grains of sand also may be classified as a heap. Although the meaning of “sufficiently small changes” depends partly on the context of application and partly on other lexical features of the applied predicates, the notion of “tolerance” has an easily understandable, intuitive content.
Sorites-susceptibility may be seen as a further symptom of predicate vagueness. One of the standard formal accounts of the sorites argument is based on simple mathematical induction:
(P1) Fx 1
(P2) ∀n(Fxn → Fxn +1)
(C) ∀n(Fxn)
Premise (P1) represents an object to which F may be unambiguously applied. Premise (P2) says, reasonably, that sufficiently small differences do not influence the applicability conditions of F. The nonsensical conclusion (C) is derived from (P1) and (P2) via mathematical induction or by repeated applications of ∀-elimination and modus ponens.Footnote 2 As an illustration of soritical reasoning, consider the following familiar example:
(P1’) A man with 1 hair on his scalp is bald.
(P2’) For all n, if a man with n hairs on his scalp is bald, then a man with n+1 hairs on his scalp is bald.
(C) A man with a million hairs on his scalp is bald.
Sorites arguments of this sort are paradoxical, because the premises are all well-motivated and the argument is apparently valid, but the conclusion is utterly nonsensical.
The crucial question arising at this point is the following: What could be said about the relationship between the properties of having borderline cases, of being tolerant, and of being susceptible to sorites paradoxes? Different accounts of predicate vagueness would certainly offer different answers to this question. Most participants of the debate would presumably agree that the primary symptoms of vagueness are semantic in character. Indeed, it is not hard to demonstrate why vague phenomena, if they exist, must inevitably appear at the level of the meaning of predicates and sentences. But the agreement would surely end as the discussion moved away from the analysis of symptoms to the analysis of the sources of vagueness. Here opinions diverge dramatically. Currently, at least three competing theories are proposed to investigate the sources of semantic vagueness.
The first kind of view locates the sources of vagueness in the world. Proponents of this view take the semantic appearances at face value, but they claim that vagueness arises where natural-language predicates designate vague objects, objects that have indeterminate or fuzzy spatio-temporal boundaries. The central idea is that if Fx qualifies as vague, then there is no fact of the matter about where the precise material boundaries of the designated x lie. It is often claimed, on this ground, that semantic vagueness may be traced back to ontological indeterminacy.Footnote 3
According to the second kind of view, semantic vagueness is a phenomenon that has its roots in the internal workings of our cognitive system. Vagueness is supposed to arise at a non-conceptual level of perceiving and relating to the external world.Footnote 4 Adherents of this conception point out that in order to analyze properly this fundamental level of cognition, certain parts of the classical framework of first-order predicate logic must be given up. Some try to invalidate the principle of bivalence; others cast doubt on the transitivity of identity or legitimate a particular type of contradiction. There are other possible strategies, but the common distinctive feature of this conception is that its success depends heavily on the intricate details of non-classical logical theories.
The third kind of view is the epistemicist approach to vagueness. Epistemicism’s central notion is “ignorance.” The view is based on the insight that whatever the real sources of semantic vagueness are, we cannot identify these sources because of our restricted cognitive abilities. The fact that we remain ignorant of the semantic difference between Fxn and Fxn+ 1 is due to our limited powers of conceptual discrimination.
Each of these three views represents an intuitively plausible way of theorizing about vagueness, but epistemicism seems to have some important advantages over its rivals. First, contrary to the other two views, epistemicism preserves the validity of classical bivalent logic and the familiar disquotational principles about truth. Second, being in harmony with the rules and principles of classical reasoning, it leaves most of our everyday vagueness-related beliefs intact. Third, epistemicism is the only view that provides a direct solution to the problem originally raised by the opaque relationship between the properties of having borderline cases, of being tolerant, and of being susceptible to sorites paradoxes.
I will argue below that although there is a set of problems surrounding the notion of ignorance, the core thesis of the epistemicist position is basically correct. In section 2, I evaluate briefly Williamson’s and Rescher’s recent versions of epistemicism. Section 3 presents some significant objections to their theory. Finally, in the last section, I make a proposal that may help to move the discussion forward.
2. Epistemicist Theories
Vagueness, as Timothy Williamson conceives it, is an epistemic phenomenon that can be characterized in terms of absence of knowledge. The notion “absence of knowledge” means here something stronger than mere lack of relevant information; it means the in-principle unknowability of the extension of vague predicates.
Introducing some terminology, let us say that our overall epistemic situation concerning a particular predicate F is exhaustively complete if we possess all the necessary background knowledge and evidence to discriminate between instances of F and not-F and know that F cannot be instantiated in any other way.
It is reasonable to suppose, then, that the extension and anti-extension of precise predicates is knowable in exhaustively complete epistemic situations. For example, given our knowledge of elementary arithmetic, we are able to know with certainty that the predicate “is greater than or equal to ten” is either true or not true of a given natural number n. This is not so in cases of vagueness. If F belongs to the class of paradigmatic vague predicates, no amount of background knowledge and evidence will help us to discriminate clearly between the cases in which Fx is true and cases in which Fx is not true.Footnote 5
Williamson explicates this epistemic anomaly in the following way.Footnote 6 Let us create a standard sorites series for a vague predicate F, and assume that the series consists of minimally differing objects. Classical logic would then suggest that there must be somewhere a cut-off point between the extreme members, that is, between the clear instance of F and the clear instance of not-F. But if the difference between the adjacent pairs of the series is sufficiently small, the precise location of the cut-off point remains unknowable due to our limited discriminative capacity.
To take an example, imagine a long sorites series of men for the predicate “is bald.” Suppose that there is a man with one hair on his scalp on the one extreme, and there is a man with a million hairs on his scalp on the other extreme. Let the difference between adjacent members of the series be only one single hair. On Williamson’s view, it is hopeless to try to divide this series into two separate parts because every potential candidate for the cut-off point would be perceptually indistinguishable from its adjacent members. If the nth man seems to be bald, then, arguably, the (n + 1)th and the (n – 1)th men seem also to be bald.
The perceptual indistinguishability of adjacent members may provide an explanation of why we cannot know where the cut-off point is. Williamson’s argument begins with the claim that knowledge requires a reliable basis. If the nth man is bald, then our knowledge of this fact should rest on some kind of evidence. Now, suppose that the nth man is bald and we know this fact on the basis of our observational evidence and our semantic competence with the predicate “is bald.” As we have seen, comparing the nth man with the (n + 1)th man, we cannot discover any significant difference concerning the instantiation of the property of being bald.Footnote 7 So it is natural and tempting to classify the (n + 1)th man as bald too. But it should not be forgotten that the correctness of this classification depends on the above-mentioned two preconditions: the close similarity of the (n + 1)th man to the nth man and our evidence-based knowledge of the baldness of the nth man. In our current example, both preconditions are satisfied. It follows, then, that the (n + 1)th man also is bald, whether or not we have sufficient evidence to know this.
In connection with this argument, Williamson emphasizes that where our discriminative capacities are limited, knowledge is governed by margin for error principles: in reasoning about perceptually indistinguishable objects, a certain kind of epistemic uncertainty becomes inevitable. A margin for error principle for the predicate “is bald” may be construed in the following form:
(MEP) If xn is similar enough to xn +1, and xn is known to be bald, then xn +1 must also be bald.
(MEP) helps to clarify why we remain necessarily ignorant about the location of the cut-off point. Let us hypothesize that the cut-off point for baldness is at 1,010 hairs. In this case, we would have to know both that the man with 1,010 hairs on his scalp is bald and that the man with 1,011 hairs on his scalp is not bald. However, (MEP) prevents us from knowing these facts simultaneously. In order to acquire reliable knowledge of the first fact, an appropriate margin for error is required. The same is true of the second fact. But because of the close similarity of these facts, our knowledge would require the same margin for error in both cases, and this is clearly impossible.
The general moral is that our overall epistemic situation with respect to the extension and anti-extension of paradigmatic vague predicates cannot be exhaustively complete. If F is vague, our epistemic resources prove to be insufficient to obtain reliable knowledge about the cut-off point between the clear instances of F and the clear instances of not-F.
Nicholas Rescher has recently provided a slightly different argument for epistemicism.Footnote 8 Rescher seems to agree with Williamson at least on two basic claims. The first common claim is that in a fruitful theory of vagueness, the advantages of classical logic should be preserved. The second common claim is that all theories that deny the real existence of cut-off points are doomed to failure. The novelty of Rescher’s own approach consists in his analysis of vague predication. So, let us discuss what this means in a little more detail.
According to Rescher, there is an epistemic incongruity inherent in vague predication.Footnote 9 The incongruity is said to stem from the fact that vague predicates are applicable but non-instantiable. In respect of the possibilities of correct application, vague predicates do not differ significantly from precise ones. In both cases, the possibility of correct application depends in large part on such contingent matters as the features of the context and the communicative competence of language users. Differences arise only when we attempt to find specific exemplifications. A precise predicate must have at the very least one exemplifying instance. For example, the precise predicate “is a finite ordinal between 12 and 14” has exactly one exemplifying instance, namely 13.
Rescher’s main contention is that for vague predicates such instances cannot be identified. It should be stressed once again that the problem does not pertain to the applicability conditions of these predicates. The difficulty lies instead in the knowability of their specific instances.
An important insight here is that Fx allows two readings: the indefinite “something is an F” and the specific “this is an F.” If Fx is interpreted according to the first reading, it expresses the general thought that there exist some objects to which F may be applied. That is, it is expressed that there is such a property as F-ness. On the other hand, if Fx is interpreted according to the second reading, it expresses the singular thought that there exists a particular object to which F may be applied. That is, it is expressed that a particular object is an F.
While the first reading is a viable option in cases of vagueness, the second reading leads unavoidably to incoherence. We cannot express the singular thought that a particular object is an F, since the very make-up of the specification of F makes this impossible. Vague predicates are in this respect similar to incoherent descriptions. Consider the following examples:Footnote 10
# a person who has passed into total oblivion
# a never-formulated question
# an idea no one any longer mentions
Indefinite descriptions of this sort are incoherent in the sense that they express properties that cannot be instantiated.
As one would expect, there arises a corresponding duality at the level of knowledge. The first reading of Fx implies generic knowledge: it is known that something has F. In contrast, the second reading induces specific knowledge: one knows of a particular object that it has F. The formal equivalents of these two kinds of knowledge may be stated in terms of quantificational structure. The contrast lies between K[∃x(Fx)] and ∃x[K(Fx)].Footnote 11 And again, the first structure seems to be unproblematic, but the second may lead to unsolvable difficulties.
With this contrast at hand, let us turn back to the hypothesized cut-off point for baldness. According to Rescher, K[∃xn(Bxn & ∼Bxn +1)] symbolizes that we know, in general, that there is a cut-off point in the sorites series for the predicate “is bald.” However, this form of knowledge should not be confused with ∃xn[K(Bxn & ∼Bxn +1)] that symbolizes that we know of a particular number n, say 1,010, that it represents the precise location of the cut-off point. This latter knowledge claim is unacceptable, because it rests on the false presupposition that “is bald” has identifiable instances in the borderline region.Footnote 12
Following this line of thought, Rescher arrives at his version of epistemicism that admits the existence of cut-off points, but denies their identifiability. Accordingly, Rescher’s position culminates in the statement that vagueness should be viewed as a product of epistemically problematic predicates. I think this is another way to say that exhaustively complete epistemic situations are not possible with respect to the extensions and anti-extensions of vague predicates.
3. What is Wrong with Epistemicism?
Having set out the fundamentals of these two types of epistemicism, let us now turn to potential objections.
One objection that may be raised against Williamson’s version of epistemicism concerns the methodological plausibility of margin for error principles. As Williamson remarks, the details of such principles are forced to remain unspecified, for the required degree of similarity may vary as the epistemic situations of speakers change. There is, however, a meta-level principle which holds in all situations: Where our cognitive capacities are unable to yield reliable knowledge, some margin for error principle must be applied.Footnote 13
It is not at all obvious, however, whether Williamson’s meta-level principle is generally valid. In particular, it might be objected that in some soritical circumstances there is no room for any margin for error principle to operate. If Williamson’s view is right, we remain necessarily ignorant about the precise location of the cut-off points of sorites series because we would have to make significant distinctions between objects that are perceptually similar to each other, and our cognitive means are not adequate to this end. But even if we have only limited cognitive resources, there are a number of ways in which we may cope with the problem of perceptual similarity.
Consider, for example, a completely bald man and a man with only one hair on his scalp. We can detect the difference between them by simply looking at their heads. Consider, then, a man with only one hair on his scalp and a man with two hairs on his scalp. Again, the difference is easily observable by naked eye. And when we reach a point in the series where one-hair differences cannot yet be ascertained on the basis of available observational evidence, we can still apply the method of counting. By careful counting we can distinguish even between 1,010 and 1,011 hairs. So, if the exact location of the cut-off point would be at 1,010 hairs, then we would know that the man with 1,010 hairs on his scalp is bald, and we would also know that the man with 1,011 hairs on his scalp is not bald. The procedure may be repeated in the same manner for all other possible cut-off points.
Of course, we do not know where is the cut-off point. But, as the example above illustrates, Williamson is wrong in saying that the source of our ignorance lies in our limited discriminative capacities. It is always an open possibility that we may go beyond the limits of our perceptual abilities by exact counting, measuring, or weighing. Consequently, we may contend that Williamson’s meta-level principle is methodologically problematic: margin for error principles such as (MEP) cannot and should not be applied in all soritical circumstances.Footnote 14
Williamson might reply that the argument above is only partially correct, for the most relevant cases where our discriminative capacities prove to be limited are conceptual rather than perceptual. Suppose again that, in accordance with the sharp cut-off hypothesis, a borderline bald man with 1,010 hairs on his scalp belongs to the extension of “is bald” and a borderline bald man with 1,011 hairs on his scalp belongs to the anti-extension of “is bald.” So someone whose beliefs are consistent with this supposition may judge truly that a man with 1,010 hairs on his scalp is bald. But the statement “A man with 1,010 hairs on his scalp is bald,” though unquestionably true, does not express knowledge since the belief on which it rests is not reliable enough. Had things been very slightly different in our overall use of “is bald,” then though we had the same belief it might not have produced the same statement. For instance, if our overall use had been slightly shifted in such a way that a man with 1,010 hairs on his scalp would already belong to the anti-extension of “is bald,” then the statement “A man with 1,010 hairs on his scalp is bald” would have been false. Such shifts in use are so small that they must remain undetectable to us: our conceptual capacities are not sensitive enough to discriminate between slightly altered extensions. But then we cannot detect shifts in truth value either: no one can ascertain whether the borderline statement “A man with 1,010 hairs on his scalp is bald” is true or false. The final consequence, obtained by semantic descent, is that it cannot be known whether a man with 1,010 hairs on his scalp is bald or not bald.
At first sight, this reply seems to be appropriate for defending (MEP) at the conceptual level. There is nothing wrong in saying that the extension of vague predicates depends, at least partly, on how we use them. It seems also plausible to claim that slight changes in use may invole slight changes in extensions. At closer inspection, however, the same methodological difficulty arises as before. The essential point of the reply is that small shifts in use and the corresponding shifts in extensions must be thought of as undetectable. Nevertheless, it is not immediately clear why this should be necessarily so. Undoubtedly, it would be an extremely complicated task to gather all the relevant fact of usage concerning the predicate “is bald.” A more complicated task would be to detect all the minute changes in its extension. But, as far as I can see, Williamson offers no independent argument to show why the performance of these tasks is in principle impossible. He does not even try to explain what prevents us from applying our conceptual resources in an impoverished way that would allow us to detect the changes in question. Then, from a purely methodological point of view, neither the perceptual nor the conceptual argument for the defense of (MEP) can be said to be entirely convincing.
A second objection against epistemicism concerns the way in which Rescher distinguishes different readings of Fx. As we have seen, Rescher establishes a difference between the indefinite “something is an F” and the specific “this is an F.” It is allegedly the second reading that may give rise to epistemic difficulties. When F is vague, we know that there is such a property as F-ness, but we cannot find any concrete object that exemplifies this property.
Here again, we are faced with a methodological problem. To see the difficulty, let us try to apply Rescher’s distinction to our standard example “is bald.” No one would deny that the indefinite description “someone is bald” may express or convey generic knowledge about baldness. If someone is bald and there is appropriate evidence to know this generic fact, then we must accept the real existence of the property of baldness. Rescher’s view in this respect is entirely correct. But what should we say about the other side of the distinction? What should we think about the argument showing that the singular fact expressed by “this man is bald” is impossible to know? Rescher seems to maintain that it is impossible to acquire knowledge about a particular bald man, because the property of baldness cannot have concrete instances.Footnote 15
I think Rescher is mistaken on this latter point. Now, let us imagine that there is a man in front of us with zero hair on his scalp. If the epistemic conditions were good enough, then it is hard to see why we would not be able to know that this man instantiates baldness. Knowing such a singular fact is not a very complicated matter: we have only to rely on the evidence of our senses and our semantic competence.
It is more important to note, however, that a man with zero hair on his scalp represents a clear case of baldness. The notion of “clear case” is central to any theorizing about vague predicates. In fact, it might be argued that without the notion of “clear case of baldness,” no notion could we have of either “borderline case of baldness” or “cut-off point for baldness.” Or, in other words, for an individual x to be borderline case of baldness or cut-off point for baldness, it is required that there exists an individual y being clear case of baldness.
If true, this requirement for the notional and explanatory priority of clear cases has an important implication for Rescher’s theory. What can be shown is that in order to understand the behaviour of vague predicates, it is mandatory that we have a firm grasp of the clear cases of these predicates. That is, we must know that, for all vague F, there is at least one object x to which F is unambiguously applicable and this specific knowledge must have the form ∃x[K(Fx)]. This suggests that the second part of Rescher’s methodological proposal is not entirely adequate as it stands.
4. Reflecting on Epistemic Ecumenism
The objections above are methodological in nature, but they are related to a deeper issue. When epistemicists try to explain why we are ignorant about the cut-off points of vague predicates, they commonly agree that the roots of the problem lie in our restricted cognitive abilities. Williamson, as we have seen, supposes that the unknowability of the cut-off points can be explained by our limited discriminative capacities. In a rather similar way, Rescher conjectures that the ignorance in question is nothing else than a lack of a certain type of knowledge.
It is far from clear, however, what we are ignorant of when we are confronted with borderline cases. Perhaps the most natural approach for an epistemicist is to say that if x is a borderline case of F, then we are ignorant of certain facts. Namely, we are ignorant either of the fact that x is F or of the fact that x is not-F. And because it is not known which of those is indeed a fact, we are prevented from knowing which fact we are ignorant of.Footnote 16 In a short form:
(i) It is not known that x is F, and it is not known that x is not-F.
But as a moment’s reflection reveals, this cannot be the whole story. Epistemicists like Williamson and Rescher often make the mistake of thinking that the epistemic status of borderline cases depends exclusively on the existence of unknowable facts. It is not too difficult to explain what is wrong with this conception. The explanation may be formulated in terms of compatibility between borderline cases and knowable facts. That is, it must be realized that the epistemic status of a borderline object x for a vague predicate F is compatible with the following two knowledge claims:
(ii) It is known that x is F.
(iii) It is known that x is not-F.
If a man with 1,010 hairs on his scalp counts as a borderline case for the predicate “is bald,” then we are entitled to say either (i) that he is known to be bald, or (ii) that he is known to be not bald. The entitlement in question is purely epistemic. It derives from the circumstance that there is no available evidence that would enable us to decide which of these knowledge claims is true. Remember, the epistemicist cannot exclude the possibility that the cut-off point for baldness lies exactly at 1,010 hairs. So, if it is in principle possible for the cut-off point to be at 1,010 hairs, then the epistemicist cannot exclude the possibility that a man with 1,010 hairs on his scalp might be known to be bald. Likewise, if it cannot be excluded that the the cut-off point lies at 1,009 hairs, then nothing rules out the possibility that a man with 1,010 hairs on his scalp might be known to be not bald.
The admission of merely possible facts, of course, does not constitute knowledge. The epistemicist may rightly say, however, that if we want to give an overall picture of the epistemic status of borderline cases, then these facts should also be taken into consideration.
According to the present view, we remain necessarily ignorant of the truth value of the claim “It is known that a man with 1,010 hairs on his scalp is bald,” because our epistemic situation is not exhaustively complete with respect to the predicate “is bald.” For the same reason, we remain necessarily ignorant of the truth value of the claim “It is known that a man with 1,010 hairs on his scalp is not bald.” In spite of this, such claims have an essential role in characterizing the epistemic status of borderline baldness. They demonstrate that our cognitive relation to borderline cases is weakly compatible with three different claims of epistemic possibility.Footnote 17 Borderline instances of vague predicates may be treated not only with explicit ignorance as in (i), but also with a positive epistemic attitude as in (ii) and (iii). The nature of this phenomenon, which might be called epistemic ecumenism, is captured by the following plausible principle:
The Principle of Epistemic Ecumenism: If an object x counts as a borderline case of the vague predicate F, then the epistemic status of Fx is weakly compatible with the following three knowledge claims: (i) it is not known that x is F, and it is not known that x is not-F, (ii) it is known that x is F, and (iii) it is known that x is not-F.
If valid, this principle implies an important revision of the epistemicist account of vagueness. Recent advocates of epistemicism argue, quite rightly, that vagueness rules out a certain kind of factual knowledge: we cannot know where is the boundary between the Fs and the not-Fs. But, on the other hand, epistemicist theories do not take into consideration the permissive aspect in the behaviour of vague phenomena. What is lacking in these theories is a careful reflection on the epistemic status of borderline cases. In particular, neither Williamson’s margin for error principles nor Rescher’s distinction between different readings of Fx is able to account for the fact that positive knowledge claims concerning the classification of borderline instances of vague predicates have the epistemic possibility of being true.
In a rather weak, epistemic sense of “entitlement,” we are entitled to make both negative and positive knowledge claims with respect to the same borderline situation. This does not embroil us in truth value contradictions, for the epistemic possibility that a man with a certain number of hairs on his scalp might be known to be bald (or not bald) does not imply that it is in fact true that he is bald (not bald). The existence of these possibilities reflects only how things might turn out to be in the borderline area. If it is not excluded by any existing piece of knowledge that a man with a certain number of hairs on his scalp is bald, then we are entitled to say that it is epistemically possible to know that he is bald; and similarly for not bald.
Thus, by realizing that the unknowability of the truth value of borderline Fs is weakly compatible with both negative and positive knowledge claims, we make a first step toward a better understanding of the relationship between the basic properties of vague predicates.Footnote 18
