Epistemicism and Semantic Plasticity[1]

John Hawthorne

(Very much work in progress)



I shall endeavor to make vivid a kind of puzzle that arises when Timothy Williamson’s epistemicist machinery is applied to borderline cases of (i) personhood and (ii) semantic properties. My aim will be to raise some concerns about Williamson’s development of the epistemicist view, and then to explore an alternative way of thinking about epistemicism. What follows is very much a progress report on unfinished business, but I hope there is enough progress to warrant the report.


Consider a Sorites series in which a subject S has his hair removed, one hair at a time, beginning with a full head of hair, ending with no hair at all. At the beginning, S is clearly not bald. At the end, S is clearly bald. However, there will be times when S is neither clearly bald, nor clearly not bald: at those times, S is a borderline case of baldness. Williamson’s epistemicism combines the following theses about borderline cases of baldness:

(1)   The relevant instances of excluded middle hold. Supposing that our subject is now a borderline case of baldness, it is nevertheless true that


S is bald or ~ S is bald


(2)   Bivalence holds for any baldness ascription to S. Thus, whether or not S is now a borderline case of baldness


S is bald’ is true or ‘S is bald’ is false


(3)   If S is a borderline case of baldness, then we are unable to know whether or not S is bald.

(4)   Not all ignorance is due to vagueness. In the case of borderline cases, vagueness has a distinctive source, namely: if we had used the word ‘bald’ ever so slightly differently, we would have picked out a different property by ‘bald’. We are insensitive to the ways that slight differences in usage make a different to the semantic value of our terms. When ignorance is due to that kind of insensitivity, we have ignorance that is due to vagueness.


How does the kind of insensitivity in (4) make for ignorance? Well suppose ‘bald’ is true of S, but it is also true that if we had used the term ‘bald’ ever so slightly differently, ‘bald’ would have been false of S. If we are insensitive to the ways that slight differences in usage makes for a difference in semantic value, then insofar as we actually believe S is bald, there will be close worlds in which we make a mistake in a relevantly similar situations. Thus even if we believe that S is bald and get it right, we do not know that S is bald.

In a borderline case of baldness, then (a) there are a plurality of candidate semantic values, where a semantic value is a candidate for ‘bald’ insofar as, for all we are able to know, it is the actual semantic value of the term ‘bald’ (b) one of the candidates is the actual semantic value, which in turn determines the actual truth value of the relevant baldness ascription and (c) some of the candidates hold of the case at hand, others not.

(A bit of terminology: Let us say that a term is “semantically plastic” when (a) slight differences in usage make for differences in semantic value and (b) we are insensitive to the ways in which difference in usage makes for a difference in semantic value. )


Assuming that persons are material beings, the term ‘person’ appears to be vague.

(i) To begin: there is vagueness as to whether a person comes into existence. Even if, for example, one is convinced that a person begins with conception, there will be vagueness at the beginnings of personhood owing to vagueness as to when conception actually takes place (think of the trajectory of sperm, slowly approaching and entering the egg. It is clearly a vague matter when conception occurs). In general, all views about the beginnings of personhood use vague predicates in the favored criterion – even granting any given one of those views, vagueness along the temporal dimension will not disappear.

(ii) And unless one believes that persons enjoy life (or at least existence) everlasting, there will be vagueness as to when a person’s existence comes to an end. Once one remembers that such predicates as ‘dies’ and ‘is braindead’ and so on are vague, the relevant thesis about persons should be obvious enough.

(iii) Further, there is vagueness as to where the spatial boundaries of a person lie. There are for example, certain atoms in the vicinity of my surface such that it is a vague matter whether or not they are parts of me.

(And there is also vagueness as to whether various less sophisticated beings counts as people. And so on.)


Let us pick on a case. Suppose that there is a speck of dirt – call it Tony—such that it is a vague matter whether or not the sentence ‘The person sitting down has Tony as a part’ (hereafter S) is true. Let us try to accommodate the vagueness of the case using epistemicist machinery. What we should say, it seems, is that there are a multitude of candidate semantic values for the term ‘person’ such that the truth value of S differs according to which of those candidates is adopted as the interpretation of ‘person’. The point presumably extends to personal pronouns. It will thus presumably also be a vague matter whether or not the sentence ‘He has Tony as a part’ (pointing at the person in the chair) is true (hereafter S2). And this will because there are a range of candidate semantic values for ‘He’ such that the sentence differs in truth value according to which semantic value is adopted. Suppose S2  is true. There will be a meaning that could very easily have been given to ‘He’ such that S2 is false.

This semantic picture invites us to posit a plentitude of overlapping objects in the vicinity of the chair. Only one of them falls within the extension of ‘person’. Only one of them is the referent of the personal pronoun ‘He’. Insofar as the object in question has a Tony as a part, then S1 and S2 are true. But owing to semantic plasticity, there are a variety of candidate semantic values (v1… vn) of ‘person’ that each associate a particular object with the definite description ‘The person in the chair’ (that is, relative to any candidate interpretation of ‘person’, a particular object will count as satisfying the definite description ‘The person in the chair’). While some of the candidate semantic values associate an object containing Tony as a part with the definite description ‘The person in the chair’, others will not.

Let us focus on two of the candidates, one containing Tony as a part, the other not. Call them Grubby and Clean. Suppose Grubby and not Clean falls within the actual extension of the term ‘person’ in English (though of course we would be unable to know this). Then S1 and S2 are both true. But there is a possible tribe that uses the word ‘person’ ever so slightly differently such that Clean and not Grubby falls within the extension of the term ‘person’ in their mouth and S1 and S2 in their mouth are false. Such a tribe might even be actual. Pretend that there exists a tribe of Twinglish speakers that uses ‘person’ in such a way that ‘person’ is true of Clean and not Grubby. Then when a Twinglander says ‘The person sitting in the chair has Tony as a part’ he will express a false proposition even though we say something true.


There are a variety of overlapping object on the chair. Only one of them is a person. That is, only one of them falls under the extension of the actual semantic value of ‘person’. Same, presumably, for such predicates as ‘thinks’ ‘talks’ and so on. Only one of the objects thinks. Only one of them talks. The one that is the person is also the one that thinks and talks. Other of the overlapping objects fall within the extension of candidate semantic values for ‘thinks’ ‘talks’ and ‘is a person’ (in the sense of candidacy explained.). Suppose Grubby thinks and talks. If our use of ‘think’ and ‘talk’ had been ever so slightly different in certain ways then ‘think’ and ‘talk’ would apply to Clean. Let us say that an object thinks* iff it falls within the extension of one of the candidate semantic values for ‘thinks’. We could similarly introduce the predicates ‘person*’ ‘thinks*’ and ‘talks*’. There are many objects that on the chair that are persons*, which think* and which talk*. But only one of them thinks, talks and is a person.

Why do I insist that only one of the object in is a person? Well I take it that a feature of our usage we do not allow that many objects at a time are people when those objects mereologically overlap almost entirely. If semantic values are going to respect that aspect of usage then each candidate semantic value for ‘person’ will only allow one of the objects on the chair to fall within its extension.[2]

What makes, say, Grubby and not Clean the thinker? If we could know the answer to that question then (says the epistemicist) it would not be a vague matter whether Tony is part of the person. We cannot know what it is about our use of ‘thinks’ that determines one of the candidate semantic values to be the actual one. Our knowledge of semantic relations is incapable of extending that far. And that is precisely why ignorance arises in the case at hand.

Call the approach just sketched the ‘Simple Epistemicist Treatment of Persons’



There is a problem for the simple epistemicist treatment. Let me illustrate it by an example. Suppose a Twinglander is sitting in a chair. Suppose that the Twinglander uses ‘thinks’ and ‘person’ ever so slightly differently (and is otherwise very much like an ordinary English speaker), so that the semantic value for ‘thinks’ and ‘person’, as used by the Twinglander, is different from ours. In particular, let us suppose that Grubby and Clean are both on that chair, and our use of ‘person’ is such that it is true of Grubby and not Clean and the Twinglanders use of ‘person’ is such that it is true of Clean and not Grubby.

Here are some very obvious truths

(1)   The Twinglander is the person sitting on the chair

(2)   The person sitting on the chair is only thing on the chair that is able to talk and think.

(3)   When the person sitting on the chair says ‘I’ the person is referring to himself.

(4)   If the person sitting on the chair says something of the form ‘a is F’ then that claim is true iff the predicate ‘F’ is the mouth of that person is true of the thing referred to by ‘a’.


Let us add to these obvious truths the added facts provided by our epistemicist-driven description of the scenario

(5)   Grubby is the person sitting on the chair (and Clean is not)

(6)   ‘is a person’ in the mouth of the Twinglander is true of Clean and not Grubby.


Suppose the Twinglander says ‘I am a person’.  We can deduce (i)  that it is Grubby that says ‘I am a person’ (ii) that nothing else on the chair says ‘I am a person’. (iii) that Grubby is referring to himself (iv) that ‘I am a person’ as uttered by Grubby is true iff ‘is a person’ in the mouth of Grubby is true of Grubby (v) ‘is a person’ in the mouth of Grubby is not true of Grubby . All of this has us conclude that when the Twinglander says ‘I am a person’, the Twinglander expresses a false proposition and that nothing says something true The same argument, mutatis mutandis, could have been run for ‘I think’ and ‘I talk’. We should now conclude further that if our use of ‘person’ and ‘think’ had been ever so slightly different, then the sentences ‘I think’ and ‘I talk’ would be false. Worse, if some of the candidate semantic values for ‘think’ are such that ‘I think’ comes out false, then we do not know whether or not ‘I think’ in our mouths is true. (At close worlds we make a mistake, which on a safety-driven conception of knowledge, undermines knowledge at the actual world.) Clearly, something has gone terribly wrong.

Let us get a bit clearer about the source of the problem. As things have been set up, it is our standards for ‘person’, ‘thinks’, ‘talks’ and so on that determine which of the candidate objects is an object that is self-referring, but it is an object’s own, potentially different, standards that determines the extension of ‘is a person’, ‘thinks’ ‘talks’ and so on in its own mouth. It is our own standards that determine which objects are objects that are capable of engaging in the activity of drawing boundaries. In short, our standards determine which objects are boundary drawers. But it is the potentially different standards of the boundary drawer that determines where the extension of ‘thinks’ and ‘person’ in its mouth are to fall. Supposing the set of boundaries corresponding to ‘person’ that are drawn by a boundary drawer do not include its own boundary. Than were that boundary drawer ever to self-ascribe that predicate to itself it would make a mistake.

Let we us say that a person uses a close variant of ‘person’ iff the semantic value of ‘person’ in the mouth of that person is a candidate semantic value for ‘person’ in English. Assuming that ‘person’ is semantically plastic, it seems very easy for a variant of ‘person’ in the mouth of a boundary drawer to be false of the boundary drawer. That is just to say it is easy for a person to be such that they use a close variant of ‘person’ in a way that is false of that person. Same for ‘thinks’. The trouble is that we do not want to say that there are close variants of ‘thinks’ and ‘person’ such that those predicates are falsely self-ascribed. For it is all too easy to self-ascribe those predicates – ‘I’ thoughts will do the trick.

The simple epistemicist model needs supplementation or revision. I shall explore three approaches.






Strategy A:  One response is to deny that self-ascription is easy for close-variants of English. One might insist that when the Twinglander says ‘I think’, the Twinglander is not referring to himself by ‘I’ but is referring to, say, Clean. On this model ‘I think’ in the mouth of the Twinglander is much more like ‘He thinks’ than it may first appear. In brief: whenever a person uses a close variant of ‘thinks’ in such a way that the person does fall within the extension of ‘thinks’ then the person will have no readily available device for self-reference and in particular will not self refer when that person uses what would naturally be taken as a cognate of ‘I’.

This view may not be as bad as it first appears. It is here worth bearing in mind a potential symmetry between myself and the Twinglander. For on this view the Twinglander may well say ‘He cannot refer to himself’ where ‘He’ in the Twinglander’s mouth refers to something that mereologically overlaps me, but which is not identical to me, and that satisfies ‘person’ ‘talker’ ‘thinker’ and so on in his mouth. So the proponent of this view can do something to deflate the suggestion that we are really special by being able to self-refer.

We should also bear in mind that, almost inevitably, the epistemicist is going to have to learn to live with a mass of strange counterfactuals. Suppose I do not want to be bald. It seems at first blush to be true that if I had gone to the pub last night, I would still wanted not to be bald. But suppose that if I had gone to the pub the evening’s conversation would have induced slight differences in use that shifted the meaning of ‘bald’, ever so slightly. Then it would seem that, strictly speaking, the counterfactual is false, since at the closest world where I go to the pub I do not stand in the desire relation to the proposition actually expressed by ‘I am not bald’.

All that said, I remain unimpressed by the view that self-reference fails in close variants of English. There is something exceedingly strange about a view according to which at close worlds, many people (perhaps most people), do not have linguistic devices of self-reference. Relatedly, it is extremely natural to think that if a pronominal device has the conceptual role of the first person pronoun in a person’s cognitive life, then that pronoun will be a device of self-reference. The thought is a little rough and ready, owing to the rough and ready nature of the concept of “conceptual role”, but has some force nevertheless. “I” thoughts in the Twinglanders belief-box will have stereotypical roles in practical reason and so on that make it utterly natural to suppose that they are devices of self-reference.  While the meaning of ‘bald’ may be fragile, ‘I’ does not seem to be so easily purged of the character that Kaplan described for it.


Strategy B: Return to the case of the case of the Twinglander. Suppose that by the standards of the Twinglander, ‘person’ is true of Clean. In short, Clean is the object that counts as the utterer of ‘I’- talk by the standards of the Twinglander. Perhaps the concept of a person is distinctive in that it always defers to the self-conception of people: an object can only count as a boundary drawer insofar as its draws its own boundaries at its own boundaries.[3] On the hypothesis that Grubby is a person, Grubby counts Clean but not Grubby as the referent of ‘I’. That counts as a reductio of the idea that Grubby is a person at all.

Example: Suppose there is an Englander and a Twinglander in a chair. The intrinsic environment is pretty much the same for each. There are, inter alia, two objects GrubbyE and CleanE in the Englander’s chair and two objects GrubbyT and CleanT in the Twinglander’s chair. Suppose the Englander self-descriptions (without him knowing it) privileges GrubbyE and the Twinglander self-descriptions (without him knowing it) privileges  CleanT. Then ‘person’ in the Twinglander’s mouth is true of GrubbyE  and CleanT. Same for the Englander. Each defers to the other’s self-description as the prime semantic determinant of which object is the referent of ‘I’ thoughts and in term for which object counts as the extension of ‘person’. (I am here abstracting away from issues connected to what it may or may not take to be a person beyond being a  thinker that can self-refer) .

Our simple epistemicist position claimed, in effect, that people could very easily have used the term person in slightly different ways such that they did fall under the extension of ‘person’ in their mouth. Our revised position denies this.


Let us turn to the diachronic case. Suppose I begin life with the self-conception of a Twinglander and towards the end of my life move towards the self-conception of an Englander. (Of course, the shift my not be epistemically obvious to me: in fact if the shift is around borderline cases, it will not be.) Suppose my earlier self is sitting in a chair and my earlier self’s usage privileges Grubby (where now let Grubby be an object that always has Tony as a part)  and not Clean (which never has Tony as a part) as the referent of ‘I’ thoughts. My later self however privileges Clean but not Grubby as the referent of ‘I’ thoughts. The natural application of the deferential conception is the following: My earlier self refers to Grubby with his ‘I’ thoughts and my later self refers to Clean with his ‘I’ thoughts. But this can’t be right. My earlier self IS my later self. But Grubby is not identical to Clean. When I look back on my earlier self, I want to say ‘I was referring to myself when I said ‘I am hungry’. If I am Clean then it cannot be that my earlier self refers to Grubby. The problem is analagous to the earlier one. The ascribee’s usage puts semantic pressure to count one thing as the referent of its ‘I’ thoughts, whereas the ascriber’s standards on who is to count as a thinker in the first place puts semantic pressure towards a different thing to count as as the referent of ‘I’ thoughts. Where the target and the ascriber take themselves to be one and the same person, the ‘to each his own’ deferential strategy cannot be made to work. At this point the epistemicist needs to appeal to a new, anti-individualistic, theme.

Return to the case of ‘bald’. It is wrong to think that the extension of ‘bald’ in my mouth is simply a matter of how I use ‘bald’. My own use creates various semantic pressures, but I am a member of a linguistic community. The usage of others also contributes to the extension of the term in my mouth. Indeed, one reason – though perhaps not the only one – as to why I cannot know which value is the semantic value of ‘bald’ in my mouth is the fact that I am not privy to all the details of others’ usage.[4] Now what goes for my relation to others in the linguistic community may go for my later or earlier self. The extension of ‘person’ in my mouth – and relatedly, the referent of ‘I’ may be constitutively determined by the usage of my later or earlier self. Suppose, to simplify, a community consisted of two individuals A and B. A’s usage of ‘bald’ may favor a cutoff at 17 hairs. Roughly speaking: if A was the only member of the community, then the extension of ‘bald’ in his mouth would include all and only people with 17 hairs or less. Suppose, meanwhile, B’s usage favors a cutoff at 15 hairs. This does not mean that B and A have different semantic values for ‘bald’. The fact that they translate each other homophonically puts semantic pressure against such a resolution. If God were to interpret them he would likely say that their terms have the same semantic value, adopting some appropriate weighting of the various semantic pressures at play. We cannot know how the weighting would proceed of course – our lack of knowledge of the details of the relevant laws of semantics  is what gives rise to the phenomenon of vagueness (says the epistemicist). Similar remarks apply, mutatis mutandis, to the case at hand. If earlier usage favors Grubby, the later Clean, but there is considerable semantic pressure for uniformity of semantic value across uses, then the matter is resolved by some (we know not what) weighting of semantic pressures to yield uniform semantic value. Perhaps the usage that favors Grubby loses out to the usage that favors Clean. In that case, both the early and later self use ‘person’ in a way that the extension of ‘person’ includes Clean and not Grubby. In that case no utterance of ‘I am a person’ gets to be false: on both occasions, it it Clean that makes the utterance. Grubby is not a person at any time. Both utterances come out true since Clean is always, Grubby never, included in the extension of ‘person’.


I have sketched a picture. Does it fit with the barebones of epistemicist I began with? It is of course perfectly consistent with excluded middle, with bivalence, and with the thesis that borderline cases are beset by ignorance. What is less clear on this picture is that there is semantic plasticity in borderline cases. More specifically, it is not clear that the intension associated with ‘person’ could easily have been different. To make this vivid, consider a world W containing nothing but a Twinglander sitting in a chair. If ‘person’ is deferential in the way described and the Twinglander’s usage favors Grubby, then I should say that Grubby is the person in that world. Could the intension of ‘person’ have easily been such that it did not deliver Grubby as the extension taking W as its argument? (I am operating here with a standard conception of intensions as functions from worlds to extensions.) It would seem that a word that was not deferential in the way described would not be a close variant of ‘person’. But that inclines me to think that there is no close variant of ‘person’ that does not deliver the set containing Grubby as value taking W as argument. We want to say that it is a vague matter whether the person in W has Tony as a part. But it is not at all clear that the ignorance can be explained as a matter of semantic plasticity.



(An aside: Suppose one adopts the epistemicist picture sketched thus far. Given the plenitudinous ontology in the background and the willingness to defer to self-conceptions as the determinants of boundaries, it is at least natural to extend the picture to allow that some people have wildly different boundaries simply on account of wildly different self-conceptions. To make this vivid, consider a community of Eggers. An Egger believes that he or she came into existence as a human Egg. An Egger will say that fertilization made him or her gendered (and in general a lot more interesting). But on an Egger’s self conception, his or her existence began with an egg. Embrace the plenitudinous ontology suggested by epistemicism and it is very natural to at least suppose that there is an object – don’t ask yet whether it is a person – which comes into existence at the time that the egg does, which endures for 70 or 80 years and which has all the intrinsic requirements for thought later in life. Run with the deferential perspective and I think it very natural to say that these are the objects that the Eggers refer to by their ‘I’ thoughts. Some people, then, come into existence prior to fertilization. Not us perhaps. But the Eggers do. Is this a reductio of the deferential conception of persons? I myself don’t think so. Perhaps some will.)



Strategy C[5]: One might say in these cases that vagueness in spatio-temporal temporal boundaries does not, despite first appearances, have its source any vagueness associated with the term ‘person’ (and in the associated personal pronouns), but rather in vagueness associated with other pieces of vocabulary. Suppose it is a vague matter when I came into existence. It might be claimed that there is some particular object o, such that it is definite that ‘person’ is true of o, and definite that ‘I’ in my mouth refers to o, but that the location of o is indefinite owing to vagueness associated with the term ‘occupies’. Now in the case of mereologically complex objects such as myself (assuming the falsity of Cartesianism), it seems clear that the location of the whole is derivative upon the location of the small parts. It would thus be strange to suppose it is a sharp matter as to whether any given atom  is or is not part of a mereologically complex thing at each time but that the location of the thing is vague, unless there is vagueness in the location of particular atoms. The latter is not plausibly the source of indeterminacy in my boundaries. So as far as I can tell, then, the most promising development of the current strategy will require positing vagueness in the concept of parthood: it is vague for various pairs xy that, at some given time, x is is part of y. Assuming semantic plasticity, this will require that we posit a variety of candidate semantic values for ‘is a part of’, that differ intensionally.

Some will consider this an extremely radical tack, almost as radical as the idea that ‘exists’ is vague. I think that this reaction can likely be traced to a tendency to think of mereology as part of logic, and thus a tendency to think that ‘part of’ enjoys purity of the kind possessed by existence and identity. This will be reinforced insofar as one adopts an extensionalist mereology according to which x is identical to y iff x is part of y and y is part of x. For then it will be particularly difficult to maintain vagueness for ‘part of’ in combination with precision for ‘is identical to’.

I do not find such abstract logical considerations very pressing myself: I am not particularly drawn to extensionalism, nor to a logicist thesis about mereology. There is a more local problem however. For consider a case in which it is vague whether or not a person comes into existence at all (imagine the room blowing up just as the sperm is entering the egg). In that case it seem extremely difficult to explain the vagueness of the case in terms of vagueness in ‘part of’, since here it seems that it is vague whether ‘person’ is true of anything at all. Similarly, consider a world where there an explosion that renders it vague whether I come into existence. (Imagine the world is just like the actual world up to the freak explosion.) If it can be vague whether I exist at that world then there had better be something that exists there that is a candidate. Let us use ‘Johnny’ as a precise name for that object. It now appears that ‘I am Johnny’ is indefinite. But it is hard to see in this case how the vagueness can be blamed on ‘part of’.

I can think of only one promising escape. In other works, Timothy Williamson has defended a necessitarianism about objects according to which any object exists eternally and necessarily.[6] Adopt that view and it is no longer coherent to suppose that there is a world where it is vague whether I exist at all. It is of course coherent to suppose that there is a world where, at some time, it is vague whether I am concrete. Further, it is coherent to suppose that there is a world where it is vague whether I am ever concrete. But this kind of vagueness can, perhaps, be blamed on ‘part of’. Suppose for one of the candidates of ‘part of’, there is a world w where I never have atoms as parts, but that on another precisification of ‘part of’, I do have atomic parts at w. On this conception, there are worlds where, while I definitely exist, it is vague whether or not I ever, so to speak,  dip my toe in the void that is filled with atoms.  Acknowledging a restricted use of  ‘exists’ to mean ‘concretely exists’, there is a sense it which it is vague whether I exist at that world, though the vagueness can be placed squarely on the shoulder of ‘part of’.

I recommend strategy B to non-necessitarian epistemicists. However, strategy C may very well turn out to be preferable if we assume a necessitarian setting. Strategy B makes trouble for the semantic plasticity component of Williamson’s view. Strategy C does not. Considerations of personal identity thus appear to forge an interesting bridge from his version of epistemicism to necessitarianism. (Not that one needs to rely on such a bridge in order to motivate necessitarianism.)


The issues I have raised are far from unique to epistemicism. Consider the main competing theory of vagueness,[7] supervaluationism. The supervaluationist uses supertruth and superfalsity as her primary concepts of semantic evaluation. A sentence is supertrue iff it is true on all precisifications, superfalse iff false on all precisifcations. A borderline case is one where the relevant sentence is true on some precisifications, false on others. Suppose ‘He has Tony as a part’ is borderline (where ‘Tony’ is precise). That will be because there are various precisifcations of ‘He’, some of which contain Tony as a part, others not. Here too we have a plenitude of objects required by the semantics. Suppose further we embrace the analogue of semantic plasticity: small shifts in use generate small shifts in the range of acceptable precisfications of a term. Suppose now a Twinglander has a slightly different set of acceptable precisifcations for ‘person’ than I do. Now the competing pressures described above will arise. On the one hand, I want to say that the Twinglander self-refers by ‘I’. This encourages me to treat all and only the acceptable precisifications of ‘The Twinglander’ in my mouth as acceptable precisifications of ‘I’ in the Twinglander’s mouth. But suppose some of the acceptable precisifcations of ‘The Twinglander’ in my mouth are not acceptable precisifications of ‘person’ in the Twinglander’s mouth. Then there is a threat that ‘I am a person’, in the mouth of the Twinglander, will not come out supertrue. (Same for ‘I think’). Such a sentence will have to be reckoned borderline. This result is intuitively unacceptable. The same theme is in play, namely competing pressures on the semantics of ‘I’ in the mouth of another, generated on the one hand by my conception of a person and on the other by the self-conception of the other. It is thus relatively straightforward to recast the issues just discussed within the alternative semantic framework of supervaluationism. A shift from epistemicism to supervaluationism will thus not make the problems go away, so long as semantic plasticity is maintained.

            (I also note that one currently popular approach to vagueness does particularly badly in connection with our puzzle cases. Delia Graf and Scott Soames have advocated a “shifting sands” approach to vagueness[8], according to which the boundary associated with a vague term actually shifts during the process of moving up and down a Sorites series (never mind the details). Suppose I am confronted with a  Sorties series in which I am asked of a series of particles whether each is part of Hartry Field, where each case is further from Hartry’s center of mass than the previous one, and where some of the cases are in the vicinity of Hartry’s boundary. The shifting sands view would apparently have us believe that the extension of ‘part of Harty Field’ shifts during the series of judgments. Apparently, then, ‘Hartry Field’ in my mouth, may well denote a larger object by the time that I am done! We thus appear to have surprising capacities to make people grow.)




I now turn to the case of semantic predicates. Williamson is happy to suppose that his picture extends to such predicates as ‘refers’ and ‘is true’. Thus, in a reply to a commentary by Stephen Schiffer, he writes:

. . . semantic ascent preserves vagueness. For example, since it is clear that something is bald if and only if it is in the extension of ‘bald’, ‘bald’ has the same borderline cases as ‘in the extension of “bald”’.

My general explanation of the ignorance that constitutes vagueness extends to semantic terms. Although someone may judge truly ‘Baldness is the property of having fewer than 3832 hairs on one’s scalp’, the judgement does not express knowledge, for whatever produced a judgment in those words could very easily have done so even if the overall use of ‘bald’ had been very slightly shifted (as it could very easily have been) in such a way that it referred to the property of having fewer than 3831 hairs on one’s scalp, in which case the judgement then made in those words would have been false. What produces the judgement does not produce true judgments reliably enough to produce knowledge. …To extend this explanation of our non-semantic ignorance to an explanation of our semantic ignorance, note that in the envisaged counterfactual circumstances the sentence ‘“Bald” refers to baldness’ naturally still commands assent (clearly, ‘bald’ refers to baldness). In those circumstances, the false judgement in the words ‘Baldness is the property of having fewer than 3,832 hairs on one’s scalp’ goes with a false judgment in the words “’Baldness’ refers to the property of having fewer than 3,832 hairs on one’s scalp’. Although someone may use the latter words to make a true judgment in the actual circumstances, the judgement does not express knowledge, for what produces it does not produce true judgements reliably enough to produce knowledge. Thus the account explains equally why we are not in a position to know that ‘baldness’ refers to the property of having fewer than 3,832 hairs on one’s scalp.[9]


I think that this passage betrays a confusion that is easy to overlook. Let us distinguish semantic plasticity – the phenomenon whereby the intension associated with a term could easily have been different – from extensional plasticity – the phenomenon whereby the extension of a term could easily have been different. Suppose I am moody. I fall under the extension of ‘happy at noon, April 15, 2004 ’. Being moody, I could very easily have fallen under the extension of ‘not happy’. Thus, there are close worlds where the extension of ‘happy at noon, April 15, 2003’ is different to what it actually is. But that phenomenon obviously does not by itself indicate that ‘happy’ is semantically plastic. Now if ‘bald’ is semantically plastic in the way that Williamson envisages, that certainly means that, say,  ‘expresses’ (conceived of as a relation between a noise token and a property) is extensionally plastic. A noise token may in this world express a property P, baldness, and yet at a close world that noise token not express that property. Similarly, if we allow ‘c means p by noise type n’ to express a three place relation between a phonetic or graphemic type, a community and a  property, then the semantic plasticity of ‘bald’ will make for extensional plasticity with regard to that ternary predicate. Moreover, it is plausible enough to think that such extensional plasticity, coupled with our insensitivity to the ways that slight shifts in use make for differences in semantic value, will undermine the possibility of knowledge of the propositions expressed by the relevant metalinguistic claims in borderline cases. The words “’Baldness’ refers to the property of having fewer than 3,832 hairs on one’s scalp’ could easily have expressed a falsehood. And given our insensitivity to the shifts in semantic value, one who accepted that sentence would not plausibly express knowledge thereby. But none of this shows that semantic terms are themselves semantically plastic (as well as extensionally plastic). And thus none of this shows that ignorance in borderline cases can be traced in every case to semantic plasticity besetting some term figuring in the indeterminate claim.




We have seen that the considerations adduced by Williamson in the quoted paragraph do not demonstrate that semantic terms are semantically plastic. But is there any positive reason to think that they are not? Interestingly, the puzzle adduced earlier can be reproduced here. We generated havoc earlier by allowing our standards for what counts as a person to draw boundaries in different places to cognate terms (terms used at close worlds with almost indistinguishable conceptual roles) used by various counterfactual people themselves. What happens if we allow, say, the predicate ‘true’ to draw boundaries in a way that fail to match the boundaries drawn by people at close worlds who use a term with a conceptual role that bears the hallmark of our use of ‘true’? Havoc similarly results.

Thus let us suppose that ‘true’ is semantically plastic, so that its intension at close worlds differs from its actual intension. Here, I am thinking of ‘true’ as a predicate of utterance tokens, so that its semantic value will be a function from worlds to sets of utterance tokens.[10] Suppose then, that at a nearby world W the semantic value of ‘true’ was slightly different, so that each utterance of a particular sentence S by a particular community C fell under the extension of the semantic value expressed by our term ‘true’ (given W as argument), but did not fall under the extension of the semantic value of ‘true’ as used at the nearby world. Let us assume, as required by the Williamson picture, that at that world the use of ‘true’ is only ever so slightly different, so that the fundamental features of its conceptual role at our world – in particular the behavior gestured at by the “T-schema” – are intact.

Consider now an utterance of


‘S’ is true iff S


made by an inhabitant of the nearby world under consideration. By hypothesis the right hand side of the biconditional is true. How about the left hand side? By hypothesis, ‘S’ does not fall within the extension of ‘true’ as used by the community at that world. Thus, the community would be saying something false by the left hand side. Thus, if the community were to utter the relevant biconditional, the left hand side would be false, the right hand side true. The biconditional would be false. Assuming semantic plasticity, we have been led to conclude that at close worlds, certain counterparts of the T-schema are false! This seems just as bad as conceding that at close worlds people do not self-refer by ‘I’. Note in both cases we supposed that a certain conceptual role is accompanied by a certain semantic achievement: a pronoun with the conceptual role of ‘I’ has self reference, a predicate with the conceptual role of ‘true’ will yield true instances of the associated T-schema. In both cases, semantic plasticity induces a detachment between the relevant conceptual role and the associated semantic achievement.[11]

As in the case of ‘I’, one might try to soften the blow. “After all,” it may be said, “while counterparts of the T-schema are false at close worlds, they are true*, where the property of being true* is the property expressed by ‘true’ at close worlds.” But I take it that this is not satisfactory. Truth is the norm by while we evaluate both our actual and counterfactual selves. The response requires us to think that at nearby worlds truth doesn’t really matter. As such, it is not acceptable. (Consider an analogous conversation in ethics, where one tries to let one’s counterpart of the hook by combining a concession that he is cruel with the observation that he is not cruel*, where cruel* is what he means by ‘cruel’.)

Untoward results can also be reproduced for ‘refers’ ‘expresses’ ‘designates’ and so on. Suppose, say, that ‘refers’ is semantically plastic, so that while tokens of some counterfactual name n refer to x, the pair <n,x> does not fall under the extension of ‘refers’ as used at that (nearby) world. Consider now the claim


‘n’ refers to n


as used at that counterfactual world. That claim is true just case the pair picked out by the flanking singular terms falls under the extension of the binary predicate. The referent of ‘‘n’’ is the name itself. By hypothesis the referent of ‘n’ is x. By hypothesis the pair <n, x> does not fall under the extension of ‘refers’, as used by members of the counterfactual community under consideration. Thus certain instances of the “disquotational schema for reference” come out false at nearby counterfactual worlds. Once again, an intolerable result.

The lesson, I take it, is that we should be very cautious about positing semantic plasticity for semantic vocabulary. Not only does the quoted passage by Williamson fail to provide any reason for embracing it; there are powerful reasons for rejecting it.




Insofar as we are sympathetic to epistemicism, we are left with a residual problem. What exactly is distinctive of the ignorance due to vagueness?  I have argued that it is implausible that our ignorance concerning the boundaries of personhood can be traced to semantic plasticity. But this does not seem to be a good reason for denying that in some very reasonable sense, ‘person’ is vague. We should similarly allow that in some very reasonable sense, certain claims of the form


‘bald’ is true of people with less than N hairs


are borderline, even though the vagueness of such claims cannot be traced to the fact that certain terms occurring in them are semantically plastic. (Even if ‘bald’ is semantically plastic, that does not mean that ‘’bald’’ is.) Now we have noted, of course, that semantic terms may well be beset by extensional plasticity in borderline cases. But we cannot say that in general, ignorance due to extensional plasticity makes for the kind of ignorance associated with borderline cases. Suppose a particle moves rapidly between point A and B, so rapidly that we cannot in principle discern whether, at a given time, the particle is at A or B. Consider the claim ‘The particle is at A at noon’. There is extensional plasticity, sure enough. Suppose ‘is at A at noon’ is true of the particle. That predicate could easily have been false of the particle. But this case does not in any way have the feel of a case in which there is ignorance due to vagueness.

So let us reexamine the question as to what the epistemicist should say about the ignorance that is distinctive of borderline cases.  Let us begin with what might be called a picture of the epistemicist metaphysics of semantics. It would be very strange indeed to deny that semantical facts (and propositional attitude facts) supervene on a groundfloor comprised of a certain distribution of fundamental properties across space-time (which will be microphysical, assuming that some broad naturalistic picture is correct). The epistemicist is thus happy to believe that there is some sort of function from fundamental distributions to semantical facts. Call that function F. Meanwhile, semantical ignorance about a certain noise type may have at least two different sources. On the one hand, we may be ignorant of various facts about the groundfloor which serve as input to F. Such facts will, let us suppose (or, if you prefer, pretend), straightforwardly encode this or that fact about how the noise type is used some member of the community, by fellow members of the community, the causal relations of that noise type to this or that feature of the world and so on. Let us call this source of ignorance about semantic facts groundfloor ignorance. On the other hand, we may have an rather incomplete grasp of F itself, so that even if one were (idealizing now) to have a full grip on the array of fundamental facts, one would still not be in a position to discern the semantical facts on the grounds that one’s grasp of how the latter depends on the former is radically incomplete. Let us imagine that the nature of F could be captured by a set of semantical laws that describe how semantical facts depend on the groundfloor. Insofar as we didn’t know what the semantical laws were, we would have ignorance not traceable to groundfloor ignorance. Let us call the second kind of ignorance semantico-nomic ignorance.[12]

Now it is quite clear that in a borderline case, Williamson supposes the ignorance not merely to be rooted in groundfloor ignorance: even if one knew all of the relevant groundfloor facts, one would not be able to make the ignorance go away. Supposing that the groundfloor facts are captured by P and the relevant semantic facts about the extension of some predicate Q. The problem is not merely that we do not know that P. It is that we are in no position to know that P É Q, even though that material conditional is presumably a necessary truth. Groundfloor omniscience would not remove our insensitivity to the true semantic mechanisms.

Notice now that this picture provides a plausible epistemicist account of ignorance due to vagueness that does not proceed by way of semantic plasticity: in cases which we call “ignorance due to vagueness”, we have a sentence that expresses a proposition P such that our principled inability to know whether P that is rooted in semantico-nomic ignorance. Even if, say, some claim of the form


(1) Tokens of ‘big number’ as used by community C are true of any number great than 154


 are not semantically plastic, we may have a principled ignorance of their truth value that is rooted in semantico-nomic ignorance. Hence our ignorance of (1) will count as “ignorance due to vagueness”.

One might worry that the picture just sketched disrupts a safety-based conception of knowledge according to which belief is knowledge just in case there is no danger or error – that is, no error at “close worlds”.[13] Suppose someone dogmatically believed some claim S of the form (1) above. Clearly such a person would not know that S even if it were true. Williamson provides us with a vision of how semantic plasticity explains ignorance in borderline cases: Suppose someone were to dogmatically accept a borderline claim S. Even if S is true, then, owing to semantic plasticity, S would express a falsehood at “close worlds”. Thus, at close worlds, the dogmatist would make a mistake. His actual belief thus turns out not to be safe and so he The dogmatist does not know S. Eschew plasticity for S  and no similar explanation is available.  Someone who dogmatically accepted S would, it seems express a truth at close worlds and so, by a “safety” theoretic test, counts as knowing that S is true. 

Is this a real problem for the current brand of epistemicism? I don’t think so. To use the preceeding line of thought against that account is to presuppose an all too crude safety-theoretic account of knowledge (one that advocates of safety-based account – including Williamson – will be at pains to distance themselves from). We all know that if someone dogmatically cleaves to Goldbach’s conjecture, that will not in itself secure knowledge. But, given that there seems to be no semantic plasticity in the relevant mathematical language, such a dogmatist would not be in error at close worlds. In that case, we are hardly inclined to use a crude safety based account as grounds for admitting that the dogmatist knows Goldbach’s conjecture after all (assuming that it is true). We instead refine our conception of what knowledge comes to. Similar remarks apply to semantico-nomic ignorance, mutatis mutandis.




It is not clear that the semantic plasticity gloss on ignorance due to vagueness has to be jettisoned altogether. Return to


(1), viz:


 Tokens of ‘big number’ as used by community C are true of any number great than 154


I have been suggesting that (1) is not semantically plastic. But (1) is plausibly about a term that is semantically plastic. When it comes to evaluating sentences in our own language, semantical claims like (1), made about ourselves, will be evaluated by way of sentences that do not contain semantical vocabulary. Thus,


(2) ’154 is a big number’ is true


in my language, will be evaluated by me by way of


(3) 154 is a big number.


Suppose, as I have been suggesting (3) is semantically plastic but (2) isn’t. If (3) is vague, then so is (2) on the grounds that indefiniteness transmits across known equivalences.[14] We cannot, then, quite say that indefinite sentences are semantically plastic. But there is a thesis that is still arguably defensible, call it the Modified Plasticity Thesis, according to which, when a sentence in our own language is vague, the canonical means for evaluating it will be via a plastic sentence.

            Note, though, that if we adopt strategy B above for handling personal identity issues, then it is not clear that even the Modified Plasticity Thesis is correct. And that is because, on that view, ‘person’ is vague without being semantically plastic. On that view then, sentences like ‘Some person has Clean as a part’ are not semantically plastic.[15] Someone who adopted that strategy can endorse the view that vagueness turns on semantico-nomic ignorance, but cannot endorse even the Modified Plasticity Thesis.





Suppose I introduce a predicate ‘is a dommal’ by a pair of stipulations: Let ‘is a dommal’ be true of dogs and false of non-animals.[16] [17]My stipulations do not settle whether


(2) Cats are dommals


Williamson accepts bivalence here. On his preferred picture, the truth value is determined by a default principle, the two candidates being, roughly,


P1: A sentence is false unless one has done enough to secure its truth


P2: A sentence is true unless one has done enough to secure its falsehood.


Williamson seems to think that he knows which default principle is the true one[18], but such an epistemic stance does not seem very plausible to me. Do we really have access, a priori or otherwise, to the relative merits of P1 and P2? I reckon it better to combine bivalence with an admission of principled ignorance: we do not know (2) owing to semantico-nomic ignorance.

Imagine a Sorites sequences of cases in which, at one end,[19] a community are very firmly in favor of P1, and thus find ‘Cats are dommals’, as introduced by a member of their community, obviously false, and at the other end, a community that is very firmly in favor of P2, and thus finds ‘Cats are dommals’, as introduced by a member of that community, very obviously true. Imagine that we are somewhere in the middle, accounting for the borderline status of ‘Cats are dommals’ in our mouth. [20]

Do we at last have a case where it is plausible to think that the meaning of ‘true’ is intensionally shifty? Should we say that the community at one end means one thing by ‘true’ (governed by P1) and that the community at the other end means another thing by ‘true’ (governed by P2). Such a reaction would conflate intensional with extensional considerations. For it is altogether compelling here to think that each community ought to be deferential to the other. Suppose I am in the community that applauds P1. Suppose I look at a community that is all in favor of P2. Should I, using ‘false’ in my mouth, say that ‘Cats are dommals’ is false, even as uttered by the community of P2 followers? This seems like the wrong reaction. It seems much more natural to suppose that when I say ‘Cats are dommals’, that is false, and expresses one proposition, but when the community of P2 lovers says ‘Cats are dommals’, they express a quite different proposition. And that is because the practices of my community determine ‘dommal’ in my mouth to have the same intention as ‘dog’, but the practices of that community determine ‘dommal’ in their mouths to have the same intension as ‘animal’. Thus the community naturally thought of as P1 lovers should only endorse a principle like P1 when it is suitably restricted to their linguistic locale. No one in the Sorites sequence should think that P1, unrestricted, might be a necessary truth. On this conception, which I take to be the most plausible way of thinking about the case, the intension of ‘true’ is invariant along the Sorites sequence. (It is interesting to notice the structural affinities between this case and the discussion of deference above.)





Let us turn to the phenomenon of semantico-nomic ignorance itself. The picture is one according to which semantic mechanisms transcend our grasp of them in a deep and principled way. Some will find this deeply intolerable. It is interesting, here, to note a contrast between our attitudes towards mathematics and semantics. In the realm of mathematics, the view that there are evidence-transcendent features of this or that mathematical structure, while hotly debated, is not regarded as extreme or bizarre. Yet analogous views about semantics are apt to strike readers as somewhat outrageous. This reaction is at least in part rooted in a reluctance to recognize semantical properties as natural kinds, joints in nature with distinctive real essences. This “hypeinflationary” conception of semantical properties would not, of course, suffice to establish the current brand of epistemicism. But it would render the idea of semantico-nomic ignorance rather more palatable and thus help to make my favored version of epistemicism a going concern.

Let me thus offer a few preliminary motivating remarks in support of hyperinflationism.

 Consider first the following frequently voiced concern about epistemicism:


For any predicate, there are every so many functions from use to extension that “fit” the use of that predicate. What on earth could it be that makes one of those functions special in such a way that ‘true of’ should be specially associated with it? Shouldn’t we instead make every attempt to do justice to the though that each of the functions provides an equally good candidate extension?[21]


The concern needs refinement. Recalling Kripkenstein,[22] none of us (or hardly any of us) think that quus is an equally good candidate semantic value for ‘plus’ in the mouth of our earlier selves as plus (where plus and quus are functions that differ only with regard to pairs of natural numbers whose sum we are unable to entertain due to considerations of our finitude.) But both candidates “fit” use in some fairly obvious sense, since each interpretation is equally charitable with regard to our actual and counterfactual use of ‘plus’ (so long as suitable compensating adjustments are made in the interpretation of other pieces of arithmetical vocabulary in which generations about additions are stated.) The lesson generalizes to non-arithmetical vocabularly. Bizarre interpretations can be concocted to “fit” use which none of us are very inclined to think are acceptable interpretations.

            In response to all this, some will go the way of Kripkenstein’s skeptical solution,[23] combining a suitably disquotationalist story about our truth predicate with a recognition that there are no deep objective constraints on the acceptability of a translation. The Quine-Field development of this view would have us believe that an ascription of truth to some utterance made by my earlier self (or some interlocutor) has to be relativized to a translation scheme. Semantico-nomic ignorance will have no place in that framework. Indeed, the chasm between such theorists and the current brand of epistemicism is far too vast for me to hold out much hope of closing it here. (For many of us, it is cost enough for that view that it relinquishes all hope of salvaging straightforward truth for claims made by our earlier selves and our fellows.)

Of more interest to me here is the Lewisian reaction to Kripkenstein,[24] one which allows the distinction between natural and gerrymandered properties to do work in the foundations of semantics. Roughly speaking, the picture maintains there are two desiderata on interpretation, namely: (a) The Requirement of Charity:  ceteris paribus, interpret us so that our claims come out true so interpreted; (b) The Requirement of Eligibility: ceteris paribus, interpret us so that our predicates get assigned more rather than less natural properties as their semantic values.

On the Lewis picture, naturalness of reference is what explains there being a fact of the matter as to what something refers to: more specifically it is the comparative naturalness of one candidate over others that explains why there is a fact of the matter to the effect that a term refers to that candidate. What makes a plus-interpretation more acceptable than a quus interpretation? Well, while both interpretations may do equally well on the score of charity, one interpretation scores far higher with regard to eligibility. Thus the quus interpretation can be discounted. Consider by contrast the case of ‘bald’. Each “candidate” semantic value is, intuitively, equally natural. So neither charity nor eligibility can break the tie.

There is a second role that naturalness plays in the Lewisian account, viz: naturalness begets semantic stability. If we refer to a highly natural property by some term t, which more natural than properties in the vicinity, then the semantic value of t is will often remain stable despite quite significant shifts in use.

It seems as if the epistemicist has little to gain from a Lewisian distinction between natural and non-natural properties.[25]  Such reaction would be far to hasty, I think. Let me explore a few themes in that connection, making vivid those ways in which a Lewisian distinction between natural and non-natural properties can serve as a springboard for hyperinflationism.

(i) Suppose one were to embrace the Requirement of Eligibility, along with an objective distinction between natural and unnatural properties. This is to already to recognize the existence of deep principles about semantics that transcend the ken of ordinary folk. Perhaps one might think that we semantic theorists can appreciate the plausibility of such a requirement and even know that it is probably true. But it would be outrageous to suggest that ordinary linguistic competence brings with it knowledge of any such principle. To claim that such a principle is known “implicitly” by ordinary folk is to court further confusion: to claim that those principles that describe how terms refer are automatically known implicitly by people simply on account of their ability to refer is, in truth, no more plausibly than the claim that those principles that describe how we maintain our balance are automatically known implicitly by people simply on account of their ability to maintain their balance. The eligibility requirement does not govern the semantics of ordinary folk not by being known implicitly by them. Rather, it governs the semantics of ordinary folk (if it does so at all) by virtue of being a correct (if partial) account of the nature of semantical relations. Accepting the eligibility requirement is, obviously, not yet to accede to epistemicism. But to accept it is to embrace the existence of fundamental semantic mechanisms that are beyond the ken of ordinary folk, a move that should provide real encouragement indeed to the epistemicist.

(ii)The version of epistemicism I am interested in is best served by a metaphysic according to which semantic properties – reference, truth and so on – are themselves natural kinds, joints in nature. Where a property marks a natural kind, we are open to the thought that it has a “real essence” that transcends our ordinary understanding of it, even one that in some respects transcends our cognitive capacities. The fundamental metaphysical task of my epistemicist, then, is to render plausible the picture of semantic properties as joints in nature. Does the Lewisian metaphysics help or hinder in this way?

First, some preliminaries. Lewis embraces a plentitude of properties, some of which are metaphysically “haloed”, that is natural. More precisely, there is a continuum from more to less natural properties, with perfectly natural properties at one end, and increasing “gruesomeness” as one moves along the continuum. Which are the natural properties? Even supposing that we think that everything supervenes on physics, the issue is not settled. For if we accept a natural property framework, we must choose between an austere physicalism on the one hand and what might be called an “emergentist” framework on the other. According to the austere physicalist, the perfectly natural properties  will only be found at the microphysical groundfloor, relative naturalness being a matter of definitional distance from the perfectly natural properties: to calibrate the naturalness of a property, see how complicated the definition of that property would be in a “canonical” language in which each predicate corresponded to a perfectly natural property.[26] From such a perspective, the property of, say, being a chair will likely turn out hopelessly unnatural, far less natural than, say, the disjunctive property of being either a hydrogen atom or being fifteen feet from a quark. Indeed, it wouldn’t be surprising if, lacking functional predicates, the canonical definition of a chair was infinitary. The “emergentist” by contrast, believes that naturalness is not a matter of mere definitional distance from the microphysical groundfloor. Perhaps being a cat is far more natural than certain properties far more easily definable in Lewis’ canonical language. On the emergentist conception of things, there is no algorithm is available for calibrating naturalness in terms of a perfect microphysical language.

Now Lewis’ own development of the eligibility view certainly provides a hindrance to the picture of semantical joints – joints delineated by semantic predicates themselves, since his physicalism is an austere one. Though lacking the space to develop the point here, I suspect that we find hereabouts a fundamental tension in his world view: on the one hand he wishes the eligibility requirement to do dispel the specter of rampant indeterminacy presented by Kripkenstein and Quine. Yet on the other hand, he offers us an austere physicalist account of what eligibility comes to. It does not seem that the two perspectives can be reconciled. How can the eligibility requirement provide some reasonable measure of determinacy for ‘gavagai’ if the property of being a rabbit turns out to be hopeless gruesome? Far better, it seems to me, to opt for an emergentist physicalism, in which semantical joints remain a live option.

We can go further. As we have noted, the eligibility framework offers a useful perspective on the presence and absence of semantic plasticity. Suppose there is a highly natural property that distinguishes itself among the properties that reasonably well “fit” the use of a predicate. The other “candidates” are far less natural and so the highly natural property easily wins the semantic competition. Even if the use of the term had been slightly different, the highly natural property would win the competition, since even a slightly lower score vis a vis a gruesome property on the score of charity would be trumped by a far higher score in naturalness. Using language that has recently become popular: the highly natural property serves as a reference magnet.But we have seen above that semantical predicates are not semantically plastic. The reasonable conclusion seems to be that semantical properties are reference magnets and therefore highly natural themselves. Epistemicism is not yet forced upon us, but a suitable metaphysical underpinning for such a view – one replete with semantic magnets -- is now in place.

To sum up: I identified two themes in Lewis’ own use of natural properties in semantic theory. First, we found the idea that there being a fact of the matter about reference typically requires there being a highly eligible referent. The epistemicist metaphysic I am envisaging denies this. There is a fact of the matter concerning the reference of ‘bald’, but it is not explained by the naturalness of the referent of ‘bald’. There is, of course, a way in which one might still be able to pay lip service to the idea that reference to a particular property is begotten by naturalness. For there being a fact of the matter as to what ‘bald’ refers to is, on this picture, explained by the eminent naturalness of the reference relation itself.

Second, we found the idea that eligibility begets semantic stability. This idea can be taken on board, pretty much as it stands, by the current brand of epistemicism. ‘Bald’ is plastic, on account of the non-eligibility of its referent. ‘refers’ is stable, on account of the high eligibility of its referent.[27]





Do I suppose myself to have offered absolutely decisive arguments for the ‘magnet’ version of epistemicism over Williamson’s? I do not.

            Let me end by sketching what I think is the best strategy to pursue for one who repudiates semantic magnets and wishes to stick closely to Williamson’s original vision.

I will make a number of simplifying assumptions. Let us imagine that there is one community to a world, that sentences are worldbound, that each community speaks a language that is not context dependent, and that languages contain their own truth predicate.[28]

Let us distinguish domestic ascriptions of a close variant of ‘true’ – which are ascriptions of the predicate to sentences in one’s own language, from foreign ascriptions, which are ascriptions to sentences in a language that isn’t one’s own. Suppose communities C1 in alpha, and C2 in Beta, are close variants of each other. Let the home extension of a variant of the truth predicate be defined as follows:


Home extension: A sentence S is part of the home extension of a close variant v of the truth predicate iff S is in the language that v belongs to and v is true of S.


We can now state a thesis of domestic stability for ‘True’.


 ‘True’ (as we use it) is domestically stable just in case a sentence is part of the home extension of a close variant of ‘True’ iff ‘True’ is true of it.


Suppose, now that ‘True’ is domestically stable. At some close variant community C, if our predicate ‘True’ is true of a sentence S in the language of C, then the variant of ‘True’ in C will be true of  S as well. But this is not yet semantic stability. For it is quite compatible with domestic plasticity that the variant of ‘True’ that is in C is false of some sentence of our language that ‘True’ is true of.  Consider for example, the following scenario: there is a sentence of our language S1, a sentence of their language S2, such that: (i) The intension of our term ‘true’ includes both S1 and S2, but the intension of their variant of ‘true’ includes S2 but not S1 (and indeed that the intension of their variant of ‘false’ while not including S2 does include S1). This scenario is quite compatible with domestic stability but not with semantic stability. Now, crucially, domestic stability is enough to ensure that close variants of instances of the T-schema are true. So it turns out that one can save the original Williamson approach from the argument given earlier by combining domestic stability with semantic plasticity.


            Let ‘true*’ be the close variant of ‘true’, and assume the scenario just described. It is false that, for any variant of ‘true*’ something is part of the home extension of that variant iff ‘true*’ is true of it. After all ‘true’ is a close variant of ‘true*’ and while S1 is part of the home extension of ‘true’, ‘true*’ is not true of it.

Does this mean that while we would be quite correct in uttering the thesis of semantic stability, the nearby community would be making a mistake when uttering the counterpart of that thesis? And wouldn’t that make us incredibly lucky?  To think so is to miss out on the fact that ‘true of’ may not be semantically stable.

Let ‘true’, ‘true of’ name our expressions, ‘true*’ and ‘true of*’ a pair of close variants . Let being true of* be the relation that is meant by ‘true of*’. In the scenario described, ‘true’ is true of S1  and S2, ‘true*’ true of S2 but not S1. Supposing for simplicity that these are all the pertinent facts, we can say that ‘true’ is domestically stable but that a domestic stability thesis for ‘true*’ isn’t true.

Suppose, however that ‘true’ is true of* S1 but not S2, and that ‘true*’ is true of* S1 and S2.  Let some sentence S be part of the home extension* of a variant of ‘true*’ iff S is in the language of the variant and the variant is true* of it. Suppose the community of true* users utters the sentence that we use to express domestic plasticity. We can now allow that they would be saying something true (and not merely true*), since they would be saying that if something is a close variant of ‘true*’, then a sentence is part of the home extension* of that variant iff ‘true*’ is true* of it.[29] Each community gets to speak the truth by the variants of the domestic plasticity thesis owing to the fact that shifts in ‘true’ get compensating shifts in ‘true of’.

Similar considerations can be raised for a community with ‘true’ and ‘says that’ as their basic ideology. We certain want close variants of the schema:


if ‘s’ says that S then ‘s’ is true iff S


to have only true instances. If we now allow that ‘true’ is not domestically stable, then we are in trouble. For we certain want disquotational instances of ‘S’ says that S to be true at close variant communities. Suppose, then, that S2 is true but that ‘’S2’ is true*’ is false, and that S2 belongs to the language of ‘true*’. Then


If ‘S2’ says that S2 then (‘S2’ is true* iff S2’)


will have a true antecedent and false consequent. Assuming, then, that we want disquotational instances of ‘‘S’ says that S’ to come out true at close variant communities, we must think of ‘true’ as domestically stable.

But we do not have to think of ‘true’ as semantically stable. Let ‘Talking donkeys are impossible’ and ‘Talking donkeys are impossible*’ be S1 and S2 respectively.  Suppose further:


‘Talking donkeys are impossible*’ says that talking donkeys are impossible



Suppose  ‘Talking donkeys are impossible’ and ‘Talking donkeys are impossible*’ are both true. Compatibly with all this I can allow that the variant community to speak the truth by


‘Talking donkeys are impossible’ is false*




‘Talking donkeys are impossible*’ is true*



Suppose the following was true.


‘Talking donkeys are impossible’ says that* talking donkeys are impossible*


Then there will be a false instance of the schema:


If a sentence S says that* that P then S is true* iff P.


For given that ‘Talking donkeys is impossible*’ is true, and ‘‘Talking donkeys is impossible’ is  true*’ is false, then


‘If ‘Talking donkeys are impossible’ says that* talking donkeys are impossible*, then (‘Talking donkeys are impossible’ is true* iff talking donkeys are impossible)’


would contain a true antecedent and a false consequent. But we can allow the schema in question to have all and only true instances by allowing for compensating adjustments in the meaning of ‘says that’, and in particular that


‘‘Talking donkeys are impossible’ says that* talking donkeys are impossible*’


a sentence in the variant language, is false. Once again, we can perfectly well combine domestic stability with semantic plasticity.





I have, in effect, tried to develop two competing versions of epistemicism. It should be clear that I have a mild preference for the metaphysically inflationary version. But my main aim has been to map out the conceptual terrain, rather than to advance a view with any confidence. After all, the discussion will seem inevitably simpleminded once one takes stock of the literature on semantic paradox. It may turn out to be obligatory in that setting to  complicate the picture of a few simple semantic magnets, especially if one wishes for a treatment that is compatible with a commitment (for any given truth predicate) to bivalence.[30] I am also open to the suggestion that considerations of higher order vagueness may end up militating in favor of one approach over the other. That said, I hope that the discussion here will serve as a useful preliminary to future treatments of both semantic plasticity and epistemicism.  
















[1] I am grateful for conversions with Cian Dorr, Kit Fine, Hud Hudson, David Manley, Ted Sider, Ryan Wasserman, Peter Van Inwagen, Dean Zimmerman, audiences at Syracuse and Oxford, and especially Timothy Williamson.

[2] Even one goes against that aspect of usage and counts many of the objects sitting on the chair as a person, some of the issues that follow will arise. For even if there are many persons on the chair, there will presumably be certain objects for which it is vague whether or not those objects are persons at all.

[3] This is not to say, of course, that a person cannot radically misdescribe himself. He may think himself an immaterial being when in fact he is material. And so on. The point is that we must allow questions of who the person is to march in step with questions about what is picked out by “I” in the person’s mouth. Thus once we concede that, overall, the pattern of usage in the person’s mouth privileges x over y as the reference of ‘I’, it is no longer an option to nevertheless reckon y as the person/thinker/utterer.

[4] Williamson ref.

[5] Thanks to Mark Johnston, Ted Sider and Timothy Williamson for urging me to consider this strategy.

[6] ref

[7] Or at least until recently the main competitor. The contemporary landscape is more complicated, in no small measure thanks to some people here today.

[8] refs

[9] ‘Reply to Commentators’ PPR December 1997, Vol LVII No. 4, p. 947-8

[10] The points that I am making can, however, be recast in terms of other different semantical frameworks, including, for example, one that takes ‘true’ as a predicate of utterance types -- where an utterance type is individuated by a combination of phonemic/phonetic considerations and a specified community of users. 

[11] Some people have suggested to me a salvage that goes by way of reinterpreting ‘iff’ at close worlds. It does not seem plausible that logical operators are semantically plastic in the relevant respect. Moreover one can redo the puzzle so that instead of considering a statement ‘‘S’ is true iff S’, we consider an inference from ‘‘S’ is true’ to ‘S’ (where the paradoxical result is that at close worlds the inference is invalid).

[12] There may be other kinds of ignorance that do not fall into the two categories I have just described. Suppose that there are higher order natural kinds and fundamental but inscrutably principles about how they depended on the groundfloor. Suppose, for example, phenomenal properties are like that. Our ignorance about whether bats have a certain quale might not then depend either on groundfloor ignorance or ignorance about how semantical laws work. It would then depend on ignorance of some principles of psycho-physics that hold of necessity. (Thanks to Hartry Field here).

[13] I am grateful for conversations with Williamson here.

[14] Assuming that Definiteness satisfies the distribution principle of normal modal logic andthat knowing P implies definitely P, then it just won’t do to say (2) is definite but (3) is indefinite. For it is known by me that (2) implies (3). So it is definite that (2) implies (3). By distribution, Definitely (2) implies Definitely (3). By contraposition, Indefinitely (3) implies indefinitely (2).

[15] Thanks to Nico here.

[16] ref

[17] Another interesting kind of case to consider is where symmetric constraints are at work. Suppose I stipulate that F’s at to be dommals and Gs are to be gommals, that everything is to be either a dommal or a gommal, and that nothing is to be both a dommal and a gommal. Saying that in such cases the terms have no semantic value is dangerous, since similar phenomena may be prevalent in subtler form in natural language. In the pure case, I would suppose that the epistemicist should say that one of the stipulations fails to hold (it being unknowable which).  I shall not pursue the matter here.

[18] ref

[19] Thanks to Tim Williamson and a member of an Oxford audience for bringing Sorites series of this sort to my attention.

[20] An aside: Suppose I define a ‘tulower’ by the pair of stipulations: if something is a tulip it is a tulower, and if something is not a flower it is not a tulower. Consider now (2*) ~ Daisies are tulowers. Similar considerations as above incline us to the view that (2) is vague. However, it seems very clear that if P1 is in force, it applies to both (2) and (2*) and that if P2 is in force it applies to both (2) and (2*). It thus seems that we can know that the conjunction of (2) and (2*) is false, and hence it turns out that the conjunction of (2) and (2*) is determinately false. 

[21] The relevant notion of “fit” deployed in such arguments in typically left unexplained, but I take it that talk of an interpretation “fitting” use of some term is tantamount to a claim that some interpretation provides some reasonably charitable interpretation of our settled dispositions to use a term.

[22] ref

[23] ref

[24] ref

[25] Relatedly, Lewis would certainly have it that there is no uniquely best way to weight charity versus eligibility when it comes to assigning semantic value. If one interpretation scores slightly better on charity, another slightly better on eligibility, then both will likely stand as acceptable interpretations that can be supervaluated over.

[26] ref

[27] Suppose one says that ‘true’, as a predicate of utterances, is semantically stable. Here is a possible problem. I write down a string of marks. There is a question of the boundaries of the utterance. Suppose some bit of ink, call it Ink, is such that it is borderline whether or not it is part of the utterance. It seems that it may be that while ‘true’ in my mouth expresses a property that applies to a thing with Ink, a nearby community may go a different way. The problem is not obviously solved by shifting to a picture according to which the fundamental semantical relations concern mental representations, since their boundaries can be vague as well. The concern is not decisive. The earlier discussion is not irrelevant. For example, think back to strategy C above. That can be replayed here. Perhaps the vagueness in this case turns on the vagueness of ‘part of’. In that case, there may be no vagueness concerning which object ‘true’ applies to, only vagueness concerning which are the parts of that object.


[28] If these assumptions are relaxed, the relevant points can still be made, but are a little harder to both state and see.

[29] In the toy scenario described, something is a close variant of ‘true*’ iff it is a close variant of ‘true*’. So nothing is being exploited along that axis. (Though that is not to say in other cases, extra complications will arise via that fact that the close variant relation is not transitive.).

[30] Though that is still not to say that one need to allow for some variants ‘true’ to be false of a sentence that ‘true’ is true of. One might thus complicate the picture by (at a first pass) allowing for a range of ‘truth’ magnets of increasingly wider intensions, but none with competing intensions (in the sense that the intension  of some variant of ‘true’ never includes a possible sentence that is also delivered by the intension of some variant of ‘false’.)