Category Archives: Logic

“Supervaluationism”: the word

I’ve got progressively more confused over the years about the word “supervaluations”. It seems lots of people use it in slightly different ways. I’m going to set out my understanding of some of the issues, but I’m very happy to be contradicted—I’m really in search of information here.

The first occurrence I know of is van Fraassen’s treatment of empty names in a 1960’s JP article. IIRC, the view there is that language comes with a partial intended interpretation function, specifying the references of non-empty names. When figuring out what is true in the language, we
look at what is true on all the full interpretations that extend the intended partial interpretation. And the result is that “Zeus is blue” will come out neither true nor false, because on some completions of the intended interpretation the empty name”Zeus” will designate a blue object, and others he won’t.

So that gives us one meaning of a “supervaluation”: a certain technique for defining truth simpliciter out of the model-theoretic notions of truth-relative-to-an-index. It also, so far as I can see, closes off the question of how truth and “supertruth” (=truth on all completions) relate. Supervaluationism, in this original sense, just is the thesis that truth simpliciter should be defined as truth-on-all-interpretations. (Of course, one could argue against supervaluationism in this sense by arguing against the identification; and one could also consistently with this position argue for the ambiguity view that “truth” is ambiguous between supertruth and some other notion—but what’s not open is to be a supervaluationist and deny that supertruth is truth in any sense.)

Notice that there’s nothing in the use of supervaluations in this sense that enforces any connection to “semantic theories of vagueness”. But the technique is obviously suggestive of applications to indeterminacy. So, for example, Thomason in 1970 uses the technique within an “open future” semantics. The idea there is that the future is open between a number of currently-possible histories. And what is true about is what happens on all these histories.

In 1975, Kit Fine published a big and technically sophisticated article mapping out a view of vagueness arising from partially assigned meanings, that used among other things supervaluational techniques. Roughly, the basic move was to assign each predicate with an extension (the set of things to which it definitely applies) and an anti-extension (the set of things to which it definitely doesn’t apply). An interpretation is “admissible” only if it assigns an set of objects to a predicate that is a superset of the extension, and which doesn’t overlap the anti-extension. There are other constraints on admissibility too: so-called “penumbral connections” have to be respected.

Now, Fine does lots of clever stuff with this basic setup, and explores many options (particularly in dealing with “higher-order” vagueness). But one thing that’s been very influential in the folklore is the idea that based on the sort of factors just given, we can get our hands on a set of “admissible” fully precise classical interpretations of the language.

Now the supervaluationist way of working with this would tell you that truth=truth on each admissible interpretation, and falsity=falsity on all such interpretations. But one needn’t be a supervaluationist in this sense to be interested in all the interesting technologies that Fine introduces, or the distinctive way of thinking about semantic indecision he introduces. The supervaluational bit of all this refers only to one stage of the whole process—the step from identifying a set of admissible interpretations to the definition of truth simpliciter.

However, “supervaluationism” has often, I think, been identified with the whole Finean programme. In the context of theories of vagueness, for example, it is often used to refer to the idea that vagueness or indeterminacy arises as a matter of some kind of unsettledness as to what precise extensions are expressions pick out (“semantic indecision”). But even if the topic is indeterminacy, the association with *semantic indecision* wasn’t part of the original conception of supervaluations—Thomason’s use of them in his account of indeterminacy about future contingents illustrates that.

If one understands “supervaluationism” as tied up with the idea of semantic indecision theories of vagueness, then it does become a live issue whether one should identify truth with truth on all admissible interpretations (Fine himself raises this issue). One might think that the philosophically motivated semantic machinery of partial interpretations, penumbral connections and admissible interpretations is best supplemented by a definition of truth in the way that the original VF-supervaluationists favoured. Or one might think that truth-talk should be handled differently, and that the status of “being true on all admissible assignments” shouldn’t be identified with truth simpliciter (say because the disquotational schemes fail).

If you think that the latter is the way to go, you can be a “supervaluationist” in the sense of favouring a semantic indecision theory of vagueness elaborated along Kit Fine’s lines, without being a supervaluationist in the sense of using Van Fraassen’s techniques.

So we’ve got at least these two disambiguations of “supervaluationism”, potentially cross-cutting:

(A) Formal supervaluationism: the view that truth=truth on each of a range of relevant interpretations (e.g. truth on all admissible interpretations (Fine); on all completions (Van Fraassen); or on all histories (Thomason)).
(B) Semantic indeterminacy supervaluationism: the view that (semantic) indeterminacy is a matter of semantic indecision: there being a range of classical interpretations of the language, which, all-in, have equal claim to be the right one.

A couple of comments on each. (A) of course, needs to be tightened up in each case by saying which are the relevant range of classical interpretations quantified over. Notice that a standard way of defining truth in logic books is actually supervaluationist in this sense. Because if you define what it is for a formula “p” to be true as it being true relative to all variable assignments, then open formulae which vary in truth value from variable-assignment to variable assignment end up exactly analogous to formulae like “Zeus is blue” in Van Fraassen’s setting: they will be neither true nor false.

Even when it’s clear we’re talking about supervaluationism in the sense of (B), there’s continuing ambiguity. Kit Fine’s article is incredibly rich, and as mentioned above, both philosophically and technically he goes far beyond the minimal idea that semantic vagueness has something to do with the meaning-fixing facts not settling on a single classical interpretation.

So there’s room for an understanding of “supervaluationism” in the semantic-indecision sense that is also minimal, and which does not commit itself to Fine’s ideas about partial interpretations, conceptual truths as “penumbral constraints” etc. David Lewis in “Many but also one”, as I read him, has this more minimal understanding of the semantic indecision view—I guess it goes back to Hartry Field’s material on inscrutability and indeterminacy and “partial reference” in the early 1970’s, and Lewis’s own brief comments on related ideas in his (1969).

So even if your understanding of “supervaluationism” is the (B)-sense, and we’re thinking only in terms of semantic indeterminacy, then you still owe elaboration of whether you’re thinking of a minimal “semantic indecision” notion a la Lewis, or the far richer elaboration of that view inspired by Fine. Once you’ve settled this, you can go on to say whether or not you’re a supervaluationist in the formal, (A)-sense—and that’s the debate in the vagueness literature over whether truth should be identified with supertruth.

Finally, there’s the question of whether the “semantic indecision” view (B), should be spelled out in semantic or metasemantic terms. One possible view has the meaning-fixing facts picking out not a single interpretation, but a great range of them, which collectively play the role of “semantic value” of the term. That’s a semantic or “first-level” (in Matti Eklund‘s terminology) view of semantic indeterminacy. Another possible view has the meaning-fixing facts trying to fix on a single interpretation which will give the unique semantic value of each term in the language, but it being unsettled which one they favour. That’s a metasemantic or “second-level” view of the case.

If you want to complain that second view is spelled out quite metaphorically, I’ve some sympathy (I think at least in some settings it can be spelled out a bit more tightly). One might also want to press the case that the distinction between semantic and metasemantic here is somewhat terminological—what we choose to label the facts “semantic” or not. Again, I think there might be something to this. There are also questions about how this relates to the earlier distinctions—it’s quite natural to think of Fine’s elaboration as being a paradigmatically semantic (rather than metasemantic) conception of semantic supervaluationism. It’s also quite natural to take the metasemantic idea to go with a conception that is non-supervaluational in the (A) sense. (Perhaps the Lewis-style “semantic indecision” rhetoric might be taken to suggest a metasemantic reading all along, in which way it is not a good way to cash out what’s the common ground among (B)-theorists is). But there’s room for a lot of debate and negotiation on these and similar points.

Now all this is very confusing to me, and I’m sure I’ve used the terminology confusingly in the past. It kind of seems to me that ideally, we’d go back to using “supervaluationism” in the (A) sense (on which truth=supertruth is analytic of the notion); and that we’d then talk of “semantic indecision” views of vagueness of various forms, with its formal representation stretching from the minimal Lewis version to the rich Fine elaboration, and its semantic/metasemantic status specified. In any case, by depriving ourselves of commonly used terminology, we’d force ourselves to spell out exactly what the subject matter we’re discussing is.

As I say, I’m not sure I’ve got the history straight, so I’d welcome comments and corrections.

Logically good inference and the rest

From time to time in my papers, the putative epistemological significance of logically good inference has been cropping up. I’ve been recently trying to think a bit more systematically about the issues involved.

Some terminology. Suppose that the argument “A therefore B” is logically valid. Then I’ll say that reasoning from “A” is true, to “B” is true, is logically good. Two caveats (1) the logical goodness of a piece of reasoning from X to Y doesn’t mean that, all things considered, it’s ok to infer Y from X. At best, the case is pro tanto: if Y were a contradiction, for example, all things considered you should give up X rather than come to believe Y; (2) I think the validity of an argument-type won’t in the end be sufficient for for the logically goodness of a token inference of that type—partly because we probably need to tie it much closer to deductive moves, partially because of worries about the different contexts in play with any given token inference. But let me just ignore those issues for now.

I’m going to blur use-mention a bit by classifying material-mode inferences from A to B (rather than: “A” is true to “B” is true”) as logically good in the same circumstances. I’ll also call a piece of reasoning from A to B “modally good” if A entails B, and “a priori good” if it’s a priori that if A then B (nb: material conditional). If it’s a priori that A entails B, I’ll call it “a priori modally good”.

Suppose now we perform a competent deduction of B from A. What I’m interested in is whether the fact that the inference is logically good is something that we should pay attention to in our epistemological story about what’s going on. You might think this isn’t forced on us. For (arguably: see below) whenever an inference is logically good, it’s also modally and a priori good. So—the thought runs—for all we’ve said we could have an epistemology that just looks at whether inferences are modally/a priori good, and simply sets aside questions of logical goodness. If so, logical goodness may not be epistemically interesting as such.

(That’s obviously a bit quick: it might be that you can’t just stop with declaring something a priori good; rather, any a priori good inference falls into one of a number of subcases, one of which is the class of logically good inferences, and that the real epistemic story proceeds at the level of the “thickly” described subcases. But let’s set that sort of issue aside).

Are there reasons to think competent deduction/logically good inference is an especially interesting epistemological category of inference?

One obvious reason to refuse to subsume logically good inference within modally good inferences (for example) is if you thought that some logically good inferences aren’t necessarily truth-preserving. There’s a precedent for that thought: Kaplan argues in “Demonstratives” that “I am here now” is a logical validity, but isn’t necessarily true. If that’s the case, then logically good inferences won’t be a subclass of the modally good ones, and so the attempt to talk only about the modally good inferences would just miss some of the cases.

I’m not aware of persuasive examples of logically good inferences that aren’t a priori good. And I’m not persuaded that the Kaplanian classification is the right one. So let’s suppose pro tem that the logically good inference are always modally, a priori, and a priori modally, good.

We’re left with the following situation: the logically good inferences are a subclass of inferences that are also fall under other “good” categories. In a particular case where we come to believe B on the basis of A, where is the particular need to talk about its logical “goodness”, rather than simply about its modal, a priori or whatever goodness?

To make things a little more concrete: suppose that our story about what makes a modally good inference good is that it’s ultra-reliable. Then, since we’re supposing all logically good inferences are modally good, just from their modal goodness, we’re going to get that they’re ultra-reliable. It’s not so clear that epistemologically, we need say any more. (Of course, their logical goodness might explain *why* they’re reliable: but that’s not clearly an *epistemic* explanation, any more than is the biophysical story about perception’s reliability.)

So long as we’re focusing on cases where we deploy reasoning directly, to move from something we believe to something else we believe, I’m not sure how to get traction on this issue (at least, not in such an abstract setting: I’m sure we could fight on the details if they are filled out.). But even in this abstract setting, I do think we can see that the idea just sketched ignores one vital role that logically good reasoning plays: namely, reasoning under a supposition in the course of an indirect proof.

Familiar cases: If reasoning from A to B is logically good, then it’s ok to believe (various) conditional(s) “if A, B”. If reasoning from A1 to B is logically good, and reasoning from A2 to B is logically good, then inferring B from the disjunction A1vA2 is ok. If reasoning from A to a contradiction is logically good, then inferring not-A is good. If reasoning from A to B is logically good, then reasoning from A&C to B is good.

What’s important about these sort of deployments is that if you replace “logically good” by some wider epistemological category of ok reasoning, you’ll be in danger of losing these patterns.

Suppose, for example, that there are “deeply contingent a priori truths”. One schematic example that John Hawthorne offers is the material conditional “My experiences are of kind H > theory T of the world is true”. The idea here is that the experiences specified should be the kind that lead to T via inference to the best explanation. Of course, this’ll be a case where the a priori goodness doesn’t give us modal goodness: it could be that my experiences are H but the world is such that ~T. Nevertheless, I think there’s a pretty strong case that in suitable settings inferring T from H will be (defeasibly but) a priori good.

Now suppose that the correct theory of the world isn’t T, and I don’t undergo experiences H. Consider the counterfactual “were my experiences to be H, theory T would be true”. There’s no reason at all to think this counterfactual would be true in the specified circumstances: it may well be that, given the actual world meets description T*, the closest world where my experiences are H is still an approximately T*-world rather than a T-world. E.g. the nearest world where various tests for general relativity come back negative may well be a world where general relativity is still the right theory, but it’s effects aren’t detectable on the kind of scales initially tested (that’s just a for-instance: I’m sure better cases could be constructed).

Here’s another illustration of the worry. Granted, reasoning from H to T seems a priori. But reasoning from H+X to T seems terrible, for a variety of X. (So: My experiences are of H + my experiences are misleading in way W will plausibly a priori supports some T’ incompatible with T). But if we were allowed to use a priori good reasoning in indirect proofs, then we could simply argue from H+X to H, and thence (a priori) to T, overall getting an a priori route from H+X to T. the moral is that we can’t treat a priori good pieces of reasoning as “lemmas” that we can rely on under the scope of whatever suppositions we like. A priori goodness threatens to be “non-monotonic”: which is fine on its own, but I think does show quite clearly that it can completely crash when we try to make it play a role designed for logical goodness.

This sort of problem isn’t a surprise: the reliability of indirect proofs is going to get *more problematic* the more inclusive the reasoning in play is. Suppose the indirect reasoning says that whenever reasoning of type R is good, one can infer C. The more pieces of reasoning count as “good”, the more potential there is to come into conflict with the rule, because there’s simply more cases of reasoning that are potential counterexamples.

Of course, a priori goodness is just one of the inferential virtues mentioned earlier: modal goodness is another; and a priori modal goodness a third. Modal goodness already looks a bit implausible as an attempt to capture the epistemic status of deduction: it doesn’t seem all that plausible to classify the inferential move from A and B to B as w the same category as the move from this is water to this is H2O. Moreover, we’ll again have trouble with conditional proof: this time for indicative conditionals. Intuitively, and (I’m independently convinced) actually, the indicative conditional “if the watery stuff around here is XYZ, then water is H2O” is false. But the inferential move from the antecedent to the consequent is modally good.

Of the options mentioned, this leaves a priori modal goodness. The hope would be that this’ll cut out the cases of modally good inference that cause trouble (those based around a posteriori necessities). Will this help?

I don’t think so: I think the problems for a priori goodness resurface here. if the move from H to T is a priori good, then it seems that the move from Actually(H) to Actually(T) should equally be a priori good. But in a wide variety of cases, this inference will also be modally good (all cases except H&~T ones). But just as before, thinking that this piece of reasoning preserves its status in indirect proofs gives us very bad results: e.g. that there’s an a priori route from Actually(H) and Actually(X) to Actually (T), which for suitably chosen X looks really bad.

Anyway, of course there’s wriggle room here, and I’m sure a suitably ingenious defender of one of these positions could spin a story (and I’d be genuinely interested in hearing it). But my main interest is just to block the dialectical maneuver that says: well, all logically good inferences are X-good ones, so we can get everything we want having a decent epistemology of X-good inferences. The cases of indirect reasoning I think show that the *limitations* on what inferences are logically good can be epistemologically central: and anyone wanting to ignore logic better have a story to tell about how their story plays in these cases.

[NB: one kind of good inference I haven’t talked about is that backed by what 2-dimensionalists might call “1-necessary truth preservation”: I.e. truth preservation at every centred world considered as actual. I’ve got no guarantees to offer that this notion won’t run into problems, but I haven’t as yet constructed cases against it. Happily, for my purposes, logically good inference and this sort of 1-modally good inference give rise to the same issues, so if I had to concede that this was a viable epistemological category for subsuming logically good inference, it wouldn’t substantially effect my wider project.]

CEM journalism

The literature on the linguistics/philosophy interface on conditionals is full of excellent stuff. Here’s just one nice thing we get. (Directly drawn from a paper by von Fintel and Iatridou). Nothing here is due to me: but it’s something I want to put down so I don’t forget it, since it looks like it’ll be useful all over the place. Think of what follows as a bit of journalism.

Here’s a general puzzle for people who like “iffy” analyses of conditionals.

  • No student passes if they goof off.

The obvious first-pass regimentation is:

  • [No x: x is a student](if x goofs off, x passes)

But for a wide variety of accounts, this’ll give you the wrong truth-conditions. E.g. if you read “if” as a material conditional, you’ll get it coming out true if all the students goof and succeed! What is wanted, as Higgenbotham urges, is something with the effect:

  • [No x: x is a student](x goofs off and x passes)

This seems to suggest that under some embeddings “if” expresses conjunction! But that’s hardly what a believer in the iffness of if wants.

What the paper cited above notes is that so long as we’ve got CEM, we won’t go wrong. For [No x:Fx]Gx is equivalent to [All x:Fx]~Gx. And where G is the conditional “if x goofs off, x passes”, the negated conditional “not: if x goofs off, x passes” is equivalent to “if x goofs off, x doesn’t pass” if we have the relevant instance of conditional excluded middle. What we wind up with is an equivalence between the obvious first-pass regimentation and:

  • [All x: x is a student](if x goofs off, x won’t pass).

And this seems to get the right results. What it *doesn’t* automatically get us is an equivalence to the Higgenbotham regimentation in terms of a conjunction (nor with the Kratzer restrictor analysis). And maybe when we look at the data more generally, we’ll can get some traction on which of these theories best fits with usage.

Suppose we’re convinced by this that we need the relevant instances of CEM. There remains a question of *how* to secure these instances. The suggestion in the paper is that rules governing legitimate contexts for conditionals give us the result (paired with a contextually shifty strict conditional account of conditionals). An obvious alternative is to hard-wire in CEM into the semantics, as Stalnaker does. So unless you’re prepared (with von Fintel, Gillies et al) to defend in detail fine-tuned shiftiness of the contexts in which conditionals can be uttered then it looks like you should smile upon the Stalnaker analysis.

[Update: It’s interesting to think how this would look as an argument for (instances of) CEM.

Premise 1: The following are equivalent:
A. No student will pass if she goofs off
B. Every student will fail to pass if she goofs off

Premise 2: A and B can be regimented respectively as follows:
A*. [No x: student x](if x goofs off, x passes)
B*. [Every x: student x](if x goofs off, ~x passes)

Premise 3: [No x: Fx]Gx is equivalent to [Every x: Fx]~Gx

Premise 4: if [Every x: Fx]Hx is equivalent to [Every x: Fx]Ix, then Hx is equivalent to Ix.

We argue as follows. By an instance of premise 3, A* is equivalent to:

C*. [Every x: student x] not(if x goofs off, x passes)

But C* is equivalent to A*, which is equivalent to A (premise 2) which is equivalent to B (premise 1) which is equivalent to B* (premise 2). So C* is equivalent to B*.

But this equivalence is of the form of the antecedent of premise 4, so we get:

(Neg/Cond instances) ~(if x goofs off, x passes) iff if x goofs off, ~x passes.

And we quickly get from the law of excluded middle and a bit of logic:

(CEM instances) (if x goofs off, x passes) or (if x goofs off, ~ x passes). QED.

The present version is phrased in terms of indicative conditionals. But it looks like parallel arguments can be run for CEM for counterfactuals (Thanks to Richard Woodward for asking about this). For one of the controversial cases, for example, the basic premise will be that the following are equivalent:

D. No coin would have landed heads, if it had been flipped.
E. Every coin would have landed tails, if it had been flipped.

This looks pretty good, so the argument can run just as before.]

Must, Might and Moore.

I’ve just been enjoying reading a paper by Thony Gillies. One thing that’s very striking is the dilemma he poses—quite generally—for “iffy” accounts of “if” (i.e. accounts that see English “if” as expressing a sentential connective, pace Kratzer’s restrictor account).

The dilemma is constructed around finding a story that handles the interaction between modals and conditionals. The prima facie data is that the following pairs are equivalent:

  • If p, must be q
  • If p, q


  • If p, might be q
  • Might be (p&q)

The dilemma proceeds by first looking at whether you want to say that the modals scope over the conditional or vice versa, and then (on the view where the modal is wide-scoped) looking into the details of how the “if” is supposed to work and showing that one or other of the pairs come out inequivalent. The suggestion in the paper is if we have the right theory of context-shiftiness, and narrow-scope the modals, then we can be faithful to the data. I don’t want to take issue with that positive proposal. I’m just a bit worried about the alleged data itself.

It’s a really familiar tactic, when presented with a putative equivalence that causes trouble for your favourite theory, to say that the pairs aren’t equivalent at all, but can be “reasonably inferred” from each other (think of various ways of explaining away “or-to-if” inferences). But taken cold such pragmatic explanations can look a bit ad hoc.

So it’d be nice if we could find independent motivation for the inequivalence we need. In a related setting, Bob Stalnaker uses the acceptability of Moorean-patterns to do this job. To me, the Stalnaker point seems to bear directly on the Gillies dilemma above.

Before we even consider conditionals, notice that “p but it might be that not p” sounds terrible. Attractive story: this is because you shouldn’t assert something unless you know it to be true; and to say that p might not be the case is (inter alia) to deny you know it. One way of bringing out the pretty obviously pragmatic nature of the tension in uttering the conjunction here is to note that asserting the following sort of thing looks much much better:

  • it might be that not p; but I believe that p

(“I might miss the train; but I believe I’ll just make it”). The point is that whereas asserting “p” is appropriate only if you know that p, asserting “I believe that p” (arguably) is appropriate even if you know you don’t know it. So looking at these conjunctions and figuring out whether they sound “Moorean” seems like a nice way of filtering out some of the noise generated by knowledge-rules for assertion.

(I can sometimes still hear a little tension in the example: what are you doing believing that you’ll catch the train if you know you might not? But for me this goes away if we replace “I believe that” with “I’m confident that” (which still, in vanilla cases, gives you Moorean phenomena). I think in the examples to be given below, residual tension can be eliminated in the same way. The folks who work on norms of assertion I’m sure have explored this sort of territory lots.)

That’s the prototypical case. Let’s move on to examples where there are more moving parts. David Lewis famously alleged that the following pair are equivalent:

  • it’s not the case that: if were the case that p, it would have been that q
  • if were that p, it might have been that ~q

Stalnaker thinks that this is wrong, since instances of the following sound ok:

  • if it were that p, it might have been that not q; but I believe if it were that p it would have been that q.

Consider for example: “if I’d left only 5 mins to walk down the hill, (of course!) I might have missed the train; but I believe that, even if I’d only left 5 mins, I’d have caught it. ” That sounds totally fine to me. There’s a few decorations to that speech (“even” “of course” “only”). But I think the general pattern here is robust, once we fill in the background context. Stalnaker thinks this cuts against Lewis, since if mights and woulds were obvious contradictories, then the latter speech would be straightforwardly equivalent to something of the form “A and I don’t believe that A”. But things like that sounds terrible, in a way that the speech above doesn’t.

We find pretty much the same cases for “must” and indicative “if”.

  • It’s not true that if p, then it must be that q; but I believe that if p, q.

(“it’s not true that if Gerry is at the party, Jill must be too—Jill sometimes gets called away unexpectedly by her work. But nevertheless I believe that if Gerry’s there, Jill’s there.”). Again, this sounds ok to me; but if the bare conditional and the must-conditional were straightforwardly equivalent, surely this should sound terrible.

These sorts of patterns make me very suspicious of claims that “if p, must q” and “if p, q” are equivalent, just as the analogous patterns make me suspicious of the Lewis idea that “if p, might ~q” and “if p, q” are contradictories when the “if” is subjunctive. So I’m thinking the horns of Gillies’ dilemma aren’t equal: denying the must conditional/bare conditional equivalence is independently motivated.

None of this is meant to undermine the positive theory that Thony Gillies is presenting in the paper: his way of accounting for lots of the data looks super-interesting, and I’ve got no reason to suppose his positive story won’t have a story about everything I’ve said here. I’m just wondering whether the dilemma that frames the debate should suck us in.

Degrees of belief and supervaluations

Suppose you’ve got an argument with one premise and one conclusion, and you think its valid. Call the premise p and the conclusion q. Plausibly, constraints on rational belief follow: in particular, you can’t rationally have a lesser degree of belief in q than you have in p.

The natural generalization of this to multi-premise cases is that if p1…pn|-q, then your degree of disbelief in q can’t rationally exceed the sum of your degrees of disbelief in the premises.

FWIW, there’s a natural generalization to the multi-conclusion case too (a multi-conclusion argument is valid, roughly, if the truth of all the premises secures the truth of at least one conclusion). If p1…pn|-q1…qm, then the sum of your degrees of disbelief in the conclusions can’t rationally exceed the sum of your degrees of disbelief in the premises.

What I’m interested in at the moment is to what extent this sort of connection can be extended to non-classical settings. In particular (and connected with the last post) I’m interested in what the supervaluationist should think about all this.

There’s a fundamental choice to be made at the get-go. Do we think that “degrees of belief” in sentences of a vague language can be represented by a standard classical probability function? Or do we need to be a bit more devious?

Let’s take a simple case. Construct the artificial predicate B(x), so that numbers less than 5 satisfy B, and numbers greater than5 fail to satisfy it. We’ll suppose that it is indeterminate whether 5 itself is B, and that supervaluationism gives the right way to model this.

First observation. It’s generally accepted that for the standard supervaluationist

p &~Det(p)|-absurdity;

Given this and the constraints on rational credence mentioned earlier, we’d have to conclude that my credence in B(5)&~Det(B(5)) must be 0. I have credence 0 in absurdity; and the degree of disbelief in the conclusion of this valid argument (1) must not exceed the degree of disbelief in its premises.

Let’s think that through. Notice that in this case, my credence in ~Det(B(5)) can be taken to be 1. So given minimal assumptions about the logic of credences, my credence in B(5) must be 0.

A parallel argument running from ~B(5)&~Det(~B(5))|-absurdity gives us that my credence in ~B(5) must be 0.

Moreover, supervaluational entails all classical tautologies. So in particular we have the validity: |-B(5)v~B(5). The standard constraint in this case tells us that rational credence in this disjunction must be 1. And so, we have a disjunction in which we have credence 1, each disjunct of which we have credence 0 in. (Compare the standard observation that supervaluational disjunctions can be non-prime: the disjunction can be true when neither disjunct is).

This is a fairly direct argument that something non-classical has to be going on with the probability calculus. One move at this point is to consider Shafer functions (which I know little about: but see here). Now maybe that works out nicely, maybe it doesn’t. But I find it kinda interesting that the little constraint on validity and credences gets us so quickly into a position where something like this is needed if the constraint is to work. It also gives us a recipe for arguing against standard supervaluationism: argue against the Shafer-function like behaviour in our degrees of belief, and you’ll ipso facto have an argument against supervaluationism. For this, the probablistic constraint on validity is needed (as far as I can see): for its this that makes the distinctive features mandatory.

I’d like to connect this to two other issues I’ve been working on. One is the paper on the logic of supervaluationism cited below. The key thing here is that it raises the prospect of p&~Dp|-absurdity not holding, even for your standard “truth=supertruth” supervaluationist. If that works, the key premise of the argument that forces you to have degree of belief 0 in both an indeterminate sentence ‘p’ and its negation goes missing.

Maybe we can replace it by some other argument. If you read “D” as “it is true that…” as the standard supervaluationist encourages you to, then “p&~Dp” should be read “p&it is not true that p”. And perhaps that sounds to you just like an analytic falsity (it sure sounded to me that way); and analytic falsities are the sorts of things one should paradigmatically have degree of belief 0 in.

But here’s another observation that might give you pause (I owe this point to discussions with Peter Simons and John Hawthorne). Suppose p is indeterminate. Then we have ~Dp&~D~p. And given supervaluationism’s conservativism, we also have pv~p. So by a bit of jiggery-pokery, we’ll get (p&~Dp v ~p&~D~p). But in moods where I’m hyped up thinking that “p&~Dp” is analytically false and terrible, I’m equally worried by this disjunction. But that suggests that the source of my intuitive repulsion here isn’t the sort of thing that the standard supervaluationist should be buying. Of course, the friend of Shafer functions could just say that this is another case where our credence in the disjunction is 1 while our credences in each disjunct is 0. That seems dialectically stable to me: after all, they’ll have *independent* reason for thinking that p&~Dp should have credence 0. All I want to insist is that the “it sounds really terrible” reason for assigning p&~Dp credence 0 looks like it overgeneralizes, and so should be distrusted.

I also think that if we set aside truth-talk, there’s some plausibility in the claim that “p&~Dp” should get non-zero credence. Suppose you’re initially in a mindset where you should be about half-confident of a borderline case. Well, one thing that you absolutely want to say about borderline cases is that they’re neither true nor false. So why shouldn’t you be at least half-confident in the combination of these?

And yet, and yet… there’s the fundamental implausibility of “p&it’s not true that p” (the standard supervaluationist’s reading of “p&~Dp”) having anything other than credence 0. But ex hypothesi, we’ve lost the standard positive argument for that claim. So we’re left, I think, with the bare intuition. But it’s a powerful one, and something needs to be said about it.

Two defensive maneuvers for the standard supervaluationist:

(1) Say that what you’re committed to is just “p& it’s not supertrue that p”. Deny that the ordinary concept of truth can be identified with supertruth (something that as many have emphasized, is anyway quite plausible given the non-disquotational nature of supertruth). But crucially, don’t seek to replace this with some other gloss on supertruth: just say that supertruth, superfalsity and gap between them are appropriate successor concepts, and that ordinary truth-talk is appropriate only when we’re ignoring the possibility of the third case. If we disclaim conceptual analysis in this way, then it won’t be appropriate to appeal to intuitions about the English word “true” to kick away independently motivated theoretical claims about supertruth. In particular, we can’t appeal to intuitions to argue that “p&~supertrue that p” should be assigned credence 0. (There’s a question of whether this should be seen as an error-theory about English “truth”-ascriptions. I don’t see it needs to be. It might be that the English word “true” latches on to supertruth because supertruth what best fits the truth-role. On this model, “true” stands to supertruth as “de-phlogistonated air” according to some, stands to oxygen. And so this is still a “truth=supertruth” standard supervaluationism.)

(2) The second maneuver is to appeal to supervaluational degrees of truth. Let the degree of supertruth of S be, roughly, the measure of the precisifications on which S is true. S is supertrue simpliciter when it is true on all the precisifications, i.e. measure 1 of the precisifications. If we then identify degrees of supertruth with degrees of truth, the contention that truth is supertruth becomes something that many find independently attractive: that in the context of a degree theory, truth simpliciter should be identified with truth to degree 1. (I think that this tendancy has something deeply in common with the temptation (following Unger) to think that nothing that nothing can be flatter than a flat thing: nothing can be truer than a true thing. I’ve heard people claim that Unger was right to think that a certain class of adjectives in English work this way).

I think when we understand the supertruth=truth claim in that way, the idea that “p&~true that p” should be something in which we should always have degree of belief 0 loses much of its appeal. After all, compatibly with “p” not being absolutely perfectly true (=true), it might be something that’s almost absolutely perfectly true. And it doesn’t sound bad or uncomfortable to me to think that one should conform one’s credences to the known degree of truth: indeed, that seems to be a natural generalization of the sort of thing that originally motivated our worries.

In summary. If you’re a supervaluationist who takes the orthodox line on supervaluational logic, then it looks like there’s a strong case for a non-classical take on what degrees of belief look like. That’s a potentially vulnerable point for the theory. If you’re a (standard, global, truth=supertruth) supervaluationist who’s open to the sort of position I sketch in the paper below, prima facie we can run with a classical take on degrees of belief.

Let me finish off by mentioning a connection between all this and some material on probability and conditionals I’ve been working on recently. I think a pretty strong case can be constructed for thinking that for some conditional sentences S, we should be all-but-certain that S&~DS. But that’s exactly of the form that we’ve been talking about throughout: and here we’ve got *independent* motivation to think that this should be high-probability, not probability zero.

Now, one reaction is to take this as evidence that “D” shouldn’t be understood along standard supervaluationist lines. That was my first reaction too (in fact, I couldn’t see how anyone but the epistemicist could deal with such cases). But now I’m thinking that this may be too hasty. What seems right is that (a) the standard supervaluationist with the Shafer-esque treatment of credences can’t deal with this case. But (b) the standard supervaluationist articulated in one of the ways just sketched shouldn’t think there’s an incompatibility here.

My own preference is to go for the degrees-of-truth explication of all this. Perhaps, once we’ve bought into that, the “truth=degree 1 supertruth” element starts to look less important, and we’ll find other useful things to do with supervaluational degrees of truth (a la Kamp, Lewis, Edgington). But I think the “phlogiston” model of supertruth is just about stable too.

[P.S. Thanks to Daniel Elstein, for a paper today at the CMM seminar which started me thinking again about all this.]

Supervaluations and logical revisionism paper

Happy news today: the Journal of Philosophy is going to publish my paper on the logic of supervaluationism. Swift moral. It ain’t logical revisionary; and if it is, it doesn’t matter.

This previous post gives an overview, if anyone’s interested…

Now I’ve just got to figure out how to transmute my beautiful LaTeX symbols into Word…

Supervaluations and revisionism once more

I’ve just spent the afternoon thinking about an error I found in my paper “supervaluational consequence” (see this previous post). I’ve figured out how to patch it now, so thought I’d blog about it.

The background is the orthodox view that supervaluational consequence will lead to revisions of classical logic. The strongest case I know for this (due to Williamson) is the following. Consider the claim “p&~Determinately(p)”. This (it is claimed) cannot be true on any serious supervaluational model of our language. Equivalently, you can’t have both p and ~Determinately(p) both true in a single model. If classical reductio were an ok rule of inference, therefore, you’d be able argue from ~Determinately(p) to ~p. But nobody thinks that’s supervaluationally valid: any indeterminate sentence will be a counterexample to it. So classical reductio should be given up.

This is stronger than the more commonly cited argument: that supervaluational semantics vindicates the move from p to Determinately(p), but not the material conditional “if p then Determinately(p)” (a counterexample to conditional proof). The reason is that, if “Determinately” itself is vague, arguably the supervaluationist won’t be committed to the former move. The key here is the thought that as well as things that are determinately sharpenings of our language, their may be interpretations which are borderline-sharpenings. Perhaps interpretation X is an “admissible interpretation of our language” on some sharpenings, but not on others. If p is true at all the definite sharpenings, but false at X, then that may lead to a situation where p is supertrue, but Determinately(p) isn’t.

But orthodoxy says that this sort of situation (non-transitivity in the accessibility relation among interpretations of our language) does nothing to undermine the case for revisionism I mentioned in the first paragraph.

One thing I do in the paper is construct what seems to me a reasonable-looking toy semantics for a language, on which one can have both p and ~Determinately p. Here it is.

Suppose you have five colour patches, ranging from red to orange (non-red). Call them A,B,C,D,E.

Suppose that our thought and talk makes it the case that only interpretations which put the cut-off between B and D are determinately “sharpenings” of the language we use. Suppose, however, that there’s some fuzziness around in what it is to be an “admissible interpretation”. For example, an interpretation that places the cut-off between B and C, thinks that both interpretations placing the cut-off between C and D, and interpretations placing the cut-off between A and B, are admissible. And likewise, an interpretation that place the cut-off between C and D think that interpretations that place the cut-off between B and C are admissible, but also thinks that interpretations that place the cut-off between D and E are admissible.

Modelling the situation with four interpretations, labelled AB, BC, CD, DE, for where they place the red/non-red cut-off, we can express the thought like this: each intepretation accesses (thinks admissible) itself and its immediate neighbours, but nothing else. But BC and CD are the sharpenings.

My first claim is that all this is a perfectly coherent toy model for the supervaluationist: nothing dodgy or “unintended” is going on.

Now let’s think about the truths values assigned to particular claims. Notice, to start with, that the claim “B is red” will be true at each sharpening. The claim “Determinately, B is red” will be true at the sharpening CD, but it won’t be true at the sharpening BC, for that accesses an interpretation on which B counts as non-red (viz. AB).

Likewise, the claim “D is not red” will be true at each sharpening, but “Determinately, D is not red” will be true at the sharpening BC, but fails at CD, due to the latter seeing the (non-sharpening) interpretation DE, at which D counts as red.

In neither of these atomic cases do we find “p and ~Det(p)” coming out true (that’s where I made a mistake previously). But by considering the following, we can find such a case:

Consider “B is red and D is not red”. It’s easy to see that this is true at each of the sharpenings, from what’s been said above. But also “Determinately(B is red and D is not red)” is false at each of the sharpenings. It’s false at BC because of the accessible interpretation AB at which B counts as non-red. It’s false at CD because of the accessible interpretation DE at which D counts as red.

So we’ve got “B is red and D is not red, & ~Determinately(B is red and D is non-red).” And we’ve got that in a perfectly reasonable toy model for a language of colour predicates.

(Why do people think otherwise? Well, the standard way of modelling the consequence relation in settings where the accessibility relation is non-transitive is to think of the sharpenings as *all the interpretations accessible from some designated interpretation*. And that imposes additional structure which, for example, the model just sketch doesn’t satisfy. But the additional structure seems to me totally unmotivated, and I provide an alternative framework in the paper for freeing oneself from those assumptions. The key thing is not to try and define “sharpening” in terms of the accessibility relation.).

The conclusion: the best extant case for (global) supervaluational consequence being revisionary fails.

Probabilistic multi-conclusion validity

I’ve been thinking a bit recently about how to generalize standard results relating probability to validity to a multi-conclusion setting.

The standard result is the following (where the uncertainty of p is 1-probability of p):

An argument is classically valid
for all classical probability functions, the sum of the uncertainties of the premises is at least as great as the uncertainty of the conclusion.

It’ll help if we restate this as follows:

An argument is classically valid
for all classical probability functions, the sum of the uncertainties of the premises + the probability of the conclusion is at least 1.

Stated this way, there’s a natural generalization available:

A multi-conclusion argument is classically valid
for all classical probability functions, the sum of the uncertainties of the premises + the probabilities of the conclusions is greater than or equal to 1.

And once we’ve got it stated, it’s a corollary of the standard result (I believe).
It’s pretty easy to see directly that this works in the “if” direction, just by considering classical probability functions which only assign 1 or 0 to propositions.

In the “only if” direction (writing u for uncertainty and p for probability)

Consider A,B|=C,D. This holds iff A,B,~C,~D|= holds by a standard premise/conclusion swap result. And we know u(~C)=p(C), u(~D)=p(D). By the standard result, the sum of uncertainties of the premises of a single-conclusion argument must be greater than that of the conclusion. That is, the single-conc argument holds iff u(A)+u(B)+u(~C)+u(~D) is greater than equal to 1. But by the above identification, this holds iff u(A)+u(B)+p(C)+p(D) is greater than or equal to 1. This should generalize to arbitrary cases. QED.

Supervaluational consequence again

I’ve just finished a new version of my paper supervaluational consequence. A pdf version is available here. I thought I’d post something explaining what’s going on therein.

Let’s start at the beginning. Classical semantics requires, inter alia, the following. For every expression, there has to be a unique intended interpretation. This single interpretation will assign to each name, a single referent. To each predicate, it will assign a set of individuals. Similarly for other grammatical categories.

But sometimes, the idea that there are such unique referents, extensions and so on, looks absurd. What supervaluationism (in the liberal sense I’m interested in) gives you is the flexibility to accommodate this. Supervaluationism requires, not a single intended interpretation, but a set of interpretations.

So if you’re interested in the problem of the many, and think that there’s more than one optimal candidate referent for “Kilimanjaro”; if you’re interested in theory change, and think that relativist and rest mass are equi-optimal candidate properties to be what “mass” picks out; if you are interested in inscrutability of reference, and think that rabbit-slices, undetached rabbit parts as well as rabbits themselves are in the running to be in the extension of “rabbit”; if you’re interested in counterfactuals, and think that it’s indeterminate which world is the closest one where Bizet and Verdi were compatriots; if you think vagueness can be analyzed as a kind of multiple-candidate indeterminacy of reference; if you find any of these ideas plausible, then you should care about supervaluationism.

It would be interesting, therefore, if supervaluationism undermined the tenants of the kind of logic that we rely on. For either, in the light of the compelling applications of supervaluationism, we will have to revise our logic to accommodate these phenomena; or else supervaluationism as a theory of these phenomena is itself misconceived. Either way, there’s lots at stake.

Orthodoxy is that supervaluationism is logically revisionary, in that it involves admitting counterexamples to some of the most familiar classical inferential moves: conditional proof, reductio, argument by cases, contraposition. There’s a substantial hetrodox movement which recommends a hetrodox way of defining supervaluational consequence (so called “local consequence”) which is entirely non-revisionary.

My paper aims to do a number of things:

  1. to give persuasive arguments against the local consequence heterodoxy
  2. to establish, contra orthodoxy, that standard supervaluational consequence is not revisionary (this, granted a certain assumption)
  3. to show that, even if the assumption is refused, the usual case for revisionism is flawed
  4. to give a final fallback option: even if supervaluational consequence is revisionary, it is not damagingly so, for it in no way involves revision of inferential practice.

It convinces me that supervaluationists shouldn’t feel bad: they probably don’t revise logic, and if they do, it’s in a not-terribly-significant way.

Eliminating singular quantification

I’ve been thinking idly about plural quantification over the last day or so (the things one does on ones holidays…).

The general idea is that we can go beyond standard first-order predicate logic, by adding distinctively plural quantification. So, in addition to quantifying by saying “there is something such that it is Mopsy”; we may also say “there are some things such that Mopsy is one of them”. Oystein Linnebo has a really nice summary of plural logics in the Stanford Encyclopedia.

The setting I was thinking of is less expansive than the systems that Oystein concentrates on (those he calls PFO and PFO+). The way it is less expansive is this. the languages of PFO and PFO+ includes both singular quantification/singular terms and plural quantification/plural terms in its primitives. I want a system that has only plural quantification/terms as primitive. This means that rather than taking the relational primitive “is one of”, holding between singular terms and plural terms, as primitive, I’ll take “are among”, which holds between pairs of plural terms. The payoff may be this: singular terms, variables and predication may turn out to be “dispensible”, in the same sense that Russell’s theory of descriptions showed that individual constants were dispensible. This may well be stuff that is already covered by the literature (or just obvious). If so, I’d be very happy to get references!

I will be taking it that the language of plurals contains predicates of plural terms. In this way, we follow what Linnebo calls L_PFO+ rather than L_PFO. Now generally we can distinguish between plural predicates that are distributive; and those that are non-distributive. Linnebo’s examples are: the distributive predicate “is on the table” (if some things are on the table, then each one of those things is individually on the table); and the non-distributive predicate “forms a circle” (Some things can form a circle, even though there is no sense in which each individually forms a circle). Linnebo says that he does this to allow for non-distributive predications; but part of my motivation is to allow also for distributive plural predications. Syntactically, we need not pay attention to this (though if the semantic treatment of distributive and non-distributive plural predicates is to differ, we might want to differentiate them syntactically: introducing two sets of predicates. I’m going to ignore such refinements for now.)

Here’s the language:

1. L_Plural has the following plural terms (where i is any natural number):

* plural variables xxi;
* plural constants aai.

2. L_Plural has the following predicates:

* a dyadic logical predicate <. (to be thought of as are among);
* non-logical predicates Rni (for every adicity n and every natural number i).

3. L_Plural has the following formulas:

* Rni(t1, …, tn) is a formula when Rni is an n-adic predicate and tj are plural terms;
* t < t’ is a formula when t and t’ are plural terms;
* ~φ and φ&ψ are formulas when φ and ψ are formulas;
* (Ev)v.φ is a formula when φ is a formula and vv a plural variable.
* the other connectives are regarded as abbreviations in the usual way.

What I’m interested in is whether we can develop a natural logic of plurals on the basis of this language: and if so, what its expressive power would be.

An immediate task would be to reintroduce singular quantification. The intuitive thought is that singular quantification can be thought of as a special case of plural quantification, where we somehow ensure that there is just one of them. The trick is to show how this can be done without circular appeal to singular quantification.

My thought (roughly) is to treat this as the following restricted quantifier [Exx : (yy)(if yy < xx then xx < yy].

Why will this play the role of singular quantification? Well, just because if you’ve got a plurality of things, which is such that every subplurality is also a superplurality, it’s got to be a plurality consisting of just one thing (I’m assuming that there are no “null” pluralities). Now, of course, L_plural doesn’t contain restricted quantifiers. But it’s easy enough to find things that play the role of restricted quantifiers (formally, we’ll define a paraphrase from L_PFO+ into L_plural that’ll play this role). In parallel fashion, we can get a paraphrase of sentences containing singular terms, and paraphrase them into something that only uses plural vocabulary.

E.g. “(Ex)Elephant(x)” may go to: “(Exx)((yy)(if yy < xx then xx < yy)& Elephant(xx))”. And “Runs(Susan)” may go to: “(yy)(if yy < Susan then Susan < yy)& Runs(Susan) )

Now, it seems to me that there are some interesting questions of detail about how best to formalize the “intuitive” logical theory for L_plural that I’ve been working with. But let me leave the this for now. Question is: does the above elimination of singular quantification and terms in favour of plural quantification and terms seem tenable? Does the paraphrase work on the “intuitive” reading of L_plural. Can people see any obstacles to formalizing this intuitive logic for L_plural?