Saturday, September 02, 2006

Speakers Use Their Actual Language

This is a fairly trivial point, but it's worth noting that the words we use have particular meanings, and these actual semantic facts are unchanged by imagining counterfactual languages. So it's not really true that "whether something is good or bad depends on your definition of 'good' and 'bad'." The actual disquoted use of 'good' refers to a particular property, goodness, and whether something possesses that property is an entirely language-independent fact.

If 'right' meant wrong, then the sentence "murder is right" would be true, but that wouldn't make murder right. For my latter (disquoted) use of the word 'right' is spoken in my actual language. It picks out a property that murder lacks no matter what we call it. The earlier, quoted mentions of the word 'right' (i.e. my uses of ''right'') are not used to express this property. They instead refer to the word itself. And if this word were to pick out the property of wrongness, i.e. the property that is actually denoted by our word 'wrong', then the former word would extend to murder. That is, if 'right' meant wrong, then 'right' would pick out a property, namely wrongness, that murder has. But again, that obviously wouldn't make murder right. Murder would not gain the property that we actually refer to using 'right'. It would merely become describable using the word just mentioned -- a word that would then be used to mean the opposite of what it means in actual use.

So much for counterfactual languages. What about counter-actual ones, i.e. other languages considered as actual, say using indicative conditionals? To borrow an example from Yablo, compare:

(1) If 'tail' had meant wing, then horses would not have had tails.
(2) If 'tail' means wing, then horses do not have tails.

Clearly (1) is false, for the reasons explained above. But what about (2)? Yablo argues that it's true:
If 'tail' as a matter of fact means wing, then to say that horses have tails is to say that they have wings. Horses do not have wings. So if 'tail' means wing, then horses do not have tails. ('Coulda, Woulda, Shoulda', p.450)

Here's a puzzle: our actual use of 'horses do not have tails' is (near enough to) synonymous with 'horses lack post-trunk extensions'. But now consider:

(2*) If 'tail' means wing, then horses lack post-trunk extensions.

That claim is surely false. And substituting synonyms shouldn't alter the truth value of a sentence, so (2) should be false too. But perhaps this is an exception to that general rule, rather like:

(3) 'tail' has four letters.
(3*) 'post-trunk extension' has four letters.

But the obvious problem with (3*) is that the substitution occurs within quote-marks, wherein the word 'tail' is being mentioned rather than used. (What's being used is the string ''tail'', and the replacement string ''post-trunk extension'' is not synonymous with that. Rather than referring to one and the same body part, these two terms refer to two different words. Whereas 'tails' refers to tails, ''tails'' merely refers to 'tails'. Don't let the superficial similarity fool you!) The substitution from (2) to (2*) is not illegitimate in this way. So if it is illegitimate at all, we need some further explanation why. The mere fact that the word is mentioned earlier in the sentence does not suffice. For compare:

(4) 'Four' has four letters.
(4*) 'Four' has 3+1 letters.

There's clearly nothing wrong with moving from (4) to (4*), despite the early mention of the word that gets substituted when used later in the sentence.

If we accept the equivalence of (2) and (2*), then we will reject Yablo's first premise: "If 'tail' as a matter of fact means wing, then to say that horses have tails is to say that they have wings." This seems a tough bullet to bite, but perhaps it is defensible. After all, that horses have tails is the same proposition as that horses have post-trunk extensions. To say one is to say the other, as they are one and the same thing. But to say that horses have post-trunk extensions (tails), no matter what particular language one says this in, is not to say that they have wings.

My argument here is a little suspicious, since it seems like I'm not giving the antecedents of Yablo's indicative conditionals due weight -- it might be suggested that they undermine my synonymy claims, thereby blocking my response. I'm not too sure what to think of that. I guess there must be something wrong with the substitutions after all, but I'd welcome any clear insights into what exactly the rule broken by (2*) is...



  1. Isn't the problem with:
    (2) If 'tail' means wing, then horses do not have tails.
    that it is the sentence in the actual language? As such, I don't think it has power to change its own meaning.
    If we on other side clearly devide the sentences, and in which language they are used, this kind of problems are not present:
    (2+) if in the language A 'tail' means wing, then the sentence in A - 'horses do not have tails' would be true
    Or... even:
    (2++) From now on, when I say 'tail', I will mean 'wing'... Horses do not have tails.
    This also seems OK, as it is clear that the first sentence is in the actual language. And the second is in the redefined language.

  2. It seems to me that (2*) is trivially true (in our actual world) for the simple reason that the antecedent is false. 'Tail' does not in fact mean wing.

    On the other hand, if you postulate in advance that the four-letter sequence 'tail' refers to a wing, then you can't replace occurences of 'tail' with 'post-trunk extension'.

    That is, only in worlds in which 'tail' does not refer to wing are you justified in making the substitution from (2) to (2*). And, of course, in these worlds the antecendent of the conditional is false. If you were to be instead in a world in which the antecedent is true then you couldn't make that substitution.

  3. What is happening in (2) seems similar to identifying the relevant meaning of a homonym. Compare:

    (2b) If 'bank' is used to mean the land bounding a river, banks do not loan people money.

    But our actual use of "banks do not loan people money" typically is (roughly) synonymous to "establishments for the exchange of money do not loan people money." So consider:

    (2b*) If 'bank' is used to mean the land bounding a river, establishments for the exchange of money do not loan people money.

    I'd say that (2b) is true and (2b*) is false, and the move from (2b) to (2b*) is illegitimate (just as Yablo would say about 2 & 2*). I think the general principle identifying what makes the move illegitimate is that, if a term 'x' can take on multiple meanings, then it is incorrect to replace it with a phrase that is synonymous to one of those meanings unless that is the relevant meaning in this context. Locutions like "If 'x' means y," or "If 'x' is used to mean y," or "Using 'x' to mean y" establish what meaning of 'x' is relevant in a given context, whether that meaning is a conventionally accepted one (in the homonymy case) or not (in the counter-actual case).

    An alternative explanation of the problem with going from (2) to (2*) (if you don't want to allow polysemy to combine actual and counter-actual meanings) is that, counter-actually (taking 'tail' to mean wing), our use of 'horses do not have tails' is not synonymous with 'horses lack post-trunk extensions'. And need this claim about the counter-actual world, not the claim about the actual world that you made, in order to legitimately turn (2) into (2*). There would be no coherent way to think counter-actually if you could just insert any fact about the actual world into your counter-actual hypothetical (since you could lose or contradict anything contrary to actual).

  4. Tanasije - yeah, that's the kind of worry I have here. Though many philosophers wouldn't consider (2) to be intuitively problematic, which is why I explored some more indirect arguments...

    Cecil - I'm not taking these as material conditionals, so a false antecedent does not suffice for their truth. But if you don't like that, you can easily reinterpret my demand for "truth" as instead for "non-vacuous truth". Yablo thinks (2) is non-vacuously true, but (2*) clearly isn't, so we still need some explanation of this difference, whatever you want to call it.

    I like Blar's suggestions. One response to his first point is that disambiguating locutions cannot be used to establish a meaning that 'x' does not actually have. Unless, as in Tanasije's (2++) example, we take the consequent to be using the revised/counteractual language, rather than our actual language. But that seems odd. (Like the post title says: "Speakers Use Their Actual Language"!)

    On the second point, I agree that you can't "just insert any fact about the actual world into your counter-actual hypothetical". But I didn't take my substitution to involve inserting any facts at all (though it clearly does depend on the actual fact of synonymy). Assuming semantic compositionality -- and perhaps this is a counterexample to it -- (2*) simply means the exact same thing as (2). So nothing is added. We've simply restated one and the same claim. Or if we haven't, then the challenge is to explain exactly why they are distinct claims. Presumably the antecedent somehow sets up an opaque context, but this is pretty weird...

  5. I find it hard to conceive of the meaning of the antecedent "If 'tail' means 'wing'" as an ordinary proposition (but not a counterfactual) without being in some state of uncertainty about the meaning of the word 'tail'.

    Which is to say, if-counterfactually I were in a state of uncertainty about my language such that 'tail' meaning 'wing' would be plausible, my stating the sentence "If 'tail' means 'wing', then horses do not have tails." would seem meaningful and true. On the other hand, whether the substitution to 'post-trunk extensions' is valid depends further on my epistemic state:

    If I know for a fact that 'tail' and 'post-trunk extension' are synonyms then by transitivity I must be in uncertainty about the meaning of 'post-trunk extension' also - it must be plausible to me that 'post-trunk extension' could mean 'wing'. In that case it would seem valid and true for me also to say "If 'tail' means 'wing', then horses do not have post-trunk extensions." (consider "If 'tail' means 'wing', then 'post-trunk extension' means 'wing'.").

    However, consider the opposite case where I am in no uncertainty about 'post-trunk extension' (say, that I believe the dictionary definition). Then in my language 'tail' and 'post-trunk extension' aren't obviously synonymous, because of the uncertainty related to the former. This is true even if they refer to the same thing, and become synonymous once my uncertainty is resolved. So I wouldn't be confident substituting "If 'tail' means 'wing', then horses do not have post-trunk extensions." and expecting it to have the same meaning and truth value.

    Leaving all that aside, I also have some sympathy for the 'odd' hypothesis that the antecedent is not an ordinary proposition, but introduces a modified language as in (2++).


Visitors: check my comments policy first.
Non-Blogger users: If the comment form isn't working for you, email me your comment and I can post it on your behalf. (If your comment is too long, first try breaking it into two parts.)

Note: only a member of this blog may post a comment.