I've been spending much time recently in coffee shops with colleagues talking about the stuff that's coming up in the fantastically named RIP Being conference (happening in Leeds this weekend). Hopefully I won't be treading on toes if I draw out one strand of those conversations that I've been finding particularly interesting.
(continued below the fold)
The story for me begins with an old paper by Hartry Field. His series of papers in the 70's is one of the all-time great runs: from "Tarski's theory of truth" through "Quine and the correspondance theory", "Theory Change", "Logic, meaning and conceptual role", "Conventionalism and Instrumentalism in semantics" and finishing off with "Mental representation". (All references can be found here). Not all of them are reprinted in his collection Truth and the absence of fact, which seems a pity. The papers I mentioned above really seemed to me to lay out the early Fieldian programme in most of the details. Specifically, in missing out the papers "Logic, meaning ..." and "Conventionalism and instrumentalism...", you miss out on the early-Field's take on how the cognitive significance of language relates to semantic theory; and the most interesting discussion I know of concerning what Putnam's notorious "just more theory" argument might amount to.
The "just more theory" move is supposed to be the following. It's familiar that you can preserve sensible truth conditions, by assigning wildly permuted reference-schemes to language (see my other recent posts for more details and links). But, prima facie, these permuted reference schemes are going to vitiate some plausible conditions of what it takes for a term to refer to something (e.g. that the object be causally connected to the term). Now, some theorists of meaning don't build causal constraints into their metasemantic account. Davidson, early Lewis and the view Putnam describes as "standard" in his early paper, are among these (I call these "interpretationisms" elsewhere). But the received view, I guess, is to assume that some such causal constraint will be in play.
Inscrutability argument dead-in-the-water? No, says Putnam. For look! the permuted interpretation has the resources to render true sentences like "reference is a relation which is causally constrained". For just as, on the permuted interpretation "reference" will be assigned as semantic value some weirdo twisted relation Reference*, so on the same interpretation "causation" will be assigned some weirdo twisted relation Causation. And it'll turn out to be true that Reference* and Causation* match up in the right way. So (you might think), how can metasemantic theories tell you rule in favour of the sensible interpretation over this twisted one? For whichever no matter which of these we imagine to be the real interpretation of our language, everything we say will come out true.
Well, most people I speak to think this is a terrible argument. (For a particularly effective critique of Putnam---showing how badly things go if you allow him the "just more theory" move---see this paper by Tim Bays.) I'll take it the reasons are pretty familiar (if not, Lewis's "Putnam's paradox" has a nice presentation of a now-standard response). Anyway, what's interesting about Field's paper is that it gives an alternative reading of Putnam's challenge, which makes it much more interesting.
Let's start by granting ourselves that we've got a theory which really has tied down reference pretty well. Suppose, for example, that we say "Billy" refers to Billy in virtue of appropriate causal connections between tokenings of that word and the person, Billy. The "Wild" inscrutability results threatened by permutation arguments simply don't hold.
But now we can ask the following question: what's special about that metasemantic theory you're endorsing? Why should we be interested in Reference (=Causal relation C)? What if we tried to do all the explanatory work that we want semantics for, in terms of a different relation Reference*? We could then have a metasemantic* theory of reference*, which would explain that it is constrained to match a weirdo relation causation*. But, notice, that the relation "S expresses* proposition* p" (definable via reference*) and "S expresses proposition p" (definable via reference*) are coextensional. Now, if all the explanatory work we want semantics to do (e.g. explaining why people make those sounds when they believe the world is that way) only ever makes appeal to what propositions sentences express, then there just isn't any reason (other than convenience) to talk about semantic properties rather than semantic* ones.
The conclusion of these considerations isn't the kind of inscrutability I'm familiar with. It's not that there's some agreed-upon semantic relation, which is somehow indeterminate. It's rather that (the consideration urges) it'll be an entirely thin and uninteresting matter that we choose to pursue science via appeal to the determinate semantic properties rather than the determinate semantic* properties. You might think of this as a kind of metasemantic inscrutability, in contrast to the more usual semantic inscrutability: setting aside mere convenience, there's no reason why we ought to give this metasemantic theory rather than that one.
Now, let's turn to a different kind of inscrutability challenge. For one reason or another, lots of people are very worried over whether we can really secure determinate quantification over an absolutely unrestricted domain. Just suppose you're convinced that there are no abstracta. Suppose you're very careful to never say anything that commits you to their existence. However, suppose you're wrong: abstracta exist. Intuitively, when you say "There are no abstracta, and I'm quantifying over absolutely everything!" you're speaking falsely. But this is only so if your quantifiers range over the abstracta out there as well as the concreta: and why should that be? In virtue of what can your word "everything" range over the unrestricted domain? After all, what you say would be true if I interpreted the word as ranging over only concreta. I'd just take you to be saying that no concreta exist (within your domain; and that you were quantifying over absolutely everything in your domain. Both of these are true, given that your domain happens to contain only concreta!
Bring in causality doesn't look like it helps here; neither would the form of reference-magnetism that Lewis endorsed, which demands that our predicates latch onto relatively natural empirical kinds, help. Ted Sider, in a paper he's presenting at the RIP conference, advocates extending the Lewis point to make appeal to logical "natural kinds" (such as existence) at this point. However, let me sketch instead a variant of the Sider thought that seems more congenial to me (I'll sketch at the end how to transfer it back).
My take on Lewis's theory is the following. First, identify a "meaning building language". This will contain only predicates for empirical natural kinds, plus some other stuff (quantifiers, connectives, perhaps terms for metaphysically basic things such as mereological notions). Now, what it is for a semantic theory for a natural language to be the correct one, is for there to be a semantic theory phrased in the meaning-building language, which (a) assigns to sentences of the natural language truth-conditions which fit with actual patterns of assent and dissent; and (b) is as syntactically simple as possible. (I defend this take on what Lewis is doing here).
Now, clearly we need to use some logical resources in constructing the semantic theory. Which should we allow? Sider's answer: the logically natural ones. But for the moment let's suppose we don't want to commit ourselves to logically natural kinds. Well, why don't we just stipulate that the meaning building language is going to contain this, that, and the next logical operator/connective? In the case of predicates, there's the worry that our meaning-building theory should contain all the empirical kinds there are or could be: since we don't know what these are, we need to give a general definition such as "the meaning building language will contain predicates for all and only natural kinds". But there seems no comparible reason not simply to lay it down that "the meaning building language will contain negation, conjunction and the existential quantifier).
Indeed, we could go one further, and simply stipulate that the existential quantifier it contains is the absolutely unrestricted one. The effect will be just like the one Sider proposes: this metasemantic proposal has a built-in-bias towards ascribing truly unrestricted generality to the quantifiers of natural language, because it is syntactically simpler to lay down clauses for such quantifiers in the meaning-building language, than for the restricted alternatives. You quantify over everything, not just concreta, because the semantic theory that ascribes you this is more eligible than one that doesn't, where eligibility is a matter of how simple the theory is when formulated in the meaning-building language just described.
Ok. So finally finally I get to the point. It seems to me that Field's form of Putnam's worries can be put to work here too. Let's grant that the metasemantic theory just described delivers the right results about semantic properties of my language; and shows my unrestricted quantification to be determinate. But why choose just that metasemantic theory? Why not, for example, describe a metasemantic theory where semantic properties are determined by syntactic simplicity of a semantic theory in a meaning building language where the sole existential quantifier is restricted to concreta? Maybe we should grant that our way picks out the semantic properties: but we've yet to be told why we should be interested in the semantic properties, rather than the semantic* properties delivered by the rival metasemantic theory just sketched. Metasemantic inscrutability threatens once more.
(I think the same challenge can be put to the Sider-style proposal: e.g., consider the Lewis* metasemantic theory whereby the meaning-building language contains expressions for all those entities (of whatever category) which are natural*: i.e. are the intersection of genuinely natural properties (emprical or logical) with restricted domain D.)
I have suspicians that metasemantic inscrutability will turn out to be a worrying thing. That's a substantive claim: but it's got to be a matter for another posting!
(Major thanks here go to Andy and Joseph for discussions that shaped my thoughts on this stuff; though they are clearly not to be blamed..).