After the wonderful REC conference, I’ve been thinking some about the relationship between philosophy of language and various accounts of our use of the term ‘knows’ and its cognates. What follows are some thoughts about the connections between semantical theory as discussed in Jason’s nice earlier post here, and my own attractions to Kaplanian structure for semantical theories. What I’m trying to grasp is a certain kind of logical space for epistemological theories, and so I’ll abstract away from specifics of semantical theories to the major categories I see. And maybe those more versed in semantics than I can provide further, or alternative, illumination…
Logically perfect languages require very little complexity in a semantical theory, compared with what is needed for messier, natural languages. For a logically perfect language, we need a function from sentences to truth-values, and though the account of the function itself might be quite complicated, that’s about all we need.
But natural languages have various features that logically perfect languages don’t.
A first distinctive feature is ambiguity. In order to account for ambiguity we need a function from sentences to meanings, and once we get that far, we’ll need a function to get us to truth-values as well. If ambiguity were the only troublespot with natural languages, that’d be enough.
But there is more complexity than this to natural language, as is evidenced by the existence of indexicals and demonstratives. Given a broadly Kaplanian framework, we need a further function from character to content.
There is, of course, one further issue, and that is whether content can be specified exactly enough that content yields a truth value once a given world is specified, and if we suppose even further complexity, we need a function from contents to truth values even within a world.
Given all this, we get three functions, which can be schematically represented as follows:
F1(sentences) -> character
F2(character) -> content
F3(content) -> truth value
Thought of in this way, we get a variety of linguistic approaches to epistemological metatheory. If ‘knows’ is ambiguous, that tells us something about F1 (think here of Goldman’s claim that there is a weak use of ‘knows’ that is just true belief and a stronger use that is what epistemologists typically focus on). If ‘knows’ is contextual, that tells us something about F2 (think here of standard contextualists, such as DeRose, Cohen, Lewis, et. al.); and if subject-sensitive invariantism is true, that tells us something about F2 as well (think of Hawthorne, Stanley, and maybe McGrath&Fantl here). Telling us something about F3, on this taxonomy, gives us a different picture entirely, and I’ll use the label ‘relativism’ for views that arise at this stage. I’m stipulating at least a little bit a sense of ‘relativism’ here, and in defense will say this: the troubling kind of relativism to most of us is the kind where, once you have the proposition in hand, the world alone is not sufficient to yield a truth value–perhaps you have to know what culture you’re in as well, for example.
Fixing the taxonomy will help in a variety of ways, so I’m interested in the extent to which others find this structure problematic. But if we fix on this taxonomy, there are a couple of implications I find interesting and important. Actually, they amount to mistakes that can be made in defending one of these approaches. The first is by creating alliances with opponents to make one’s view look better. For example, a relativist may wish to hide from criticism by trying to identify with contextualism, or a contextualist with an ambiguity theorist. The second is by masking the work done by different functions. The easiest to detect would be a relativist claiming that relativism is nothing more than ambiguity writ large. Harder to detect is the move from, say, ambiguity to contextualism. When contextualists and invariantists talk primarily about high stakes and low stakes contexts, it can appear that their view is nothing more than a generalization of the ambiguity to cover the possibility that there are multiple ambiguities to account for. In a sense, that’s right, and in a sense, it’s wrong, and noticing the difference between F1 and F2 should prevent using arguments for ambiguity as arguments for contextuality.
The harder question is how to sort the data: which data is relevant to which function? That’s a harder question, so I’ll pass on it for now at least.