Switcharoos

We normally think that the content of judgments are dictated by the content of the concepts used inside of them. This feels especially right for artificial languages:

• Take “If it’s a \(P\), then it’s a \(Q\).”

  • Or, logically: \(\forall x, P(x)\implies Q(x)\)

• The meaning of this statement seems to depend on what concepts \(P\) and \(Q\) we substitute in. E.g. with \(P \mapsto {\rm red\ thing},\ Q \mapsto {\rm colored\ thing}\), it’s a good a judgment, whereas \(Q \mapsto {\rm rectangular\ thing}\) would no longer be a good judgment.

However, Kant turned this around (see here). In this view, judgments are prior (in the order of understanding) to concepts.

This is also related to a switch of priority made by Sellars: we normally think of logically-valid inferences (e.g. \(A \land B \implies B \lor C\)) as something we understand prior to particular inferences (e.g. “If it’s red and triangular, then it’s triangular or heavy”). Sellars calls these particular inferences material_inferences and argues that it is only through understanding them that we could understand logically-valid inferences.

One argument for our initial intuition is that the logically-valid inferences are a priori, whereas the particular inferences are a posteriori.¹ However, Achilles and the Tortoise feels relevant for arguing against this point of view: the logically-valid inferences exist a priori as abstract mathematical/syntactical objects, but without any practical experience of actually making inferential moves, we don’t have access to them qua inferences.

Tarski’s theory of truth says: if you take meanings for granted, I can give you a truth predicate.

Davidson flipped this on its head, saying: Tarski provided the meanings of a class of terms given how they interact with other propositions.

It feels natural to try to reduce concepts to combinations of sense impressions (this is what Locke/Hume do). This is a mistake that comes from the fact that sense impressions are causally prior to our concepts. This does not mean they are conceptually prior.

Sense data itself is not contentful/conceptual in nature and cannot stand in relations of justification/implication.