Your approach seems (to me anyways, I may be misreading!) to treat counterfactuals as strict conditionals, where "would" operates as a modal necessity operator quantifying over all relevant contexts and scoping over a material conditional.
This view, however, appears unsatisfactory. A semantic account of counterfactuals should ideally allow for defeasibility. That is, it should not always be the case that P → Q implies P ∧ R → Q. Yet, suppose it is true in all relevant contexts that P → Q; within each of those contexts/worlds, P → Q entails the truth of P ∧ R → Q for some arbitrary R. Therefore, it holds across all relevant contexts that P ∧ R → Q.
To provide a satisfactory semantic account of counterfactuals, you would need to specify structural constraints on the selection of relevant contexts.
Yes, that's exactly right (although I'll reserve 'contexts' for contexts of utterance, instead of cases / worlds). This account allows for strong logical properties — strengthening the antecedent, as you point out, and also things like contraposition; we don't bake any of these limitations into the semantics, but leave them to the pragmatics. For instance, it seems clear that asserting 'If it were the case that P and R, then it'd be the case that Q' makes relevant some worlds where P and R hold (if there are any); so, we do predict that this conditional is true in the old context, but admit that asserting it almost invariably forces a change in context. Notice that 'If Sydney came over, I'd be happy' sounds fine out of the blue, but sounds bad after 'If Sydney came and she were severely injured, I wouldn't be happy'; this is just what we'd expect if it were a contextual effect (and this isn't as neatly explained if the constraints are in the semantics).
>(and this isn't as neatly explained if the constraints are in the semantics).
I’m inclined to disagree. If we model a counterfactual A → B as a true statement in some context u only if B is true in a set of contexts where A is also true, determined by a selection function, then by imposing certain constraints on that selection function, our semantics for counterfactuals allows for defeasibility.
It seems to me a more straightforward and natural explanation that “If Sydney came over, I’d be happy” sounds true, but “If Sydney came and she were severely injured, I wouldn’t be happy” sounds false if the truth conditions of a counterfactual are such that the former can be true while the latter can be false. On your view, the truth of the former entails the truth of the latter, which seems odd. While I agree that contextual shifts could explain this, such an explanation isn't as developed nor as straightforward as a semantic account based on the truth conditions of counterfactuals without further elaboration. In fact, analyzing our selection function as a similarity relation between contexts seems like a more direct and substantive way to ground our intuition that there are these implicit ceteris paribus clauses in evaluating conditionals that make facts like the strengthening of the antecedent unappealing. This approach is (in my opinion) more satisfying in that appeals to contextual effects, which, though plausible, require additional theoretical elaboration. Simply attributing it to pragmatics, while somewhat explanatory, leaves out the further theoretical work that needs to be developed.
I think the important point in my reply above was that the contextual view almost immediately predicts that reverse Sobel sequences (uttering the weak-antecedent conditional one after the strong-antecedent conditional) should sound bad, which is a big point in its favor — even if it feels less straightforward at first blush, it's much more robust to slight variations of the test case. But to reply to what you've added here — it might be a matter of taste, but I don't think the contextual prediction is especially unstraightforward. Your selection function had better consider context as well, allowing either 'If Caesar were prosecuting the Korean war, he'd use nukes' or 'If Caesar were prosecuting the Korean war, he'd use catapults' based on context. But now the selection function just looks like an unnecessary intermediary, and making it less prone to counterexamples (given all the ways counterfactuals can vary) requires adding ugly epicycles — better to just drop it!
Did this work? I mean it worked on me so clearly it worked at least a little bit, but did this post get you more subscribers than the others? I bet it did.
Your approach seems (to me anyways, I may be misreading!) to treat counterfactuals as strict conditionals, where "would" operates as a modal necessity operator quantifying over all relevant contexts and scoping over a material conditional.
This view, however, appears unsatisfactory. A semantic account of counterfactuals should ideally allow for defeasibility. That is, it should not always be the case that P → Q implies P ∧ R → Q. Yet, suppose it is true in all relevant contexts that P → Q; within each of those contexts/worlds, P → Q entails the truth of P ∧ R → Q for some arbitrary R. Therefore, it holds across all relevant contexts that P ∧ R → Q.
To provide a satisfactory semantic account of counterfactuals, you would need to specify structural constraints on the selection of relevant contexts.
Yes, that's exactly right (although I'll reserve 'contexts' for contexts of utterance, instead of cases / worlds). This account allows for strong logical properties — strengthening the antecedent, as you point out, and also things like contraposition; we don't bake any of these limitations into the semantics, but leave them to the pragmatics. For instance, it seems clear that asserting 'If it were the case that P and R, then it'd be the case that Q' makes relevant some worlds where P and R hold (if there are any); so, we do predict that this conditional is true in the old context, but admit that asserting it almost invariably forces a change in context. Notice that 'If Sydney came over, I'd be happy' sounds fine out of the blue, but sounds bad after 'If Sydney came and she were severely injured, I wouldn't be happy'; this is just what we'd expect if it were a contextual effect (and this isn't as neatly explained if the constraints are in the semantics).
>(and this isn't as neatly explained if the constraints are in the semantics).
I’m inclined to disagree. If we model a counterfactual A → B as a true statement in some context u only if B is true in a set of contexts where A is also true, determined by a selection function, then by imposing certain constraints on that selection function, our semantics for counterfactuals allows for defeasibility.
It seems to me a more straightforward and natural explanation that “If Sydney came over, I’d be happy” sounds true, but “If Sydney came and she were severely injured, I wouldn’t be happy” sounds false if the truth conditions of a counterfactual are such that the former can be true while the latter can be false. On your view, the truth of the former entails the truth of the latter, which seems odd. While I agree that contextual shifts could explain this, such an explanation isn't as developed nor as straightforward as a semantic account based on the truth conditions of counterfactuals without further elaboration. In fact, analyzing our selection function as a similarity relation between contexts seems like a more direct and substantive way to ground our intuition that there are these implicit ceteris paribus clauses in evaluating conditionals that make facts like the strengthening of the antecedent unappealing. This approach is (in my opinion) more satisfying in that appeals to contextual effects, which, though plausible, require additional theoretical elaboration. Simply attributing it to pragmatics, while somewhat explanatory, leaves out the further theoretical work that needs to be developed.
I think the important point in my reply above was that the contextual view almost immediately predicts that reverse Sobel sequences (uttering the weak-antecedent conditional one after the strong-antecedent conditional) should sound bad, which is a big point in its favor — even if it feels less straightforward at first blush, it's much more robust to slight variations of the test case. But to reply to what you've added here — it might be a matter of taste, but I don't think the contextual prediction is especially unstraightforward. Your selection function had better consider context as well, allowing either 'If Caesar were prosecuting the Korean war, he'd use nukes' or 'If Caesar were prosecuting the Korean war, he'd use catapults' based on context. But now the selection function just looks like an unnecessary intermediary, and making it less prone to counterexamples (given all the ways counterfactuals can vary) requires adding ugly epicycles — better to just drop it!
> You would subscribe if you were cool.
Did this work? I mean it worked on me so clearly it worked at least a little bit, but did this post get you more subscribers than the others? I bet it did.
For a quicker version, see part of what?
Part of a Williamson paper... (I've fixed it now, thanks!)