I finally got around to reading Jeff King’s paper on syntactic evidence for semantic theories, and I was struck by one of the examples he uses. At first I thought what he said was obviously wrong, but on reflection I think it might not be so obvious. (Most of the paper seems right to me, at least on a first reading through, but I didn’t have anything detailed to say about those parts. Well, except to note that debates in philosophy of language are getting pointed nowadays.)

Anyway, here was the point that I think I disagree with. Jeff wants to argue that syntactic data can be sometimes used to support semantic theories. One example of this (not the only one) involves negative polarity items (NPIs). Examples of NPIs are ever and any when it is meant as an existential quantifier. It seems these words can only appear inside the scope of a negation, or in a context that behaves in some ways as if it were inside the scope of a negation.

Simplifying the argument a little bit, Jeff seems to suggest that the following argument could be used to provide support for its conclusion.

(1) NPIs are licenced in the antecedents of conditionals

(2) NPIs are only licenced in downwards entailing contexts

(3) The antecedent of a conditional is a downwards entailing context

A ‘downwards entailing context’ is (roughly) one where replacing a more general term with a more specific term produces a logically weaker sentence. So while (3a) does not entail (3b), thus showing ordinary contexts are not downwards entailing, (3c) does entail (3d), showing negated contexts are not downwards entailing.

(3a) I will be given a birthday cake tomorrow.

(3b) I will be given a poisonous birthday cake tomorrow.

(3c) I will not be given a birthday cake tomorrow.

(3d) I will not be given a poisonous birthday cake tomorrow.

(I assume here that poisonous birthday cakes are still birthday cakes. I do hope that’s true, or all my examples here will be no good.)

(2) was first proposed (to the best of my knowledge) in William Ladusaw’s dissertation in I think 1979, and it has been revised a little since then, but many people I think hold that it is something like the right theory of when NPIs are licenced. But it does have one striking consequence: it implies (3). To give you a sense of how surprising (3) is, note that it implies that (4) entails (5).

(4) If I am given a birthday cake tomorrow, I will be happy.

(5) If I am given a poisonous birthday cake tomorrow, I will be happy.

Now, many people think that (4) could be true while (5) is false. It is certainly the case that there are contexts in which one could say (4) and not say (5). Perhaps the best explanation for that is pragmatic. Those who think that indicative conditionals are either material or strict implications will hold that it is pragmatic. But perhaps it is semantic. Officially, I think it is semantic, though I think the other side of the case has merit.

Here’s where I think I disagree with Jeff. Imagine I am undecided about whether (4) really does entail (5). I think that the argument (1), (2) therefore (3) has no force whatsoever towards pushing me to think that it does. Rather, I think that only evidence to do with conditionals can tell in favour of the entailment of (5) by (4), and if that evidence is not sufficient to support the entailment claim, all the worse for premise (2).

At least, that was what I thought at first. On second thoughts, I think maybe I was a little too dogmatic here. On further review, though, I think my dogmatism was in the right direction. To test this, try a little thought experiment.

Imagine you think that all the evidence, except the evidence about conditionals, supports (2), or some slightly tidied up version of it. (This is not too hard to imagine I think, (2) does remarkably well at capturing most of the data.) And imagine that you think that while there are pragmatic explanations of the apparent counter-examples to If Fa then p entails If Fa and Ga then p, you think those explanations are fairly weak. (Again, not too hard to imagine.) Does the inductive evidence in favour of (2), which we acknowledge is substantial, and the obvious truth of (1) give you reason to take those pragmatic explanations more seriously, and more generally more reason to believe that If Fa then p does entailIf Fa and Ga then p? I still think no, but I can see why this might look like dogmatism to some.

I sometimes play at being a semanticist, but at heart I’m always a philosopher. And one of the occupational hazards of being a philosopher is that one takes methodological questions much more seriously than perhaps one ought. So at some level I care more about the methodological question raised in the last paragraph than I care about the facts about conditionals and NPIs. At that level, I’m rather grateful to Jeff for raising this question, because it’s one of the harder methodological questions I think I’ve seen for a while.