But more plausibly, he may not have the concept of a meaning. I don’t think its necessary to have this concept in order to mean things by words. If he doesn’t, I think it’s hard to say these are even tacit beliefs. And if he doesn’t have them because he doesn’t have the concept, that isn’t withholding in any interesting sense.

Having said that, I agree that there must be some way to “opt-out” of a social practice, and maybe something like consciously withholding these beliefs is a way of opting-out. I guess I just think that opting out requires more conceptual sophistication than it takes to have the concept HORSE.

And I think even if I grant you everything you say here, it only barely lets Williamson back in the game. He’d still have to show (a) that these metalinguistic beliefs of S are knowledge, no easy task I’d think, and (b) this gives S more knowledge in the salient sense.

One last point. The principle of charity starts to behave very oddly I think when we are changing what S knows by changing the interpretation of words that aren’t used in the sentence, like in “I mean the same thing mommy and daddy mean by ‘horse’”. This isn’t an objection, just that it’s not the way I think the principle of charity is usually used. There are probably some tricky issues here to be worked through.

The key to your argument is that S is, even if a young child, a member of a community with a shared language in which ‘horse’ means horse, instead of horsey-thing.

Of course, it’s POSSIBLE for a person in a community like that to use the word ‘horse’ to mean horsey-thing. I can use words to mean whatever I want. I could stipulate right now, for instance, that my word ‘horse’ means horsey-thing, explicitly saying that I don’t care whether that’s what other members of my community mean by ‘horse’. Obviously, your subject isn’t like that.

What I’m wondering is whether certain (maybe tacit) beliefs are necessary for inheriting the social meaning of the shared word. I guess I think it’s plausible that some might be. These, for instance:

mommy means horse by ‘horse’
daddy means horse by ‘horse’
I mean the same thing mommy and daddy mean by ‘horse’

Both of the following seem plausible to me: First, ordinary children, even unreflective ones, believe and even know many propositions like these. Second, such beliefs/knowledge may be constitutive of the sorts of social practices that give rise to social externalism.

I think the first claim is pretty hard to deny. The second strikes me as less obvious, mostly because people who lack these beliefs are pretty weird. But my thinking is, to deny, or even withhold, on the claim that the members of my community mean what I mean with some word is just what it takes to divorce oneself from the community in the way gestured at by my story about stipulation above.

If both of these claims are true, then Williamson is back in the game.

In general, I think most of the arguments for social externalism will be problems for Williamson, because he makes meaning depend on the speaker’s mental states. Wide states, to be sure, but still just the speaker.

And even if I let S have some of those beliefs, we need to show two things before Williamson is off the hook.

1) The beliefs are knowledge on the natural interpretation.
2) Interpreting S that way produces more knowledge, by the relevant weighting.

As you say, (2) really isn’t obvious, and Williamson doesn’t say much to make us believe it. (There’s really a lot of reliance on what a weighting of knowledge might show, without anything looking like a proof.) But note that (1) isn’t trivial either. It’s true that the natural interpretation makes these beliefs true. But I don’t think it is clear it makes them knowledge. If the ‘weighting’ referred to in (2) is close, then the metalinguistic belief you mentioned may not be knowledge.

I’m not sure if I have a solid grip on just what your subject S is like with respect to his ‘equine’. I wonder whether the agnosticism restrictions you put on him force him to be more alien than it seems.

I take it that you mean to deny that S believes any of the following:

‘equine’ picks out a natural kind.
‘equine’ is the name of a species.
‘equine’ means the same thing that the person who taught me the word ‘equine’ meant by ‘equine’.
sometimes I might mistake a non-equine for an equine, because my eyesight is fallible.

If he believes any of these things, then Williamson can point to a respect in which knowledge maximization favors the HORSE interpretation. (I agree that there is a worry about how to manage trade-offs.)

So I guess S has to fail to believe any of these things, in order for your counterexample to work. He must not believe any of these things even tacitly. (I am assuming that many of our beliefs, and much of our knowledge, is at most times tacit.) It’s one thing to be unreflective — I’m having a harder time getting my head around someone who doesn’t even have such tacit beliefs.

Once we’ve stipulated so many natural beliefs away, I’m not sure it’s obvious that ‘equine’ in S’s idiolect means HORSE. Can you say a little more about why we should favor that interpretation?