In a discussion of Aquinas' account of religious language, Copleston compares describing a dog as intelligent and describing God as intelligent. Following Aquinas, he calls both descriptions "analogical", apparently indicating by this that the amount of intelligence necessary for a thing to be intelligent is somehow relative to what sort of thing it is. (He also seems to think that, since the two descriptions are analogical, neither assigns "intelligent" its ordinary, literal semantic value. That's weird, but needn't detain us.) He notes a further similarity between the dog's intelligence and God's intelligence. Both of these facts are somehow helpfully elucidated by pointing to the material effects they have had - in the former case, on the dog's behavior; in the latter case, on the (putative) goodness and orderliness of Creation. Copleston's attitude seems to be that this elucidation is semantic. By enumerating more and more of (certain of?) the effects of the dog's intelligence or God's intelligence, we characterize with greater and greater precision what "the dog is intelligent" or "God is intelligent" means, or perhaps what people mean when they call the dog or God intelligent.
Still, Copleston holds that, at least in the case of God, this sort of elucidation can't yield an "adequate positive explanation" of (the meaning of) any sentences describing God as intelligent. This called to mind Goodman's work on dispositional terms. Take an uncontroversially dispositional term like "flammable". One of Goodman's ideas (a little roughly) is that "flammable" picks out whatever property a thing has in virtue of which, in certain relevant circumstances, it lights on fire. Importantly, we can call a thing flammable without knowing exactly what that property is, or being able to give an (adequate) positive explanation of what it is in other terms. It would be too harsh, in such a case, to say that we don't know the meaning of "x is flammable", or what people mean when they utter sentences of that form. Discovering that a certain substance contains hydrogen would certainly furnish something we might call an "adequate positive explanation" of its flammability, but that doesn't mean that our flammability-talk, prior to that discovery, was in any sort of semantic error.
There is nothing special about "flammable" here. Roughly, for any set of truths T, and any thing x, we can define a dispositional predicate "F", such that "F(x)" means that x has a certain property F, such that each element of T is an effect of x's having that property. Now, this will raise all sorts of problems if it turns out that the elements of T obtain for reasons that have nothing to do with x. But, in many cases, this way of introducing a dispositional predicate is totally harmless.* I think Copleston should say that God-talk is just such a case. If, like Copleston and Aquinas, we think we know that all sorts of facts about the world are demonstrably effects of God's general nature and particular actions (perhaps logically following from that nature), then, for any class of such facts, we can define a healthy dispositional predicate for God. If, furthermore, we can specify all (or certain interesting subsets) of the effects that elucidate "God is intelligent", then we can give a decent dispositional semantics for that sentence in line with the schema just described. No problems here, as far as I can see.
I think all this has confirmed a suspicion I already had. If God-talk is non-cognitive, it probably isn't because we can't make sense of the predicates we apply to God. The problem, rather, is probably that God is a difficult thing to refer to, or that the predicate "is a god" in particular (at least in modern Judaeo-Christian discourse) doesn't contribute to the truth-conditions of sentences or utterances in any obvious way. It certainly isn't obvious that the name "God" or the predicate "is a god" are susceptible of the same analysis we have given of (certain) theological predicates.
* - It may turn out that our introduction of F fails to cut nature at the joints. This will be the case when the property of x in virtue of which all of the elements of T obtain is highly disjunctive - if several more fundamental or primitive properties of x are individually responsible for several subsets of T. But F-ness doesn't have to be fundamental in order for "F(x)" to be true.