on their competence, their reflections may have contained mistakes. Perhaps reflection of sufficient length and depth on one’s competence would lead one to the correct answer to the original question. But the capacity for such more or less philosophical reflection is not a precondition of semantic competence. Philosophers should resist the professional temptation to require all speakers to be good at philosophy.
We can distinguish two levels of reflection, the logical and the metalogical. In response to the original question, logical reflection involves reasoning with terms of the kind in which the question is phrased; the aim is to reach a conclusion that answers the question. For example, one might conclude by classical logic that Mars was always either dry or not dry; one might conclude by fuzzy logic that it is indefinite whether it was always one or the other. The logical level is not purely mechanical. When the reasoning is complex, one needs skill to select from the many permissible applications of the rules one sequence that leads to an answer to the question. When the reasoning is informal, one needs good judgment to select only moves that really are permissible applications of the rules. But one is still thinking about whatever the question was about. One starts only at the metalogical level of reflection to think about the semantics of the logical connectives and other expressions one employed at the logical level. For example, at the metalogical level one may assert or deny that the sentence “Mars was always either dry or not dry” is a logical truth. The rules used at the logical level are articulated only at the metalogical level.
It must be possible to think logically without thinking metalogically, for otherwise by the same principle thinking metalogically would involve thinking metametalogically, and so ad infinitum: our thinking never goes all the way up such an infinite hierarchy. What can prompt ascent to the metalogical level are hard cases in which one feels unclear about the permissibility of a given move at the logical level. One’s mastery of the language and possession of concepts leave one quite uncertain how to go on. In the case of the original question, a salient line of classical reasoning leads to a positive answer: it persuades some competent speakers while leaving others unconvinced. Even to discuss the contentious reasoning we must semantically ascend. We cannot hope to resolve the dispute undogmatically if we never leave the lower level.
3
The argument so far has reached two conclusions at first sight hard to reconcile with each other. First, the original question is not about thought or language. Second, to answer it adequately one must assess rival theories of vagueness in thought and language. How can that way of reaching an answer be appropriate to the original question? We might, therefore, find ourselves tempted back to the idea that somehow the original question was surreptitiously about thought or language.
On further reflection, the combination of the two conclusions is less surprising. Many non-philosophical questions that are not about thought or language cannot be resolved without inquiry into thought or language. Suppose that a court of law must decide whether Smith killed Jones. The question is not who said or thought what. Nevertheless, the crucial arguments may be over whether to trust the witnesses’ testimony. How is what they say now related to what they think now or thought then? How is what they think now or thought then related to what actually happened? Are they lying or sincere? Are their memories confused or clear? Those are questions about their thought and speech. They hold the key to whether Smith killed Jones, even though that question is not about thought about language.12 Of course, the questions about the thought or talk are not about it in isolation from what it is thought or talk about: they are relevant because they concern the relation between the thought or talk and what it is about.
The court must decide the issue on the evidence before it. In a criminal case, does the evidence put it beyond reasonable doubt that Smith killed Jones? In a civil case, does the evidence make it more probable than not? If the court is really deciding a question about testimonial evidence, that is already a question about talk.13 But the question about the evidence arises in virtue of its bearing on the primary question, whether Smith killed Jones. Indeed, the question about the evidence is exactly a question about its bearing on the primary question. So the point stands.
Historians are often in a similar position. They want to know what happened. The way to achieve that is largely by considering documents, linguistic accounts of what happened – not in isolation, but in relation to what they represent. Most obviously, historians want to know whether the documents accurately represent what happened, but to answer that question they must in turn ask about their provenance: who produced them, when and why? Thus the history of the events of primary interest requires a history of thought and talk about those events. Those histories typically overlap, for thought or talk about some part of a complex human event is often another part of the same complex event.
Something analogous occurs in the methodology of the natural sciences. We wish to know the value of some physical quantity. We must devise apparatus to measure it. We may find ourselves in disputes over the functioning of different devices. Although the primary question was not about those measuring devices, we cannot answer it adequately without considering them. We need a theory about the relation between the value of the quantity and the representations of it we record when we use our instruments. The scientific investigation of the physical quantity widens to include the scientific investigation of its interaction with our experimental equipment. After all, our apparatus is part of the same natural world as the primary topic of our inquiry.
These analogies make it less surprising that when we try to answer the original question, which is not a question about thought or language, our main task is to adjudicate between rival theories of vague thought and language. A theory of vagueness validates some deduction that concludes with an answer to the original question. That deduction uses but does not mention vague thought or language. It is formulated at the logical level, like the original question itself, not at the metalogical level. But discursively to justify trusting that deduction, rather than one that reaches another conclusion by other rules, one must assess the rival theories of vagueness.
That theories of vagueness conflict in their answers to the original question shows that they are not confined to claims about thought and talk. Theories such as epistemicism and supervaluationism which employ classical logic have ‘Mars was always either dry or not dry’ as a theorem, once they are formulated in a suitably expressive language. To reiterate, that theorem is not about thought or talk.
For the three-valued and fuzzy approaches, the matter is only slightly more complicated. Their proponents assert:
(C) It is indefinite whether Mars was always either dry or not dry.
On those approaches, C does not count as about thought or language. Strictly speaking, however, C does not follow from the three-valued or fuzzy theory of vagueness itself; for all the theory implies, there was never any liquid on Mars, in which case it would always have been either dry or not dry, even by three-valued or fuzzy standards, and so would not have been indefinite. The theory implies only a conditional theorem:
(P1) If it was once indefinite whether Mars was dry then it is indefi- nite whether Mars was always either dry or not dry.
Three-valued or fuzzy theorists can combine P1 with what they regard as an empirical truth about Mars:
(P2) It was once indefinite whether Mars was dry.
From P1 and P2 they use the rule of modus ponens (from “If P then Q” and “P” infer “Q”) to infer C, the answer to the original question. Although their theorem P1 does not answer the question by itself, it is no more about thought or language than C is. Their theories are just as committed as classical ones to making claims that are not about thought or language.
In principle, just as the considerations relevant to adjudicating the dispute between theories of vagueness are relevant to answering
