Reconstructing Davidson's Truth-Conditional Approach to Metaphor — Epoche C1
"Achilles is a lion" is false: Achilles was a man, and lions are large cats of the genus Panthera . The sentence nonetheless tells a hearer something, and the question this note addresses is whether that something belongs inside a theory of meaning or outside it. Donald Davidson's answer, in his essay "What Metaphors Mean" (1978), is that it belongs outside: the sentence means exactly what its words literally mean and carries no second, metaphorical content. The proposal here is that the cognitive work metaphor plainly does can be given a precise place — as a normative inferential move that fixes the standards against which later literal assertions are evaluated — and that this reconstruction, unlike the theories Davidson attacked, does not require adding any meanings to his semantics. Making that case requires first setting out what a truth-conditional theory of meaning is, since the whole dispute is about what such a theory may and may not contain. What a truth-conditional semantics is, and why metaphor looks like a threat A truth-conditional theory of meaning holds that to know what a sentence means is to know under what conditions it would be true. The technical shape of such a theory comes from Alfred Tarski (1944), who was addressing not natural language but the problem of defining truth for a formal language without paradox. Tarski laid down an adequacy condition, Convention T: a definition of truth for a language $L$ counts as adequate only if it entails, for every sentence $s$ of $L$, a biconditional of the form $$s \text{ is true-in-}L \ \Longleftrightarrow \ p$$ where $s$ is named on the left and $p$ is its translation into the language in which the theory is written. The familiar instance is: "Snow is white" is true in English if and only if snow is white. The instances look trivial one at a time; the theory is not trivial, because it must generate all of them from a finite base. Davidson (1967) proposed taking this apparatus as a theory of meaning for natural language. His argument for doing so is worth stating, because it is what makes metaphor a problem rather than a curiosity. A speaker of English can understand sentences she has never encountered, and there is no upper bound on how many. If understanding a sentence required learning its meaning separately, a finite creature could not do this. So the meanings of complex expressions must be generated recursively from a finite stock of words and a finite stock of composition rules — the principle of compositionality. A Tarski-style truth theory does exactly this: finitely many axioms for words, finitely many rules for combining them, infinitely many truth conditions entailed. Now the threat. If a metaphorical utterance carried, in addition to its literal truth conditions, a distinct metaphorical content, that content would have to be generated too, on pain of being unlearnable. And no one has produced rules that generate it. There is no function taking the literal meaning of "lion" to the metaphorical meaning it bears when applied to Achilles, and applying reliably to the next metaphor. The problem is not that metaphors are hard; it is that they are not systematic in the way the compositional argument requires. Davidson's verdict, stated accurately Davidson's response in 1978 is the most economical available: there is no metaphorical meaning to generate. Metaphors mean what their words literally mean, and nothing further; what is special about a metaphor lies in what a speaker does with a sentence, not in what the sentence means. Metaphor, on this account, belongs to the theory of use rather than to the theory of meaning, and the effect it produces — bringing a hearer to notice a likeness, to attend to the subject in a new way — is not the transmission of a proposition. He calls metaphor "the dreamwork of language" (Davidson 1978), which signals precisely this: something that works on us, and whose interpretation reveals as much about the interpreter as about the speaker. Two features of this position are commonly misreported, and the earlier version of this note misreported both. First, Davidson is not claiming that metaphor is ornamental or unimportant. His view is that metaphor does substantial cognitive work and that this work is not the conveying of content; the claim is about classification, not about value. To read him as demoting metaphor to decoration is to answer a position he did not hold. Second, it is not the case on his view that metaphor is constituted by literal falsehood. Patent falsity is a common signal that a metaphorical reading is intended, but it is neither necessary nor sufficient: some metaphors are literally true, some literal falsehoods are simply mistakes, and a theory that keys metaphor to falsity therefore has the wrong extension. What does the work is the hearer's recognition that a literal reading is not what the occasion calls for. The middle position, and why Davidson rejects it Between "metaphors have a second meaning" and "metaphors have no cognitive content" lies the position most philosophers of language occupy, and it must be on the table before the reconstruction can be assessed. Paul Grice's framework, collected in Studies in the Way of Words (1989), distinguishes what a sentence means from what a speaker means in uttering it, and gives an account of how a hearer computes the second from the first: conversation is governed by an expectation of cooperation, and when a speaker flouts that expectation at the literal level — by saying something obviously false — the hearer infers a different intended message. Metaphor is then a flouting of the maxim of quality, and the "metaphorical meaning" is a conversational implicature: something meant but not said. Max Black (1954) had earlier given a richer account of what gets conveyed. On his interaction view, a metaphor has a focus (the metaphorical word) and a frame (the rest of the sentence); understanding it consists in applying to the primar