Against Magnetism

Abstract

Magnetism in meta-semantics is the view that the meaning of our words is determined in part by their use and in part by the objective naturalness of candidate meanings. This hypothesis is commonly attributed to David Lewis, and has been put to philosophical work by Brian Weatherson, Ted Sider and others. I argue that there is no evidence that Lewis ever endorsed the view, and that his actual account of language reveals good reasons against it.

Keywords:

• meta-semantics
• reference magnetism
• naturalness
• meaning
• David Lewis

Notes

1 Throughout, Lewis's papers are cited by the date of original publication, but page references, where applicable, are always to the reprinted versions.

2 ‘New Work for a Theory of Universals’ [1983b] and ‘Putnam's Paradox’ [1984] are not on this list, for reasons given in §5.

3 This is obviously only a rough sketch of Lewis's account. Among other things, I have ignored non-assertoric speech acts, implicatures, vagueness, and the ‘two-dimensionalist’ distinction between truth-conditions in the present sense (‘meaning₂’ in Lewis [1975: 173]) and truth-conditions in the sense of Kaplanian content (‘meaning₁’). My sketch matches what Lewis himself took to be the core of his meta-semantics [Lewis 1975; 1986b: 40–50; 1992], and the complications would not make a significant difference to the topic of this paper.

4 Lewis's argument in his [1983b] uses Jeffrey's [1965] ‘evidential’ notion of expected utility. It is known that in Jeffrey's decision theory, unlike in Savage's, hypothetical choices (or preferences) do not determine a unique assignment of beliefs and desires [Joyce 1999: 122–7]. Lewis [1981] himself endorsed a version of Savage's theory. However, he presumably rejected Savage's permissive conception of acts as arbitrary functions from ‘states’ to ‘outcomes’, which is crucial for the proof of Savage's representation theorem.

5 Here I ignore Lewis's [1996] puzzling suggestion that our evidence consists of propositions about our brain states.

6 Admittedly, at other places Lewis is not as clear about the distinction between fundamentality and undefinability as he should have been. The importance of this distinction is emphasized in Busse [2009].

7 Contrary to widespread reports, Lewis never suggested that the abstract mathematical function of addition is objectively more natural than quaddition. Sider [2004: 15] cites this as a precedent for extending naturalness to logical and mathematical notions.

8 Note 6 of Lewis [1993] might also be cited, although it only concerns thought, not language. In n. 2 of his [2004], Lewis mentions the hypothesis that even though there are many candidate referents for paradigmatically vague terms like ‘the cloud’, one of these candidates is the true referent in virtue of being ‘a mighty reference magnet’, but he does not appear to endorse this idea.

9 Analogous problems arise for Global Descriptivism as a meta-semantics for propositional attitudes—a form of ‘conceptual role semantics’ attributed to Lewis in Stalnaker [2004b] (see n. 10 below). Here the idea would be that agents have a certain set of sentence-like representations in their heads, whose content is given by whatever interpretation makes most of them come out true. Again, this ignores many facts that are clearly relevant to the content of an intentional state, such as the conditions under which the state typically occurs, the (non-verbal) behaviour it typically causes, etc. It also makes expressive redundancy a precondition of mental representation, and it radically underdetermines the content of mental states.

10 Robert Stalnaker [2004b] reads the passage just quoted from ‘Putnam's Paradox’ as indicating that Lewis endorsed magnetized Global Descriptivism as his theory of mental content. This is a very strange interpretation. It not only ignores the attached footnote, it is also completely at odds with Lewis's numerous writings on intentionality both before and after 1984. Stalnaker notes that his interpretation does not square with things Lewis says elsewhere, but takes this to be a problem for Lewis. How could he think that ‘Putnam's Paradox’, of all places, contains Lewis's final theory of mental content? (Why not consult, say, the section ‘Content’ in ‘Reduction of Mind’ [Lewis 1994]?) The answer, I suspect, is that Stalnaker didn't fully distinguish two notions of narrow content. Lewis always maintained that mental content is narrow in the sense that functional duplicates in the same world share their beliefs and desires, no matter whether they live on Earth, on Twin Earth, or in a vat [1979a: 142f; 1994: 312–14]. Stalnaker disagrees and holds that the contents of our beliefs are in part determined by the objects to which we are actually causally related—H₂O, for example, and not XYZ. He seems to assume that the only way to avoid the consequence that content is in this sense wide is to assume that mental content is determined without any recourse to external causes and effects [Stalnaker 1993; 2004a]. This would indeed only leave room for something like phenomenalism or a mentalese version of Global Descriptivism. But Lewis's mental content is not narrow in the sense of being determined independently of external causes and effects. On the contrary, such causal connections play a big role in Lewis's folk psychological analysis.

11 I am not the first to suggest that the application conditions of words sometimes involve objective naturalness: see e.g. Lewis [1994: 313] and Jackson [1998a: 95].

12 This paper is an updated and extended version of a manuscript circulated in 2006 under the title ‘Lewisian meaning without naturalness’. Thanks to Karl Schafer, Brian Weatherson, J. Robert G. Williams and the metasemantics reading group at the ANU for comments and discussion.