Truth conditions for the counterfactual with the Limit Assumption

In document Context-indexed counterfactuals and non-vacuous counterpossibles (Page 79-82)

The counterfactual 𝐴 > 𝐵 is true at world w (according to $) if and only if There is no A-permitting sphere in $𝑤: ⋃$𝑤∩ [𝐴] ≠ ∅


B holds at every A-world in the smallest A-permitting sphere: ⋂{𝑆 ∈ $𝑤: 𝑆 ∩ [𝐴] ≠ ∅} ⊆ [𝐵]

Or even simpler:

The counterfactual 𝐴 > 𝐵 is true at world w (according to $) if and only if B holds at every closest (most similar) 𝐴-world: [𝐴] ⊆ [𝐵]

(Where [𝐴] ∶= ⋂{𝑆 ∈ $𝑤: 𝑆 ∩ [𝐴] ≠ ∅} ∩ [𝐴] and no world outside of ⋃$𝑤 is closest).

Lewis observes that this assumption can’t always be made, and consequently decides against adding it to the conditions that characterize $. He argues that because there may be cases where LA is simply false, truth conditions from Definition 2.22 would give the wrong

analysis. Consider the following argument: I’m thinking (at this moment) of a line that is half a unit in length (in Euclidian 3-space). But let us suppose, counterfactually, that I thought (just then) of a line that was longer than a unit. That is, the counterfactual antecedent is ‘I’m thinking of a line that is longer than a unit’, abbreviated with ‘1 < 𝐿’. Next, we see that {𝑆 ∈ $@: 𝑆 ∩ [1 < 𝐿] ≠ ∅} is the set of all (1 < 𝐿)-permitting spheres, i.e. spheres containing worlds where I’m thinking about a line that is longer than a unit. Now, here’s a key step in Lewis’ argument: the worlds where I’m thinking of a line that is 1½ units long are more similar to the actual world than worlds where I’m thinking of a line that is 2 units long, and

89 (Lewis 1973, §1.4, pp.19-20)

90 The Limit Assumption is explicit in Stalnaker’s truth conditions for the counterfactual – which I discuss in the next section – so this argument can be viewed as being indirectly aimed at Stalnaker’s analysis.

likewise the worlds where I’m thinking of a line that is 1¼ units long are more similar to the actual world than worlds where I’m thinking of a line that is 1½ units long, and so on, ad infinitum.91 The shorter we make the line (above 1 unit), the more similar we make it to the length that I actually thought about, so presumably, the closer we come to the actual world.92 But how long is the line that I’m thinking about in the closest (most similar) (1 < 𝐿)-worlds? The short answer is that there is no such length. 93 Since there is no smallest length that is greater than one unit, there are no worlds where I’m thinking about a line with such length, and consequently no smallest sphere containing all and only such worlds. Hence,

⋂{𝑆 ∈ $𝑤: 𝑆 ∩ [1 < 𝐿] ≠ ∅} = ∅.94

On the truth conditions given in Definition 2.22, the counterfactual ‘If I thought just then about a line that is longer than one unit, then…’ would be vacuously and erroneously evaluated as true for any consequent. In particular, the following counterfactual would be erroneously evaluated as true: ‘Had I been thinking about a line that is longer than one unit, then I would have been thinking about a line that is not longer than one unit’. But it would be evaluated vacuously as true not because the antecedent does not express an entertainable proposition (clearly it does, e.g. take worlds where I was thinking of the line being 1½ units long), but because there are no closest (most similar) accessible worlds where it is true.

There have been a number of replies to Lewis’ argument against LA.95 Some (Pollock 1976,

Stalnaker 1980) demonstrate questionable implications of denying LA,others (Hájek 2014) argue that the implications about comparative similarity stemming from Lewis’ formulation of the argument against LA commit him to the truth of clearly false counterfactuals. Another

91 Lewis uses a different example – one involving a physical line (that is actually less than an inch long and the counterfactual supposition is that it is more than 1’’ long), printed on a page, but admits that such examples aren’t decisive, since given the printing process, there would be a limit on how long a line can be printed, being contingent on the printing process (e.g. the number of ink molecules on the page being finite). To avoid this, I have chosen a to exemplify an idealized (mathematical) line, which isn’t subject to these kinds of physical restrictions. To make it unambiguous and precise, say that I’ve been actually thinking about the half-unit vector 0.5𝒊 in the representation of Euclidian 3-space ℝ3 (where 𝒊 is the unit vector in the 𝑥-coordinate direction), and in the counterfactual supposition I’m thinking of some 𝑙𝒊 where 1 < 𝑙 ∈ ℝ.

92 This part of Lewis’ argument gets him in trouble (a point I’ll return to shortly) and seems to be at odds with what he says earlier: “And so it goes; respects of similarity and difference trade off. If we try too hard for exact similarity to the actual world in one respect, we will get excessive differences in some other respect” (Lewis 1973, p.9).

93 This follows from the property of real numbers that for any two real numbers 𝑥 < 𝑦 there is a third real number 𝑧 such that 𝑥 < 𝑧 < 𝑦.

94 Based on Lewis (1973, §1.4, p.20).

family of replies (Stalnaker 1980, Brogaard & Salerno 2013) involve appeal to a conception of comparative similarity – one that emphasizes the respects of similarity that are relevant to the context in which a counterfactual is being considered – which can diffuse the problematic consequence of accepting LA, identified by Lewis, and avoid the aforementioned issue raised by (Hájek 2014). I adopt a version of this conception of comparative similarity – one that is consistent with S1+W – and devote part of Chapter 4 to motivating and defending it.For now, we turn to an example of a theory that has adopted the limit assumption.

2.2.7 Stalnaker’s theory and conditional excluded middle

Having discussed various conceptions of comparative similarity of worlds in the previous sections, we have only hinted at the character of Stalnaker’s theory. In this section we examine carefully the systems of spheres corresponding to the account of the conditional given by Stalnaker (1968, 1970), which, aside from the limit assumption places another significant restriction on $. It’s yet another addition to the kinds of assumptions about comparative similarity that result in logics that are misaligned with the intended logic of counterfactuals.

I won’t discuss the formal account of Stalnaker’s (1968, 1970) original model theory, other than saying that he had originally based his semantics on Kripke frames, augmenting them with a world selection function 𝑓: 𝐹𝑜𝑟 × 𝑊 ⟶ 𝑊 that selects for each antecedent-world pair (𝐴, 𝑤) a single world 𝑓(𝐴, 𝑤) regarded as the most similar 𝐴-world to 𝑤. Lewis (1973, §3.4) shows how such selection-function models can be equivalently expressed in terms of his own similarity sphere models (given appropriate restrictions on $) – and that’s the manner in which I choose to talk about Stalnaker’s theory in this chapter, which after all is devoted to Lewis’ semantics.96

Stalnaker’s truth conditions for the counterfactual:

“Consider a possible world in which A is true, and which otherwise differs minimally from the actual world. ‘If A, then B’ is true (false) just in case B is true (false) in that possible world.”97

96 For an account of conditional logics in terms of the selection functions, and a comprehensive comparison of Stalnaker’s and Lewis’ theories in terms of selection functions see (Priest 2008, §5) or (Lewis 1973, §2.7, §3.4). For a general discussion of the selection-function semantics for conditional logics see (Chellas 1975) or (Nute 1980, §3.2).

Apparently, in addition to the limit assumption – that for any world w and entertainable antecedent A there is at least one A-world that is most similar to w – Stalnaker makes a stronger assumption, i.e. that there is a unique world like this.98 It leads to an important difference between Lewis’ and Stalnaker’s accounts of the counterfactual. Adding this assumption to S1 validates Conditional Excluded Middle (abbr. CEM), i.e. ‘(𝐴 > 𝐵) ∨ (𝐴 > ~𝐵)’.99 Lewis calls CEM the “principal virtue and the principal vice” of Stalnaker’s theory,

presumably because although it may appeal to the intuitions of ordinary language users, nevertheless it’s hardly true of the subjunctive conditional – as I’ll shortly demonstrate via Lewis’ argument against CEM.

Definition 2.23: Stalnaker’s uniqueness assumption: for every world 𝑤 and antecedent 𝐴, that

In document Context-indexed counterfactuals and non-vacuous counterpossibles (Page 79-82)