**Definition 5.7: Let (C) be a condition predicable of an ordering frame (e.g like the**

**5.4 The question of comparability of impossible worlds**

**5.4.1 Weiss’ objection**

First, I’ll outline Weiss (2017) objection and show that (5.10) is valid on the strongest system 𝐂𝐒4∗, characterized by ordering frames that satisfy the stronger totality condition (T2) whereby

all worlds are totally preordered. Then I’ll give a detailed account and critical analysis of Weiss’ alleged counterexample to (5.10), and finally I’ll show that (5.10) is invalidated on systems where (T2) is replaced by (T1).

Weiss (2017) objects to the inference rule (5.10), which holds for all CS logics characterized
*by ordering frames based on preorderings that are total.*

189_{ Notably Pollock (1976) and Kratzer (1981) effectively argue in favour of what would correspond to partial }
orderings on ordering semantics. For a good discussion of the various approaches see Lewis (1981, §3-5).
190_{ Comparability is the basic assumption about comparative similarity of worlds that states: any two worlds x }*and y are comparable to each other in terms of their similarity relative to the world of evaluation z. Totality of *
*preorders captures comparability for ordering frames, and nesting captures this for systems of spheres. *

(5.10) 𝐴 > 𝐵, 𝐵 > 𝐴 ⊨ (𝐴 > 𝐶) ≡ (𝐵 > 𝐶).191_{ }

To be exact, the objection is actually addressed to systems of spheres candidate semantics for
*a non-vacuist account of counterpossibles, and Weiss correctly identifies the nesting *

*condition (see §2.2.3), fundamental for those systems, as responsible for the rule’s validity.*192
*Indeed (5.10) is characteristic of systems of spheres that are nested, and therefore all systems *
of spheres as defined by Lewis (1973). I’ll mirror the discussion in terms of CS* models,
*noting that (5.10) is characteristic of ordering frames based on preorderings that are total, and *
therefore all CS models.193_{ That is, comparability takes the form of nesting on systems of }

*spheres and the form of totality on ordering frames based on preorders. *

The objection is set up via what Weiss takes to be a counterexample to (5.10), formulated in
terms of counterpossibles, and – the argument goes – because all systems of spheres satisfy
nesting, a successful counterexample to (5.10) amounts a counterexample to sphere semantics
in general. This objection extends to all other formulations that encode the intuitions about
*comparative similarity in terms of conditions corresponding to comparability – the basic *
intuition regarding comparative similarity of worlds. Therefore, in particular it is also an
*objection to ordering semantics based on total preorderings. However, Weiss’ conclusion is *
too strong. Surely the alleged counterexample alone, even if correct, doesn’t justify

abandoning comparability altogether, but rather at most justifies lifting the nesting condition for spheres containing impossible worlds, or correspondingly in ordering semantics, lifting the totality condition from impossible worlds. And such a much weaker conclusion is not as damaging to similarity semantics.194

**Proposition 5.2: **𝐴 > 𝐵, 𝐵 > 𝐴 ⊨𝐂𝐒4∗ (𝐴 > 𝐶) ≡ (𝐵 > 𝐶)

191_{ The axiomatic counterpart of (5.10) is CSO: [(𝐴 > 𝐵) ∧ (𝐵 > 𝐴)] ⊃ [(𝐴 > 𝐶) ≡ (𝐵 > 𝐶)], see Nute (1980, }
§3.1).

192_{ Weiss’ (2017) sphere models for non-vacuism are based on sphere models equivalent to S models (see }
chapter 2), whose domains are extended to include non-normal worlds and the truth conditions at non-normal
worlds are extended much in the same manner as I have modified CS models to yield CS* models (definition 5.4
and 5.4).

193_{ Nesting and totality}_{are each other’s counterparts on S frames and CS frames, respectively. For a formal }
*proof of that correspondence see lemmas A.1.0.1 and A.1.0.2 in the Appendix. *

194_{ An axiom, characteristic of all Lewis-Stalnaker logics of counterfactuals and closely related to (5.10), has }
been objected to before by Gabbay (1972) more generally, i.e. even in cases where the antecedent of the
counterfactual doesn’t express an impossibility. Gabbay objects to ((𝐴 > 𝐵) ∧ (𝐵 > 𝐴) ∧ (𝐴 > 𝐶)) ⊃ (𝐵 > 𝐶),
by providing an insightful counterexample to what he believes as illustrating some relevance violating features
of that inference. Note that its failure implies the failure of (5.10).

*Proof : First, I’ll prove that if *𝑖 ⊩ {𝐴 > 𝐵, 𝐵 > 𝐴} for any 𝐂𝐒4∗ model (𝑊, 𝑁, {≲𝑖: 𝑖 ∈ 𝑁}, 𝜌), 𝑖 ∈

𝑊, and 𝐴, 𝐵 ∈ 𝐹𝑜𝑟, then ∃𝑘 ∈ [𝐴] ∩ [𝐵] such that ↓.𝑘 ∩ [𝐴] = ↓.𝑘 ∩ [𝐵]. Assuming 𝑖 ⊩ 𝐴 > 𝐵

implies ∃𝑘 ∈ [𝐴] such that ↓.𝑘 ∩ [𝐴] ⊆ [𝐵], and 𝑖 ⊩ 𝐵 > 𝐴 implies ∃𝑘′ ∈ [𝐵] such that ↓.𝑘′ ∩

[𝐵] ⊆ [𝐴]. Note that in both cases 𝑘, 𝑘′ ∈ [𝐴] ∩ [𝐵]. Either 𝑘 ≲_{𝑖} 𝑘′ or 𝑘′ ≲𝑖 𝑘, by (T2). Suppose

𝑘 ≲𝑖𝑘′. Hence ↓.𝑘 ⊆ ↓.𝑘′. Now, ↓.𝑘 ∩ [𝐴] ⊆ [𝐵] implies ↓.𝑘 ∩ [𝐵] ⊆ ↓.𝑘′ ∩ [𝐵] ⊆ [𝐴]. Hence, ↓.𝑘 ∩

[𝐵] ⊆ [𝐴]. Hence finally, 𝑘 ∈ [𝐴] ∩ [𝐵] and ↓.𝑘 ∩ [𝐴] = ↓.𝑘 ∩ [𝐵]. A similar argument shows

that ↓.𝑘′ ∩ [𝐴] = ↓.𝑘′ ∩ [𝐵] when 𝑘′ ≲𝑖 𝑘. Let us denote such a world, which is guaranteed by

𝑖 ⊩ {𝐴 > 𝐵, 𝐵 > 𝐴} with 𝑘*. Now we will show that 𝑖 ⊩ 𝐴 > 𝐶 implies 𝑖 ⊩ 𝐵 > 𝐶, for any 𝐶 ∈ 𝐹𝑜𝑟. Assuming 𝑖 ⊩ 𝐴 > 𝐶 implies ∃𝑘 ∈ [𝐴] such that ↓.𝑘 ∩ [𝐴] ⊆ [𝐶]. Next, by totality,

either 𝑘* ≲𝑖 𝑘 or 𝑘 ≲𝑖 𝑘*. Now, suppose 𝑘* ≲𝑖 𝑘. Hence ↓.𝑘* ⊆ ↓.𝑘, and we note that ↓.𝑘*∩

[𝐵] = ↓.𝑘*∩ [𝐴] ⊆ ↓.𝑘 ∩ [𝐴] ⊆ [𝐶]. Hence, 𝑖 ⊩ 𝐵 > 𝐶. Next, suppose 𝑘 ≲𝑖 𝑘*. Therefore ↓.𝑘 ⊆

↓.𝑘*, which implies ↓.𝑘 ∩ ↓.𝑘*∩ [𝐴] = ↓.𝑘 ∩ ↓.𝑘*∩ [𝐵]. In conjunction with the hypothesis this

implies ↓.𝑘 ∩ [𝐵] = ↓.𝑘 ∩ [𝐴] ⊆ [𝐶]. Hence, 𝑖 ⊩ 𝐵 > 𝐶. The proof in the other direction is

similar. So, 𝑖 ⊩ 𝐴 > 𝐶 iff 𝑖 ⊩ 𝐵 > 𝐶, as required. □

Now I’ll focus on the alleged counterexample itself given by Weiss (2017), which arms the aforementioned general objection to most similarity accounts of counterfactuals and

counterpossibles. Weiss (2017) formulates it as a variation on Williamson’s (Hempel

Lectures 2006, and Williamson 2007) objection to a non-vacuist account of counterpossibles, presented and discussed earlier in Brogaard and Salerno (2013, pp.649-50). I’ll argue that the context-shift resulting from allowing (as true) certain premises opens the door to formulating other counterexamples that undermine inferences that are valid on all systems Weiss (2017) endorses as alternatives to similarity accounts to counterpossible analysis. But those premises are required for the counterexample to work.

The argument against (5.10) goes as follows:

Fred asks George what 5+7 is, and George mistakenly responds 13. Fred snidely remarks, “if 5 + 7 were 13, you would have answered correctly.” This is true. What else might be the case if 5 + 7 = 13? Plausibly, 5 + 6 = 12. Conversely, if 5 + 6 = 12, it would seem reasonable to expect that 5+7 = 13. From [5.10] and the truth of Fred's initial remark, we can infer “if 5+6=12, George would have answered correctly,” which is not obviously true. (Weiss 2017, p.390)

Let us denote the relevant counterpossibles.

(1) If 5+7 were 13, then George would have answered correctly. (2) If 5+7 = 13, then 5+6 = 12.

(3) If 5+6 = 12, then 5+6 = 13.

(4) If 5+6 were 12, then George would have answered correctly.

Both (2) and (3) seem true enough, although not as obviously as (1) does. Weiss discounts all potential objections attacking the soundness of the argument as question begging on the basis of the intuitive truth of (2) and (3). This riposte has some merit but seems a little too quick, and as such introduces problems of its own. One way of arguing against their truth is to say that contexts where we want (1) to be true, need not always be ones where we’d be also willing to admit (2) and (3) as true. Indeed, there seem to be many ways of arguing against the truth of (2) and (3) in contexts where (1) is true, however I agree that it doesn’t seem obvious that there should be no context at all where we would allow all three to be true, thereby admitting the counterexample as legitimate.

However, caution should be exercised when accepting a general strategy for generating counterexamples that admits the truth of premises whose relevance to the pertinent context can be questioned, because this may pave the way to invalidating more than one has

bargained for. Finally, Weiss discounts all potential counter-objections that would defend the truth of (4), by asserting that it is intuitively false. But to me it doesn’t seem all that much less acceptable than what is already taken on-board when admitting both (2) and (3) as true. As a matter of fact, by admitting those additional premises, (4) doesn’t seem as odd as it would be in their absence.195

The first thing to note is that (1) bares very close resemblance to a statement of

counterpossible identity, and as such is intuitively true. That is, it appears to mean no more and no less than:

(1.a) If 5+7 were 13, then George answering ‘13’ to the question what ‘5+7’ is, would have answered correctly.

195_{ Weiss uses a Sorites kind of reasoning, which employs numerous applications of (5.10), to amplify the }
salience of the falsehood of (4) further (or rather, diminish any intuitive claim to truth that (4) may have) and
derive an arguably much less plausible version (4’). But a similar objection can be set against it, i.e. that the
amplified conclusion (4’) is no less obvious than the assumed stability of reasoning involved in deriving it and the
truth of all the intermediate steps required to arrive at (4’).

Or even more explicitly:

(1.b) ‘If 5+7 were 13, then George saying ‘5+7 is 13’ would be telling the truth’ We’re dealing with a counterpossible whose consequent’s impossible content is not

ampliative relative to the antecedent, i.e. no greater than the content of (1)’s antecedent. The consequent can be said to very naturally follow from the antecedent. Or, yet to put it another way, which (1.b) intends to highlight, (1) is closely related to a relatively safe counter- factual/possible:

*(1.c) If A were true, then saying ‘A’ would be to speak truly. *

Therefore, worlds where the consequent is true would certainly seem like scenarios no
stranger than those corresponding to whatever (impossibility) is expressed in the antecedent
alone. However, this doesn’t seem to be the case for the admission of the truth of (2), which,
on top of the impossibility expressed in the antecedent requires us to accept additional
*assumptions about arithmetic in such impossible situations, which feels stranger than the *
scenario envisaged in (1). That would speak in support of an argument that the context has
indeed shifted – but we’re allowing for that, so let’s continue. In other words, granting the
*truth of (2), and then (3) would seem to be stretching the strangeness of a world w that *
suffices to make (1) true. So, the antecedent world or worlds required to make (2) and (3)
*true, should at least be distinct from the antecedent world w that makes (1) true and certainly *
*no more similar to the actual world than w. Those would seem to be the correct and weakest *
comparative similarity requirements fitting this scenario. With these assumptions about
comparative similarity in place, the task is to salvage the truth of (1), (2), and (3) without
committing to the truth of (4). This is impossible on a notion of comparative similarity of
worlds with unrestricted comparability of worlds, i.e. characterized by total preorders. But
perhaps it could be argued that the antecedent world (or worlds) required for the non-vacuous
*truth of (2) and (3) is not so much stranger than the antecedent world(s) required for the non-*
*vacuous truth of (1) but strange in a different way. This interpretation of comparative *

similarity/dissimilarity of worlds could potentially serve as intuitive motivation for abandoning comparability over impossible worlds.