Skip to main content

The representational limits of possible worlds semantics


This paper evaluates Stalnaker’s recent attempt to outline a realist interpretation of possible worlds semantics that lacks substantive metaphysical commitments. The limitations of his approach are used to draw some more general lessons about the non-representational artefacts of formal representations. Three key conclusions are drawn. (1) Stalnaker’s account of possible worlds semantics’ non-representational artefacts does not cohere with his modal metaphysics. (2) Invariance-based analyses of non-representational artefacts cannot capture a certain kind of artefact. (3) Stalnaker must treat instrumentally those aspects of possible worlds formalism governing the interaction between quantification and modality, under any analysis whatsoever of non-representational artefacts.

This is a preview of subscription content, access via your institution.


  1. Assumption: what’s possibly necessary is actual; \(\Diamond \Box A \rightarrow @A\).

  2. I am careful about use/mention only when necessary to avoid misunderstanding. ‘\(A\)’ and ‘\(B\)’ are usually metalinguistic variables ranging over sentences, though I sometimes use them schematically. Formal expressions are usually, but not always, mentioned.

  3. Both sections draw on and elaborate (Stalnaker 2012, chs1–2).

  4. (Stalnaker 2012, pp. 42–43) endorses this thesis, arguing that it’s required by the actualist thesis that actuality exhausts reality.

  5. I extract commitment to NSW from (Stalnaker 2012, pp. 13–20, 28).

  6. A singular witness for an existential proposition (or fact) of the form something is such a singular proposition (or fact) of the corresponding form \(x\) is such that.... An individual \(x\) is a witness for the existential iff \(x\) is such that.... (Complete the dots in the same way each time.)

  7. Recall that mere possibilia are simply actual individuals that occupy a certain formal semantic role: they’re elements of one of a model’s point-domains not in its privileged point-domain. They are not merely possible objects.

  8. I use fact-talk in a lightweight sense: for there to exist a fact that \(p\) just is for it to be that \(p\). An ontology of facts is not what’s at issue. Fact-talk is merely an expository convenience.

  9. This application draws heavily on (Williamson 2013, 189–190).

  10. A variant approach employs a separate assignment of qualitative characters to intensions, independently of any language and valuation.

  11. Alternatively: \(I\) represents a qualitative character related to those represented by other intensions in the appropriate way.

  12. Because our object language lacks constant terms, elements of the privileged domain cannot receive representational import from the object language directly, as intensions do.

  13. Better: the existence of mere possibilia and elements of \(D(w_@)\) standing in a certain network of relations to one another should be invariant under \(\approx\).

  14. Although models with a single mere possibilium are problematic, they threaten only the letter of the view, rather than its spirit. Employing the actuality-fixing isomorphisms of (Williamson 2013, note 49 on p192) in place of \(\sigma\)-functions in the definition of \(\approx\) would resolve the problem. Actuality-fixing isomorphisms are like \(\sigma\)-functions, except allowing the sets of mere possibilia and points to vary (with fixed cardinality).

  15. Williamson’s primary concern here lies not with \((\exists \hbox {con})\) but with the Barcan Formula \(\Diamond \exists x A \rightarrow \exists x \Diamond A\). The issues are closely connected. Every countermodel to the Barcan Formula satisfies \((\exists \hbox {con})\). And every model that satisfies \((\exists \hbox {con})\) is either a countermodel to the Barcan Formula, or differs from one only over the valuation function.

  16. Objection: \(o\) is a mere possibilium of all models bearing \(\approx\) to \(m\); so that’s representational on Stalnaker’s approach; so \(m\)’s representational content isn’t silent about \(o\). Response: modify the definition of \(\approx\) by replacing \(\sigma\)-functions with actuality-fixing isomorphisms. See note 14.

  17. Thanks to a referee for bringing this to my attention.

  18. For details, see (Williamson 2013, Sect. 7 of ch5). Williamson’s semantics interprets object-language quantifiers as having the same (unrestricted higher-order analogue of a) domain at each point. This needs complicating to accommodate the proposal in the text. Each point needs associating with a higher-order analogue of a domain, and the point-relative satisfaction-conditions—captured using Williamson’s predicate ‘TRUE’—for quantified sentences restricting accordingly. I won’t go into details here.

  19. The assumption is reasonable given reality’s infinitude.

  20. Example: sentences featuring empty names plausibly lack truth-conditions. On this approach, the variant model lacks accuracy conditions, unlike the original model \(m\).

  21. Objection: the set of mere possibilia is invariant under actuality-fixing permutations of the outer domain; so the membership of that set should be representational; so models represent modal reality as fixing which non-actual things there could be, leaving it open which qualitative roles they could occupy. Reply: liberalise Stalnaker’s definition of \(\approx\) by employing actuality-fixing isomorphisms in place of \(\sigma\)-functions, as suggested in note 14. Note that the argument in the text survives this modification.

  22. On measurement theory, see (Krantz et al 1971, esp. 9–12); on the semantic conception of theories, see (Suppes 2002, esp. ch.4).

  23. The example is inspired by a similar one in Cook (2002).

  24. Certain forms of representational redundancy, or overdetermination of content, may even allow representors to vary between representationally equivalent models. I won’t examine that here.

  25. More cautiously: the nature and intrinsic structure of that truth-condition aren’t represented; only its relationships to other truth-conditions are represented.

  26. Rejection of the Barcan Formula \(\Diamond \exists x A \rightarrow \exists x \Diamond A\) is also required. Defenders of \((\exists \hbox {con})\) must reject the Barcan Formula anyway; otherwise \((\exists \hbox {con})\) entails the actually inconsistent: \(\exists x \Diamond @ \forall y (x \ne y)\). Note 15 outlines the intimate model-theoretic connection between \((\exists \hbox {con})\) and the Barcan Formula.


  • Cook, R. T. (2002). Vagueness and mathematical precision. Mind, 111, 225–247.

    Article  Google Scholar 

  • Krantz, D. H., Luce, R. D., Suppes, P., & Tversky, A. (1971). Foundations of measurement volume 1: Additive and polynomial representations. New York: Academic Press.

    Google Scholar 

  • Stalnaker, R. (1994). The interaction of quantification with modality and identity. In W. Sinnott Armostrong & N. A. D. Raffman (Eds.), Modality, Morality and Belief: Essays in honor of Ruth Barcan Marcus. Cambridge: Cambridge University Press (Reprinted with a Postscript in (Stalnaker, 2003, ch8)).

  • Stalnaker, R. (2003). Ways a world might be: Metaphysical and anti-metaphysical essays. OUP.

  • Stalnaker, R. (2010). Merely possible propositions. In B. Hale & A. Hoffman (Eds.), Modality: Metaphysics, logic, and epistemology (pp. 21–33). (OUP. chap 1).

  • Stalnaker, R. (2012). Mere possibilities: Metaphysical foundations of modal semantics. Princeton: Princeton University Press.

    Google Scholar 

  • Suppes, P. (2002). Representation and invariance. Stanford: CSLI Publications.

    Google Scholar 

  • Williamson, T. (2013). Modal logic as metaphysics. OUP.

Download references


For comments and discussion, I’m grateful to Justin Clarke-Doane, Billy Dunaway, Peter Fritz, Anil Gomes, Bob Hale, Rory Madden, Ian Phillips, Scott Sturgeon, Lee Walters, and Al Wilson. Thanks also to an audience in Nottingham, and the members of the Hossack-Textor WiP group. I’m especially grateful to Bob Stalnaker, for detailed and instructive written comments.

Author information

Authors and Affiliations


Corresponding author

Correspondence to Nicholas K. Jones.

Appendix: A fully realistic semantics?

Appendix: A fully realistic semantics?

An appendix to Mere Possibilities presents a variant semantics. Stalnaker describes it thus:

It is technically feasible to do the compositional semantics in a way that mixes reference to real things with artefacts of the model and then use the equivalence relations to filter out the artifacts at the end, but one might hope to do the semantics more directly, where all the values of a complex expression...are entities that exist in the actual world. Our Tarskian semantics makes reference to infinite sequences of possible individuals, sequences that may include “merely possible individuals” that are artifacts of the model. But this was just a notational convenience. Using the methods sketched in appendix B, we could do the compositional semantics directly, where all intermediate values in the composition are actual things. (Stalnaker 2012, pp. 124–125)

However, the semantics described in appendix B addresses not the problems above, but a slightly different issue. To avoid confusion, I now explain what Stalnaker’s appendix B does and doesn’t address.

The semantic theories of appendix B and Sect. 3 differ in two relatively trivial ways.

First way. Stalnaker does not relativise satisfaction to variable assignments. Assignments are needed to handle quantification. So Stalnaker takes an alternative approach to quantification. Rather than quantifying over assignments, he quantifies over extensions of the valuation function to new constant terms. The new clause for \(\exists\) is:

  • \(m, w \,\Vdash\, \exists v A\) iff, for some \(d \in D(w) : m^{[d/a]}, w \,\Vdash\, A^{[a/v]}\).

\(a\) is a new constant, not present in the original language. \(A^{[a/v]}\) is the sentence obtained by substituting \(a\) for all occurrences of \(v\) bound by the initial quantifier in \(\exists v A\). \(m^{[d/a]}\) is a model just like \(m\), except for its valuation \(V^{[d/a]}\). \(V^{[d/a]}\) is just like \(m\)’s valuation \(V\), except that \(V^{[d/a]}(a)=d\). The resulting satisfaction-condition for, say, \(\Diamond \exists x A\) involves quantification over all elements of the outer domain and consideration of all the associated extensions of the valuation function. So this is almost a notational variant of the assignment-based approach in Sect. 3. The only real difference is that Stalnaker’s assignments are defined only for variables appearing in the sentence, rather than for all variables in the language. That could be mimicked with partially defined assignments.

Second way. Stalnaker’s valuations of constant terms are point-relative. For a constant term \(a\), \(V_w(a)\) is the same object \(d\) at every point \(w\) whose domain contains \(d\), and undefined otherwise. We can mimic this with point-relative assignments subject to the same constraint. For example, the new clause for atomic predication is:

  • \(m, w, \alpha \,\Vdash\, \varPhi v_{1},\ldots ,v_{n}\) iff \(\langle \alpha _w(v_1), \ldots , \alpha _w(v_n) \rangle \in V(\varPhi )(w)\).

What are the gains of this approach? Think of the semantic value of an open sentence (of one free variable) as a function that takes, at a world \(u\), each individual \(x\) to a singular proposition about \(x\). This function should be defined at \(u\) only for individuals that exist at \(u\): since there are no other individuals at \(u\), there are no other arguments to consider at \(u\), and no singular propositions about them to return as value. Functions defined on objects that don’t exist at \(u\) themselves don’t exist at \(u\). Treating assignments as insensitive to worlds forces the (formal representatives of) semantic values of open sentences to be defined relative to whatever there could possibly be (since they’re defined relative to all mere possibilia). We can avoid this by using world/point-relativised assignments subject to the same constraint as Stalnaker places on valuations: \(\alpha _w(v)\) is the same object \(d\) at every point \(w\) whose domain contains \(d\), and undefined otherwise.

The quote above contrasts Stalnaker’s semantics with one that uses infinite sequences to determine assignments. On that approach, the semantic value of an open sentence at a world is defined relative to all such sequences. So it’s defined at some worlds relative to sequences containing things that don’t exist at that world. Interpreting this realistically will result in inaccuracy. Stalnaker’s focus on valuations for extensions of the language by a single constant, together with his world-relativisation of valuations, avoids this (Stalnaker 2012, pp. 143–147). The semantic theory in Sect. 3 does not involve infinite sequences; it quantifies over variable assignments directly. Employing partially defined, point-relative assignments as just described allows us to mimic Stalnaker’s approach, thereby avoiding the problems arising from satisfaction by infinite sequences.

None of this bears on the issues discussed above. The satisfaction-condition for \(\Diamond \exists x A\) still requires a witness for the satisfaction-condition of \(A\). Since its truth-condition doesn’t require a witness for the truth-condition of \(A\), the mismatch between satisfaction-conditions and truth-conditions remains. Stalnaker’s appendix B does not resolve and (as I read Stalnaker) is not intended to resolve, the problems with existential contingency that I have been discussing.

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Jones, N.K. The representational limits of possible worlds semantics. Philos Stud 173, 479–503 (2016).

Download citation

  • Published:

  • Issue Date:

  • DOI:


  • Stalnaker
  • Contingent existence
  • Possible worlds semantics
  • Invariance
  • Artefacts