Abstract
We investigate two different broad traditions in the abstract valuational model theory for nontransitive and nonreflexive logics. The first of these traditions makes heavy use of the natural Galois connection between sets of (many-valued) valuations and sets of arguments. The other, originating with work by Grzegorz Malinowski on nonreflexive logics, and best systematized in Blasio et al. (Bull Sect Log 46(3/4): 233–262, 2017), lets sets of arguments determine a more restricted set of valuations. After giving a systematic discussion of these two different traditions in the valuational model theory for substructural logics, we turn to looking at the ways in which we might try to compare two sets of valuations determining the same set of arguments.
Similar content being viewed by others
Notes
Throughout we will use \(\phi , \psi \) and other lowercase Greek letters as schematic letters for formulas, and \(\Gamma , \Delta , \Sigma \) and other uppercase Greek letters for sets of formulas.
Note that this third claim, unlike the second, is not a biconditional. As occasion arises we will refer to the classes of valuations implicitly delimited here as \({\mathfrak {V}}_3^r\) (for those which never assign the value ), \({\mathfrak {V}}_3^t\) (for those which never assign the value \({\mathop {\mathchoice{*\displaystyle }{*\textstyle }{*\scriptstyle }{*\scriptscriptstyle }}}\)). In the interests of completeness, the class of all valuations which never assign a formula either of \({\mathop {\mathchoice{*\displaystyle }{*\textstyle }{*\scriptstyle }{*\scriptscriptstyle }}}\) or will be referred to as \({\mathfrak {V}}_2\).
Since the valuation \(v_\top \) that assigns \(\top \) to every formula is a counterexample to no Set-Fmla argument, it can always be added to or removed from a set of valuations without affecting the resulting consequence relation. The same goes for any valuation that assigns only values from \(\{\top , {\mathop {\mathchoice{*\displaystyle }{*\textstyle }{*\scriptstyle }{*\scriptscriptstyle }}}\}\).
One particularly well-known case of this involves the Set-Fmla consequence relation of classical propositional logic. This set of arguments is determined by the usual set of all two-valued Boolean valuations (using the value \(\top \) for truth and \(\bot \) for falsity); but the same consequence relation is also determined by the set of valuations that adds \(v_\top \)—which is not Boolean—to this usual set. It is cases like this which are used in Carnap (1943) to motivate a shift to what is essentially the Set-Set framework. For further discussion and references on this and related issues, see Humberstone (2012, p.101ff).
Galois connections come in antitone and monotone versions; we are here using the (original) antitone version. These are essentially the same thing, however: a monotone Galois connection between S and T is exactly an antitone Galois connection between S and the order-dual of T. For helpful discussion on this difference, see Dunn (1991).
A closure operation on a partially-ordered set \(\langle S, \le \rangle \) is an operation C such that for every \(X, Y \in S\): 1) \(X \le C(X)\); 2) if \(X \le Y\), then \(C(X) \le C(Y)\); and 3) \(C(C(X)) \le C(X)\). (Equivalently, such that for every \(X, Y \in S\): \(X \le C(Y)\) iff \(C(X) \le C(Y)\).) An \(X \in S\) is closed wrt C iff \(X = C(X)\).
Note that Proposition 2 does not hold in cases where \(\Gamma \) is nonempty and explosive according to A in the sense that for all \(\phi \) we have \([\Gamma \triangleright \phi ]\in A\). In this case \(m_{A}^{\Gamma }\) is just a characteristic function for \(\Gamma \), assigning \(\top \) to all formulas in \(\Gamma \) and \({\mathop {\mathchoice{*\displaystyle }{*\textstyle }{*\scriptstyle }{*\scriptscriptstyle }}}\) to everything else. But in this case every valuation counterexamples every A-invalid argument whose premises are \(\Gamma \) (there are none!), so in particular the \(\sqsubseteq \)-least valuation \(v_{{\mathop {\mathchoice{*\displaystyle }{*\textstyle }{*\scriptstyle }{*\scriptscriptstyle }}}}\) which assigns \({\mathop {\mathchoice{*\displaystyle }{*\textstyle }{*\scriptstyle }{*\scriptscriptstyle }}}\) to every formula does so. But as in this case (since \(\Gamma \) is nonempty) \(m_{A}^{\Gamma }\) is not \(v_{{\mathop {\mathchoice{*\displaystyle }{*\textstyle }{*\scriptstyle }{*\scriptscriptstyle }}}}\) it is also not \(\sqsubseteq \)-below it.
This is not the definition given there, but amounts to the same and is more convenient for our purposes here.
That is to say \(Mod({A}) = \{ V | A = {\mathcal {A}}(V)\}\).
For reference, this means that A is the following set of arguments \(\{[p \triangleright p], [q \triangleright q], [r \triangleright r], [p, q \triangleright p], [p, q \triangleright q], [p, r \triangleright p], [p, r \triangleright r], [q, r \triangleright q], [q, r \triangleright r], [p, q, r \triangleright p], [p, q, r \triangleright q], [p, q, r \triangleright r], [p, q \triangleright r]\}\).
It’s perhaps worth noting that \(V_3\) consists only of Malinowski valuations for A: it is \(\{m_{A}^{\{p\}}, m_{A}^{\{q\}}, m_{A}^{\{p, q\}}\}.\) (This is how this example was found.) However, it is not \({\mathcal {M}}A\), since it omits \(m_{A}^{\emptyset } = v^\bot _\bot \).
An anonymous referee called our attention to the similarity between this definition and the definition of the function called \(\flat \) in Blasio et al. (Blasio et al. 2017, §5). These are different functions put to different uses: \(v_{a}\) is a function from the language to values, determined by a particular argument; while \(\flat \) is a function from other possible sets of values to our familiar four values, determined by a particular ‘\({\mathsf {B}}\)-matrix’ [see Blasio et al. (2017) for definition]. We suspect the referee is onto something, and there may be value in pursuing the analogy between formulas and arguments on the one hand, and other value spaces and \({\mathsf {B}}\)-matrices on the other.
Indeed, \({\mathcal {V}}_4A \not \subseteq {\mathfrak {V}}_3^t\) unless \({\mathcal {V}}_4A\) is empty, but so long as A is monotonic we have \({\mathcal {A}}{\mathcal {V}}_4A = A\).
The full strength of complete transitivity wasn’t needed here; the weaker property (Shoesmith and Smiley 1978) calls ‘cut for formulas’ is enough. But since we’re only going to apply Proposition 11 in cases where we’re also assuming compactness, and since in the presence of compactness cut for formulas suffices for complete transitivity, it wouldn’t be worth stating the slightly stronger formulation of Proposition 11 that its proof makes possible.
References
Bimbó, K., & Dunn, J. M. (2008). Generalized Galois logics: Relational semantics of nonclassical logical calculi. Number 188 in CSLI lecture notes. Stanford: CSLI Publications.
Birkhoff, G. (1967). Lattice theory (3rd ed.). New York: American Mathematical Society.
Blasio, C., Marcos, J., & Wansing, H. (2017). An inferentially many-valued two-dimensional notion of entailment. Bulletin of the Section of Logic, 46(3/4), 233–262.
Brink, C. (1993). Power structures. Algebra Uniersalis, 30, 177–216.
Carnap, R. (1943). Formalization of logic. Cambridge, Massachusetts: Harvard University Press.
Davey, B. A., & Priestley, H. A. (2002). Introduction to lattices and order. Cambridge: Cambridge University Press.
Dunn, J. M. (1991). Gaggle theory: An abstraction of Galois connections and residuation with applications to negations and various logical operators. In Logics in AI, Proceedings of European workshop JELIA 1990 (pp. 31–51). Berlin: LNCS.
Dunn, J. M., & Hardegree, G. M. (2001). Algebraic methods in philosophical logic. Oxford: Oxford University Press.
Erné, M., Koslowski, J., Melton, A., & Strecker, G. (1993). A primer on Galois connections. Annals of the New York Academy of Sciences, 704(1), 103–125.
Frankowski, S. (2004). Formalization of a plausible inference. Bulletin of the Section of Logic, 33, 41–52.
Frankowski, S. (2008). Plausible reasoning expressed by p-consequence. Bulletin of the Section of Logic, 37(3–4), 161–170.
French, R., & Ripley, D. (2019). Valuations: Bi, tri, and tetra. Studia Logica, 107(6), 1313–1346.
Hardegree, G. M. (2005). Completeness and super-valuations. Journal of Philosophical Logic, 34(1), 81–95.
Humberstone, L. (1988). Heterogeneous logic. Erkenntnis, 29, 395–435.
Humberstone, L. (2012). The connectives. Cambridge, Massachusetts: MIT Press.
Malinowski, G. (1990). Q-consequence operation. Reports on Mathematical Logic, 24, 49–59.
Ore, O. (1944). Galois connexions. Transactions of the American Mathematical Society, 55, 493–513.
Ripley, D. (2018). Blurring: An approach to conflation. Notre Dame Journal of Formal Logic, 59(2), 171–188.
Scott, D. (1974). Completeness and axiomatizability in many-valued logic. In L. Henkin (Ed.), Proceedings of the Tarski symposium (pp. 411–436). Providence: American Mathematical Society.
Shoesmith, D. J., & Smiley, T. J. (1978). Multiple-conclusion logic. Cambridge: Cambridge University Press.
Acknowledgements
We would like to thank the audience at the 2018 Australasian Association for Logic meeting, and two anonymous referees for their helpful discussion and comments on this material. David Ripley’s contribution was partially supported by the project “Logic and Substructurality”, grant number FFI2017-84805-P, Ministerio de Economía, Industria y Competitividad, Government of Spain.
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendix: Least sets of valuations in the Set-Set framework
Appendix: Least sets of valuations in the Set-Set framework
Recall that, given a set A of arguments, there is a set \(Mod({A}) = \{V : {\mathcal {A}}V = A\}\). Consider \(Mod({A})\) as ordered by \(\subseteq \). No matter whether we’re working Set-Set or Set-Fmla, and no matter whether we’re considering \({\mathfrak {V}}_4\), \({\mathfrak {V}}_3^t\), \({\mathfrak {V}}_3^r\), or \({\mathfrak {V}}_2\), \(Mod({A})\) has a greatest element: \({\mathcal {V}}A\). This follows from Theorem 1. What we want to answer in this appendix is the following question: under what conditions does it have a least element? In the body of the paper we dealt with the four Set-Fmla cases, and in this appendix we’ll look at the remaining four Set-Set cases.
As above, we begin by looking at what happens when we are considering bivaluations, since this is the best-known and best-explored area, and we can largely answer our question by appealing to or adapting existing results. Second, we turn to the general situation involving tetravaluations. We follow up by considering in turn both reflexive, as well as transitive trivaluations.
1.1 Bivaluations
It is known [for example (Dunn and Hardegree 2001, p. 202)] that for any monotonic reflexive completely transitive set of Set-Set arguments A there is exactly one set V of bivaluations such that \({\mathcal {A}}_{\textsc {ss}}V = A\). That suffices to answer our question for this case: there is always a least such V, since there is always exactly one such V.
1.2 Tetravaluations
Things get trickier when we go to the tetravaluational case. Here, there can be multiple distinct Vs with \({\mathcal {A}}_{\textsc {ss}}V = A\). For example, for any nonempty V, \({\mathcal {A}}_{\textsc {ss}}V = {\mathcal {A}}_{\textsc {ss}}(V \cup \{v_{\mathop {\mathchoice{*\displaystyle }{*\textstyle }{*\scriptstyle }{*\scriptscriptstyle }}}\})\), where \(v_{\mathop {\mathchoice{*\displaystyle }{*\textstyle }{*\scriptstyle }{*\scriptscriptstyle }}}\) is the valuation assigning \({\mathop {\mathchoice{*\displaystyle }{*\textstyle }{*\scriptstyle }{*\scriptscriptstyle }}}\) to every formula. (This is because \(v_{\mathop {\mathchoice{*\displaystyle }{*\textstyle }{*\scriptstyle }{*\scriptscriptstyle }}}\) is a counterexample only to the empty argument , and every valuation is a counterexample to this argument.)
We will show that when A is a compact monotonic set of Set-Set arguments, then there is a least set V of valuations with \({\mathcal {A}}_{\textsc {ss}}V = A\). To do this, we first show that there are certain valuations that must be in any V with \({\mathcal {A}}_{\textsc {ss}}V = A\); then we show that, so long as A is compact, these valuations alone are enough to determine A precisely. The needed valuations are the exact counterexamples to those arguments that are maximally out of A. We define each of these notions in what follows, proving needed results along the way. (We earlier defined the notion of exact counterexample for Set-Fmla arguments, but here we need it for Set-Set.)
Definition 9
Given a Set-Set argument , its exact counterexample \(v_{a}\) is the valuation such that \(v_{a}(\phi ) = \)
-
\(\top \) iff \(\phi \in \Gamma {\setminus } \Delta \),
-
\(\bot \) iff \(\phi \in \Delta {\setminus } \Gamma \),
-
iff \(\phi \in \Gamma \cap \Delta \), and
-
\({\mathop {\mathchoice{*\displaystyle }{*\textstyle }{*\scriptstyle }{*\scriptscriptstyle }}}\) iff \(\phi \not \in \Gamma \cup \Delta \).Footnote 13
Proposition 8
For any arguments a, b, we have iff \(b \sqsubseteq a\).Footnote 14
Proposition 9
For any argument a and any valuation v, we have iff \(v_{a} \sqsubseteq v\).
Proof
Unpacking definitions, in both cases. \(\square \)
It follows from each of these results that ; an argument’s exact counterexample is indeed a counterexample. Proposition 8 gives us one sense in which this counterexample is ‘exact’: it is a counterexample to all and only subsequents of a. Proposition 9 gives us a different sense: it is information-least among counterexamples to a.
Now, to arguments that are maximally out:
Definition 10
An argument is maximally out of a set A of arguments iff: it is not in A and any proper superargument of it is in A.
Lemma 6
If A is a set of Set-Set arguments, and the argument c is maximally out of A, then for any set V of valuations with \({\mathcal {A}}V = A\) it must be that \(v_{c} \in V\).
Proof
Take any such c, A, V, to show \(v_{c} \in V\). Since \(c \not \in A\) and \({\mathcal {A}}V = A\), there must be some \(v \in V\) with . By Proposition 9, \(v_{c} \sqsubseteq v\). Suppose towards a contradiction that \(v \ne v_{c}\). Then there must be some formula \(\phi \) receiving a different value in v than in \(v_{c}\). Since \(v_{c} \sqsubseteq v\), there are five possibilities:
-
(1)
\(v_{c}(\phi ) = {\mathop {\mathchoice{*\displaystyle }{*\textstyle }{*\scriptstyle }{*\scriptscriptstyle }}}\) and \(v(\phi ) = \top \)
-
(2)
\(v_{c}(\phi ) = {\mathop {\mathchoice{*\displaystyle }{*\textstyle }{*\scriptstyle }{*\scriptscriptstyle }}}\) and \(v(\phi ) = \bot \)
-
(3)
\(v_{c}(\phi ) = {\mathop {\mathchoice{*\displaystyle }{*\textstyle }{*\scriptstyle }{*\scriptscriptstyle }}}\) and
-
(4)
\(v_{c}(\phi ) = \top \) and
-
(5)
\(v_{c}(\phi ) = \bot \) and
But on any of these possibilities, v is a counterexample to some proper superargument of c: in the first, third, and fifth cases, the argument adds \(\phi \) to the conclusions of c, while in the first, second, and fourth, it adds \(\phi \) to the premises. These are indeed proper superarguments: Definition 9, plus what we know about \(v_{c}(\phi )\) in each case, suffices for this. But since c is maximally out of A, this superargument is in A, and so \({\mathcal {A}}V \ne A\), which is a contradiction. Thus, \(v = v_{c}\), and so \(v_{c} \in V\). \(\square \)
So every \(V \in Mod({A})\) must include all the exact counterexamples to those arguments maximally out of A. This matters when A is compact because there are enough arguments maximally out of it:
Proposition 10
If A is compact, then every argument \(a \not \in A\) is contained in some argument c that is maximally out of A.
Proof
Take a compact A and some \(a \not \in A\). Consider the set \( B = \{b \;|\; a \sqsubseteq b \; \& \; b \not \in A\}\) of all superarguments of a that are not in A. Every \(\sqsubseteq \)-chain in B has an upper bound in B: if the chain is finite, its maximum member will do; and if it is infinite, its sequent join will do. (Note that Compactness is needed at this step to ensure that these joins are \(\not \in A\)). But then by Zorn’s lemma, B has a maximal element c. Since \(c \in B\), we know \(a \sqsubseteq c\) and \(c \not \in A\).
To show that c is maximally out of A, it remains only to show that any proper superargument d of c is in A. But if there were some where \(d\not \in A\), then d would have to have been in B, and so c would not be maximal in B after all. \(\square \)
This is enough now for the theorem.
Theorem 13
If A is a compact monotonic set of Set-Set arguments, then there is a least \(V \in Mod({A})\).
Proof
By Lemma 6, any \(V \in Mod({A})\) must be such that \(V_{0} = \{ v_{c}~|~c \text { is maximally}~\text {out of } A \} \subseteq V\). So if \({\mathcal {A}}V_0 = A\), then \(V_0 \in Mod({A})\) and we’re done. Showing this has two phases: that \({\mathcal {A}}V_0 \subseteq A\) and that \(A \subseteq {\mathcal {A}}V_0\).
First, that \({\mathcal {A}}V_0 \subseteq A\). Take any \(a \not \in A\). By Proposition 10, there is some c maximally out of A with \(a \sqsubseteq c\). Since c is maximally out of A we have \(v_{c} \in V_0\), and by one direction of Proposition 8 we have . So \(a \not \in {\mathcal {A}}V_0\).
Second, that \(A \subseteq {\mathcal {A}}V_0\). Take any \(a \not \in {\mathcal {A}}V_0\); this has some counterexample \(v_{c} \in V_0\). By the other direction of Proposition 8, \(a \sqsubseteq c\). But since c is maximally out of A, it is at least out of A; and since A is monotonic, \(a \not \in A\). \(\square \)
1.3 Reflexive trivaluations
If A is a reflexive set of arguments, then the general story carries over immediately, owing to Lemma 5. So when A is reflexive and compact, the least set of tetravaluations determining A (which exists by Theorem 13) is a set of reflexive trivaluations by Lemma 5, and thus a least set of reflexive trivaluations determining A.
1.4 Transitive trivaluations
For transitive trivaluations, the reasoning is not so immediate, because there is no result analogous to Lemma 5 available for complete transitivity and transitive trivaluations.Footnote 15 But we can still make our way to the corresponding result; restricting our attention to completely transitive sets of arguments gives us extra tools to work with.
Proposition 11
If A is monotonic and completely transitive, and is maximally out of A, then \(\Gamma \cup \Delta = {\mathcal {L}}\).
Proof
Suppose A is monotonic and completely transitive, and is maximally out of A, but that \(\Gamma \cup \Delta \ne {\mathcal {L}}\). Then there must be some \(\phi \in {\mathcal {L}}\) with \(\phi \not \in \Gamma \cup \Delta \). So both and are proper superarguments of . Since is maximally out of A, both of these superarguments must be in A. But then since A is completely transitive, ; contradiction.Footnote 16\(\square \)
This is now enough to proceed.
Theorem 14
If A is a monotonic, completely transitive, and compact set of Set-Set arguments, then there is a least set V of transitive trivaluations determining it.
Proof
As in the proof of Theorem 13, the desired set is the set of exact counterexamples to those arguments maximally out of A. As we have seen in that proof, this set is the least set of tetravaluations determining A. So as long as it is itself a set of transitive trivaluations, we’re done. By Proposition 11, every maximally out of A is such that \(\Gamma \cup \Delta = {\mathcal {L}}\). Consulting Definition 9 reveals that the exact counterexample to any such argument cannot use the value \({\mathop {\mathchoice{*\displaystyle }{*\textstyle }{*\scriptstyle }{*\scriptscriptstyle }}}\), and so is a transitive trivaluation. \(\square \)
Rights and permissions
About this article
Cite this article
French, R., Ripley, D. Two traditions in abstract valuational model theory. Synthese 198 (Suppl 22), 5291–5313 (2021). https://doi.org/10.1007/s11229-019-02447-8
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s11229-019-02447-8