* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Download Suszko`s Thesis, Inferential Many-Valuedness, and the
Analytic–synthetic distinction wikipedia , lookup
Truth-bearer wikipedia , lookup
Structure (mathematical logic) wikipedia , lookup
Willard Van Orman Quine wikipedia , lookup
Fuzzy logic wikipedia , lookup
Foundations of mathematics wikipedia , lookup
Meaning (philosophy of language) wikipedia , lookup
Tractatus Logico-Philosophicus wikipedia , lookup
Quantum logic wikipedia , lookup
Combinatory logic wikipedia , lookup
First-order logic wikipedia , lookup
Curry–Howard correspondence wikipedia , lookup
Modal logic wikipedia , lookup
Mathematical logic wikipedia , lookup
Lorenzo Peña wikipedia , lookup
Jesús Mosterín wikipedia , lookup
History of logic wikipedia , lookup
Natural deduction wikipedia , lookup
Axiom of reducibility wikipedia , lookup
Propositional formula wikipedia , lookup
Intuitionistic logic wikipedia , lookup
History of the function concept wikipedia , lookup
Propositional calculus wikipedia , lookup
Law of thought wikipedia , lookup
Suszko's Thesis, Inferential Many-Valuedness, and the Notion of a Logical System Author(s): Heinrich Wansing and Yaroslav Shramko Reviewed work(s): Source: Studia Logica: An International Journal for Symbolic Logic, Vol. 88, No. 3 (Apr., 2008), pp. 405-429 Published by: Springer Stable URL: http://www.jstor.org/stable/40268956 . Accessed: 07/02/2013 10:10 Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp . JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected]. . Springer is collaborating with JSTOR to digitize, preserve and extend access to Studia Logica: An International Journal for Symbolic Logic. http://www.jstor.org This content downloaded on Thu, 7 Feb 2013 10:10:04 AM All use subject to JSTOR Terms and Conditions Suszko's Thesis, Inferential Many-valuedness, and the Notion of a Logical System H. Wansing Y. Shramko Abstract. Accordingto Suszko'sThesis, there are but two logical values, true and false. In this paper, R. Suszko's,G. Malinowski's,and M. Tsuji's analysesof logical two valuednessare criticallydiscussed. Anotheranalysisis presented,which favorsa notion of a logicalsystem as encompassingpossibly more than one consequencerelation. Keywords:Suszko'sThesis, inferentialmany-valuedness,many-valued logic, bivaluations, algebraicvalues, logicaltruth values. [A] fundamental problem concerning many-valuedness is to know what it really is. [13, p. 281] 1. Introduction Many-valued logic is one of the oldest branches of modern formal nonclassical logic. It has been developed in very influential works by Jan Lukasiewicz, Emil Post, Dmitri Bochvar, and Stephen Kleene in the 1920s and 1930s and is now a fully established and flourishing research program, see, for instance, [22] and [26]. In the 1970s, however, the prominent logician Roman Suszko called into question the theoretical foundations of manyvalued logic. Suszko distinguishes between the logical values truth and falsity (true and false) on the one hand, and algebraic values on the other hand. Whereas the algebraic values are just admissible referents of formulas, the logical values play another role. One of them, truth, is used to define valid semantic consequence: If every premise is true, then so is (at least one of) the conclusion(s). If being false is understood as not being true, then, by contraposition, also the other logical value can be used to explain valid semantic consequence: If the (every) conclusion is not true, then so is at least one of the premises. Suszko declared that "Lukasiewicz is the chief perpetrator of a magnificent conceptual deceit lasting out in mathematical logic to the present day" [39, p. 377] and he claimed that "there are but two logical values, true and false" [9, p. 169]. This challenging claim Presented by Jacek Malinowski; Received May 26, 2007 Studia Logica (2008) 88: 405-429 DOI:10.1007/S11225-008-9111-Z This content downloaded on Thu, 7 Feb 2013 10:10:04 AM All use subject to JSTOR Terms and Conditions ©Springer 2008 406 H. Wansingand Y. Shramko is now called Suszko's Thesis.1 It is has been given a formal contents by the so-called Suszko Reduction, the proof that every structural Tarskian consequence relation and hence also every structural Tarskian many-valued propositional logic is characterized by a bivalent semantics.2 Suszko's Reduction calls for an analysis not only because it seems to undermine many-valued logic, but also because it invites a re-consideration of the notion of a many-valued logic in particular and the notion of a logical system in general. According to Marcelo Tsuji [40, p. 308], "Suszkothought that the key to logical two-valuedness rested in the structurality of the abstract logics (for Wojcicki's theorem was the cornerstone of his reduction method)". Wojcicki's Theorem states that every structural Tarskian consequence relation possesses a characterizing class of matrices.3 As pointed out in [9] and [10], Suszko's Reduction can, however, be carried out for any Tarskian consequence relation, and also da Costa et al. [13] emphasize that the assumption of structurality is not needed for a reduction to two-valuedness. Jean-Yves Béziau [6] noticed that a reduction to a bivalent semantics can be obtained even if a much weaker and less common notion of a logical system is assumed. In the light of this observation, Tsuji [40] criticizes Grzegorz Malinowski's analysis of Suszko's Reduction [24], [25], [27]. Malinowski takes up Suszko's distinction between algebraic many-valuedness and logical two-valuedness and highlights the bi-partition of the algebraic values into designated ones and values that are not designated, a division that plays a crucial role in the Suszko Reduction. In fact, in the literature on many-valued logics sometimes an explicit distinction is drawn between a set V+ of designatedalgebraic values and a set V~ of antidesignated algebraic values, see, for example, [21], [22], [26], [31]. This distinction leaves room for values that are neither designated nor antidesignated and for values that are both designated and antidesignated, and therefore amounts to replacing logical two-valuedness (understood as a bipartition of the set of algebraic values) by logical four-valuedness in general, and by logical three-valuedness if it is postulated that D+nP" = 0or that 1SometimesSuszko'sThesis is stated in more dramaticterms. Tsuji [40, p. 299], for example, explains that "Suszko'sThesis maintainsthat many-valuedlogics do not exist at all". 2Bivalent interpretationsof many-valued logics have also been presentedby Urquhart [41],Scott [34],and da Costa and later Béziau, see [6], [13],and referencestherein. 3Wdjcickiproved his theorem for consequenceoperations, but the presentationsin terms of consequencerelationsand operationsare trivially interchangeable.Moreover,it should be pointed out that the entailmentrelationdefinedby a matrix in this case takes into accountonly valuationswhich are homomorphisms,cf. Section 2. This content downloaded on Thu, 7 Feb 2013 10:10:04 AM All use subject to JSTOR Terms and Conditions and the Notion . . . Suszko'sThesis,InferentialMany-valuedness, 407 T>+ U V~ is the set of all algebraic values available. The first condition is imposed by Malinowski [24], [25], [26], [27] and Gottwald [22], and the second condition may be used to define systems of paraconsistent logic.4 In order to provide a counterexample to Suszko's Thesis, Malinowski defined the notion of a single-conclusion quasi-consequence (^-consequence) relation. The semantic counterpart of ç-consequence is g-entailment. Singleconclusion g-entailment is defiiied by requiring that if every premise is not antidesignated, then the conclusion is designated. Malinowski [24] proved that for every structural g-consequence relation, there exists a characterizing class of g-matrices, matrices which in addition to a subset of designated values D+ comprise a disjoint subset of antidesignated values V~ .5 Not every g-consequence relation has a bivalent semantics. Moreover, a g-consequence relation need not be reflexive, i.e., it may be the case that a formula is not a q-consequence of a set of formulas to which it belongs, Tsuji [40] observed that reflexivity characterizes the existence of an adequate set of bivaluations for any relation \-C V{£) x C (where V(C) is the powerset of the propositional language C) i.e., for any abstract logical structure in the sense of Beziau's Universal Logic [4]. In this paper, we argue that Malinowski's analysis in a sense does capture the central aspect of explaining the feasibility of Suszko's reduction and the distinction between algebraic and logical values. If the idea of logical manyvaluedness as opposed to a multiplicity of algebraic values is taken seriously, then Tsuji's [40]analysis is question-begging, because it presupposes a notion of a logical system that admits at most logical two-valuedness if the single assumed consequence relation is reflexive. With another concept of a logical system, every such system is logically A>valued for some k G N (k > 2), and, moreover, the definition of a logical system is such that every entailment relation in a logically fc-valuedlogic is reflexive. By increasing the number of logical values (considered as separate subsets of the set of algebraic values) from two to three, Malinowski did a step towards logical many-valuedness, but in a way that gives up the idea of entailment as preservation of a logical value (from the premises to the conclusion, or vice versa) and thereby at the price of violating reflexivity. In the present paper, it is suggested to take a further step and to admit an increase not only of the number of logical values, but also of the number of entailment relations. To keep this article readable and self-contained, we review Suszko's reduction method in Section 2 and there also collect some well-known facts 4In [43]it is explainedwhy it is not at all unreasonableto admit T>+DP 5Also Malinowskiconsidersonly valuationswhich are homomorphisms. This content downloaded on Thu, 7 Feb 2013 10:10:04 AM All use subject to JSTOR Terms and Conditions ^ 0. H. Wansingand Y. Shramko 408 about consequence relations. In Sections 3 and 4 we present and briefly discuss Malinowski's and Tsuji's analyses of Suszko's Thesis, and in Section 5, we present our own analysis. Finally, in Section 6, we briefly mention yet another analysis leading to higher-arity consequence relations, and in Section 7 we draw some general conclusions. 2. Suszko's reduction Suszko suggested his reduction to a bivalent semantics with respect to the standard notion of a structural Tarskian logic (structural Tarskian consequence relation) and the standard notion of an n-valued matrix (semantically defined n-valued logic). Let £ be a language in a denumerable set of sentence letters and a finite non-empty set of finitary connectives C = {ci, . . . ,cm}. A Tarskian consequence relation on £6 is a relation h Ç V(C) x C such that for every A, B e C and every A, T C C: AU{A}\-A If A h A then AuT\- (I) (Reflexivity) A (Monotonicity) If A h A and P U {A} h B, then AUThB (Cut) (II) (III) A Tarskian consequence relation h on the language C is called structural iff for every A e C, every A Ç £, and every uniform substitution function a on C (every endomorphism of the absolutely free algebra (£, ci, . . . , cm)) we have A h A iff <j(A) h a(A) (Structurally), (IV) where <r(A) = {&{B) | B e A}. It is standard terminology to call a pair (£, h) a Tarskian (structural Tarskian) logic iff h is Tarskian (structural and Tarskian). An n-valued matrix based on £ is a structure 2Jt = (V, 2?,{/c : c G C}), where V is a non-empty set of cardinality n > 2, V is a non-empty proper subset of V, and every fc is a function on V with the same arity as c. The elements of V are usually called truth values (or truth degrees), and the elements of V are regarded as the designated truth values. In Suszko's terminology, V is the set of algebraic values, whereas V and its complement represent the two logical truth values. A valuation function v in 9Jtis 6 Often, consequencerelationsare definedfor arbitrarysets. In this paper we restrict our interest, however,to consequencerelationsdefinedon propositionallanguages. This content downloaded on Thu, 7 Feb 2013 10:10:04 AM All use subject to JSTOR Terms and Conditions and the Notion . . . Suszko'sThesis,InferentialMany-valuedness, 409 a function from C into V. Also a structure 971= (V,£>, {fc : c G C}) may be viewed as a logic, because the set of designated truth values determines a Tarskian (semantical) consequence relation \=$ji Ç V(C) x C by defining A f=3HA iff for every valuation v in 9Jt, t>(A) Ç V implies v(A) G X>, where v(A) = {v(-B) | B G A}. Usually, only truth-functional valuations are considered. A valuation ?; in 971is truth-functional (structural) iff it is a homomorphism from (£, ci, . . . , Cm) into (V, /C1, . . . , /Cm). A pair .M = ((V,2?,{/c : c G C»,v), where 9Jt = (V,P,{/C : c G C}) is an n-valued matrix and v a valuation in 9JÎ,may be called an n-valued model based on 3D?.A model (3D?, v) is structural iff v is a truth-functional valuation in 3D?. an n-valued model M. determines a Tarskian (semantical) conseObviously, quence relation \=m Ç V(C) x C by defining A \=m A iff v(A) Ç V implies v(A) G V. A Tarskian logic (£, h) is said to be characterized by an n-valued matrix 3D?iff h = |=sai, (£, h) is characterized by an n-valued model M. iff h = f=.M, (£, h) is characterized by a class £ of n-valued matrices iff h = PKNshI 3DIG ^}, and, finally, (£, h) is characterized by a class & of n-valued models iff h = nih^tl M E Â}. Ryszard Wojcicki [44] showed that every structural Tarskian logic is characterized by its so-called Lindenbaum bundle: {((£, {,4 G C | A h j4},C), v) | A Ç £, v is an endomorphism of £}. Theorem 2.1. (Wojcicki) Every structural Tarskian logic is characterized by a class of structural n-valued models, for some n < NoTheorem 2.2. (Suszko [39], Malinowski [26])7 Every structural Tarskian logic is characterized by a class of two-valued models. Proof. Let A = (£, h) be a structural Tarskian logic. Then, by Wojcicki's Theorem, A is characterized by a class <£aof structural n-valued models. For ((V, £>,{fc'-ce C}), v) G Ca, the function tv from C into {0, 1} is defined as follows: 1 if v(A)€V tv[A) - / 0 if v(A) <£V \ The class {(({0, 1}, {1}, {/c : c 6 C}),tv) \ ({V,V,{fc : c € C})tv)'e £A} of 2-valuedmodelscharacterizesA. 7As Caleiroet al. [10]point out, "thereseems to be no paper where Suszkoexplicitly formulates(SR)[the SuszkoReduction]in full generality!". This content downloaded on Thu, 7 Feb 2013 10:10:04 AM All use subject to JSTOR Terms and Conditions H. Wansingand Y. Shramko 410 Malinowski [27, p. 79] highlights (though in other words) that the Suszko Reduction does not establish the existence of a characterizing class of structural two-valued models. Suszko was fully aware of this fact. In [39, p. 378] he explains that the logical valuations are morphisms (of formulas to the zero-one model) in some exceptional cases, only. Thus, the logical valuations and the algebraic valuations are functions of quite different conceptual nature. The former relate to the truth and falsity and, the latter represent the reference assignments. This conception may explain why the sets V are regarded by Suszko as sets of "algebraic values". They are algebraic, because they are assigned to formulas by valuations which are homomorphisms. If the elements of V need not be assigned by homomorphisms, they may be called referential values. The semantics which emerges as a result of Suszko's Reduction is referentially bivalent and hence logically bivalent in the sense that there exist only two distinct non-empty proper subsets of the set of algebraic values, see also Section 5.2. The fact that the valuations tv need not be homomorphisms deprives the Suszko Reduction of much of its jeopardizing effect on manyvalued logic, if structurality is viewed as a defining property of a logic. It has been emphasized by da Costa et al. [13] and Caleiro et al. [9], [10], however, that the condition of structurality can be given up. Theorem 2.3. Every Tarskian logic is characterized by a class ofn-valued models, for some n < NoTheorem 2.4. Every Tarskian logic is characterizedby a class of two-valued models. The proofs of Theorem 2.2 and 2.4 are non-constructive. Caleiro et al. [9], [11] considerably improve upon Suszko's Reduction by defining an effective method for associating to each structural finitely-valued model an equivalent two-valued model (under the assumption of effectively separable truth values). 3. Malinowski's analysis of Suszko's Thesis Malinowski [27, p. 80 f.] succinctly analyses Suszko's reduction as follows: [L]ogicaltwo-valuedness ... is obviously related to the division of the universe of interpretation into two subsets of elements: distinguished This content downloaded on Thu, 7 Feb 2013 10:10:04 AM All use subject to JSTOR Terms and Conditions and the Notion ... Suszko'sThesis,InferentialMany-valuedness, 411 and others. It also turned out, under the assumption of structurality, that Tarski's concept of consequence may be considered as a "bivalent" inference operation. He then describes his response to the question "whetherlogical many-valuedness is possible at all" as giving an affirmative answer to this question by invoking a formal framework for reasoning admitting rules of inference which lead from nonrejected assumptions to accepted conclusions. This approach may be viewed as taking 'true' and 'false' to be predicates that give rise to contrary instead of contradictory pairs of sentence. As such, the pair 'true' versus 'false' is reflected by the contrary pairs 'designated' versus 'antidesignated' and 'accepted' versus 'rejected'. Admitting algebraic values that are neither designated nor antidesignated amounts to admitting in addition to the logical values true and false the third logical value neither true nor false. In other words, being false is distinguished from not being true. Whereas the algebraic values that are not designated are already given with the set of designated values V as its set-theoretical complement, the treatment of true and false as values that are independent of each other leads to distinguishing a non-empty set £>+ of designated algebraic values from a non-empty set T>~of antidesignated algebraic values. Let £ be as above. An n-valued g-matrix (quasi-matrix) based on C is defined by Malinowski as a structure M = (V, D+,P',{/c:cG C}>, where V is a non-empty set of cardinality n > 2, 2?+ and V~ are distinct non-empty proper subsets of V such that V+ DV~ = 0, and every fc is a function on V with the same arity as c. If it is not required that V+C\V~ = 0, we shall talk of generalized g-matrices. A valuation function v in 9JI is a function from C into the set of truth degrees V, and Malinowski considers only valuations which are homomorphisms (from (£, ci, . . . , Cm)into (V, fci , . . . , fcm))To obtain a kind of entailment relation that does not admit of a reduction to a bivalent semantics, Malinowski defines such a relation, called g-entailment, as depending on both sets T>+ and V~ . A q-matrix 9H = (V, X>+,2?~, {fc-c€ C}) determines a g-entailment relation (=3jiC V(C) x C by defining A |=^ A iff for every (in Malinowski's case homomorphic) valuation v in SDt,v(A) n V~ = 0 implies v(A) 6 T>+. A pair M = : c e C}) is «V,2>+,I>-,{/c : c e C»,v>, where m = (V,P+,P",{/C an n-valued g-matrix and v a valuation in 9JÎ, may be called an n-valued g-model based on 371. The relation \=mQ ^(£) x £ determined by such a model is defined by the following equivalence: A \=m A iff v(A) n V~ = 0 This content downloaded on Thu, 7 Feb 2013 10:10:04 AM All use subject to JSTOR Terms and Conditions H. Wansingand Y. Shramko 412 implies v(A) G V+ . A model (3DÎ,v) is structural iff v is a truth-functional valuation in M. If M = (V,P+,P~, {fc : c G C}) is a ç-matrix and T>+is not the complement of 2?", there is no class of functions from C into {1,0} such that A \=m A iff for every function v from that class, v(A) Ç {1} implies v(A) = I.8 Let M = {(V,V+,T>-,{fc : c G C}),v) be an n-valued g-model. Malinowski pointed out that an equivalent three-valued g-model M' = (({0, i, 1}, {1}, {0}, {fc-ce C}),tv) can be defined as follows: if v(A) e V+ \ iff v(A)eV\(v+uv-) 0 if v(A) € V~ A g-entailment relation \=wi is a special case of what Malinowski calls a g-consequence relation. A g-consequence relation on £ is a relation Ih Ç V(C) x C such that for every A G C and every A, F Ç C: If A IhA then A U T Ih A (Monotonicity) (V) A U {B | A Ih B) IhA iff A IhA (Quasi-closure) (VI) A g-consequence relation on C is called structural iff for every A G £, every AÇ£, and every uniform substitution function a on C we have A IhA iff a(A) Iha(A) (Structurality). (VII) A pair (£, Ih) is said to be a g-logic, it is structural iff Ih is structural. A g-logic (£, Ih) is said to be characterized by an n-valued q matrix 9Jt iff Ih = Hart?(£>"") is characterized by an n-valued g-model A4 iff Ih = \=Mi (£, Ih) is characterized by a class  of n-valued g-matrices (g-models) iff Ih = fKhml a« e £} (ih= f|{h^l M e &}). If A Ç £, let V\ = {A G C \ A h A} and D^ = £\(AU{A G C \ A h A}). Malinowski [24] showed that every structural g-logic is characterized by the following Lindenbaum bundle: {((£,£>£, P^,C),t>) | A Ç £, v is an endomorphism of £}. 8In [5, p. 120] it is stated that "Malinowskiconstructs (using an extended concept of a matrix) a consequencerelationwhichhas no two-valuedlogicalsemanticsbecauseit fails to obey the "identity"axiom of Tarski. Howeverit has been shown (cf. [Krause/Béziau 1997]) that we can adapt in some way two-valued logical semantics even in the case of such kind of consequencerelation". Note, that this 'adaptation'is sketchedin terms of two functionsmodi, mod,2each assigningto every formulaand every set of formulasa class of models. This content downloaded on Thu, 7 Feb 2013 10:10:04 AM All use subject to JSTOR Terms and Conditions Suszko'sThesis,InferentialMany-valuedness, and the Notion . . . 413 Theorem 3.1. (Malinowski) Every structural q-logic is characterized by a class of structural n-valued q-models, for some n < NoBy the above definition of three-valued g-models M! and by the Suszko Reduction for the case that V \ {V+ U V~) = 0, it follows that g-logics are logically two-valued or three-valued. Theorem 3.2. (Malinowski) Every structural q-logic is characterized by a class of two-valuedq-models or by a class of three-valuedq-models. Another non-standard entailment relation, based on a certain generalized ^-matrix, has been presented in [12] and is there said not to be "overly outlandish or inconceivable", although it fails to be a Tarskian consequence relation. The "tonk-consequence"relation of [12] for a given language £ and set of connectives C is the relation (= Ç V(C) x £ defined for the generalized qmatrix 971= <{0, {T}, {F}, {T, F}}, {{T}, {F, T}}, {{F}, {T, F}}, {fc:cE C}) as follows: A \= A iff either for every truth-functional valuation v in 9Jt, v(A) Ç £>+ implies v(A) G 2?+, or for every truth-functional valuation v in 3DÎ,v(A) fl V~ = 0 implies v(A) £ V~ . Since tonk-entailment is not transitive, sound truth tables for Prior's connective tonk are available such that this addition of tonk does not have a trivializing effect (but see also [42]). 4. Tsuji's analysis of Suszko's Thesis Tsuji [40, p. 305] emphasizes that Malinowski's analysis "is not wrong but that it misses the main point of logical two-valuedness." Taking up Beziau's idea of a Universal Logic [4], Tsuji assumes a very general notion of a logical system. An abstract logical structure is any pair (5, h), where S is an arbitrary set and hÇ V(S) x S. We shall again restrict our attention to relations hÇ V{£) x £, where £ is a propositional language. Theorem 4.1. (Tsuji [40]) An abstract logical structure (£,h) is characterized by a class of two-valued models iffY- satisfies (Reflexivity). Proof. Let h be reflexive and consider the class of two-valued models € = {(({0, 1}, {1}, {fc:ce C}), vA) | A Ç £, vA(A) = 1 iff A h A}. Obviously, A h A implies A \=c A. Suppose A \f A. Then vA(A) = 0. By (Reflexivity), v&(A) Ç {1}. Therefore A ^e A. Conversely, if h is not reflexive, there exist A Ç £ and A G A with A \f A. But A \=m A for any two-valued model M, since A G A. This content downloaded on Thu, 7 Feb 2013 10:10:04 AM All use subject to JSTOR Terms and Conditions 414 H. Wansingand Y. Shramko Tsuji [40, p. 308] takes this result to reveal that "the problem of logical twovaluedness has more to do with the "geometrical"properties of the" relation h of an abstract logical structure (5, h) , "than with the algebraic properties of it's set 5 or of it's Lindenbaum bundle". In the next section we shall raise doubts about this analysis. 5. 5.1. Logical n-valuedness as inferential many-valuedness What is a logical value? Suszko does not define the notion of a logical value except for stating that true and false are the only logical values, but he distinguishes the logical values form algebraic values and explains that the possibly many algebraic values are denotations of (propositional) formulas. Moreover, he claims that "any multiplication of logical values is a mad idea" [39, p. 378]. In any case the question arises, by virtue of which properties true and false are to be considered as logical values. The logical value true is given with the notion of entailment and the specification of a set of distinguished algebraic values. A formula A is entailed by a set of premises A if and only if it is the case that if every premise is true (alias designated), then so is the conclusion. Thus, truth is what is preserved in a valid inference from the premises to the conclusion. Let us refer to this notion of entailment as t-entailment. A formula A is logically true iff A is t-entailed by the empty set (iff for every assignment v of algebraic values to the formulas of the language under consideration, v(A) is designated), and A is logically false iff A t-entails the empty set (iff for every assignment v, v(A) is not designated). If we consider a bi-partition of the set of algebraic values, truth is identified with a non-empty subset T>of designated algebraic values and falsity with the complement of V. Now, it is characteristic for falsity to be preserved in the reverse direction, i.e., from the conclusion to at least one of the premises. One might wish to consider a notion of /-entailment understood as the backward preservation of values associated with falsity. Obviously, membership in the complement of V is preserved from the conclusion to the premises, but this gives the very same entailment relation. Since V is uniquely determined by its complement, and vice versa, logical two-valuedness is, in fact, reduced to logical mono-valuedness if there is just one entailment relation defined as truth preservation form the premises to the conclusion. However, if we treat 'falsity' not as a mere abbreviation for 'non-truth' and if we correspondingly distinguish not only a set P+ of designated algebraic values but also a set P~ of antidesignated algebraic values which This content downloaded on Thu, 7 Feb 2013 10:10:04 AM All use subject to JSTOR Terms and Conditions and the Notion ... Suszko'sThesis,InferentialMany-valuedness, 415 are not obligatorily complements of each other, then /-entailment may well be different from t-entailment.9 Indeed, consider, e.g., the g-matrix L3* := : c E {~,A,V,D}}), where the functions fc are ({T,0,F},{T},{F},{/c the usual operations of Lukasiewicz's three-valued logic (see, e.g., [28], [29]). Let us call L3* Lukasiewicz's ç-matrix. It is easy to see (cf. [43]) that in L3* £-entailment and /-entailment do not coincide, for A A (A D B) |=+ i?, but A A (A D B) y=~ B. In this situation it would hardly be justifiable to prefer one entailment relation over the other and to deal, e.g., only with t-entailment and to completely disregard /-entailment. And if both entailment relations are pari passu (as the sets 2?+ and T>~are), why should one not conceive of logical systems with two (or perhaps even more) entailment relations?10 If logic is thought of as the theory of valid inferences, then a logical value may be seen as a value that is used to define in a canonical way an entailment relation on a set of formulas. By a canonical definition of entailment we mean a definition of entailment as a relation that (in the single conclusion case) preserves membership in a certain set of algebraic values either from the premises to the conclusion of inferences, or from the conclusion to the premises. Such a relation will be Tarskian (since preservation of a logical value from the conclusion to the premises means that if the conclusion possesses the value, then so does at least one of the premises, whereas preservation from the premises to the conclusion means that if every premise possesses the value, then so does the conclusion). Two logical values are independent of each other iff the canonically defined entailment relations associated with these values are distinct. In addition to the term 'logical many-valuedness', Malinowski also uses the term 'inferential many-valuedness'. One might understand Malinowski's analysis as assuming (i) that logical values play a role in defining a single entailment (inference) relation and (ii) that this entailment relation need 9 Denoting t-entailment as |=+ and /-entailment as |= , we obtain the following definitions:A |=+ D iff for every valuation function v: v(A) Ç V+ implies v(B) G P+; A [=~ B iff for every valuation function v: v(B) G *D~~implies v(A) G D~, for some AeA. 10The idea of a logical system comprisingtwo consequencerelations may be detected already in [14, Chapter 6], where H.B. Curry distinguishes between deducibility and refutabiliy. Curry assumes a theory T generated by axioms and a theory 5 generated by counteraxioms. Moreover,it is assumedthat a formulaA is refutable,if a refutable formulais deduciblefrom A. Currythen introducesa notion of negationas refutationby requiring(i) that the negationof every counteraxiomis provableand (ii) that the negation -1A of a formulaA is provable,if the negationof a formulaB is provableand B is deducible from A. This content downloaded on Thu, 7 Feb 2013 10:10:04 AM All use subject to JSTOR Terms and Conditions 416 H. Wansingand Y. Shramko not be defined canonically. Truth and falsity are treated by Malinowski as logical values insofar as both sets of algebraic values are used to define g-entailment, which, however, is not defined canonically. If the idea of entailment as preservation of a logical value is given up, then entailment will not, in general, be a Tarskian relation. There are thus several issues involved: (i) the number of sets of designated values (as representing logical values), (ii) the relation of the algebraic values to the logical values, and (iii) the definition of entailment in terms of the logical values. As to the latter, we shall here join Jennings and Schotch [23, p. 89] in declaring that "we want inferability to preserve". On our conception of Malinowski's term 'inferential many-valuedness', therefore it is distinctive of a logical value that it is used to canonically define an entailment relation.11 A logic may then be said to be logically (or inferentially) k-valued if it is a language together with k canonically defined and pairwise distinct entailment relations on (the set of formulas of) this language. Each of these k entailment relations is Tarskian and hence, in particular, it is reflexive. Not only the logic of a g-entailment relation, but also classical propositional logic is obviously not logically two-valued in this sense, because these logics are defined by a language together with a single entailment relation. If there in fact is more than one logical value, each determining a separate entailment relation, and if therefore it is reasonable to think of a logic as possibly containing more than just one entailment relation (or, syntactically speaking, more than just one consequence relation), our contribution to the discussion of Suszko's Thesis is this: 1. Suszko's notion of a many-valued logic as comprising a set of designated values V but not also a set of antidesignated values that may be distinct from the complement of V amounts to assuming logical mono-valuedness instead of logical two-valuedness, when it is assumed that a logic comes with a single entailment relation. 11Thisis actually not Malinowski'sunderstandingof inferentialmany-valuedness.In [29] he explainsthat the chief featureof a q-consequenceoperationis that the repetitionrule: rep=({A},A\AeC) in general is not a rule of the operation. Moreover,^-consequenceis the central notion of a purely inferential approachin the theory of propositionallogics in the sense that "[t]heprincipalmotivationbehindthe quasi-consequence. . . stems fromthe mathematical practicewhichtreats some auxiliaryassumptionsas merehypothesesratherthan axioms". The fact that the repetition rule is not unrestrictedlyvalid allows Malinowski[29, Section 2] to define two congruencerelationson £, inferentialextensionalityand inferential intensionality,which in generalare independentof each other. This content downloaded on Thu, 7 Feb 2013 10:10:04 AM All use subject to JSTOR Terms and Conditions Suszko'sThesis,InferentialMany-valuedness, and the Notion ... 417 2. Malinowski's conception of logical many-valuedness is that logical values contribute to the definition of a single entailment relation, but there are reasons to let every logical value give rise to an entailment relation. 3. Tsuji's characterization of logical two-valuedness in terms of reflexivity, assuming Beziau's notion of an abstract logical structure containing a single consequence relation, is question begging, because with another concept of a logical system, in a logically &-valued logic every entailment relation is reflexive. Assuming three algebraic values and understanding logical truth as receiving one of these values under any valuation, da Costa et al. [13, p. 292 f.] raise the question "Can logical truth also be multivalent?" and reply: It seems that a priori there is no good philosophical argument to reject this possibility, and this is another reason why we can reject Suszko's thesis. Can one provide evidence to the effect that logical truth is multivalent? More generally, is there more than one logical value, each of which may be taken to determine its own entailment relation? It seems that Suszko did not raise fundamental doubts about algebraic many-valuedness, i.e., about assuming more than two algebraic values. One natural question then comes with the distinction between algebraic and logical values: Is every non-empty subset of a given set of algebraic values a logical value? We need not try to decide this question here, because for our present purposes it is enough to insist on falsity as a logical value not uniquely determined by truth. Instead of assuming with Frege that logic is the science of the most general laws of being true, for our purposes it is enough to assume that logic is the science of the most general laws of being true and the most general laws of being false.12 5.2. Another kind of counterexample Before developing further the concept of a logical system with more than one entailment relation, we wish to present shortly another definition of an entailment relation, a relation which is, so to say, "essentially three-valued". The key point of Malinowski's reduction (the proof that every g-consequence 12 Interpretationsof distinguishedsets of algebraicvalues need not appeal to truth or falsity. In a seriesof papers,Jennings,Schotch,and Brownhavearguedthat paraconsistent logic can be developedas a logic that preservesa degreeof incoherencefromthe premises to the conclusionof a valid inference,see [8], [23]and the referencesgiven there. This content downloaded on Thu, 7 Feb 2013 10:10:04 AM All use subject to JSTOR Terms and Conditions 418 H. Wansingand Y. Shramko relationhas a two-valuedor a three-valuedsemantics)is to presupposea tripartitionof the set of algebraicvaluesratherthan a bi-partition.As already observed,Malinowski'sç-entailmentis not reflexive.It is neverthelesspossible to definea reflexiveentailmentrelationwhich is in a sense necessarily three-valuedby employingMalinowski'sidea that entailmentshoulddepend on both designatedand antidesignatedvalueswhichdo not exhaust the set of all valuesavailable. Let us returnto the observationabovethat in Lukasiewicz's ç-matrixL3* we may In context coincident. this not t-entailmentand /-entailment are also considerthe intersectionof these two relations,whichcan be called£/entailment.Dunn [18,p. 11] introducesthese three entailmentrelationsand shows that they are all distinct. The latter point is crucialfor defininga three-valuedentailmentrelationwhich cannot be providedwith a logically two-valuedsemanticswithin the given set of algebraicvalues. Namely,for the logic based on Lukasiewicz'sg-matrixL3*such a relationis represented by t/-entailment:13 A \=^~ B iff Vv in L3*: (1) v(A) CD+^ v(B) € £>+; (2) v(B)eV-^3AeA(v(A)eV-).u consequences" Incidentally,f=£\ restrictedto the so called "first-degree (statements of the form A \=^ 5, where neither A nor B contains implications)representsthe first-degreeentailmentfragmentof Lukasiewicz's 3-valuedlogic (cf. [18,p. 15]). Now, it is not difficultto showthat everytfentailmentrelationand hence |=£'^ is Tarskian.This meansthat, according to Theorem2.4, |=jj"'*has an adequatelogically two-valuedsemantics. It turns out, however,that it is impossibleto define 1=^'^by a two-valued valuationalfunctiondefinedon the carrierof L3*. Indeed, assume such a definitionwouldbe possible. Then we need to pick fromthe set {T, 0, F} a propersubset V of designatedalgebraicvalues. Moreover,V shouldcontain T (because \=t^ *s truth-preserving). There are only three such subsets: {T}, {T,0}, and {T,F}. But in the first case [=+'* would coincidewith 13Thatis, we disregardfora momentthe t-entailmentand /-entailment takenseparately, and concentratesolely on their intersection.Note, that it is most importanthere to have a value which is neither designated nor anti-designated,otherwise t/-entailment would trivially collapseinto t-entailment(and /-entailment). 14A generalizationof this definitionto obtain a tf -entailmentrelationfor any ^-matrix 9JI based on a three-element set of algebraic values using Malinowski'sthree-valued valuationtv (as definedon p. 412) is straightforward. This content downloaded on Thu, 7 Feb 2013 10:10:04 AM All use subject to JSTOR Terms and Conditions and the Notion . . . Suszko'sThesis,InferentialMany-valuedness, 419 f=£ * (which is impossible), in the second case \=^ would coincide with f=£ * (which again, is impossible), and in the third case the intended definition for (=£'* would be simply inadequate (making it falsity preserving from the premises to the conclusion). 5.3. An example of a natural bi-consequence logic A natural example of a logic with two entailment (or consequence) relations (which can be called a bi-consequence logic) can be presented in the context of generalizing Nuel Belnap's idea of passing information about a given proposition to a computer (see [36], [37], [38], and for the seminal papers on Belnap's and Dunn's four-valued logic [1], [2], [17]). It turned out that in this context the resulting set of truth values is such that its elements can be partially ordered not only according to their truth content but also to their falsity content. These orderings are defined in such a way that in the definition of the truth (falsity) order no reference is made to the classical truth value F (falsity) (T (truth)) and the two orderings are not inverses of each other. In the simplest case, instead of the set of classical truth values 2 = {T,F} the set 16 = V(V(2)) is considered, giving the following generalized truth values: 1. N = 0 2. N = {0} 3. F ={{F}} 4. T ={{T}} 5. B = {{F, T}} 6. NF = {0, {F}} 7. NT = {0, {T}} 8. NB = {0,{F,T}} 9. FT = {{F},{T}} 10. FB = {{F} , {F, T}} 11. TB = {{T},{F,T}} 12. NFT = {0,{F},{T}} 13. NFB = {0, {F} , {F, T}} 14. NTB = {0, {T} , {F, T}} 15. FTB = {{F} , {T} , {F, T}} 16. A = {0,{T},{F},{F,T}}. On 16 a truth order <t and a falsity order <y can be defined in a natural way. Definition 5.1. For every x, y in 16 let xl = {y e x | T G y}, x~l = {y ex\T $y), xf = {y G x \ F € y], and x"/ = {y G x | F g y}. Then x <t y iff xl Ç y* and y~~*Ç x~l and x </ y iff x$ Ç y-f and y~^ Ç x~A Moreover, set-inclusion may be regarded as an information order on 16: x <i y iff x Ç y. We thereby obtain an algebraic structure that combines the three (complete) lattices (16, <*), (16, <t), and (16, </) into the trilattice SIXTEEN^ = (16, <»,<*,</). SIXTEEN^ is presented by a triple Hasse This content downloaded on Thu, 7 Feb 2013 10:10:04 AM All use subject to JSTOR Terms and Conditions 420 H. Wansingand Y.Shramko t - f~l) Figure1. TrilatticeSIXTEEN^(projection diagramin Figure 1. Meets and joints exist in SIXTEEN^ for all three partial orders. We will use n and U with the appropriatesubscriptsfor these operationsunderthe correspondingorderingrelations.Sincefromthe operationsone can recoverthe relations,the trilatticeSIXTEEN^ may also be representedas the structure (16,ni,Ut,nt,Ut,n/,U/). In what follows we will be especially interestedin the "logical"operationsn^LJ^ri/ and Uf. Since the relations<t and </ are treated on a par, the operationsDt and Ut are not privilegedas interpretationsof conjunctionand disjunction connectives.The operationU/ may as well be regardedas a conjunctionand FI/as a disjunction.In otherwords,the logicalvocabularymay be naturally split into a positivetruth vocabularyand a negativefalsityvocabulary.Also certainunaryoperationswith naturalnegation-likepropertiescan be defined in SIXTEEN^,see Table 1. Eachoperation-o (o e {£,/, i}) satisfies- o - o x = x, and x <o y implies - o y <o -ox. The algebraicoperationsfit, Ut, l~l^,U/, - $, and - / makeavailableinterpretationsfor logical connectivesin the propositionallanguageCtj defined in Backus-Naurformas follows: Ctf : A ::= p | ~t | ~/ I A*| V*| A/ | V/ This content downloaded on Thu, 7 Feb 2013 10:10:04 AM All use subject to JSTOR Terms and Conditions and the Notion . . . Suszko'sThesis,InferentialMany-valuedness, | a || -ta | -/a | N N "~~N~ N T F N F B T N B ~B~~ ~F T NF TB NF NT NT FB 421 -jd ||" a || -tg \ -fg \ -{g \ A NB FT~ FT 1PF NFT FB FB FB~ NT NFB TB NF TB TB NTB NFT NTB 1>JFB AT" FTB NFB FTB l^FT F~~ NF NTB NFT FTB T NT FTB NFB NTB B " Irr I NB I NB I NB I A | A | A | N Table1. Inversions in SIXTEEN^ Definition 5.2. Let v be a map from the set of propositional variables into 16. The function v is recursively extended to a function from Ctf into 16 as follows: 1. v (A At B) = v (A) Ht v {B) ; 4. v(AAfB) = v (A) U/ v (B) ; 2. v(A\ZtB) = v (A) Ut v (B) ; 5. v (A V/ B) = v (A) nf v {B) ; 6. v ( ~fA) - -fv (A) . 3. v ( ~t A) = -tv (A) ; The logic of SIXTEEN s may be semantically presented as a bi-consequence system, viz. the structure (£*/> f=t»N/)» where the two entailment relations |=t and [=/ are defined with respect to the truth order <t and the falsity order </, respectively. Definition 5.3. The set-to-set entailment relations \=tC V(£tf) x V(Ctj) and [=/C V(Ctf) x V(Ctf) are defined by the following equivalences: Ahr Ah/r iff Vt;ritMA) I^ e A} <t UMA) Iil e r}; iff Vi;LJ/WA) | A e r} </ ri/W^) IA e A}. What is important in the present context is that (A/? H*' H/) induces a bi-consequence system {Ctf, f=+, |=~) in the language Ctf, where the entailment relations f=+ and f=~ are defined with respect to a certain generalized (/-matrix. Definition 5.4. The structure B\q is the generalized g-matrix (16, {x G 16 | xl is non-empty}, {x G 16 | x? is non-empty}, {-ti nt) Ut, -/, LJ/,ri/}). That is, 2}+ = {a: G 16 | xl is non-empty}, and T>~ = {x G 16 | x^ is nonempty}. Moreover, for all sets of /^/-formulas A, F, semantic consequence relations |=+ and |=~ are canonically defined as follows: This content downloaded on Thu, 7 Feb 2013 10:10:04 AM All use subject to JSTOR Terms and Conditions 422 H. Wansingand Y. Shramko 1. A [=+ F iff for every valuation function v: (if for every A G A, v(A) G P+, then v(B) G P+ for some B G T); 2. A f=~ F iff for every valuation function v: (if for every AeF, then v(B) G P~ for some BeA). v(A) G 2?", Actually, B\§ viewed as the triple (A/, |=+, |=~) provides an example of what is called an harmonious many-valued logic in [43]. We believe that the motivation for (Ctf, \=t, (=/), more fully presented in [36], [38], and for (Ctf, |=+, \=~)i more comprehensively presented in [43], provides sufficient motivation for assuming a generalized notion of a many-valued logic, and hence also a generalized notion of a logical system. In the next section the concept of a logically fc-valued logic consisting of a language C together with fc entailment relations on C is introduced. 5.4. Logically n-valued logics In this section, we suggest to think of a (single conclusion) logic not as a pair (£, h) consisting of a non-empty set of formulas C and a single binary derivability relation h Ç V{£) x £, but as a k + 1-tuple consisting of a nonempty set of formulas together with k single-conclusion binary derivability relations on £, for some k > 2, k G N. Definition 5.5. A Tarskiank-dimensional logic (Tarskian fc-logic) is a fc+1tuple A = (£, hi, . . . , hfc) such that (i) £ is a language in a denumerable set of sentence letters and a finite non-empty set C of finitary connectives, (ii) for every i < fc, \~i Ç V(C) x £, and (iii) every relation \~i satisfies (Reflexivity), (Monotonicity), and (Cut). A is said to be structural iff every hj satisfies (Structurality). Definition 5.6. Let C again be a language in a denumerable set of sentence letters and a finite non-empty set of finitary connectives C. An n-valued fc-dimensional matrix (fc-matrix) is a structure 9JI = (V, Pi, . . . ,Pfc, {fc 'c G C}), where V is a non-empty set of cardinality n (2 < n), 2 < fc, every Vi (1 < i < k) is &non-empty proper subset of V, the sets T>{are pairwise distinct, and every fc is a function on V with the same arity as c.15 The sets T>iare called distinguished sets. A function from C into V is called a valuation in 9JI. 15Thenotion of a fc-matrixis not entirely new. Every /c-matrixis a ramifiedmatrix in the sense of Wôjcicki[45, p. 189]. Ramifiedmatricesare also called generalizedmatrices, see [15, p. 410 ff.]. Note, however,that Wôjcickiassociateswith a ramifiedmatrix a single entailmentrelation,namely f]{\=m | SUt= (V,T>i,{fc : c GC}), 1 < i < k}. This content downloaded on Thu, 7 Feb 2013 10:10:04 AM All use subject to JSTOR Terms and Conditions and the Notion . . . Suszko'sThesis,InferentialMany-valuedness, 423 A pair M = (SW,t;),where M = (V,2?i, . . . ,Vk,{fc : c € C}) is an nvalued fc-matrixand v a valuation in 9Jt,is called an n-valued fc-modelbased on 9DÎ. If v is a homomorphism from (£, ci, . . . , cm) into (V, /Cl, . . . , /Cm), .M is called a structural n-valued fc-model based on 9JÎ. Given an n-valued fc-modelM = ((V, 2?i, . . . , £>&,{/c '-c G C}), v), for every set Vi the semantic consequence relations \=^M Q V(C) x C and (=^ Q ^P(^) x £ are defined as follows: 1. A Kvw A iff v(A) Ç Vi implies v(i4) G A; 2. A |=^ ^1iff v(A) 6 A implies v(S) € V{ for some B € A. Obviously, the relations [=fj^ and \=fj^ are inverses of each other. We may therefore, without loss of generality, focus on the relations f=i^f • ^ we are interested in the preservation of membership in Vi from the conclusion to the premises, we may take the complement of T>{as a distinguished set of algebraic values. Moreover, the relations (=^4 are all Tarskian semantic consequence relations, whence the structure (£, 1=^1? • • • >H^.m) *s a Tarskian fc-logic. Unsurprisingly, given an n-valued fc-matrix SDÎ= (V,î>i, . . . ,2?*, {fc : c G C}), for every set P^ the entailment relations f=^ and |=f^ are defined as follows: 1. A \=um A iff for every valuation v in SDÎ:t;(A) Ç X>^implies v(A) G P<; 2. A ^fïjji A iff for every valuation v in 3JÎ: v(i4) G Vi implies v(B) G Vi for some B G A. A Tarskian fe-logic(£, hi, . . . , h fc) is said to be characterized by an n-valued k-model M (fc-matrixS0t)iff for every h», h< = \=^M (H^m)- (^C,I~i,- - . , h fc) is characterized by a class  of n-valued fc-model (fc-matrices) iff for every • , A, . . ,©*, {/c : c G C}), v) G ^} (h relation h, h = PliN^I «V,©i, = n{KTa«l<V,©i,...,2Ji,...,2?fc,{/c:ceC})6*}). The following statements follow immediately from Wôjcicki's Theorem and Theorem 2.3. THEOREM5.7. Every structural Tarskian k-logic is characterized by a class of structural n-valued k-models, for some n < NoTheorem 5.8. Every Tarskiank-logic is characterizedby a class of n-valued k-models, for some n < Ho-16 16Notethat since we definedthe entailmentrelationsfromthe k-models,the semantics of k-matricesleaves room for other definitionsof entailment. This content downloaded on Thu, 7 Feb 2013 10:10:04 AM All use subject to JSTOR Terms and Conditions 424 H. Wansingand Y. Shramko To every relation \~i in a Tarskian fc-logic one may apply the Suszko Reduction and the Béziau Reduction. But this observation does not raise any deep philosophical doubts about many-valued logic. Caleiro et al. [10, p. 3] are quite right in explaining that "thereis some metalinguistic bivalence that one will not easily get rid off: either an inference obtains or it does not, but not botK\ If falsity is taken seriously and not just dealt with as the complement of truth, we end up, in general, with inferential four-valuedness in the form of preservation of truth, preservation of falsity, preservation of being neither true nor false, and preservation of being both true and false. For this kind of logical four-valuednesswe need at least three algebraic values. Given a set V of algebraic values, the set of all available logical values is V(V) \ 0. But which non-empty subsets of a given set V should we view as logical values? This may depend on philosophical considerations, on the intuitive interpretation of the algebraic values,17 and on the intended applications of a logical system. In the previous section, we emphasized the naturalness of defining a truth order <t and a falsity order </ on 16, which led us to the generalized g-matrix B\q. In [36] it is emphasized that if we consider V = 4 = V(2) and the famous bilattice FOUR2 presented in Figure 2, then truth and falsity are not dealt with as independent of each other, because in FOUR2 it is assumed that T is at least as true as F. There is thus an interplay between choosing a set of algebraic values V and distinguished subsets *D\. . . , T>kof V. Note also that even if one is not committed to the idea of entailment as preservation of semantical values from the premises to the conclusion(s) of an inference (or vice versa), it may still be quite natural to conceive of a logical system as comprising more than just one entailment relation. Malinowski's notion of g-entailment, for example, is in a sense one-sided. The requirement that if every premise is not antidesignated, then the conclusion is designated 17ForSuszko,the relationbetweenthe algebraicand the logicalvaluesis establishedvia characteristicfunctions,and the originalintuitive interpretationof the algebraicvalues is uncoupledfromthe understandingof the logicalvalues as truthand falsity. Similarly,the understandingof the sets of designatedvalues V%is detached from the intuitive understandingof the valuesin V. It may,of course,happenthat there exists a bijectionbetween V and {X>i,. . . ,Pfc}. In the context of (/-consequencerelations, Malinowski[27, p. 83] explainsthat for some inferentiallythree-valuedlogics based on three-elementalgebras,referential assignments. . . and logical valuations... do coincide. This phenomenon delineates a special class of logics, which satisfy a "generalized"version of the FregeanAxiomidentifyingin one-onewaythree logicalvaluesand threesemantical correlates(or referents). This content downloaded on Thu, 7 Feb 2013 10:10:04 AM All use subject to JSTOR Terms and Conditions Suszko's Thesis, Inferential Many-valuedness, and the Notion ... 425 Figure 2. The bilattice FOUR2 seems not to be privilegedin comparisonto the requirementthat if every premiseis designated,then the conclusionis not antidesignated.The latter notionof entailmenthas been studiedby Frankowski[19],[20],who refersto it as p-entailment("plausibility" -entailment). For a given interpretedlanguage,p-entailment,need not coincidewith g-entailment.Againstthe backgroundof the g-matrixL3*with the followingtruth table for conjunction: /a 1 T 0 T T 0 0 0 0 F F F F F F F the formula(A AB) p-entailsA, but it is obviouslynot the case that (A A B) g-entailsA. A p-consequencerelationon a language£ is a subsetof V(C) x C whichis reflexiveand monotonie.Frankowskishowsthat foreverystructural p-consequencerelationthereexists a characterizingclassof g-matrices,where characterizationis definedin terms of p-entailmentinstead of ç-entailment. Interestingly,a relation which is both a g-consequencerelation and a pconsequencerelationis Tarskian.18 Mixed approachesto entailmentare, of course, also possible, see [43]. 18In[16, p. 27] it is stated (without proof) that the relationsof i-entailment, /-entailment, g-entailmentand p-entailmentare interconnectedin the followingway: Eq Ç Et, Eq Ç S/, Et Ç Ep, S/ Ç Ep. Here each of Ex stands for a set of pairs (F, B), such that F is a set of formulas,B is a formula,and T \=x B (wherex is respectively£, /, q ot p) holds in the ç-matrix L3*. Thus, it turns out that these four relationsproducea lattice (relativeto Ç) with Eq as the bottom and Sp as the top. This content downloaded on Thu, 7 Feb 2013 10:10:04 AM All use subject to JSTOR Terms and Conditions H. Wansingand Y.Shramko 426 In [28],Malinowskiconsidersan entailmentrelationwhich leads from nonaccepted(alias undesignated)premisesto rejected(alias false) conclusions. 6. Another analysis Instead of increasingthe numberof entailmentrelationsby associatingto everyset of designatedvaluesan entailmentrelationthat preservesmembership in this set from the premisesto the conclusion,one may increasethe numberof placesof the entailmentrelation. To everyg-matrix,for instance, one might associatea ternaryrelation\=C V(C) x V(C) x C by postulating |= (A,T, A) iff foreveryvaluationv, if v(A) C V+ and v(T) Ç V\(P+UP"), then v(A) eP+. Da Costa et al. [13, p. 292], e.g., explain: The real n-dimensionallogics (n > 2) haveto be developedby breaking downthe deepestroot of the principleof bivalence.We can easily imagine, for instance, a rule of deductionwith threepoles or more (our emphasis). Higher-aritysequent systems for many-valuedlogics have been considered in the literaturealreadyin the 1950s by Schrôter[33],and later by several other authors,see also [22]and [30]. Higher-aritysequentsfor modal logics have been investigatedby Blarneyand Humberstone[7]. The use of higherarity sequentsmay havesome technicaland conceptualmerits. However,we assume that a selection of logical values (alias non-emptysubsets of some set V) is particularlywell-motivatedif everylogicalvalue is associatedwith a dimensionaccordingto which V can be partiallyordered(truth,falsity, information,necessity,constructiveness,19 etc.). Formthis point of view, it is naturalto let each logicalvalue give rise to a binaryentailmentrelation. 7. Concluding remarks In our considerationson Suszko'sThesis we arguedthat a logicalvalue is a value that gives rise to an entailmentrelationin a canonicalway. Although for each such entailmentrelationthere exists a representationin terms of bivaluations,this does not show that it is unreasonableor unnecessaryto assumemorethan one logicalvalue. The observationthat every entailment relationhas a bivalentsemanticsdoes,however,cast doubt on the enterprise of providinga bivalent semantics seen as a researchprogramthat is not constrainedby any furtherrequirements,see also [3]. Indeed, as has been 19See, [35]. This content downloaded on Thu, 7 Feb 2013 10:10:04 AM All use subject to JSTOR Terms and Conditions and the Notion . . . Suszko'sThesis,InferentialMany-valuedness, 427 shown by Richard Routley [32] using a canonical model construction, every logic based on a A-categoricallanguage has a characterizing bivalent possible worlds semantics, where by a logic Routley understands an axiomatic system consisting of a countable set of axioms and a countable set of derivation rules leading from a finite number of premises to a single conclusion. Routley [32, p. 331] also notes that "if every logic on a A-categorial language has a twovalued worlds semantics then ipso facto it has a three-valued . . . semantics". This kind of 'reduction5 to three-valuedness certainly does not refute the claim that there are just two truth values, and likewise the Suszko Reduction does not prove that there are but two logical values. This work was supported by DFG grant WA 936/ Acknowledgements. 6-1. We would like to thank the two anonymous referees for their very valuable comments. References [1] Belnap, N.D., 'How a computer should think', in G. Ryle (éd.), Contemporary Aspectsof Philosophy,Oriel Press Ltd., Stocksfield,1977, pp. 30-55. [2] Belnap, N. D., 'A useful four-valuedlogic, in J.M. Dunn and G. Epstein (eds.), Modern Uses of Multiple-ValuedLogic, D. Reidel Publishing Company,Dordrecht, 1977, 8-37. [3] van Benthem, J., 'Possible Worlds Semantics, a ResearchProgramthat Cannot Fail?', Stadia Logica43 (1984), 379-393. J.-Y., UniversalLogic, in T. Childersand O. Majer (eds.), ProceedingsLog[4] BÉZIAU, ica' 94, Czech Academyof Sciences,Prague, 1994, pp. 73-93. J.-Y., 'What is Many-ValuedLogic?',in Proceedingsof the 27th International [5] BÉZIAU, Symposiumon Multiple-ValuedLogic, IEEE Computer Society Press, Los Alamitos/Cal., 1997, pp. 117-121. J.-Y., 'Recherchessur la logiqueabstraite: les logiquesnormales',Acta Uni[6] BÉZIAU, versitatis Wratislaviensisno. 2023, Logika18 (1998), 105-114. [7] Blamey, S., and L. Humberstone, 'A Perspectiveon Modal SequentLogic', Publications of the ResearchInstitute for MathematicalSciences, Kyoto University27 (1991), 763-782. [8] Brown, B., and P. Schotch,'Logicand Aggregation',Journalof PhilosophicalLogic 28 (1999), 265-287. [9] Caleiro, C, W. Carnielli, M. Coniglio, and J. Marcos, 'Two'scompany: "The humbugof manylogicalvalues'",in J.-Y. Beziau(éd.), LogicaUniversalis,Birkhâuser, Basel, 2005, 169-189. [10] Caleiro, C, W. Carnielli, M. Coniglio, and J. Marcos, 'Suszko'sThesis and dyadic semantics',preprint, http://www.cs.math.ist.utl.pt/ftp/pub/CaleiroC/03-CCCM-dyadicl.pdf. [11] Caleiro, C, W. Carnielli, M. Coniglio, and J. Marcos, 'Dyadicsemanticsfor many-valued logics', preprint, http://www.cs.math.ist.utl.pt/ftp/pub/CaleiroC/03-CCCM-dyadic2.pdf. This content downloaded on Thu, 7 Feb 2013 10:10:04 AM All use subject to JSTOR Terms and Conditions 428 H. Wansingand Y. Shramko [12] COOK,R., What's wrong with tonk (?)', Journal of PhilosophicalLogic34 (2005), 217-226. [13] da Costa, N., J.-Y. Béziau, and O. Bueno, 'Malinowskiand Suszko on manyvaluedlogics: On the reductionof many-valuednessto two-valuedness',ModernLogic 6 (1996), 272-299. [14] Curry, H. B., Foundationsof MathematicalLogic,McGraw-Hill,New York, 1963. [15] Czelakowski, J., ProtoalgebraicLogics, Kluwer Academic Publishers, Dordrecht, 2001. [16] Devyatkin, L., 'Non-classical definitions of logical consequence' (in Russian), Smirnov'sReadingsin Logic.Fifth Conference,Moscow(2007), 26-27. Dunn, J.M., 'Intuitive semantics for first-degreeentailment and 'coupled trees", [17] PhilosophicalStudies 29 (1976), 149-168. [18] Dunn, J.M., 'Partialityand its dual', Studia Logica66 (2000), 5-40. [19] Frankowski, S., 'Formalizationof a plausibleinference',Bulletin of the Section of Logic33 (2004), 41-52. Frankowski, S., 'p-consequenceversus (/-consequence',Bulletin of the Section of [20] Logic33 (2004), 197-207. S., MehrwertigeLogik.Eine Einfuhrungin Théorieund Anwendungen, [21] GOTTWALD, Berlin, 1989. Akademie-Verlag, A Treatiseon Many-valuedLogic,ResearchStudies Press, Baldock, Gottwald, S., [22] 2001. [23] Jennings, R., and P. Schotch, 'The Preservationof Coherence',StudiaLogica(43) 1984, 89-106. Malinowski, G., 'Q-ConsequenceOperation', Reports on MathematicalLogic 24 [24] (1990), 49-59. [25] Malinowski, G., 'Towardsthe Concept of Logical Many-Valuedness',Folia Philosophica7 (1990), 97-103. Malinowski, G., Many-valuedLogics,ClarendonPress, Oxford,1993. [26] in Jan Wolenski(éd.), Philosophical [27] Malinowski, G., 'InferentialMany-Valuedness', Logicin Poland, KluwerAcademicPublishers,Dordrecht,1994, 75-84. [28] Malinowski, G., 'InferentialParaconsistency',Logicand LogicalPhilosophy8 (2001), 83-89. [29] Malinowski, G., 'InferentialIntensionality',Studia Logica76 (2004), 3-16. R., 'On Partialand ParaconsistentLogics',NotreDame Journalof Formal [30] MuSKENS, Logic40 (1999), 352-374. [31] Rescher, N., Many-ValuedLogic,McGraw-Hill,New York, 1969. [32] Routley, R., UniversalSemantics?',Journal of PhilosophicalLogic4 (1975), 327356. [33] Schroter, K., 'Methodenzur AxiomatisierungbeliebigerAussagen-und Pràdikatenkalkiile', Zeitschriftfur mathematischeLogik und Grundlagender Mathematik1 (1955), 241-251. [34] Scott, D., Backgroundto Formalization',in H. Leblanc (éd.), Truth,Syntax and Modality,North-Holland,Amsterdam,1973, 244-273. [35] Shramko,Y., J.M. Dunn, and T. Takenaka,'The trilattice of constructivetruth values', Journalof Logic and Computation11 (2001), 761-788. This content downloaded on Thu, 7 Feb 2013 10:10:04 AM All use subject to JSTOR Terms and Conditions and the Notion . . . Suszko'sThesis,InferentialMany-valuedness, 429 [36] Shramko, Y., and H. Wansing, 'Some useful 16-valued logics: how a computer networkshould think', Journalof PhilosophicalLogic34 (2005), 121-153. [37] Shramko, Y., and H. Wansing, 'The Logic of ComputerNetworks'(in Russian), LogicalStudies (Moscow)12 (2005), 119-145. generalizedtruth values, and [38] Shramko, Y., and H. Wansing, 'Hypercontradictions, and Journal of truth falsehood', of Logic,Languageand Information15 (2006), logics 403-424. [39] Suszko, R., 'The Pregeanaxiom and Polish mathematicallogic in the 1920's',Studia Logica36 (1977), 373-380. [40] Tsuji, M., 'Many-valued logics and Suszko's Thesis revisited', Studia Logica 60 (1998), 299-309. [41] Urquhart, A., 'An interpretationof many-valued logic', Zeitschriftfur mathematische Logikund Grundlagender Mathematik19 (1973), 111-114. [42] Wansing, H., 'Connectivesstrangerthan tonk', Journal of PhilosophicalLogic 35 (2006), 653-660. [43] Wansing, H., and Y. Shramko, 'Harmoniousmany-valued propositionallogics and the logic of computernetworks',to appearin C. Dégremont,L. Keiffand H. Riickert (eds.), Festschriftdedicatedto ShahidRahman,CollegePublications,London,2008. R., 'Some Remarkson the ConsequenceOperationin SententialLogics', [44] WÔJCICKI, FundamentaMathematicae68 (1970), 269-279. [45] Wojcicki, R., Theoryof LogicalCalculi: Basic Theoryof ConsequenceOperations, KluwerAcademicPublishers,Dordrecht,1988. Heinrich Wansing Institute of Philosophy DresdenUniversityof Technology 01062 Dresden,Germany . de Heinrich . WansingQtu-dresden Yaroslav Shramko Departmentof Philosophy State PedagogicalUniversity 50086 KrivoiRog, Ukraine . ua yshramkoQukrpost This content downloaded on Thu, 7 Feb 2013 10:10:04 AM All use subject to JSTOR Terms and Conditions