Assertoric semantics and the computational power of self-referential truth (Q427223)

The paper under review is concerned with the liar paradox based on the liar sentence ``This sentence is not true.'' The liar sentence is in the paper denoted by \(\lambda\) and identified as the solution to the equation \(\lambda = [\lambda\) is not true]. Similarly, the truthteller is the sentence expressing about itself ``This sentence is true'' and satisfying the equation \(\tau= [\tau\) is true]. A widespread opinion, expressed for instance by the philosopher Ludwig Wittgenstein, is that the liar sentence is useless in language. The purpose of the paper is to disprove this conviction by showing that a language which contains both self-reference and a truth predicate has larger computational power than a language which lacks at least one of these components. To solve the problem by precise logical methods, the author introduces a truth language \(L\) containing propositional atoms, sentential constants, sentences, and sentential logical operators \(\neg, \wedge, \vee\). \(L\) also contains a truth predicate \(T\) such that if \(t\) is a constant, then \(T(t)\) is a sentence. There are two more operators in \(L\). If \(\alpha\) is a sentence, then \([\alpha]\) is a sentential constant denoting \(\alpha\). The inverse operator \(\pi\) maps any sentential constant to the sentence it denotes. Thus \(\pi([\sigma]) = \sigma\) for any sentence \(\sigma\). The logic of \(L\) is classical sentential logic enhanced with a small fragment of predicate logic. Next two further operators \(A\) and \(D\) are introduced, with the intuitive interpretations: \(A_\alpha \Leftrightarrow \alpha\) is allowed to be asserted \(\Leftrightarrow\) by asserting \(\alpha\) one does not assert a falsehood, deny a truth, or contradict oneself. \(D_\alpha \Leftrightarrow \alpha\) is allowed to be denied \(\Leftrightarrow\) by denying \(\alpha\) one does not assert a falsehood, deny a truth, or contradict oneself. It is unclear whether \(A\) and \(D\) are added to the truth language \(L\) or are meant to belong to the metalanguage of \(L\). Ten deduction rules for \(A_\alpha\) and \(D_\alpha\) in a semantic tableaux system are given, each depending on the logical structure of \(\alpha\). The semantics for the operators \(A\) and \(D\) is stated in the form of valuation functions \(V_A(\alpha, w)\) and \(V_D(\alpha, w)\) where \(w\) is a world and the values of \(V_A(\alpha, w)\) and \(V_D(\alpha, w)\) express whether \(\alpha\) is allowed to be asserted in \(w\) and \(\alpha\) is allowed to be denied in \(w\), respectively. For the liar sentence \(\lambda\), we get \((\neg A_\lambda \wedge \neg D_\lambda)\) is true in all worlds, and for the truthteller sentence \(\tau\), we have \((A_{\tau} \wedge D_{\tau})\) true in all worlds. The truth of the other possible combinations, \((A_{\alpha} \wedge \neg D_{\alpha})\) and \((\neg A_\alpha \wedge D_\alpha)\), can be obtained for a posteriori sentences \(\alpha\) and suitable worlds. The main result in the article is a theorem to the effect that a truth language like \(L\) with self-reference and a truth predicate has larger computational power than a language without at least one of these utilities, in the article called a classical language. As an example, the author considers a problem consisting in determining which one of four sentences \(p_1, p_2, p_3, p_4\) is true, given that exactly one is true. The problem is solved by asking yes/no questions to an oracle. In a classical language, the problem can be solved by asking two questions, but not by only one. In a truth language, it suffices with only one question \(\theta= ((\lambda \wedge p_1) \vee (\tau \wedge p_2)) \vee p_3\), provided that the oracle answers with the correct alternative among the four combinations \((A_\theta\wedge D_\theta), (A_\theta\wedge \neg D_\theta), (\neg A_\theta\wedge D_\theta), (\neg A_\theta\wedge \neg D_\theta)\) which imply \(p_2, p_3, p_4\), and \(p_1\), respectively. \(\lambda\) and \(\tau\) are used in \(\theta\) as markers on \(p_1\) and \(p_2\), respectively. This reduction in query complexity from two to one question is an example of what the author means by increase in the computational power in an assertoric self-referential truth language compared with a classical language. From this result, the author concludes that the liar sentence does accomplish something useful in the language and that therefore the thesis about its being useless within our language, entertained by Wittgenstein and several others (including the present reviewer), is refuted. Reviewers comments: 1. The author uses the formal language \(L\) as a model of a fragment of a natural language like English in the study of the liar sentence. A fundamental methodological principle in applied mathematics is that a model of a system is acceptable for the study of a given problem concerned with the system only if the differences between the system and the model have at most negligible consequences for the solution in the model of the problem. This principle is not respected by the author. In a formal language, the sentence is primary. The meaning of the sentence is attached to it via an interpretation in a model. In a natural, non-formal language, the meaning (a proposition) is primary and a sentence is an instrument only for the expression of the proposition. In formal languages, sentences are the primary truth bearers. In a non-formal language, propositions are the primary truth bearers while sentences inherit their truth value from the underlying proposition. Not all sentences, including the liar sentence and the truthteller sentence, have underlying propositions. Formal languages and non-formal languages have different logics as witnessed by the paradoxes by W. Cooper and A. Ross. All these differences are of paramount importance in the study of the liar paradox and cannot be neglected. A formal language is therefore not a suitable model of a fragment of a natural language when one studies the liar paradox. 2. The mathematics in the article is mostly carefully elaborated; but there are important exceptions. In the truth language \(L\), \(\lambda\) and \(\tau\) are defined by the equations \(\lambda = [\neg T(\lambda)]\) and \(\tau=[T(\tau)]\); but the identity sign is not listed among the symbols of \(L\). Worse is that, given that \(L\) is defined as the author does, it cannot be proven in \(L\) that the defining equations for \(\lambda\) and \(\tau\) have solutions. Therefore an axiom scheme, which states that each such equation involving self-reference has a unique solution, should be added. If it is the author's intention that the liar paradox should be derivable in \(L\), then also the equivalence principle (\(T\)-principle) in the form of an axiom scheme must be included: \((\pi(\sigma) \leftrightarrow T(\sigma))\), for each sentence \(\sigma\) in \(L\). With \(L\) thus amended, we get a theory \(T_L\) rather than just the language \(L\). The author gives only a partial semantics for the language \(L\). The worlds \(w\) are models of the purely sentential logical part of \(L\). Meanings must be attributed also to the predicate logical part of \(L\): the truth predicate \(T\) and the sentential constants. This is not done by the author and it cannot be done either. The existence of the liar paradox implies that the extension of the truth predicate cannot be well-defined for any formal truth language. The only way out of this dilemma is to realise that sentences cannot be primary truth bearers in a non-formal natural language and therefore not in the formal language model \(L\) of it either. 3. The author's alleged result that the 1-out-of-4 problem in \S1.4 can be solved by only one question \(\theta\) in \(L\) (or rather \(T_L\)) is not valid. This can be shown by a dilemma. Either the liar paradox can be derived in \(L\) or else it cannot. If the liar paradox can be derived, then \(L\), or rather \(T_L\), is inconsistent. Then anything is provable in \(L\), and in particular \((\tau \leftrightarrow (\beta \wedge \neg \beta))\) is provable for an arbitrary sentence \(\beta\) in \(L\). But then by asserting \(\tau\) one contradicts oneself so that \(\neg A_{\tau}\), and then \(\tau\) cannot have the unambiguous marking function needed in the question \(\theta\). If on the other hand the liar paradox cannot be derived in \(L\), then it is not possible to show in \(L\) that one contradicts oneself by asserting \(\lambda\). Hence \(A_\lambda\) is true so that \(\lambda\) loses the marking function needed in the question \(\theta\). It might be that what the author means is that the liar paradox can only be derived in the metalanguage of \(L\) and not in \(L\) itself. The objection to this is that it is not the speech acts of asserting and denying \(\lambda\) which give rise to the paradox. It is the truth and non-truth of \(\lambda\), and the truth predicate \(T\) certainly belongs to \(L\). 4. The author of the article has not succeeded in throwing new light on the liar paradox. Neither has he succeeded in showing that the liar sentence ``This sentence is not true'', or its formal counterpart \(\neg T(\lambda)\), is useful, be it in formal languages or in natural non-formal languages. The alleged proof of the main theorem in the paper under review on the increased computational power of assertoric languages with self-referential truth is invalid. Or maybe the author meant something else.