aa r X i v : . [ m a t h . HO ] J un Sense, reference, and computation
Bruno Bentzen
Abstract
In this paper, I revisit Frege’s theory of sense andreference in the constructive setting of the meaning explanations oftype theory, extending and sharpening a program–value analysis ofsense and reference proposed by Martin-L¨of building on previouswork of Dummett. I propose a computational identity criterionfor senses and argue that it validates what I see as the mostplausible interpretation of Frege’s equipollence principle for bothsentences and singular terms. Before doing so, I examine Frege’simplementation of his theory of sense and reference in the logicalframework of
Grundgesetze , his doctrine of truth values, and viewson sameness of sense as equipollence of assertions.
Frege was one of the first to dispute that all mathematical truthsare based on intuition, the cornerstone of the dominant Kantianphilosophy of mathematics, and put forward the thesis that allarithmetical truths are reducible to logical truths. The successof his logicism would eventually depend on the plausibility ofthe development of a new system of logic in which all the References to Frege’s works will be cited by name only and indicatedby page or section number in the text. I shall ambiguously refer to theformal language and formal theory of
Grundgesetze by “ideography”, unlessexplicitly stated otherwise.
Begriffsschrift , the bookwhich lays down the first systematic treatment of modern logic,the ideography, and applies the system to the mathematicaltheory of sequences. Frege hints at his logicist program in thepreface to his book, announcing that his investigations willbe continued with the logical elucidation of the concepts ofnumber, magnitude, and so forth in what he describes as animmediately following publication. Yet, it would not be untilfourteen years later that the promised book would finally appearin the form of the first volume of
Grundgesetze . Frege himselfexplains that this long delay was due to the accommodationof essentially two major changes to the ideography, a taskthat ultimately forced him to discard an almost completedmanuscript. The first change was the introduction of value-ranges of functions, new objects that are intended to generalizethe informal notion of extension of concept that he invokesin §
68 of
Grundlagen to overcome the infamous Julius Caesarproblem with an explicit definition of the concept of number.This point has been studied extensively and it will not concernus here. Instead, I would like to focus on Frege’s theory of senseand reference, the doctrine that singular terms and completesentences have a mode of presentation and denote an object,whose implementation consists in the second substantial changemade to the ideography.Unfortunately, Frege’s formal account of his distinctionbetween sense and reference leaves much to be desired, as hefailed to develop his ideography sufficiently in order to capturehis answer to the paradox of identity, namely, the problem ofexplaining the apparent difference in cognitive value betweentrue identity statements of the form a = a and a = b . This willbe the main subject of Section 2, where I shall examine the roleof the theory of sense and reference in the technical developmentof Frege’s logicist project and his views on sameness of senseas equipollence of assertions. In Section 3, I shall shift gears Grundgesetze
I, ix. For a recent survey on the topic see Bentzen (2019). and turn to an opposing philosophical position that I take topossess a more suitable semantic setting for the establishment ofa proper theory of sense and reference, namely, the constructiveconception of mathematics. The idea of reinterpreting Frege’straditional semantic distinction in a constructive setting iscertainly not new, as it has been first explored mainly byDummett (1978) and Martin-L¨of (2001). The latter author hasextended considerably the ideas of the first one by reshapingthem in the context of his type theory, a formal system inwhich every term is assigned to a type and every operationstrictly restricted to terms of a certain type. But at thesame time several aspects of Martin-L¨of’s account strike meas incomplete or problematic, especially because he does notseem to make full use of the powerful intended semantics oftype theory, an informal realizability interpretation known asthe meaning explanations (Martin-L¨of, 1982). My goal in thispaper is to revise of some of Martin-L¨of’s main views, a taskthat will be carried out in Section 4, where I carry the meaningexplanations to their ultimate logical conclusions.
In Frege’s works we find the two forms of judgment that arecurrent in modern logic, the judgment form that asserts that aproposition is true, which is written in turnstile notation as A and the judgment form that states that a sentence expresses aproposition, which is implicitly introduced for the first time in § Begriffsschrift via the content stroke A In this paper, I will be concerned with a fragment of Martin-L¨of (1982)where the only types are dependent function types, equality types, andthe natural numbers. Judgments of typehood are used instead of typeuniverses, and typehood and type membership judgments are defined viaprimitive type and term judgments, as described in Section 3.
Therein, a proposition is described as a “judgeable content”and identified with the content of a turnstile judgment.This intuitive notion of content represents the main semanticelement of the ideography at this earlier point and, in fact,logical operations such as implication, negation, and universalquantification are all exclusively applicable to contents, or,more precisely, to judgeable contents or propositions.The only exception to this structure is the relation ofidentity of content which is introduced in § Begriffsschrift asa metalinguistic relation between the expressions themselvesand not their respective contents. That is, according to thisinterpretation of equality, an equality statement a = b meansthat the expressions ‘ a ’ and ‘ b ’ have the same content. Onesemantic consequence of this is that two propositions cannot beidentified if they imply each other, for they may have differentcontents. Indeed, the propositional extensionality principle( A Ą B ) Ą ( B Ą A ) Ą ( A = B )is not only not a theorem in Begriffsschrift but also inconsistentwith its formal system, as noted in Duarte (2009). However,once the ambiguous notion of content is split into sense andreference, propositional extensionality becomes a valid principlefor references, although not for senses, since the fact that twotheorems are logically equivalent does not mean that they havethe same cognitive value. Frege’s theory of sense and reference is systematicallyintroduced for both singular terms and complete sentences inhis seminal essay
Sinn und Bedeutung . The theory seems toprovide a compelling answer to the paradox of identity withrespect to singular terms, for we are able to say that the Curiously, Frege (1980) once considered an identity criterion for sensesalong those lines in a letter to Husserl dated 9 December 1906, but thissuggestion is clearly unacceptable, since it would mean that every two truepropositions a = a would a = b express the same sense. cognitive value of a = a and a = b is generally different because a and b may differ in their sense, despite being coreferential.However, the theory becomes highly controversial when weconsider the sense and reference of sentences, which are taken toexpress a thought (in modern terminology, a proposition) andrefer to a truth value, which, since referents are always objects,are seen as the simplest representatives of logical objects. Fregeis well aware of this counterintuitive aspect of his doctrine, ashe immediately attempts to justify it by noting that what hecalls an object can only be properly understood in connectionto concepts and relations and then attempts to substantiatehis claim that the reference of a sentence is its truth value byremarking that the latter remain unchanged when a part ofthe sentence is replaced by a coreferential expression, that is,that sentences with the same truth value are interchangeable salva varitate . Later on, in the preface of Grundgesetze ,Frege remarks that the introduction of truth-values may seemstrange at first, but adds that the fact that everything becomesmuch simpler and sharper with those objects puts a greatweight in the balance in favor of his own conception. Inthe ideography, concepts and relations are treated as particularcases of functions that output truth values and Frege explicitlystates that an object is anything that is not a function.Besides the poor justifications for Frege’s adoption of hisstrage doctrine of truth values, it is disappointing that almostno considerations about sense or thoughts can be found in
Grundgesetze , at least if we assume that the real purpose ofthe distinction between sense and reference is to offer a solutionto the paradox of identity. Although Frege still seems to havea strong conviction that a criterion of identity for senses is offundamental importance, as we can judge from, for instance, aDecember 1906 letter to Husserl, where Frege writes:It seems to me that an objective criterion isnecessary for recognizing a thought again as thesame, for without it logical analysis is impossible.(Frege, 1980, p. 70) Grundgesetze
I, x.
Frege does not seem to bother to provide a logic of sense andreference in the book he intends to show once and for all thatarithmetic is nothing but logic. For this reason, we can saythat Frege failed to show that some mathematical equalitystatements have cognitive value in
Grundgesetze , because hissolution to the identity paradox is entirely dependent on whatit means for two expressions to have different senses, and noidentity criterion for senses is given in his ideography.It is possible that Frege had come to recognize that the mostimportant notion for the vindication of his logicism is that ofreference, as pointed out by Simons (1992), and that for thosepurposes all that we need to know about sense is that a sentenceneeds to express a thought in order to refer at all. In fact, nowthe horizontal stroke symbolism A which used to indicate that A is a proposition (judgeablecontent) before the division of content into sense and reference,no longer says that A expresses a proposition (thought). Insteadof being an implicit form of judgment, the horizontal is treatedas a function that refers to a truth value depending on thereference of A , yielding the true if A denotes the true andthe false otherwise. Similarly, negation does not operate oncontents anymore, for it is explicitly treated as a function A that has the opposite effect of yielding the false if A does notrefer to the true and the true otherwise. The turnstile judgmenthas kept its role, being the only judgment form of the revisedideography. It is now fully explained in terms of reference aswell, now asserting that the expression A refers to the true.There is no way to formally assert the fact that a sentenceexpresses a thought in Grundgesetze , which is to say to fulfillthe role previously played by the content stroke.In general, it would seem that Frege contradicts himselfby downplaying the value of sense and treating his doctrineof truth values as a mere technical device for the successfulaccomplishment of his logicism, as argued in Ruffino (1997)and Duarte (2009). This can be seen more clearly in light of thediscarded manuscript that was written before the introductionof the distinction between sense and reference to the ideographyin
Grundgesetze . In §
69 of
Grundlagen , immediately after theexplicit definition of the concept of number is proposed byFrege, he attempts to confirm the fruitfulness of his definitionby sketching a proof of an crucial theorem that says that thenumber that falls under the concept F is equal to the numberthat falls under the concept G iff F and G are in one-to-one correspondence (Hume’s Principle). The proof consistsin the showing that both sides of the biconditional implyeach other. But in the ideography, where a biconditional isalways represented by an equality, the formalization of theproof cannot be completed without propositional extensionality.In Grundgesetze , what makes the derivation of propositionalextensionality as a theorem possible it is the introduction ofBasic Law IV, which roughly states that the truth valuesdenoted by the sentences A and B either coincide or not ¬ ( A = B ) Ą ( A = B )But this is only allowed as an axiom in the ideography becauseof the distinction between sense and reference. If thoseobservations are correct, Frege conceived his theory of sense andreference primarily as a logical apparatus necessary to overcometechnical obstacles rather than as a solution to the paradox ofidentity as it is commonly thought. That would explain whyFrege does not do justice to the notion of sense in
Grundgesetze . What comes closer to the proposal of a criterion of identityfor senses in Frege’s writings is the equipollence principle thattwo sentences A and B express the same thought provided thatanyone who accepts A as true must also immediately accept B as true and vice-versa. Sundholm (1994) has called attention to See e.g. Ruffino (1997, §
3) and Duarte (2009, p.167). the fact that this idea has been informally suggested on manyoccasions by Frege, most notably in the opening passage of his
Kurze ¨Ubersicht meiner logischen Lehren :Now two propositions A and B can stand in such arelation that anyone who recognizes the content of A as true must thereby also recognize the contentof B as true and, conversely [...] So one has toseparate off from the content of a proposition thepart that alone can be accepted as true or rejectedas false. I call this part the thought expressed bythe proposition. (Frege, 1906, pp.197–98)I take the following equipollence to be the most natural choiceof a corresponding principle for singular terms: two singularterms a and b express the same sense if for every predicate P ,anyone who accepts P ( a ) as true must also immediately accept P ( b ) as true and vice-versa. Related to salva veritate , this ideais already hinted at in Sinn und Bedeutung
If we now replace one word of the sentence byanother having the same reference, but a differentsense, this can have no bearing upon the referenceof the sentence. Yet we can see that in such a casethe thought changes; since, e.g., the thought in thesentence ‘The morning star is a body illuminatedby the Sun’ differs from that in the sentence ‘Theevening star is a body illuminated by the Sun.’Anybody who did not know that the evening star isthe morning star might hold the one thought to betrue, the other false. (Frege, 1892, p.62)Yet, much remains to be done before both equipollenceprinciples for singular terms and sentences can be adopted assatisfactory identity criteria for senses, for we do not have arigorous account of what does it mean to “immediately” accepta proposition as true. In
Funktion und Begriff , Frege hasexpressly stated that the sense of two expressions is equal upto renaming of bound variables, that is, that we could write x − x directly as y − y without altering its sense, however,it can be very difficult to determine the extent of strictness inhis account of sameness of sense, considering that Frege alsohappens to claim that the two halves of Basic Law V expressthe same sense, but in a different way. In the remainder of this paper, I will argue that a constructiverendering of the theory of sense and reference is capable of notonly providing a precise computational interpretation to theequipollence principles criteria for singular terms and sentences,but also validating them. I will begin our discussion with a briefoutline of some important aspects of constructive semantics.In Frege’s semantics, as mentioned in the previous section,the assertion that a proposition is true is understood asdeclaring the fact that the thought expressed by a sentencedenotes the true, an interpretation that rests on a realistassumption that numbers and other sorts of mathematicalobjects are non-physical and non-mental entities that do notexist in space or time. Therefore, the most fundamental formof judgment in logic and mathematics is the truth judgment,which Frege writes as A In contrast, the constructivist tradition states that theonly legitimate way of understanding a proposition is as aspecification of a construction with certain given propertiesand, moreover, that in order to assert that a proposition is trueone has to exhibit a construction that realizes the specificationexpressed by the given proposition (Martin-L¨of, 1985). Thecentral form of judgment is therefore the one that states that a is a construction that realizes a proposition A , which, usingthe language of type theory, is often expressed as Both claims are found in Frege (1891, p.27). Frege’s views on samenessof sense have been dealt with in more detail in Sundholm (1994, p.304–307),Klement (2016, § § a : A Put differently, to assert that a proposition is realized bya construction is to declare the truth of that proposition.One thing that should be emphasized is that the existenceof a construction should not be understood as an ontologicalexistential claim that may be verified by means independentof our knowledge, but rather as the epistemic act ofconceiving such a mathematical construction. Accordingly, theconstructive conception of truth can be regarded as a form ofanti-realism with respect to mathematical objects.
In type theory, a construction is a humanly computableprocedure or program that is generally accepted as a methodfor obtaining a mathematical object. Following the so-calledpropositions-as-types correspondence (Howard, 1980), bothpropositions and sets are specifications of constructions thatare structured through the unifying concept of a type. Theelements of a type are commonly called terms, and, as expected,terms are interpreted as constructions.For a more concise semantic explanation, it is convenientto base our type theory on two equality forms of judgmentthat state that two expressions are equal types and also thattwo expressions are equal terms of a type. I shall write thoseequalities with a triple bar notation to emphasize the fact thatthey are judgmental relations, meaning that they only occur inassertions and therefore are not subject to operations such asthose determined by logical connectives A ≡ B type a ≡ b : A The reflexivity of those equality judgments can be used to definetheir more common counterparts A type and a : A , which saythat an expression is a type and that an expression is a term of atype, as A ≡ A type and a ≡ a : A . In other words, to determinetheir meaning, it suffices to determine the meaning of the type1and term equality as primitive judgments. The general strategyis that the meaning of a judgment is given via an untypedmodel of computation, an idea that derives directly from themeaning explanations (Martin-L¨of, 1982). Every meaningfulstatement is based on a notion of computation that is acceptedas a primitive concept and used to give terms a computationalbehavior, define types as term specifications, and to assignterms to types based on the values to which they compute. The whole process of meaning explanation starts with thespecification of a native computation system that takes theform of a programming language, typically the untypedlambda-calculus extended with constants for the type-formers,constructors, and eliminators of the type theory. In this paper,I will consider dependent function types Π x : A B , equality types a = A b , and the natural numbers nat . First, we specify thesyntax of the programming language var := x | y | z | ... | x ′ | y ′ | z ′ | ... expr := var Π ( var : expr ) expr | λ var . expr | expr ( expr ) expr = expr expr | refl ( expr ) | eqrec ( expr , expr , expr ) nat | | succ ( expr ) | natrec ( expr , expr , expr )Then, we endow this language with an operational semantics,a complete description of how terms are expected to computegiven in the form of a transition relation over closed terms, thatis, expressions with no occurrence of free variables, and reflexiveon certain closed terms, which are regarded as execution values.This transition relation is fully captured by the computationrules that I now symbolically describe, where a a ′ indicatesthe fact that a transitions to a ′ and a val means that a is a value,which is to say that a a is the case. I will start consideringthe rules for the dependent function type, which generalizes thenotion of universal quantification and dependent product of sets2 Q x : A B val λx.a val a a ′ a ( b ) a ′ ( b )( λx.a )( b ) a [ b/x ] λx.a ( x ) a The dependent function type Π ( x : A ) B is inhabited by functions λx.a , where a is an open term that may depend on x , andif f is a function and a a term, then f ( a ) is the applicationof f to a . The last two computation rules are of particularimportance, because they induce some obvious resemblancesbetween lambda-terms λx.f ( x ) and Frege’s value-ranges ´ ǫf ( ǫ ),since ´ ǫf ( ǫ ) ∩ a = f ( a ) is a theorem in the ideography (see §
34 of
Grundgesetze ), although ´ ǫf ( ǫ ) = f contradicts theintended semantics of the ideography since value-ranges aretaken to be objects and cannot be functions. I explore thisconnection in detail in Bentzen (2020b), which contains a morecomprehensive type-theoretic study of the ideography.The following rules provide a full computational account ofthe equality type a = A b , the type-theoretic counterpart of theusual notion of an equality proposition. Unlike the judgmentalequality a ≡ b : A , a = A b does not have any assertive force.The constructor refl ( a ) represents the reflexivity of equalityand the eliminator eqrec ( a, b ) generalizes the principle of salvaveritate that Frege borrows from Leibniz. a = A b val refl ( a ) val a a ′ eqrec ( a, b ) eqrec ( a ′ , b ) eqrec ( refl ( a ) , b ) b Finally, we have the rules for natural numbers, which giveus , a successor operator succ ( n ) and an explicit recursor3 natrec ( a, b, c ) that incorporates the principle of completemathematical induction nat val 0 val succ ( a ) val a a ′ natrec ( a, b, c ) natrec ( a ′ , b, c ) natrec ( , a, b ) a natrec ( succ ( a ) , b, c ) ( c ( a ))( natrec ( a, b, c ))As it can be seen, the basic pattern here is that all type-formersand their constructors are interpreted as values, eliminatorspreserve computation on their main argument, and eliminatorstransition to a particular information that was given to themfor when their main argument is a certain constructor.The transition relation divides terms into two classes: thosethat are themselves values are called canonical, and those thatare not themselves values but may eventually reach a value if wekeep iterating the transition process are called non-canonical.This iteration is made precise with the notion of evaluation, acomputation that halts when it founds the value that a termtransitions to after an indefinite number of transition steps. Ishall write a ↓ a ′ to mean that a evaluates to a ′ . Since we aredealing with untyped computations, the evaluation of a termwill not always terminate, but it suffices to characterize closedexpressions as programs that when executed output the valuethey evaluate to. Roughly, we explain what a type is by specifying the termsof that type following an idea that can be traced back tothe constructive conception of set advocated by Bishop (1967)and the interpretation of the intuitionistic logical constantsproposed by Heyting (1934). In fact, the first thing we doto explain what types are is distinguishing between canonicaland non-canonical types. Any expression that evaluates to a4canonical type is a type, and a canonical type is explained byprescribing what their canonical terms are.The meanings of the type and term equality judgments aremutually explained in a similar way. To account for the former,we assume that A and A ′ are closed expressions, and assert that A and A ′ are equal types provided that they evaluate to thesame canonical type up to type equality. This characterizes type equality as a relation between termsthat behave in a certain expected way. The following rulesillustrate how type equality can be explained for our canonicaltypes: dependent function types, equality types, and naturalnumber types A ↓ Q x : B C A ′ ↓ Q x : B ′ C ′ B ≡ B ′ type x : B ⊢ C ≡ C ′ A ≡ A ′ type A ↓ a = B b A ′ ↓ a ′ = B ′ b ′ B ≡ B ′ type a ≡ a ′ : B b ≡ b ′ : BA ≡ A ′ type A ↓ nat A ′ ↓ nat A ≡ A ′ type While the above stipulations endow type equality with anintensional nature (Dybjer, 2012), we often find an additionalcondition that determines a type uniquely by its terms a : A ⊢ a : A ′ a : A ′ ⊢ a : AA ≡ A ′ type Now, in order to give a full account of term equality weassume that we are given closed terms a and b as well as a type A , and evaluate them all, declaring that a and b are equal terms of type A provided that a and b evaluate to equal canonical terms of the canonical type which A evaluates to. A ↓ Q x : B C a ↓ λx.M a ′ ↓ λx.M ′ x : B ⊢ M ≡ M ′ : Ca ≡ a ′ : AA ↓ b = B b ′ a ↓ refl ( b ) a ′ ↓ refl ( b ′ ) b ≡ b ′ : Ba ≡ a ′ : AA ↓ nat a ↓ a ′ ↓ a ≡ a ′ : AA ↓ nat a ↓ succ ( M ) a ′ ↓ succ ( M ′ ) M ≡ M ′ : nat a ≡ a ′ : A The semantics given above can be inductively extended tojudgments occurring under a list of hypothesis, meaning thatthe validity of the hypothetical judgment x : A ⊢ f : B is subject to the validity of the non-hypothetical or categoricaljudgment f [ a/x ] : B for every closed term a : A . In sum,the only way a hypothetical judgment can be given meaningto is by determining what categorical judgment they result inwhen all their free variables are replaced with closed terms.Frege’s determination of the reference of functional expressionsin Grundgesetze §§ x : A, p : x = A x ⊢ p ≡ refl ( x ) : x = A x because, for a closed term a : A , a closed equality term b : a = A a will always evaluate to a canonical term of a = A a ,6which means that p ↓ refl ( x ). Since a = A a is a canonical type, c ≡ refl ( a ) : a = A a obtains. This principle is a straightforwardconsequence of the reflection rule x : A, y : A, p : x = A y ⊢ x ≡ y : A which can be easily validated using a similar line of reasoningand expresses the fact that, semantically, every propositionalequality is also a judgmental equality. It was Dummett (1978) who first noticed that Frege’s theoryof sense and reference could be rendered constructively byassuming that the sense of a singular term is related to itsreference as a program is related to its value, and that thethesis that the reference of a sentence is a truth value shouldbe entirely rejected, since, if we think of the sense of anexpression not as a mode of presentation of its reference butas an effective method for determining it, that would imply ameans of deciding whether a proposition is true or false. Although the idea that a sentence refers to a truth valuehas no place in a constructive setting, Martin-L¨of (2001) hasobserved that an interpretation of singular terms as terms andpropositions as types in the sense of the meaning explanationsallows for a more rigorous development of a theory of sense andreference for both singular terms and sentences. In this context,there is nothing better suited to mediate the passage from thesense to the reference of an expression than the evaluation ofa term to its canonical form. More concretely, the referenceof a term ( λx. refl ( x ))( ) is the canonical term it evaluates to, refl ( ). Programs are means for specifying their values and soare senses, since values are references. Expanding this idea tothe sphere of types, Martin-L¨of notes that that the reference of Moschovakis (1994) adds to this interpretation a mathematical notionof recursive algorithm that allows for a more rigorous theory of sense.Therefore, it can be seen as a forerunner of the theory of sense of referenceof Martin-L¨of (2001), which is grounded in type theory instead.
7a sentence has to be a canonical proposition, a view that fitsperfectly the meaning explanations, since an expression that isassigned to a type is interpreted as a sentence that expressesa proposition, and a type evaluates to a canonical type, whichis taken as their referent. For example, both ( λx.x )( nat ) and natrec ( , nat , λx.x ( x )) are equal types because they evaluate tothe canonical type nat . In other words, they are coreferentialtypes but specify their value in differently.If the passage from the sense to the reference of anexpression is given through evaluation, how should one interpretthe passage of an expression to its sense? I have suggestedelsewhere that an expression a comes to be known as a programwhen it is assigned to a type A . The reason for this lies inan analogy with computer programs: a piece of code is onlyrecognized as a program when it is correctly typechecked by acompiler of the programming language in question. If there isa single type mismatch error, such as an assignment of a valuebetween two variable of different types, then we did not wrotea program after all. However, if we assume that a term a hasa sense only when a : A we are at the same time excluding thepossibility that terms lacking a reference have a sense. This hasto do with the fact that a : A obtains only when a computes toa value according to our semantic stipulations, while not everyterm has a value. Consider a non-terminating term such as theinfamous Ω term ( λx.x ( x ))( λx.x ( x )) . Frege has always defended that in an exact science such asmathematics every expression must have a reference, but henever once entertained the idea that a sentence does not expressa sense if it lacks a reference. Even in our setting, it is desirableto allow expressions to have a sense when they do not have areference, for even a non-terminating term can be run just likeany other program. I shall therefore say that an expressionis recognizable as a program when it is interpretable by anoperational semantics in the style of the previous section. That See for instance Bentzen (2018, forthcoming). operational semantics evaluation
Martin-L¨of is careful to distinguish between two differentinterpretations of equality of reference, claiming that ajudgmental equality a ≡ b : A says of the senses of a and b that they are coreferential and a propositional equality a = A b ,when proven to be true, says of the references of the expressions‘ a ’ and ‘ b ’ that they are equal objects, because one is heresuppressing the equality term p that realizes the propositionalequality via the assertion of p : a = A b , the part of the judgmentthat he sees as not “referentially transparent”. Having saidthat, I cannot see the grounds for this distinction, which maybe described as confusing at best, considering that semanticallyevery propositional equality is a judgmental equality. SinceMartin-L¨of speaks of judgmental equality as an intensionalrelation, one may question if he is not simply leaving thesemantics aside and speaking of the formalism itself, since, togive an example, while it is easy to construct a term of type n + m = nat m + n via the principle of complete induction, thereis no derivation of the judgmental equality n + m ≡ m + n : nat in the theory, since both sides of the equality do not evaluate tothe same value, although they do for every particular instanceof n and m . Semantically, both propositional and judgmentalequalities say that the senses of a and b have the same reference,because they amount to the fact that the values of programs a b are one and the same.Martin-L¨of also claims that equality of sense is givenby a relation of synonymy for expressions, which includesin particular renaming of bound variables and definitionalstipulations such as π = C/d . I believe that this view ofsameness of sense is essentially correct, but it lacks a theoreticalmotivation in the setting of the meaning explanations. Thefollowing computational explanation is able to provide a morecompatible justification. Since to say that two expressions havethe same sense is to say that two terms express the sameprogram in type theory, we just have to justify an identitycriterion for programs. Clearly, it seems appropriate to identifyprograms that have the same computational content. Hardlyanyone would say that two programs are identical if one involvesmore computations than other, even though they may have thesame execution value. With that in mind, I propose that a and b express the same program if they have the exact samecomputational behavior, if to evaluate a is to evaluate b andvice-versa. It is simply natural to identify two terms up torenaming of bound variables and definition unfolding, giventhat they are evaluated in the same way. It only remains to be asked how this account of sameness ofsense can be linked to the equipollence principles that we havepreviously discussed. For the sake of argument, let us firstassume that A and B are sentences, which is to say that theyevaluate to canonical types. We have to show that A and B havethe same sense when anyone who accepts A will immediatelyaccept B and vice-versa. One interpretation of the immediacyin the latter condition is that proposed in Sundholm (1994) a : Aa : B and b : Bb : A that says that any construction that realizes A is also a realizerfor B and vice-versa. But that does not imply that A and B have the same computational behavior. Instead, it says that0they have the same reference, since from this coextensionalityit would follow that A ≡ B type . That may seem odd from aFregean perspective, but recall that sentences do not refer totruth values but to canonical types here.If we choose to interpret immediacy in a purely compu-tational way instead, then we have that anyone who accepts A will immediately accept B if, without any reference to thenotions of transition or evaluation, one is justified in an makingassertion of the truth of B from one’s assertion of the truth of A , or, equivalently, if from one’s construction of a term a : A one can construct a (possibly new) term b : B by means thatdo not involve the computation. It is plain that sameness ofsense is given by equipollence under this view, for this conditiondescribes precisely what it means for two types to have the samecomputational behavior. Finally, if we are given an open type x : A ⊢ P ( x ) type then the corresponding principle for terms follows directly fromthe equipollence of types, because to say that two terms a : A and b : A are equipollent is to explain how anyone who accepts P ( a ) will immediately accept P ( b ), which we already did.Notice that only terms of a same type can be equipollent, likeeverything else in type theory. It does not make sense to askwhether, say, Julius Caesar is equipollent to 0.If we were to interpret equality of reference for types ina strictly Fregean sense, that is, following the idea that twosentences are coreferential if they have the same truth value,then obviously the relation that we would be looking forwould be that of logical equivalence, which, type theoretically,translates to the existence of two functions f : A → B and g : B → A between the types A and B . But it is worthstressing that actual equality of reference does not follow fromlogical equivalence, for two types may imply each other withoutcomputing to the same canonical type. Put differently, if we We typically write A → B as an abbreviation for Π x : A B when the type B does not depend on x : A . U , whose terms are smaller types, then thefollowing principle of propositional extensionality( A → B ) → ( B → A ) → ( A = U B )would be false in the meaning explanations, just as it was in Begriffsschrift because A and B may have different contents. Ingeneral, we do not even have that g ( f ( a )) = A a and f ( g ( b )) = B b obtain, for every a : A and b : B . The only thing logicalequivalence tells us is that if we constructed a term of A weknow how to find another term of B and vice-versa. It comeswith no other guarantees, as Sundholm (1994) notes. Over the last few years, homotopy type theory (UFP, 2013) hasemerged as a new foundation for mathematics that unites typetheory and homotopy theory via a homotopical interpretationof type theory where a type A is viewed as a space, a term a : A as a point of the space A , an equality term p : a = A b as a pathfrom point a to point b in the space A and so on. One of the main ingredients that reinforces this inter-pretation is the univalence axiom, which offers a formaljustification for the common view among mathematicians thattwo mathematical objects are equal when they are isomorphic.Roughly, the univalence axiom implies that two types areidentical just in case they are equivalent in a technical sense. ua : A ≃ B → A = U B And this new sort of identification imposes a weakerunderstanding of propositional equality where in general aclosed term p : a = A b does not entail a ≡ b : A , becauseunivalence introduces a new canonical term ua ( e ) to the equalitytype that is not judgmentally equal to a reflexivity term,assuming that e is an equivalence. Considering that refl ( a ) For a philosophical introduction to homotopy type theory, see Ladymanand Presnell (2016) and Bentzen (2020a). a , univalenceensures the existence of non-trivial paths.Yet, this homotopical view of equality goes againstthe constructive theory of sense and reference that I haveexpounded, precisely because if propositional equality is toexpress coreferentiality, then the reference of a term cannotbe its value in homotopy type theory. In the presence of non-trivial paths, that is, if we are allowed to construct a closedterm p : a = A b such that p refl ( a ), then we have that a ↓ a ′ and b ↓ b ′ but a ′ b ′ , meaning that the interpretation ofpropositional equality as equality of values is lost. References
Bruno Bentzen.
Constructive mathematics and equality . PhDthesis, Sun Yat-sen University, 2018.Bruno Bentzen. Frege on Referentiality and JuliusCaesar in Grundgesetze Section 10.
Notre DameJournal of Formal Logic , 60(4):617–637, 2019. URL https://projecteuclid.org/euclid.ndjfl/1569830414 .Bruno Bentzen. What types should not be.
Philosophia Mathematica , 28(1):60–76, 2020a. URL https://doi.org/10.1093/philmat/nkz014 .Bruno Bentzen. Frege’s theory of types, 2020b. URL http://philsci-archive.pitt.edu/17367/ . Preprint.Bruno Bentzen. On different ways of beingequal.
Erkenntnis , forthcoming. URL https://doi.org/10.1007/s10670-020-00275-8 .Erret Bishop.
Foundations of Constructive Analysis . AcademicPress, New York, 1967.Alessandro Bandeira Duarte.
L´ogica e Aritm´etica na Filosofiada Matem´atica de Frege . PhD thesis, PUC-Rio, 2009.Michael A. Dummett. Frege’s distinction between sense andreference. In
Truth and Other Enigmas , pages 116–144.Harvard Univ. Press, Cambridge, 1978. Original workpublished in 1975.Peter Dybjer. Program testing and the meaning explanations ofintuitionistic type theory. In
Epistemology versus Ontology ,pages 215–241. Springer, 2012.3Gottlob Frege.
Begriffsschrift, eine der arithmetischennachgebildete Formelsprache des reinen Denkens . L. Nebert,1879.Gottlob Frege.
Die Grundlagen der Arithmetik: eine logischmathematische Untersuchung ¨uber den Begriff der Zahl . w.Koebner, 1884.Gottlob Frege.
Funktion und Begriff . Hermann Pohle, Jena,1891. Translated in Geach and Black (1980, pp.21–41).Gottlob Frege. ¨Uber Sinn und Bedeutung.
Zeitschriftf¨ur Philosophie und philosophische Kritik , 100:25–50, 1892.Translated in Geach and Black (1980, pp.56–78).Gottlob Frege. A brief survey of my logical doctrines.
H. Hermes, F. Kambartel, and Friedrich Kaulbach (edsand trans.). Frege: Posthumous Writings. Oxford: BasilBlackwell , 1979:197–202, 1906.Gottlob Frege.
Grundgesetze der Arithmetik, BegriffsschriftlichAbgeleitet (I/II) . Georg Olms, Hildesheim, 1962.Gottlob Frege.
Philosophical and Mathematical Correspon-dence . University of Chicago Press, Chicago, 1980.Translated by H. Kaal.P. Geach and M. Black, editors.
Translations from thePhilosophical Writings of Gottlob Frege . Blackwell, Oxford,3 edition, 1980.Arend Heyting.
Mathematische Grundlagenforschung, Intu-itionismus, Beweistheorie . Springer, Berlin, 1934.William A. Howard. The formulae-as-types notion ofconstruction. In J. P. Seldin and J. R. Hindley, editors,
Curry: Essays on Combinatory Logic, Lambda Calculus andFormalism , pages 479–490. Academic Press, London, 1980.Kevin C Klement. Grundgesetze and the sense/referencedistinction. In P. A. Ebert and M. Rossberg, editors,
Essayson Frege’s Grundgesetze . Oxford University Press, 2016.James Ladyman and Stuart Presnell. Does Homotopy TypeTheory Provide a Foundation for Mathematics.
The BritishJournal for the Philosophy of Science , 69(2):377–420, 2016.URL https://doi.org/10.1093/bjps/axw006 .4Per Martin-L¨of. Constructive mathematics and computerprogramming. In
Logic, methodology and philosophy ofscience, VI (Hannover, 1979) , volume 104 of
Stud. LogicFound. Math. , pages 153–175. North-Holland, Amsterdam,1982.Per Martin-L¨of. On the meanings of the logical constants andthe justifications of the logical laws. In
Atti degli incontridi logica matematica. Scuola di Specializzazione in LogicaMatematica , volume 2, pages 203–281. Universit´a di Siena,1985. Reprinted in
Nordic Journal of Philosophical Logic ,1(1), pp.11–60.Per Martin-L¨of. The sense/reference distinction in constructivesemantics. Paper read at Leiden (transc. by B. Jespersen),2001.Yiannis N. Moschovakis. Sense and denotation as algorithmand value. In J. V¨a¨an¨anen and J. Oikkonen, editors,
LogicColloquium ’90: Proceedings of the logic colloquium, Bristol ,volume 2 of
Lecture Notes in Logic , pages 210–249. Springer-Verlag, Berlin, 1994.Marco Ruffino. Wahrheitswerte als Gegenst¨ande und dieUnterscheidung zwischen Sinn und Bedeutung.
KritischesJahrbuch der Philosophie , 2:69–118, 1997.Peter Simons. Why Is There So Little Sense inGrundgesetze?
Mind , 101(404):753–766, 1992. URL https://doi.org/10.1093/mind/101.404.753 .G¨oran Sundholm. Proof-Theoretical Semanticsand Fregean Identity Criteria for Propositions.
The Monist , 77(3):294–314, 1994. URL https://doi.org/10.5840/monist199477315https://doi.org/10.5840/monist199477315