Issue #01. April 1995. Pp. 51-80.
Meaning Realism and the Rejection of Analyticity
Copyright (C) by SORITES and Manuel Liz
Meaning Realism and the Rejection of Analyticity
My aim in this paper is to argue that there are ways to maintain a non-holist meaning realism even though one does not accept any analytic/synthetic distinction (hereafter A/S distinction). In order to characterize with precision the kind of meaning realism that is going to be defended and the kind of analyticity that is going to be rejected, we will introduce in this section some helpful terms of art. They will be used through all our discussion. They are inspired in Boghossian (1993), but there are some important differences.
1.1.-- Minimal Meaning Realism and Meaning Irrealism.
Let us begin with a minimal characterization of meaning realism. Being minimal, this characterization will serve us to make clear what is entailed by different sorts of irrealisms with respect to meaning, and it will be also useful in order to define minimal realisms and irrealisms concerning semantical properties others than meaning.
Minimal Meaning Realism: It is constituted by the acceptance of two very simple theses, namely
1. the thesis that there exist in fact semantical properties such that particular cases of «to mean that ...» would refer to, and
2. the thesis that these meaning properties can be instantiated in our world.
Beyond these two simple but fundamental theses of minimal meaning realism, let us consider another related thesis:
3. the thesis that some of these meaning properties really are instantiated in our world.
It is clear the different force of these three theses. The third thesis is stronger than the second one, and the second thesis is stronger than the first one. The third thesis entails the second one, and the second thesis entails the first one, but converse relations of entailment would not be true. Now, in relation to the denial of each one of these three theses, we can define the following relevant positions:
Meaning Nihilism: It consists in the denial of the first thesis and, as a consequence, it implies the denial of both the second one and the third one.
Meaning Eliminativism: It consists in the denial of the second thesis and, as a consequence, it implies the denial of the third one too.
Meaning Absenteeism: It consists in the denial of the third thesis.
Although the third thesis does not properly belong to minimal meaning realism, it entails its two theses. Because of that, in order to maintain a minimal meaning realism it would be enough not to be a meaning absenteeist and to subscribe that third thesis. However, a rejection of the third thesis is compatible with the acceptance of the two theses of minimal meaning realism. One can be at the same time a minimal meaning realist and a meaning absenteeist. Really, the meaning realism we are characterizing really is minimal.
With the help of the new terms and concepts we have just introduced, we can now define meaning irrealism as follows:
Meaning Irrealism: It consists in being meaning nihilist or meaning eliminativist.
With respect to any supposed property other than meaning, we could also define a minimal realism, a nihilism, an eliminativism, an absenteeism, and an irrealism in a very similar way. Specially, that would be possible for other semantical properties like analyticity, synonymy, and so on.
In general, with respect to any supposed property X, we could define a Minimal X-Realism as constituted by the following two theses: 1) the thesis that there exists in fact a property refered by «X», and 2) the thesis that that property can be instantiated in our world. X-Nihilism would be the denial of the first thesis of that minimal X-realism, X-Eliminativism would be the denial of the second one, X-Absenteeism would be the denial of the thesis that property X really is instantiated in our world, and X-Irrealism would consist in being X-eliminativist or X-nihilist. To be X-nihilist would entail to be X-eliminativist, and to be X-eliminativist would entail to be X-absenteeist, but not the other way around. Let us follow saying that:
To maintain a Factualist Thesis about property X is to maintain the first thesis of a minimal X-realism.
To maintain a Non-Factualist Thesis about property X is to maintain an X-nihilism.
To maintain an Error Thesis about property X is to maintain an X-eliminativism.
To maintain a Non-Error Thesis about property X is to maintain the second thesis of a minimal X-realism.
We usually maintain a non-factualist thesis about supposed properties such as to be a squared circle or to be the last natural number. We usually assume that these expressions do not describe any property at all. On the other hand, even if we are factualists about properties such as to be able to go back in time or to have a temperature below absolute zero, we usually maintain an error thesis about them. We assume that these properties cannot be instantiated in our world. Finally, with respect to other properties such as to be 200 years old (for human beings), we usually are absenteeists. In fact, we accept that these properties exist and we guess that they have not been instantiated in our world, but we are neutral concerning whether they can be instantiated or not in it. Theses 1, 2, 3, and the rest of our definitions try to preserve these intuitions.
1.2.-- Absenteeism and Realism Without Determination.
With the help of the above mentioned concepts and distinctions, we have introduced the thesis of absenteeism. In contrast with Boghossian (1993)'s approach, that thesis will be very relevant for us in order to interprete the position of Quine (1951) and Putnam (1966). As it was indicated, there are important differences between meaning absenteeism and meaning irrealism. Now, these differences can be generalized saying that to be an X-absenteeist would not be the same thing as to be an X-irrealist. To be an X-irrealist necessarilly entails to be an X-absenteeist, but not the other way around. In order to be an X-realist it is enough not to be an X-absenteeist, but it is not enough to be an X-absenteeist in order to be an X-irrealist.
Related with any minimal X-realism, there would be another position that is worthy of attention. It could be characterized as maintaining some sort of X-realism without determination in the following sense:
X-Realism Without Determination: It is constituted by the acceptance of two thesis with respect to the supposed property X, namely
1. the tesis that there exists in fact a property refered by «X», and
2. the thesis that there is no cogent procedure to determine whether something has X or not.
We need to say something about the notion of cogent procedures of determination. Cogent procedures of determination would not be effective procedures. Cogent procedures of determination can be defeated. Cogent procedures of determination can be propoused and rejected, they can be orientated in a more or less empirical way, and they can be more or less accurate within certain limits. Unlike effective procedures, cogent procedures of determination sometimes can produce wrong results. But, no procedure of determination would count as a cogent procedure unless 1) it is assumed its truth conduciveness with respect to the problem in question and 2) it is assumed that that truth conduciveness can be explained as a matter of natural, conceptual, or conventional laws.
If there are cogent procedures to determine whether a property is or not instantiated, there are facts of the matter able to decide that question. Determinate properties would be properties for which there are facts of the matter concerning whether they are instantiated or not. «To have certain electrical charge», «to be made of wood», «to be in Spain», and «to be one of the members of the Spanish Parliament» are examples of descriptions that refer to determinate properties in that sense. There exist cogent procedures to determine whether something has them or not. They are truth conducive procedures, and its truth conduciveness can be explained with the help of natural, conceptual, or conventional laws. On the other hand, «to be the next winner in a horse race», «to be the more important scientific discovery in the history of humanity» or «to be events that occur simultaneously in time» would be examples of descriptions that refer to non-determinate properties. There is no cogent procedures to determine whether something has or not these properties.
A meaning realism without determination would explicitly accept the above first thesis of minimal meaning realism according to which there exist in fact semantical properties such that particular cases of «to mean that ...» would refer to. But, a meaning realist without determination rejects the existence of cogent procedures to determine whether something has or not any of these meaning properties. Such a meaning realism without determination would be a meaning realism without any way to determine whether meaning properties are or not instantiated. That very peculiar kind of meaning realism is important because it offers a possible way to make compatible some rejections of the A/S distinction with certain meaning realisms of a holist kind. In this sense, sometimes it has been sugested that the Quinean rejection of the A/S distinction would only entail a meaning irrealism concerning isolated statements, but not a meaning irrealism concerning something like the meaning of whole scientific theories of the world. One of the most crucial references for this interpretation is the following:
My present suggestion is that it is nonsense, and the root of much nonsense, to speak of a linguistic component and a factual component in the truth of any individual statement. Taken collectively, science has its double dependence upon language and experience; but this duality is not significantly traceable into the statements of science taken one by one.
The idea of defining a symbol in use was, as remarked, an advance over the impossible term-by-term empiricism of Locke and Hume. The statement, rather than the term, came with Frege to be recognized as the unit accountable to an empiricist critique. But what I am now urging is that even in taking the statements as unit we have drawn our grid too finely. The unit of empirical significance is the whole of science. (Quine, 1951)
By itself, the nonsense of distinguishing factual from linguistic components in the meaning of individual statements goes against the «Quinean» A/S distinction without entailing any meaning holism. In section 6, we will arrive to a position close to the first part of the above quote of Quine (first paragraph) trying not to be engaged in its second part (second paragraph). According to the holist interpretation of this passage there could be at least a meaning realism compatible with the «Quinean» rejection of the A/S distinction; namely, a meaning realism concerning the meaning of the whole of science. Even though it is not possible to determine whether something has or not that meaning, it must exist. That was, for instance, the main point of Acero (1993) in his commentaries to Boghossian (1993)'s arguments against the compatibility of meaning realism with the Quinean rejection of the A/S distinction.
However, the sort of meaning holism that is assumed in that interpretation only is a meaning realism without determination, and this is a very weak thesis. As we have said, it would make impossible to have cogent procedures for determination of meanings. And without being able to determine whether something has or not the property of having certain meaning, it is difficult to see how such a meaning realism could be engaged with the thesis that some meaning properties really are instantiated in our world. Anyway, it is also difficult to see how it could be engaged with the second thesis of a minimal meaning realism according to which meaning properties really can be instantiated in our world. It is plausible to argue that to maintain a non-error thesis about meaning requires to have cogent procedures to determine whether, with respect to meaning properties, the modal property of being-able-to-be-instantiated-in-our-world is or not instantiated itself in our world. And it is plausible to argue that if we have these cogent procedures, then we also have cogent procedures to determine whether something has or not those meaning properties. Because of that, it is difficult to imagine any minimal meaning realism not being a meaning realism with determination of meaning properties. Meaning realism without determination would not be a minimal meaning realism. The same would be true for whatever X-realism without determination. In general, it is plausible to argue that any minimal X-realism must be an X-realism with determination, and that no X-realism without determination would be a minimal X-realism.
Meaning realism without determination is not a meaning nihilism. As we have said, such a meaning realism accepts the first thesis of meaning realism. There is a difference between a meaning realism without determination and a meaning nihilism that denies the existence of meaning properties. However, that difference is a very tiny one. It only consists in the acceptance by the former, but not by the second, of the second order existential statement that there exist at least one property such that a particular case of «to mean that ...» would refer to. That would be the only difference. A meaning realist without determination even cannot have any cogent procedure to answer any particular case of the question «What «to mean that ...» means?».
Boghossian (1993) argues that one cannot be a minimal meaning realist rejecting at the same time the A/S distinction. Against that, we will defend in this paper the compatibilist view that it is possible to do both things. Certainly, we could assume an irreductible meaning holism maintaining this way a meaning realism without determination able to be compatible with certain rejections of the A/S distinction. However, as we have just said, that would be very weak. Also we could be meaning realists maintaining a mere absenteeism concerning the A/S distinction. But, mere absenteeism would not entail any irrealism about analyticity. The compatibilist view we want to defend involves both a minimal meaning realism and an irrealism about analyticity.
The structure of the paper is as follows. We begin in Section 2 noting that «analyticity» must be understood above all as a philosophical technical term, i.e., as a theoretical term introduced in order to explain certain phenomena. Section 3 offers a crude objection to the A/S distinction, an objection based on a direct and simple argument against the possibility of having an adequate definition of analyticity. Without such a definition, analyticity becomes a non-determinate property or, simply, a property that does not exist. The important thing is that our argument supports a non-factualist thesis, i.e., a nihilism, about analyticity that does not depend on any sort of meaning irrealism. After that, in section 4 we closely examine Boghossian (1993)'s argument against the compatibility of any minimal meaning realism with nihilism about analyticity. The general conclusion will be that even a minimal meaning realism that accepts that if the meanings of some statements are fixed then so too are their truth properties is compatible with a nihilism concerning the A/S distinction. Section 5 follows a different route. Apart from the reasons examined in preceding sections to be irrealists about analyticity, there would be also some normative reasons against it. The sort of normative reasons that together with an absenteeism one can find in Quine (1951) or Putnam (1966). As we have said, mere absenteeism by itself does not entail any irrealism about analyticity, but with the help of these normative reasons it does. The interpretation that with respect to analyticity we will offer of both authors also would be compatible with a minimal meaning realism. Finally, section 6 is about the semantical property that sentences like «all bachelor are unmarried» are supposed to have when we say that they are trivial cases of analyticity. (Note: for the sake of simplicity, we will only consider two truth values, truth and falsity, and we will not make any relevant distinction among sentences, statements, and propositions)
2.-- Analyticity as a Philosophical Technical Term
First of all, it must be noted that analyticity is not an univocal notion. The reason of that is very simple: «analyticity» is above all a philosophical technical term. Strictly, our theories about analyticity are not theories about it. They are theories about certain other phenomena, and analyticity is not among these phenomena but among the things that are intended to explain them.
Grice and Strawson (1956), and Putnam (1966), among others, held another opposite view. For them, the A/S distinction is a semantical phenomenon that does in fact exist, and the only real problem is about its nature. They are commited with a minimal realism about the A/S distinction maintaining that
where there is agreement on the use of the expressions involved with respect to an open class, there must necessarily be some kind of distinction present.
That is just the perspective that in this section I want to critizice. It seems to me radically misguided for several reasons. Firstly, what does «agreement on use» mean here? Surely, not only mere coincidence in the results of a classification (analytic statements versus non-analytic ones). Two or more classifications can lead to the same result, they can be extensionally the same, even if they are guided by quite different sets of criteria and theoretical commitments. «Agreement on use» requires something more. In our case, it would require agreement on some philosophical beliefs with respect to the A/S distinction itself. But, in this last sense, it is clear that there is no such agreement and that, therefore, the existence of a real A/S distinction can be questioned.
Really, there is some agreement. Although it is only a certain agreement on the target class of phenomena that could achieve an unified explanation through analyticity. But, very often it has happen in the history of science and philosophy that the error was just in thinking that a given class of phenomena were needed of an unified explanation. So, we must consider «analyticity» as a philosophical technical term, and we must not see any class of, let us call them, trivial cases of analyticity as proving anything about the existence of an A/S distinction well stablished in our languages.
2.1.-- «Non-Quinean» Notions of Analitycity.
The perspective we have adopted has very important consequences. Suppose, as Boghossian (1993) does, that you think of analyticity as being something like «truth by virtue of meanings». Then, surelly, specially if you wish to use that idea to explain where logical truth comes from, you will be led to the need of distinguishing two different concepts of analyticity. Using Boghossian's terminology, we can say that you will be in the need of distinguishing between «pure» analyticity and «impure» analyticity. Unlike impure analyticity, pure analyticity must have no dependence on logic. With respect to pure analyticity, facts about meaning must be sufficient for the truth, without any contribution from either empirical or logical facts. Only this way you could use analyticity to explain logical truth.
At this point, Boghossian maintains that the concept of pure analyticity only can make sense if there is some modality distinct from the logical that may be used to define the dependence of truth values on meaning that it aims to articulate. In the case of pure analyticity, Boghossian says, «by virtue of» must become some sort of metaphysical necessitation, or something like that. Really, if we work with the, let us call it «Quinean», notion of analyticity according to which
- a statement is analytic iff it is true by virtue of meanings and independently of facts,
the conclusion reached by Boghossian is compelling. As we haved said, in the case of pure analyticity, the «facts» in question must also include logical facts, and «by virtue of» must become some sort of irreducible metaphysical necessitation between meanings and truth values.
However, being «analyticity», as it is, a philosophical technical term, there are other ways to look at the phenomena that it intends to explain. The «Quinean» notion is not the only possible notion of analyticity. In relation to our actual languages, there are other very different notions of analyticity. And with respect to some of them we do not need the appeal to any sort of metaphysical necessitation between meanings and truth values. That would be so simply because there are different notions of analyticity that do not make any primary reference to things like «meaning» or «truth». Let us think, for instance, on these other notions of analyticity:
- A statement made in a language is analytic iff it is one which all speakers of that language accept and for which they cannot give any reason apart from the one consisting in the fact that they are speaking that language.
- A statement made in a language is analytic iff it is one which any speaker of that language can never give up without leaving to speak that language.
It is true that the above notions are not only semantical notions of analyticity. «To be a speaker of a language», «To accept a statement», «to give reasons», «to give up a statement», etc., have important pragmatical components. But, why must analyticity be only a semantical notion? These other notions of analyticity could be so general and powerful as the «Quinean» one can be. Moreover, the modal qualifications present in these definitions offer an alternative to the metaphysical necessitation that Boghossian is calling for.
I am not endorsing any of these, let us call them, «speaker-based» notions of analyticity. I only want to note that there are other «non-Quinean» ways to understand it. More, an analogous distinction between pure and impure analyticity could be drawn from these «speaker-based» notions, one that would not require any appeal to an irreducible methaphysical necessitation. Consider, for brevity, only the last one. Besides the lack of precision of the concepts here involved, we could define impure and pure analyticity with respect to it as follow:
- Impure analytic statements made in a language are the ones which a speaker can never give up unless the speaker gives up some of the logical statements of that language.
- Pure analytic statements made in a language are the ones which a speaker can never give up even though the speaker gives up all the logical statements of that language.
With respect to synthetic statements we would have the following:
- Synthetic statements made in a language could be defined as the ones which a speaker can ever give up without giving up any of the logical statements of that language.
Once the language were fixed, pure analyticity would be fixed too. And, once the logical statements of a language were fixed in a way or another, a distinction between analytic (both pure and impure) statements and synthetic ones would appear. The important thing is that, with these concepts at hand, we do not need any irreducible metaphysical necessitation that in the case of pure analyticity connects meaning with truth. It is true that we need some modal qualification of the possibilities and impossibilities mentioned in the above definitions. In fact, they are suppose to have some sort of pragmatical modal force. But we do not need to metaphysically connect meaning with truth because we do not have here any primary reference to meaning or truth.
What, then, about the concepts of analytic truth and logical truth? Simply the following. We could define analytic truth in a language as the class of all analytic statements (pure and impure) of that language and, given certain logic, we could define logical truth in a language as the class of all analytic statements of that language that are not pure analytic statements. From that point of view, pure analyticity could not be directly used to explain how logical truth ultimately comes from meaning. But, pure analyticity in the above sense still could be able to restrict the class of possible logical structures that are allowed for any given language. Both analytic truth and logical truth would be something derivated from the use of a language, not something derivated from meaning. Let us note that if we adopt that last perspective about analyticity, analytic truths in the language we are speaking would be unrevisable. Some of them in an absolute sense, and some of them in a sense relative to the logical structure imposed over the language. However, not every true statement would be analytic. Many true statements could be given up by the speaker of the language without any logical change being, in this way, synthetic. Notice also that we are not making any claim about the rationality of these revisions. All that is in the game is the pragmatical modal fact that nobody would speak certain languages if some very special statements that can be made in these languages are given up.
As we have said, we do not claim to endorse these alternative «speaker-based» notions of analyticity. We are only stressing the fact that in relation to our actual languages there is not only one way to look at the phenomena linked to the philosophical technical term «analyticity». Really, there are a lot of possible characterizations of analyticity others than the «Quinean» one.
3.-- A Nihilist Argument Against Analyticity
Now, we can ask «Why to accept one of these possible notions over all the other ones in order to adequately define analyticity?» «Why to accept, for instance, the `Quinean' notion of analyticity instead of some of the `speaker-based' notions?» «Which one, if any, of the multiple notions of analyticity could lead to the adequate definition of analyticity?» These and other similar questions would finally lead to another one that we are going to confront in this section, namely «What are the conditions that the adequate definition of analyticity ought to satisfy?» The analysis of some of these conditions will offer us a very direct argument against the A/S distinction, one that supporting a non-factualist thesis about analyticity will not depend on any meaning irrealism.
3.1.-- The Adequate Definition of Analyticity.
So, let us concentrate on what would be required by the adequate definition of analyticity. Any plausible candidate to define analyticity would have to adopt the following general form
(D) s is analytic iff B(s)
where «s» stands for any statement, and «B» refers to particular properties others than analyticity which must be haved by s. The adequate definition of analyticity also has to be a true statement able to cover trivial cases of supposed analyticities like
(t) «All bachelors are unmarried».
If some particular definition is the adequate definition of analyticity, then it must be a true statement such that statements like t are analytic statements in the sense defined by it. There is nothing odd up to this point.
The problem comes when we decide that a particular definition of analyticity is in fact the adequate one. The adequate definition of analyticity must be a true statement able to cover trivial cases of supposed analyticities like t. But, it must be not only that. It must also reflect some simple features haved by these trivial cases of analyticity. As we are going to see, the adequacy of a definition of analyticity in these conditions would entail the analyticity of the definition itself just in the defined sense. And, to put it in a nutshell, the problem is that it is very difficult, if not impossible, to obtain any statement of that kind.
In order to make clear what the problem is, let us call analyticity1 to the property of being analytic just in the intended sense offered by certain particular definition of analyticity, and let us call analyticity2 to the property of being analytic in the sense in which trivial statements like t are supposed to be. Now, let be Di any particular definition of analyticity. If Di is the adequate definition of analyticity, it would have to introduce analyticity1 through a true statement like
Thesis 1 (Di): s is analytic1 iff B(s).
Because the theoretical character of analyticity1, it is not necessary that to be analytic1 entails to be analytic2. But, if Di is the adequate definition of analyticity, to be analytic in a trivial sense must entail to be analytic in that defined sense. In other words, the following thesis would hold:
Thesis 2: IF s is analytic2, THEN s is analytic1.
As we have said, the adequacy of a definition of analyticity would also require to maintain some simple features haved by trivial cases of analyticity, i.e., by analyticity2. Particularly, we claim that it would require to accept at least the following three thesis:
Thesis 3: IF s is analytic1, s' is analytic1, and s iff s', THEN (s iff s') is analytic1.
Thesis 4: IF s is analytic1, THEN (s is analytic1) is analytic1.
Thesis 5: IF B(s) and s is analytic1 iff B(s), THEN B(s) is analytic1.
First thesis would consist in the clausure of analyticity1 under logical equivalence maintained among analytic1 statements. Second thesis says that to state that something is analytic1 is itself an analytic1 statement. Third thesis says that to state that something has the properties something has if and only if it is an analytic1 statement is itself an analytic1 statement.
Theses 3, 4, and 5 come from the field of analyticity2. We assume that an analogue of these three theses hold for analyticity2. Really, it is not easy to prove conclusively this point. Analyticity2 is a very fuzzy matter. However, it seems plausible to presume that these three theses reflect in fact important features haved by analyticity2. Let explain this. With respect to thesis 3, «s iff s»' would be a statement enough simple to guarantee that, being true, if it is not analytic in a trivial sense, it is because s or s' are not analytic in that sense. If «s iff s»' is true, then the trivial sense in which s and s' are analytic transmites that analyticity to «s iff s»' itself. Thesis 4 would hold no more than if a statement is analytic in a trivial sense, then to state that also must count as a trivial case of analyticity. It is important to note that thesis 4 is previous to any consideration concerning whether analyticity2 and, consequently, analyticity1 must be exclusively understood as semantical properties and not, for instance, as properties derived from the use of a language. In the field of analyticity2, it is quite unproblematic to assume that, as a matter of fact, if a language contains the predicate «is analytic», then that predicate ought to be also applied to statements saying themselves that something is analytic. In general, if it is trivial to say something, it must be also trivial to say that it is trivial.
The philosophical theory of analyticity can try to reject thesis 4, perhaps by means of a hierarchy of analyticities relativized to different levels of language. But, in that case, it would have to reject also thesis 2. Not every analytic2 statement made in a language would be analytic1. This is very important, because far from offering a better explanation of the phenomenon of analyticity such as it is supposed to be present in our languages, that revisionist move would suggest (against Grice, Strawson, and Putnam) that analyticity is a theoretical concept designed to explain certain other phenomena, and that perhaps these phenomena could be better explained without any appeal to analyticity. In other words, to accept both theses 2 and 4 is the best thing a theory of analyticity could do in order to achieve an adequate definition of analyticity.
To make clear thesis 5 in relation to analyticity2, we would need to distinguish between, on the one hand, to say that a statement is analytic in a trivial sense and, on the other hand, to say that it has the sort of properties that would make it just a trivial case of analyticity, whatever these properties may be. That distinction is not explicitly present in the context of the analytic2. But the important point is that if it were present in a way or another, then for any statement having those properties it would have to be a trivial case of analyticity to state that it has them. It would be difficult to understand how any statement could be analytic in a trivial sense because the having of some sort of properties without being analytic in that trivial sense to state that it has just these properties. In sum, the adequate definition of analyticity would need to assume these features maintaining the above theses 3, 4, and 5.
3.2.-- The Problem of the Analyticity of the Adequate Definition of Analyticity.
Now, let us offer an argument in order to show that, in these conditions, if Di is the adequate definition of analyticity, it would have to be itself analytic1.
1- Suppose any analytic2 statement like t.
2- t is analytic1. [from step 1 and thesis 2]
3- It is analytic1 to state that t is analytic1. [from step 2 and thesis 4]
4- B(t). [from step 2 and thesis 1]
5- B(t) is analytic1. [from step 4, thesis 1, and thesis 5]
6- IF it is analytic1 to state that t is analytic1, B(t) is analytic1, and t is analytic1 iff B(t), THEN it is analytic1 to state that t is analytic1 iff B(t). [an instance of thesis 3]
7- It is analytic1 to state that t is analytic1 iff B(t). [Modus Ponens from steps 3, 5, and 6]
8- In consequence, Di is analytic1. [universal quantification over step 7]
From the above argument, we finally obtain the following important thesis:
Thesis 6: If Di is a true statement making an equivalence between analyticity1 and some set of properties B, and such equivalence is able to cover analyticity2 preserving the features we have indicated through theses 2-5, then Di must be itself analytic1.
The adequate definition of analyticity would have to satisfy thesis 6. This is the final requirement. And it is a very important requirement because it entails a crucial problem if we like to accept the existence of a property that, in these conditions, can be called «analyticity». The problem is that it is not easy to offer any adequate definition able to satisfy thesis 6 and that, being «analyticity» a philosophical technical term, without any such definition we must consider the supposed property of analyticity as a non determinate property or, simply, as a property that does not exist.
Of course, neither a «Quinean» definition of analyticity nor any of the «speakers-based» definitions would be able to be themselves analytic1 in the sense expressed in thesis 6. It is plausible to argue that the only way to satisfy that requirement would be through a definition making equivalent analyticity1 and analyticity2, and such that the equivalence were itself analytic2. If analyticity were defined as the property of being a trivial case of analyticity, and that definition were itself a trivial case of analyticity, then the definition of analyticity would really be analytic in the sense defined by such definition. But, the analyticity that is required for the definition of analyticity itself cannot merely be a trivial one if to be analytic in some defined sense, i.e, to be analytic1, is a theoretical property depending on our assumptions and theories. While «analyticity» follows being a philosophical technical term refering to a concept so strongly dependent on our assumptions and theories, the definition of analyticity could not be analytic2.
As a matter of fact, statements like Di, unlike t, never are trivial cases of analyticity. Philosophers interested in analyticity try to define it just because the intended definition of analyticity is not a trivial case of analyticity. From this point of view, we could say, against Putnam (1966), that analytic statements cannot be only trivial ones. Being «analyticity» a philosophical technical term, if there are analytic statements at all, there must be some non-trivial cases of them. At least, there must be one non-trivial case of analyticity, namely, the definition of analytic itself.
So, with respect to any definition of analyticity guided by our philosophical assumptions and theories, we must think of it as being analytic just in the defined sense, i.e., as being analytic1, and we must think of it as not being a trivial case of analyticity, i.e., as not being analytic2. And the crucial problem is that we do not have any definition of analyticity able to satisfy both conditions. All of that would entail to be very skeptical about whatever notion of analyticity and to maintain a non-factualism, it is to be nihilist, with respect to the property of analyticity itself. Or, at least, it would entail to refuse analyticity as a determinate property. The lack of any adequate definition of analyticity beyond the trivial cases of supposed analyticities lead us to maintain a realism without determination about it or, simply, to maintain that there is no such a property. But, in spite of that skepticism about analyticity, we would not be committed with any skepticism about meaning. We can follow accepting the determination of meanings and to be minimal meaning realists.
4.-- Boghossian's Argument for the Incompatibility of Minimal Meaning Realism with Nihilism about Pure Analyticity
Boghossian (1993) displays a crucial argument for the incompatibility of minimal meaning realism with nihilism concerning pure analyticity. His argument can be easily generalized to affect any meaning realism able to accept, at least as a consequence, that if the meaning of a statement if fixed, then there is a fact of the matter as to whether the truth values of the statement are fixed too. Really, any minimal meaning realism with determination of meanings that also accepts truth properties as determinate properties would have to endorse that thesis. Even a minimal meaning realism that accepts a meaning absenteeism would have to endorse it.
Boghossian's argument is developed on the assumption of a «Quinean» notion of analyticity as «truth by virtue of meanings» considered as a determinate property. Boghossian tries to show that any minimal meaning realism would be incompatible with a non-factualist rejection of pure analyticity in the «Quinean» sense. Although the use of pure analyticity would be here dispensable, we prefer not to modify the original format of his argument in that respect. The argument in question is the following. According to Boghossian, the non-factualist rejection of pure analyticity would quite directly entail that
(1) for any statement, there are not facts of the matter as to whether the statement is true by virtue of its meaning.
And (1) would entails that
(2) there are not facts of the matter as to whether the statement is such that, if its meaning properties are fixed, then so too are its truth properties.
But, if meaning properties and truth properties are determinate properties, as we have assumed, there must be these last facts of the matter able to decide whether the truth values of the statement are or not fixed if its meaning properties are fixed. That is, our minimal meaning realism implies that (2) is false. And if our minimal meaning realism is correct, and (2) is false, then (1) must be false too. If we accept a minimal meaning realism, then we cannot accept the non-factualist rejection of pure analyticity. In any case, Boghossian concludes, both minimal meaning realism and the non-factualist rejection of pure analyticity, i.e., a nihilism about it, are incompatible.
It must be noted that Boghossian's argument is compatible with an absenteeism about analyticity. Even if there were facts of the matter as to whether a statement is true by virtue of its meaning, it could happen that, up to now, no statement were true by virtue of its meaning. The argument really has a strong «prima facie» plausibility. But, in spite of that «prima facie» plausibility, I think that it is possible to resist it. To put it in a nutshell, it is possible to resist Boghossian's argument because it could be false that (1) entails (2)! Let us put it in other words. «(1) entails (2)» is logically equivalent to «(not-2) entails (not-1)», and to say this is to say that
(V) IF (not-2) there are facts of the matter as to whether the statement is such that, if its meaning properties are fixed, then so too are its truth properties, THEN (not-1) there are facts of the matter as to whether the statement is true by virtue of its meaning.
Conditional (V) really has a very naive appearance. If «to be true/false by virtue of the meaning» were simply the same than «to be true/false if the meaning is fixed», then (V) would be a logically valid conditional. The realist acceptance of meaning and truth as determinate properties would logically imply the factualist acceptance of pure analyticity. That is the core idea of the argument.
But, the innocence of conditional (V) is only a superficial one. As we are going to argue, it is possible to break the entailment and to defend the possible falsity of (V) under some interpretation. The important point is that if (V) can be false in some coherent interpretation, then one could be a minimal meaning realist with determination even though one does not believes in analyticity.
4.1.-- A Short Study on «Facts of the Matter» and «by Virtue of».
To begin with, let consider the set of properties for which there are «facts of the matter» as to whether something have them or not. Conditional (V) has to do with the thesis of the closure of that set under the composition of properties through the relation «by virtue of». That closure implies the truth of the following general conditional (V*) from which (V) can be interpreted as a particular case:
(V*) (for arbitrary properties F and G) IF there are facts of the matter that decide whether x being F is G, THEN there are facts of the matter that decide whether x is G by virtue of being F.
But, it is clear to me that properties are not in general closed in that sense, and that conditional (V*) is not always true. Let be, for instance, the properties «to be a house» and «to be green». There are «facts of the matter» as to whether something being a house is green. But, it would be very odd to say that there are «facts of the matter» as to whether something is green by virtue of being a house. At first look, one is tempted to say that yes, that there are such `facts of the matter», and that these «facts of the matter» say us that nothing is green by virtue of being a house. But, that would be misleading.
In Lanzarote (one of the Canary Islands) all houses are green. It is traditional in that island to paint houses in green, even there is a law forbidding to paint houses in colours others than green. With respect to Lanzarote, it is possible to say that some things really are green by virtue of being houses. In Lanzarote, one could argue, there are social practices and legal rules able to give an unified sense to the claim that some buildings are green by virtue of being houses. These social practices and legal rules are able to establish one so special link L between the properties of being a house in Lanzarote and being green that the following would hold:
- There is a link L between to be a house in Lanzarote and to be green such that, for all x, (if Lx then (if x is a house in Lanzarote, then x is green)).
In other words, the property of being green by virtue of being a house in Lanzarote is a determinate one. There could be cogent procedures based on the social practices and legal rules that are able to stablish that link L. And according to these cogent procedures, one could decide whether in Lanzarote some things are green by virtue of being houses. In consequence, there are «facts of the matter» to decide whether something is or not green by virtue of being a house in Lanzarote.
It is very important to adequately distinguish the last thesis from the thesis consisting in that there are «facts of the matter» to decide whether houses in Lanzarote are green by virtue of the above mentioned social practices and legal rules. Of course, there would be also cogent procedures and «facts of the matter» to decide whether houses in Lanzarote are green by virtue of certain social practices and legal rules. In that case, these cogent procedures would be based on other more basic nomicities. What we are defending is just that both properties would be determinate ones and that this is the way we understand «by virtue of».
But if this is so, then the supposed «facts of the matter» by which nothing is green by virtue of being a house ought to include the relevant and important condition «not being in Lanzarote». And the problem would be that, by the same token, other conditions ought to be included. But, how to know them? There would be no way to determine them completely in an unified way, and «facts of the matter» that do not admit any clear determination are not «facts of the matter». In spite of appearances, it can be argued that there are not «facts of the matter» as to whether in general something is green by virtue of being a house. There are «facts of the matter» as to whether something is or not a house, there are are «facts of the matter» as to whether something is or not green, and there are «facts of the matter» as to whether something being a house is or not green. There are also «facts of the matter» as to whether in Lanzarote something is or not green by virtue of being a house. Of course, there can be «facts of the matter» as to whether in other times, places, or circumstances something is or not green by virtue of being a house. And there are «facts of the matter» to decide whether houses in Lanzarote are green by virtue of certain social practices and legal rules. But, it is possible to argue that in spite of so many «facts of the matter», there is no «facts of the matter» as to whether in general something is or not green by virtue of being a house. Conditional (V*) is not valid.
But, even if conditional (V*) is not necessarily true, conditional (V) has a very different scope and it could be always true. Is it the same with respect to properties like to-have-certain-meaning and to-have-certain-truth-value than with respect to properties like to be a house and to be green? We can try the same strategy with respect to the first pair of properties than with respect to the second one. In order to show that Boghossian argument fails, and that (1) does not entail (2), we only need to prove that (V) can be false under some coherent interpretation! All we need to do is to make a coherent interpretation of (V) according to which (V) is false!
Our minimal meaning realism accepts that, being truth properties and meaning properties determinate properties, there are always «facts of the matter» to decide whether the truth properties of a statement are fixed being fixed its meaning properties. Moreover, our minimal meaning realism even could accept that sometimes these «facts of the matter» are able to decide the truth for some statements! And the problem is: Why must that be enough in order to maintain a factualism about the property of analyticity considered as a determinate property, and some sort of A/S distinction? Factualism about analyticity considered as a determinate property would state that there are always «facts of the matter» able to decide whether a statement is true by virtue of its meaning. A non-error thesis about such analyticity would state that there can be at least one case in which these «facts of the matter» decide the truth. The non-error thesis would entail factualism. Now, our problem with factualism is that if it is possible that (not-2) does not entail (not-1), then a door closed for analyticity is opened for meaning realism.
Beside «facts of the matter» able to decide whether a statement is such that, if its meaning properties are fixed, then so too are its truth properties, that is (not-2), what factualism about analyticity, that is (not-1), would need are «facts of the matter» that decide whether the statement has or not the property of «being true by virtue of its meaning». But, to say this would be to say that factualism about analyticity needs «facts of the matter» to decide whether, having the statement the determinate meaning it has, it bears or not the special link (let us call it SL) between its meaning and its truth value that «by virtue of» intends to refers to.
It would be a very especial link because it would not be merely the link (let us call it ML) that there could be between meaning and truth values just when the truth of a statement is fixed if its meaning is fixed too. We could say that what «facts of the matter» in (not-2) try to detect is this last link, that is a ML, whereas what «facts of the matter» in (not-1) try to detect is the first one, that is a SL. But, more precisely, what is the difference between a SL and a ML? Let us speak in general about links L among meanings M and truth values V of statements s. Now, (not-2) can be simply interpreted as follows:
(not-2') For all statement s, there is a link L between its meaning M and its truth value V such that (if Ls then (if Ms is fixed, then too is Vs)).
On the other hand, it would be possible to consider «Vs by virtue of Ms» as a mere paraphrase of «if Ms is fixed, then too is Vs». In this way, we could interprete also (not-1) as saying that
(not-1') For all statement s, there is a link L between its meaning M and its truth value V such that (if Ls then (if Ms is fixed, then too is Vs)).
In that case, (not-2') would really entail (not-1'), and hence above conditional (V) would hold. However, as in the case of things being green by virtue of being houses, there is another possible, and perhaps more demanding, interpretation of the expression «by virtue of» that appears without any restriction in (not-1). According to that point of view, (not-1) ought to be interpreted not as (not-1') but as
(not-1») There is a link L between the meaning M of a statement and its truth value V such that, for all statement s, (if Ls then (if Ms is fixed, then too is Vs)).
Here, «Vs by virtue of Ms» is not simply a paraphrase of «if Ms is fixed, then too is Vs». As it is showed in the order of quantifiers, «by virtue of» would require the existence of an unified and very strong link between meanings M and truth values V. Now, it is clear the crucial difference between a SL and a ML. The existence of a SL would imply the existence of a ML, but the existence of a ML would not imply the existence of a SL. In other words, just because «There is a link L such that for all statement s ( ... )» entails, but it is not entailed by, «For all statement s there is a link L such that ( ... )», the minimal meaning realism from which (not-2) is a consequence would not entail any factualist thesis about analyticity considered as a determinate property, that is (not-1), when such factualist thesis is interpretated as (not-1»).
If there is the link mentioned in (not-1»), really it would be a very especial link. That link would be the property a statement has when its meaning and its truth value are so especially related that the statement has the truth value it has «by virtue of» having that meaning. And this property is not simply the property a statement has when its truth value is fixed once its meaning too is. Lots of statements can have that second property, in a limit case one different property by each statement, without having the first one.
In other words, there is a point in which the links among meanings and truth values could be so heterogeneous that we cannot give any determinate sense to the expression «to be true/false by virtue of the meaning alone». Faced with this situation, we would have to deal with two main problems. One of them would be the kind of existence that something like a SL could have. Could it be, for instance, a merely disjunctive existence? Really, I do not know. Anyway, the second problem is more important. If that supposed SL exists at all, there cannot be any «facts of the matter» able to help us to detect it. As in the case of things being green by virtue of been houses, the mere links of (not-2'), or (not-1'), would not be able to do the work because we would need to determine all of them. And there are too many. Simply, the supposed property of being such SL is a non-determinate one. In the case it exist as a very complex disjunctive property, it cannot be but a non-determinate property.
In consequence, there is a coherent interpretation of conditional (V) according to which (V) could be false. Therefore, some sort of minimal meaning realism with determination of meanings in combination with considering truth values also as determinate properties could be true without being true that, for any statement, there are «facts of the matter» as to whether it is true by virtue of its meaning. Moreover, the truth values of some statements could be fixed if their meanings are fixed too. In the limit case, there could be so many different routes between meanings and truth values as different statements of that kind can be made in a language. So, a minimal meaning realist could say, for instance, that having «bachelor» and «unmarried» the meanings they have, «all bachelors are unmarried» is a true statement. A minimal meaning realist could say that without being married with analyticity. Given the way the truth is fixed in the case of «all bachelors are unmarried» once its meaning is fixed, that truth even could have a very strong modal force. But, this is another story.
5.-- Normative Rejections of the A/S Distinction
We have examined some problems concerning both the definition of analyticity and the appeal to the general notion of «truth by virtue of the meaning» in order to reject the compatibility of minimal meaning realism with nihilism about analyticity. With respect to the first topic, I have maintained that, being «analyticity», as it is, a philosophical technical term, the adequate definition of analyticity would have to be itself analytic not being trivial (more preciselly, that it would have to be analytic1 without being analytic2), and that we do not have any idea about which definition of analyticity could have these characteristics. With respect to the second topic, I have suggested an interpretation of «truth by virtue of the meaning» according to which one could be a minimal meaning realist with determination of meanings and truth values without believing in analyticity. That interpretation of «truth by virtue of the meaning» would call for a so special property of statements, a special link SL between their meanings and their truth values, that it is not easy to imagine how we could appeal to such a property, if it exists at all, in relation to our actual languages.
Nevertheless, even if we cannot imagine how both an adequate definition of analyticity and a SL between meanings and truth values can be possible in relation to our actual languages, that does not mean that we cannot imagine other situations in which these things were available. The following is a propousal to imagine one such situation. After showing that in such a situation it would be possible to maintain a minimal realism about analyticity, we will argue that there are important normative reasons against trying to transform that imagined situation in something real. In other words, what we will argue in this section is that, even if it were conceivable (in a wide sense of «conceivable») some sort of minimal realism concerning analyticity, there would be also normative reasons to maintain an irrealism about it.
5.1.-- Imagining Analyticity Step by Step.
STEP 1: In section 3, we have maintained that the adequate definition of analyticity would have to admit theses 1-5 and, therefore, to be itself analytic in the defined sense, i.e., it would have to be analytic1. Moreover, if the definition of analyticity depends on our assumptions and theories, it would have to be analytic1 without being analytic2. The nihilism defended in that section was based on the difficulty to satisfy that claim. However, as it was indicated there, we could imagine an adequate definition of analyticity according to which analyticity1 and analyticity2 were equivalent and that equivalence were itself analytic2. That would require for analyticity to fail to be a philosophically loaded concept and to be defined in a trivial analytic way. Now, we can try to construe a definition of analyticity able to satisfy these conditions. Let be, for instance, this explicit definition of analitycity:
(EDA) An analytic statement =Def One in which it is stated 1) an explicit definition or 2) a logical consequence of explicit definitions or 3) the analyticity of an explicit definition or 4) the analyticity of an analytic statement.
Again, if it were required, we could distinguish out of EDA some kind of pure analyticity from an impure analyticity, and so on. The important thing is that EDA itself is an explicit definition and that, therefore, EDA is analytic1. EDA would be able to meet also theses 1-5 of section 3.
STEP 2: Could EDA be adopted itself as a trivial case of analyticity? Well, the trivial character of a statement is only a psychological/epistemic question relative to a subject or group. And it can change if that subject or group changes. What is non trivial for a subject or group at some time can be trivial for other subject or group at the same time, or it can become trivial for the same subject or group at another time. According to that, EDA could be adopted by some subject or group as a trivial case of analyticity, even it could become trivial for ourselves.
STEP 3: With respect to the truth of EDA, we can assume that explicit definitions are always true. Assuming also that to be an explicit definition is part of the meaning of some statements, there would always be some «facts of the matter» as to whether the truth values are fixed once meanings are fixed too. In the case of explicit definitions, that assumption would fix the truth once the meaning is fixed and we notice that we are faced with an explicit definition.
STEP 4: Now, if we want to use the notion of «truth by virtue of the meaning», the only problem would be the one coming from the difference between mere links and a special link. But, this would be only a problem refered to the way things are made. The route from mere links ML to a special link SL only depends on the kinds of links there are between meanings and truth values. The point is that we could imagine a language used in such a way that the links determining the truth values once the meanings are fixed are always of a kind easily interpretable as the extension of a single natural property. This could be so out of our decisions and conventions or by the force, let us say, of the nature of things. Moreover, we could imagine a set of possible worlds in which, given certain decisions and conventions, or given the properties and relations present in these worlds, all actual and possible languages are of that kind. Being in any of the worlds of that set, we could have a special link SL, and not only mere links ML. The existence of such a SL would have even some sort of modal force with respect to that set of worlds. It would be in some way necessary. Really, we would have something like a «caeteris paribus» analyticity restricted to that set of worlds, and that set of worlds could be extremely broad.
5.2.-- Some Classical Normative Reasons Against Analyticity.
We have been arguing that is not necessary the existence of the supposed property called «analyticity». According to our analyses of both the adequacy of the definition of analyticity and the property that could support it in a determinate way when the notion of «truth by virtue of the meaning» is introduced, analyticity would not be something necessary. But, in the situation we have just described, we would really have some sort of «caeteris paribus» analyticity with all the features required. Now, the important question is: Why not to have that specific «caeteris paribus» analyticity?
I think that there is a negative answer to that question, a negative answer based on normative reasons. But, before to see these reasons, let us consider another related question also with a negative answer based on normative reasons. The question is: Why not to have in general analytic statements?
The classical normative reasons against analyticity stress that it would not benefit the progress of knowledge, scientific progress in particular. Furthermore, in some cases analyticity would paralize knowledge and scientific development. In Quine (1951), for instance, there is a very important normative component in his rejection of the existence of analytic statements others than explicit definitions. It is not only a question of fact, but a normative question. Quine not only argues that there are not, in fact, analytic statements others than explicit definitions. He thinks that there ought not to be any analytic statement but the ones being explicit definitions. Quine maintains a certain absenteeism with respect to analyticity, but it is an absenteeism embeded in certain normative thesis. At this point, it is important to realize that he does not maintain any error thesis about analyticity involving the strong modal notion of necessity. He does not maintain that analyticity cannot be instantiated. And he does not maintain either, at least in the context of Quine (1951), any non-factualist thesis nor a factualist thesis with respect to it. Instead, he argue for some Normative Thesis like:
(NT) If there exists a property expressed by «is analytic», then it has never been instantiated by statements not being explicitly definitional, and that is what ought to happen because that property ought not to be instantiated by statements not being explicitly definitional (consequently, all tokens of the statement «s is analytic», where s is not an explicit definition, have been false up to now; and all tokens of the statement «s is analytic», where s is not an explicit definition, always ought to be false).
NT does not suppose the existence of the property of analyticity, and it does not suppose either that that property is necessarily uninstantiated. Really, NT is more than a thesis about the falsity up to now of all tokens of the statement «s is analytic». It involves also a modal notion. But, it involves only a normative one. NT says that the property of analyticity has never been, and ought not to be, instantiated.
The situation of absenteeism with respect to analyticity is such that we do not know which one of the following exclusive theses is true:
a) Analyticity has never been instantiated and ought not to be instantiated, but it could be instantiated and it is a genuine property.
b) Analyticity has never been instantiated, it ought not to be instantiated, and it could not be instantiated, but it is a genuine property.
c) Analyticity has never been instantiated, it ought not to be instantiated, and it could not be instantiated; moreover, it is not a genuine property.
Theses a), b), and c) really have much more content than NT. But, I cannot find in Quine (1951) any compromise with any of them. All that is in Quine (1951) is NT. Because of that, his position does not entail any error thesis and it does not entail any non-error thesis either, and it is compatible with both a non-factualist thesis and a factualist thesis about analyticity. With respect to the analyticity not introduced by means of explicit definitions, I think that we must interpret him simply as being neutral about these things. In other words, Quine (1951) is not a minimal realist concerning analyticity, but he is not an irrealist either. He simply is an absenteeist that endorses a normative thesis like NT.
It seems to me that Putnam also could be interpreted as maintaining a similar view. Putnam (1966) maintained that outside the field of explicit estipulations, both in formal and natural languages, if «to be analytic» really refers to a genuine property, that property has never been instantiated and it ought not to be instantiated. Up to a certain extent, both Quine and Putnam think of analyticity as entailing unrevisability, and their worries are about the unrevisable character that analytic statements must have. It would block our knowledge, our scientific knowledge in particular, to declare as analytic, and therefore as immune from revision, any statement other than explicit definitions. Only truths estipulated by means of explicit definitions ought to be analytic and immune from revision.
Now, let us go back to our previous question «Why not to have in general analytic statements?». Analitical statements, we can read in Putnam (1966), could provide the advantage of brevity, intelligibility, capability of prediction of some linguistic uses, and so on. But, explicit definitions in formal and natural languages can do all that important economic work. What is more important, once it is assumed that analytic statements must be unrevisable, no other kind of analyticity would have these benefits without the danger of being an obstacle to the progress of knowledge.
We cannot confuse the thesis that analyticity entails unrevisability with the converse thesis. Actually, unrevisability does not entail analyticity. We could have, for instance, some sort of «a-priory» knowledge able to be unrevisable without being based on analyticity. More, all true statements ought to be considered unrevisable. And we cannot confuse either unrevisability with our knowledge of it. We can always be wrong about the unrevisable character of any statement. But, even accepting these things, analyticity would entail unrevisability. Hence, no revisable statement could count as analytic.
Nevertheless, the relevant point concerning analyticity and revisability lays in other place. In order to show that there is no analytic statement other than explicit definitions because all other supposed analytic statements could become revisable, it is necessary that the revisable character of these supposed analytic statements cannot be merelly interpretable as a change of the meanings involved. That it is always plausible such a «change-of-the-meaning» interpretation was one of the main theses of Grice and Strawson (1956) against Quine (1951). On the other side, Putnam (1966) was one of the main opponent to that «change-of-the-meaning» thesis. It would be convenient to have a look at the arguments. Roughly, the argument for the revisable character of any supposed analytic statement, other than explicit definitions, was as follows. Suppose that
(1) «All things being A are B»
is considered an analytic statement not introduced by means of an explicit definition. Now, suppose that we become to have good reasons to maintain that
(2.1) «All things being C are A, and only things being A are C», and
(2.2) «Some things being C are not B»
are true statements. Suppose also that, in a heuristic and theoretical sense, being C becomes more relevant that to be B. In that situation, the argument says, the good epistemic policy would be to reconsider the supposed analytic character of (1) and to modify its truth value. So, we could say that (2.1) and (2.2) in that situation suggest that
(3) «There are things that are A without being B»,
and that, in consequence, (1) cannot be a true analytic statement.
Against that argument, the «change-of-the-meaning» interpretation would reply that the revisability and, hence, the rejection of the analytic character of (1) is only apparent. The predicate «being A» appearing in (1) has not the same meaning that it has in (3). To admit (2.1) and (2.2) does not entail the falsity of (1), but a change of meaning in the predicate «being A». If (1) really is a true analytic statement, (3) cannot refute it. (1) and (3) would be simply talking about different things.
Curiously, that «change-of-the-meaning» interpretation of revisability favours conventionalism. Its defense of analyticity leads to conventionalist views of knowledge and science. In our case, the consequence would be that the choice between (1) and (3) could not be guided by our beliefs about «being A». No improvement in our beliefs about «being A» would entail to be in a better epistemic position in order to decide between (1) and (3). It is only a question of choice of a particular meaning (or language, or conceptual scheme) instead of another one.
The move of Putnam (1966) in order to avoid that «change-of-the-meaning» interpretation is worth of attention. Putnam maintains that when a statement like (1) is not introduced by means of an explicit definition, as it is by assumption, the predicate «being A» is always, as a matter of fact, a law-cluster concept. The meaning of law-cluster concepts is constitued by a cluster of laws. Supposed analytic statements not introduced by means of explicit definitions are no more than one of these laws in the meanings of the law-cluster concepts involved. A statement like (1) itself would be one of these laws for the meaning of «being A». The important point is that any of these laws can be abandoned without destroying the identity of the law-cluster concept. Just in the same way, says Putnam, as a man can be irrational from birth, or can have a growth of feathers all over his body, without ceasing to be a man. So, the meaning of «being A» would not have changed enough from (1) to (3) to affect what we are talking about. (1) simply is one of the various laws that constitute the law-cluster concept «being A», and we know that any of these laws can be abandoned preserving «being A» the meaning it has.
5.3.-- Why not to Adop EDA and Give Reality to the Situation in Which EDA Is an Adequate Definition of Analyticity?
We have remembered a very important story about why not to have analyticity in general outside the field of the analyticity introduced by means of explicit definitions. Now, it is time to come back to the problem of why not to have even the kind of specific analitycity offered by explicit definitions. In other words, why not to adopt EDA and give reality to the situation described through STEPS 1-4? Why not to transform our imagined situation in something real, achieving the economical benefits of analyticity?
Let put aside the problem of going from mere links to a special link. Any definition of analyticity that wants to use the notion of «truth by virtue of the meaning» will have to deal with that problem. Really, it is a crucial problem, but it only depends on the way things are made, or can be made. To my view, it is very unplausible that it can exist a special link out of those mere links, even if we do our best efforts to get it (see again STEP 4). If I am right, even if we are not nihilists concerning the property of analyticity itself, and even if we accept that there exists a genuine property such that «analyticity» refers to, we would have to be eliminativists. But this is another question. The point I want to emphatize now is that, besides the problem posed by that special link, there would be normative reasons against the adoption of EDA as a trivial case of analyticity, and therefore as an adequate definition of analyticity. These reasons are related with the story we are just remembered, and with the robust character of the meaning of law-cluster concepts.
In short, the predicate «to be analytic» that we find in EDA also ought to be considered as refering to a law-cluster concept, to a concept with a meaning that is not exclusively determined by EDA itself. Thus, we would have to refuse that EDA ought to be really understood as an explicit definition, with the consequence that EDA would not be analytic1 because it does not satisfy the definition of analyticity proposed.
The answer to the question «Why not to have an explicit definition of «to be analytic» such as the one offered by EDA in the situation above described?» would be, therefore, the same than for Quine and Putnam was the answer to the question «Why not to have analyticity in general?». We ought to refuse the situation in which EDA could be an adequate definition of analyticity because that situation would block our knowledge. To consider that analyticity is not a law-cluster concept and that we can explicitly define «to be analytic» would block our knowledge of what the supposed property called «analyticity» can be. Curiously, it could even be an obstacle to the discovery that in fact there is no such a property.
In consequence, with independence of the problem whether there can be or not special links between meanings and truth values, and not only mere links, we ought not to be in the situation described by STEPS 1-4. Apart from the reasons examined in preceding sections in order to be irrealist about analyticity, there would also be normative reasons against it.
6.-- Trivial Cases of Analyticity and How to Interpret Them
We have been maintaining a nihilism concerning analyticity. Also, we have argued that our nihilism in compatible with a minimal meaning realism, even with one that accepts that sometimes truth values are fixed once meanings are. The existence of a special link among meanings and truth values is very doubtful. We have also noted that even if it is possible to imagine situations in which an adequate definition of analyticity is possible, there would be normative reasons to resist being in them. Furthermore, even if we were in such situations, we would follow having problems with respect to the possible existence in our languages of a special link able to suport a notion of analyticity as «truth by virtue of the meaning». So, even if we fall to be nihilists about analyticity, we would have to be eliminativists.
To put it in a nutshell, for someone who loves analyticity, the only available options are to consider analyticity as a non-determinate property, maintaining this way a realism without determination about it, or simply to be absenteeist. And the important thing is that none of these options supports any minimal realism with respect to it.
However, we cannot forget that there are trivial cases of supposed analyticities. It is time to say something about them. They are the linguistic phenomena over which theories about analyticity are proposed, and our skepticism about analyticity has to adopt a position about them. «All bachelors are unmarried», for instance, is very often adopted as one of these trivial cases of supposed analyticity in our natural languages. Really, lexical definitions and explicit definitions made in scientific and not scientific contexts offer us lots of cases of such supposed analyticities.
We cannot explicitly define analyticity saying that analytic statements are no more that explicit definitions (and logical consequences of explicit definitions, and so on). We have just argued that that kind of definition would not be an adequate one. But, this would be compatible with the fact that some statements with the structure of explicit definitions could be, in some sense, analytic. At least, they could be analytic just in the sense of being analytic2. Really, if there is analyticity at all, statements with the structure of explicit definitions are the most plausible candidate for analyticity.
The problem is that there are very few, if any, pure lexical or explicit definitions able to institute an analyticity that cannot be revisable. That is, definitions where the definiendum is not a law-cluster concept. There are few, if any, cases of trivial analyticities that cannot be defeated. So, the same problem we have seen with respect to the explicit definition of analyticity emerges with respect to an overwhelming majority of cases of supposed pure lexical or explicit definitions that have the appearance of trivial cases of analyticity.
6.1.-- «All Bachelors Are Unmarried».
Take, for instance, the statement «all bachelors are unmarried». Putnam (1966) maintained that some statements of our natural languages, statements like that one, really are analytic. Among the reasons to maintain that there was a crucial one: «bachelor» is not a law-cluster concept. It is not a law-cluster concept, says Putnam (1966:59), because there are not, and there will not be, exceptionless laws containing the term «bachelor». So, we could consider «all bachelors are unmarried» as very close to an explicit definition of «bachelor» and, therefore, as a trivial case of analyticity.
Putnam would be right if we understand «law» only in some narrow sense. But that narrow sense in no way is the only sense that could be relevant here. That «bachelor» is not, nor will be, a law-cluster concept is true only if the class of laws we are thinking about does not contain nomicities like legal laws, social rules, and so on. Certainly, these nomicities are not physical or natural ones, but they can be so exceptionless as natural laws are, and they are a really very important part of the way we understand terms like «bachelor».
Suppose, for instance, a society in which there is a fundamental, even exceptionless, legal law saying that bachelors and only bachelors are exempt from pay certain marriage-tax that is obligatory for married people. The law is very basic or fundamental in the sense that it has achived a great importance in that society, and it maintains strong relationhips with a lot of other legal laws, social rules, etc. Suppose, as a matter of fact, that in such a society marriage has gradually adopted a great plurality of, civil and religious, forms so that at the present time it is not easy, for instance, to distinguishing married people from unmarried people that live together. There are also marriage ceremonies which are absolutely private, and so on. In general, it becomes very dificult to tell married from unmarried people. More, some people being in fact married try to keep out of sight their condition of being married in order to not paying the tax. Suppose that it is possible to do that in a lot of ways, so that not to pay the marriage-tax becomes the more relevant criterium to be bachelor, perhaps the only really operative criterium. Faced with this situation, we have two known options:
1- To say that some bachelors according to the law are not really bachelors or, in other words, that there is a change of meaning in the term «bachelor».
2- To say that some bachelors are in fact married.
With respect to the «change-of-the-meaning» option, it must be noted that not to pay the marriage-tax is a really important element of the concept of being a bachelor, an element that has not changed through the changes carried in the ways of being married. Because of that, it is plaussible to argue that «bachelor» has not changed its meaning. If not to pay the marriage-tax has become the more relevant criterium to be a bachelor and that criterium was always present in the society, then first option has not much sense. Second option is the more plausible one. But, second option entails that the statement «all bachelors are unmarried» would be false in that situation and, therefore, that it cannot be considered as an analytic statement.
The moral is that even statements like «all bachelors are unmarried» have a kind of analyticity2, i.e., a kind of trivial analyticity, that could be revised. The importance of all those legal laws, social rules, and so on, is here decisive in order to reconsider that certain concepts really are law-cluster ones, and that every one of the supposed analytic2 truths in wich they appear really could be given up.
But, this is not all. With relative independence of the above argument, there is another way to see meaning that also goes against the supposed trivial analyticity of «all bachelors are unmarried». If we consider seriously the fact that there can be concepts with a meaning determined by certain prototypes and some associated similarity conditions, it would be necessary to modify, or enlarge, our cluster analysis so that some concepts are not, or not only, law-cluster ones, but also prototype-cluster concepts. It is plausible to think that, for instance, our concepts of «chair», «window», «book», «mother», etc., are prototype-cluster concepts. And, perhaps, concepts like «bachelor» also really are prototype-cluster ones. Now, the important point is that even if we say of some of the prototypes Pi of a prototype-cluster concept A that «Pi is A», that statement could be false. It would be false if that prototype Pi comes to fail being a prototype for the concept A. For prototype-cluster concepts, particular prototypes could change without a change of meaning. If A is a prototype-cluster concept, then even if Pi is one of these prototypes it cannot be analytic the statement «Pi is A».
There is an important consequence of the above remarks. If most of our concepts are law-cluster or prototype-cluster ones, theories of analyticity of the sort provided by Kats (1972), based on the classical «kantian» idea that analyticity consists in some kind of redundant predication, would not work either. When law-cluster or prototype-cluster concepts are involved, redundant predication could be always false.
Among the multiple cases of supposed analyticities, the cases in which new symbols are introduced in a language through some intended explicit definitions occupy an important place. This happens very often in scientific and legal contexts. It is a common place, for instance, in the formal languages of mathematics and logic to make and use explicit definitions. The problem is «Are they really pure explicit definition?» «Are, for instance, the usual explicit definitions of logic pure explicit definition?» I think that these questions lack any definite answer. Not because, and this is here the crucial point, we do not have the relevant knowledges to decide these questions, but because there are not cogent procedures to determine in general whether something is or not a pure explicit definition. It depends on how each particular mathematical or logical simbol is related with mathematical and logical concepts, and on how these concepts get their meanings. To be a pure explicit definition is not a determinate property. That problem is even more evident when we go from mathematics or logic to the above mentioned legal contexts. Are the explicit definitions that one can find in legal codes really pure explicit definitions? There is no definite answer. In any case, the field of pure explicit definitions would be very narrow. Moreover, it is important to emphasize that even if there are cases of pure explicit definitions, and even if some statements really have the sort of analyticity that they are supposed to have when we say that they are analytic2, we would have only mere links between their meaning and their truth values, not a special link. The supposed analyticity of the trivial cases of analyticity is only a trivial analyticity. It is not the kind of analyticity that philosophers were looking for. It consist only in the fact that, for some statements, their truth is fixed once their meanings are.
6.2.-- Against Linguistic Arbitrariness.
Language is always conceptually motivated and engaged with reality. There are very few cases, if any, in which we come to use a new term only as the result of a pure explicit definition, without any other conceptual or factual contribution but the one that already is present in the old terms. There are always conceptual reasons to use the words, sentences, and languages in the way we use them. These conceptual reasons have to do with the rest of our beliefs and knowledges about the language we are speaking and about the world. Also, there are always externalist components of meaning that make very difficult to explain how it can exist a property exhibiting the classical features attributed to analyticity (for instance, unrevisability), and how we could know and detect analyticity through a knowledge of the meaning of the involved expressions. There are few cases, if any, in which different terms do not involve different sets of statements, theories, criteria of attribution, prototypes, or different externalist components. (With respect to what an externalist account of meaning would imply for the truth conditions of «s is analytic», see the brief but interesting paper of Pretri, 1992).
In spite of that, different terms can have the same meaning, and it can have sense to speak of synonymy and translation if meaning is not reducible to any of these particular statements, theories, criteria of attribution, prototypes or externalist components. The background of meaning is very plural and heterogeneous. This the reason why it is so difficult to obtain an unified adequate definition of analyticity from the particular philosophical senses proposed for analyticity. Neither the «Quinean» notion of analyticity as «truth by virtue of the meaning», nor any of the various «speaker-based» notions of analyticity, nor an analyticity understood as no more than explicit definition, nor a «kantian» notion of analitycity as redundant predication, etc., would be able to support such an unified adequate definition of analyticity with the intended modal force and generality.
Any supposed analytic2 statement is no more that an element of the background of the meaning of the terms involved, an element that very often can change without a change of these meanings. Because meaning has that plural and heterogeneous character, it is so questionable the existence of a special link able to institute the analyticity that some philosophers are looking for out of the mere links that, in fact, can exist among meanings and truth values in the cases where the truth is fixed once meanings are. The error was just in thinking that a given class of phenomena, linguistic phenomena where the truth values of some statements are fixed being fixed their meanings, would require an unified explanation in terms of a theoretical property called analyticity.
That approach has important consequences concerning synonymy (and, therefore, translation). From a classical perspective that is adopted by Quine, the notion of analyticity depends on the notion of synonymy. In other words, as a minimum, synonymy entails analyticity. If we follow adopting that perspective, our rejection of analyticity would entail a rejection of synonymy too. Nevertheless, there is a sense of synonymy according to which it would be still possible to have synonymy without having analyticity. It would be possible if all cases of «x means that ...» and «x has the same meaning that y» are always understood as statements that can be defeated. The situation would be one in which even if it is assumed the synonymy of terms T and T', that synonymy would not entail the philosophically intended analyticity of «All and only things being T are T»'. Assuming for T and T' the meanings they are supposed to have, we could consider fixed the truth of that statement, even it could be considered analyticly2 true. But, if that statement can be in fact false, it cannot be analyticly1 true. And it can be false if determination of meanings and determination of the sameness of meaning is always made through cogent procedures of determination that can be defeated.
All supposed analytic statements must be placed in the background of meaning. We have seen that, against the «change-of-the-meaning» option, meaning cannot be completely identified with that background. Very often, the background can change without any change in the meaning. There would be only a kind of supposed analytic statements for which changes in their truth values would entail direct changes in the meaning; namelly, statements stablishing among several terms that kind of defeasible synonymy. Only in these cases the «change-of-the-meaning» strategy seems to be directly applied. We can accept that. However, it must be noted that to recognize that role for some supposed analytic statements does not entail to accept the kind of synonymy able to make analyticity possible. These statements would be only trivial cases of analyticity, i.e, analytic2 statements. So, it would be possible to have a very useful concept of defeasible synonymy without being engaged in analyticity.
As we said above, conventionalism likes analyticity and the «change-of-the-meaning» strategy to protect analytic statements. The choice among different analytic statements would be only a matter of stipulation, it would be never a rational business. Conventionalism is a kind of relativism. But, there are also other kinds of widespread relativisms that have their roots in a «change-of-the-meaning» strategy without being conventionalists. Nowadays, these relativisms are very popular, specially in the so called «continental philosophy». According to them, whatever change in our beliefs could be reintepreted as a change of the meaning. There is not progress in knowledge, but proliferation of meanings. Here, like in conventionalism, the choice among different ways to speak is never a rational business. But, unlike conventionalism, these relativisms do not see that proliferation of meanings as a matter of stipulation. Proliferation of meanings is the effect of other causes for which we do not have any epistemic responsability. With respect to epistemic subjects, meanings are out of control. Mere stipulation is impossible, and so it is impossible analyticity too.
In that sense, conventionalism would be a relativism with analyticity. The difference between conventionalism and other relativisms without analyticity would be only one of emphasis on the control an epistemic subject can exert on the meanings its words and statements can have. The points of view we have defended go against both conventionalism and these other relativisms. The rejection of analyticity and the rejection of the «change-of-the-meaning» strategy that we have defended in the context of a non-holist minimal meaning realism with determination accept the first part of the quote of Quine (1951) made in section 1 without being engaged in its second part. And that compatibilist view would entail to be in a better position against these conventionalist and relativist moves. In a word, the rejection of analyticity is no more than the rejection of linguistic arbitrariness.
With analyticity there is something trivial and something non-trivial. About what is trivial, there is no much to discuss. Almost all of us do accept it. Once it is assumed that «bachelor» and «unmarried» have the meanings they are supposed to have in our actual languages, all of us do accept that «bachelors are unmarried» is a true statement. This is the trivial side of analyticity. But, analyticity is not that. These things are the phenomena that analyticity in the intended philosophical sense would have to to explain. That is the non-trivial side of analyticity. But it is a wrong side. The error was just in thinking that a given class of linguistic phenomena would require an unified explanation in terms of something called «analyticity».
ACERO, J. (1993) «Analyticity, Semantical Realism and the Strategy of `Two Dogmas ...' «, Sixth Conference of the Ibero-American Philosophical Society, Tenerife, Spain.
BOGHOSSIAN, P. (1993) «Analyticity», Sixth Conference of the Ibero-American Philosophical Society, Tenerife, Spain (to appear in Philosophical Issues.
GRICE and STRAWSON (1956) «In defense of a dogma», The Philosophical Review, LXV.
KATS, J. (1972) Semantic Theory, New York, Harper and Row.
PRETI, C. (1992) «Opacity, Belief and Analyticity», Philosophical Studies, 66, 3.
PUTNAM, H. (1966) «The analytic and the Synthetic», Minnesota Studies in the Philosophy of Science, vol. III.
QUINE (1951) «The two dogmas of empiricism», in From a Logical Point of View, Cambridge, Harvard Univ. Press, 1953.
This work has been partialy supported by DGICYT (Spanish Ministry for Education and Science) as a contribution to the research «Cognitive Models Applied to Pragmatical Aspects of Scientific Systems». I owe deep thanks to the valuable interest and comments of many colleages and friends, especially to those of Paul Boghossian, Fernando Broncano, Bruno Maltrás, Lorenzo Peña, Paco Salto, and Jesús Vega.
University of La Laguna (Canary Islands, Spain)