In this paper I will discuss one of the aspects of the problem of explanation: how is it possible to increase our knowledge of the world and how it is linked to the problem of understanding the meaning of linguistic expressions? Such an approach proceeds from an assumption (though debatable) that with every new explanation, whether it is good or bad, our knowledge does increase, at least in the way that there is a new text. Key words here are "knowledge of the world" - what can it mean? I will analyze this epistemological problem in terms of the philosophy of language. Comparing some related distinctions - similar but not identical - I will try to show their role in the enlargement of descriptiveness leading to explanation, and thus to the increasing of knowledge.
Let us formulate the initial question:
how is possible and how is carried out an enlargement of descriptiveness which means in fact increasing knowledge of the world?
Let us set up definitions.
Since we consider both description and more developed textual characteristics of an object - for example, explanation - being different kinds of verbal characterization, then the transition from one kind of characterization to another, in turn, can be described. The analysis of transition from description to explanation assumes an establishment of some criteria, being guided by which we can consider such transition as done. Therefore we should speak about the delimitation of description, or establishment of distinction between description and explanation. The task of differentiation between description and explanation will be in such case a task of construing foundations of their opposition.
For example, we say: "Water is H2O". Do we indicate in such a way the property of water "to be a substance, which molecule consists of two atoms of hydrogen and one atom of oxygen" or do we give by this statement an explanation (i.e., in this case, theoretical characterization) of molecular structure of the substance "water"? Obviously, both: it will depend on a context, in which the statement "Water is H2O" will be placed. It can answer a general question: "What is water?" or that special one: "Which is the molecular structure of water?" The description, unlike the explanation, can assume no question at all or most general question: "Which is this subject?" The explanation is governed by more special questions - for example (more often): "Why this subject is such (as it is)?" - and, accordingly, is aimed at establishing such interrelations between various kinds of characterizations of the examined subject, by virtue of which such and such knowledge about it takes place. In this sense, for example, Bas van Fraassen argues that a theory of explanation must be a theory of why-questions [1].
Thus, we have to admit that the description is a description by virtue of certain properties of the context, i.e. property of descriptiveness is, at least with this regard, contextually dependent.
Let us try to clarify the character of this dependence. The application of the method of descriptions results in well-known difficulties related to the identification of the referent in referentially opaque contexts, where co-referring expressions cannot be mutually replaced salva veritate. These opaque contexts can be classified in different ways. Philosophers usually distinguish two kinds of opaque context: intensional contexts, which allow the substitution of necessarily co-referring expressions but not contingently co-referring expressions; and hyper-intensional contexts, which do not even allow the substitution of necessarily co-referring expressions. Such distinctions can be, in general, as semantically narrow as one wishes: on that account, Esa Saarinen distinguishes five kinds of ambiguity at quantification in propositions with intensional predicates [2].
For our purposes we will hold that such contexts can be of two kinds. Let us consider examples.
Thus, if we say that the description is a description by virtue of certain properties of the context, it means that description appears to be relativized to some system of characterization, or conceptual field. We could see it in both examples, but in obviously different ways.
In the first case such conceptual field is constituted by a set of widespread assumptions and elements of knowledge, i.e. some (scientific) picture of world. The defeat of Napoleon at Waterloo is a historical fact, and there is a common consent about it. We do not doubt it, because we think we know the way in which such factual statements are usually made, and we believe this way worth trusting. The impossibility to identify a referent with a description of that kind would prove the low quality or incomplete validity of the knowledge possessed by the recipient of the message. Then the context can be elucidated by the coordination of the recipient's individual conceptual scheme with the standard one, which is shared by all the participants of the communicative community. The standard conceptual scheme acts here as a certain (in this respect, one can say limiting) horizon of knowledge, and the coordination would consist here in the approximation to it - the recipient should read the textbook or to come to a lesson, and simply to learn, who has been won at Waterloo. The recipient's individual conceptual scheme (the invalid one) would thus be corrected according to the valid sample - the standard scheme.
In the second case the context is referentially opaque by virtue of the conflict of two social conceptual schemes fixed by natural languages and existing in linguistic communities. The coordination of such schemes would have another form than training in history of an ignorant recipient. It would require not an approximation to similarity with something given, aiming at having no difference with it, but an establishment of conformity between two classes of concepts. And we talk here not only - and, certainly, not mostly - about the interpretation of facts, but rather about properties and elements of linguistic pictures of world. For example, in some African languages there are only two terms for the characteristic of color scale: a "warm" color and a "cold" one; so, within the limits of such conceptual scheme, the rainbow can have a description "a bi-colored phenomenon seen in the sky", i.e. such description, on which a bearer of any Indo-European language, into which the description will be translated, would never identify the referent.
Natural language thus acts not as some empirical "dogma" (language, which all Frenchmen actually speak), but as some basis of formalization (e.g., language, the set of all trivially true sentences of which defines in a unique way the extension of the concept of truth for a speaker of this language).
Similarly we can interpret problems of modal and indirect-speech contexts, where reference is impossible because the principle of the substitution of identicals does not work. For example, the expressions "number of planets in the Solar system" and "9" are two substitutions of the number 9, but the former cannot replace the latter one in the statement "9 is necessarily more than 7" or to be placed instead of one of the figures of nine in "9 is necessarily equal to 9". When we estimate the truthfulness of the sentence "the number of planets in the Solar system is necessarily equal to 9", we proceed from assumptions, according to which there is no law of nature regulating the number of planets with necessity, and there is no omnipotent being, which could imply such modality. However, in the framework of a mythological picture of the world statements like "the number of planets is necessarily equal to 9" could be understood as true. "Quantified variables" in this example of Quine (N(x)>7 or N(x)=9) are not in referential positions for us not so much because the proposition is modal (i.e. "it could not be otherwise"), but rather because they refer to certain (possibly conflicting) states of affairs, which are obviously different. Therefore, instead of saying that variables here are not in referential positions, it would be probably more correct to say that such contexts do not give opportunities for the comparison of different states of affairs or for the outlining of relations between them: we cannot conclude from this context about relations which can exist between different states of affairs and their elements.
To handle these difficulties with the substitution of co-referring expressions, notions of communicative intention, illocutive force of the statement and so on are involved; these tools being heterogeneous enough, they still do correlate in the respect that they form a certain "utterer's meaning" or "speaker's reference", governing the semantic reference. Their common raison d'ˆtre can be interpreted in the following way: as the direct indication on objects in referentially opaque contexts remains impossible and additional identifying procedures are required, the problem of reference appears to be linked to the problem of individuation of a subject. So is the individuation by means of descriptions nevertheless possible, and what requirements should answer such descriptions?
The principle of identifying descriptions is summarized in the "descriptive metaphysics" of P. Strawson, who states that names have no meanings without being supported by descriptions [3]. According to this principle the knowledge of a number of descriptions provides an opportunity to allocate "the basic individual" - the referent of the proper name - in a unique way. So, the opportunity to recognize the man named "Napoleon" is achieved if we understand definite descriptions "defeated at Waterloo", or two descriptions: "the great commander" and "the emperor of France", or three descriptions, etc. However, this requirement of definiteness of understanding of descriptions providing individuation involves the problem of an intercontextual multivalence: the descriptive way of indication in many situations does not guarantee the allocation of unique referent satisfying available descriptions. So what descriptions are most relevant for individuation?
According to traditional views, attributed to Russell and Frege, the object of reference has properties, which should be fixed by appropriate descriptions. Names secure reference to particulars through a descriptive content, which is associated with the name. That is, a name has associated with it an equivalent (synonymous) set of descriptions, and the name refers to whatever satisfies all (or most) of the descriptions. From such a point of view, descriptions give the meaning of a name, defining an object of reference, which is anything that fits (or best fits) the descriptions. A name refers to its bearer by virtue of some descriptive information, attached to the name, which concerns exclusively the bearer of the name. So, a proper name is a singular expression of reference, which is used to ensure reference to a unique (and, besides, known) individual or subject. On such an approach, the possibility of reference is determined by some level of knowledge about the object, and the problem of reference is put as an epistemological problem.
It may well be asked, how can we know that such and such description is a true one?
When a speaker asserts something having the form "N is P", where N is a name, and P is a predicate, the truth conditions look as follows.
The assertion is true iff
a) there is a certain (object) item connected definitely to the use of the name "N";
b) this item has a property named "P".
In what way should the identifying features of an object be related to the use of the name "N", so that this name could specify this object? An answer is provided by the truth-conditional semantics - an enterprise trying to coordinate the central concepts of logical and linguistic semantics.
To build up the theory of meaning, Donald Davidson involves the "Convention T" of Tarski [4], according to which, as he puts it, a satisfactory theory of truth for a language L should assume that for each sentence s from L there is a theorem of the form "s is true, if and only if p", where "s" is replaced with a canonical description of a sentence of L, and "p" is replaced by that sentence itself (or, suppose, by translation of s on the language, where the set of all trivially true sentences uniquely determines the extension of the concept of truth for its bearer, if this is not the case with L) [5]. Diverting from actual definition of truth, the Convention T embodies our intuitions of how the concept of truth should be used with reference to linguistic expressions. Then the requirement to the semantic theory of a language L consists in the following: without referring to any further semantic concepts, the theory imposes on the predicate "is T" restrictions, which should be sufficient to obtain from the scheme T all sentences, where s is replaced by the structural description of the sentence, and p - by the sentence itself. The list of T-sentences gives the complete description of meanings of the object language.
The aim of truth-conditional semantics is to give an account of the meanings of each sentence by telling what the truth conditions of that sentence are, i.e. meaning can be given by a theory of truth conditions of sentences in a language. For each sentence in the object language, the theory tells us under what conditions that sentence is true; that is, it gives the truth conditions for each sentence in the object language. The list of T-sentences constitutes a complete description of the meaning of the object language. T-sentences actually relate significant facts about what our sentences mean. They do not simply say that something is true just when it is true; what they do is to give the meaning of a sentence by telling us under what conditions it is true.
What can be objected here is the following. A list of true T-sentences for an object language should constitute an account of what every object language sentence means. But it does not, if we admit that truth conditions were assigned accidentally. One might as well have assigned any true truth condition to any true object language sentence so far as a weak criterion is concerned; one could even have assigned a single true truth condition (for example, that snow is white) to every true object language sentence, and a single false one (for example, that snow is purple) to every false object language sentence. The theory then has to have an additional complexity designed to save it from this absurd consequence.
The contrargument here would lead us to the feature which we had put earlier in the basis of the distinction of different kinds of intensional contexts: to the real way, in which natural languages exist. To see why such a theory would not result in this absurdity it is important to understand that real natural languages contain an infinite number of sentences. Obviously, then, a complete list of T-sentences for such languages cannot be given, because we could not know such an infinite list. Then we have to admit that what we know is a finite list of words and ways of combining them, such that we can figure out the meanings of an infinite number of sentences combining them. In other words, what we know is in effect an axiomatic theory with a finite number of axioms, which potentially yield as theorems every one of the T-sentences in a language. Thus our understanding of the sentences of our own language consists in our forming a theory that yields T-sentences.
The point is that such a theory is capable of constructing T-sentences for all the sentences in a language: such a theory can be considered as "mistaken" with regard to some "reality", but it will still account for any finite collection of evidence. At any rate, the theory that attempts to explain the meanings of sentences by their truth conditions within the representationist approach, relying on reference and satisfaction of predicates, will work only to the extent that the language is extensional.
According to Davidson, his theory of meaning stays within the framework of Tarski's standard theory of truth - or, at least, is very well compliant with it. Indeed, the requirement of the deductibility of T-sentences put forward by Davidson formally coincides with the requirement formulated by Tarski for the concept of truth in formalized languages. At the same time (as Davidson admitted later [6]) in truth-conditional theories of meaning T-sentences are supposed to play a role somewhat opposite to the role, which they play in the theory of truth of Tarski. Tarski aimed at giving a significantly adequate and formally correct definition of truth for the formalized languages. And on the opposite, in a truth-conditional theory of meaning the predicate "is true" is considered as initial one, and not as a concept determined within the framework of the theory. While Tarski analyzes the concept of truth appealing (in Convention T) to the concept of meaning, Davidson consideres truth to be the central primitive concept, and tries, "by detailing truth's structure, to get at meaning". Assuming that the concept of truth already is given previously, Davidson uses the construction of Tarski for the formulation of requirements to a theory of meaning: if a sentence S of a language L is given, then a statement about its meaning like "S means P" can be replaced by the corresponding T-statement.
However, the language, with which Tarski deals is a formal language, not a natural one, and, accordingly, its use is regulated by an ad hoc linguistic convention, and not by a total one, i.e. produced between all members of linguistic community. The latter obviously differs by form from the former one: it was not concluded explicitly, is not limited in time etc. Therefore it may well be asked: what happens to the Convention T at using a Tarskian definition of truthfulness for the definition of meaning in a natural language?
If a semantic theory should have the form of a theory determining conditions of truthfulness for analyzed sentences, then knowledge of the semantic concept of truth for the language L will consist in the knowledge of what means for the sentence S of the language L to be true. Instead of an exact definition the truth is characterized by a finite set of axioms. The theory of meaning is thus considered as a system of statements intended to answer questions on the relations between linguistic expressions, whereas the theory of truth represents itself as a theory of reference, i.e. system of statements intended to answer questions on the relations of linguistic expressions to the world.
Thus, in a truth-conditional theory of meaning the criterion of externality to linguistic system comes on the immediate prospect. In a certain sense, this criterion is extremely formal, because we cannot speak about objects of designation as entities, which are internally inherent to the linguistic system, or about certain properties of designation. It is hardly ddebatable that reference to something different, or direction on another object is an intrinsic property of the linguistic sign. Due to this constituting property the sign is itself, instead of being something other (say, does not belong to some class of purely physical or mental phenomena, or metaphysical concepts).
However, if we try to explain with the help of truth-conditional theories of meaning the way, in which expressions of natural language specify their referents, we will meet difficulties, which we discussed above while talking about referentially opaque contexts. Let us characterize the source of these difficulties once again: according to these assumptions, we deal with linguistic expressions in such a manner that they specify to us certain states of affairs, events, facts, situations belonging to a reality, distinct from the reality of sentences and systems of sentences. In other words, the truth, which we are talking about here is a correspondence truth. Truth-conditional theories of meaning based on the correspondence concept of truth may or may not operate with independent concepts like "sense" or "meaning". But, since it is supposed that their theorems give correct references to signs - and this is equivalent to a requirement that they demonstrate meanings of expressions, - so far they contain an explicit or implicit commitment to "the way in which the referent is given" (Frege). Such theories explicitly or implicitly assume that the T-theories can give something more than just conditions of truthfulness of expressions, namely, that they can give conditions of truthfulness in such aspect, which "displays" or "reveals" meanings of expressions. Therefore it can be assumed that while two expressions have the same referent, they have different meanings, as the expressions have different modes, or algorithms of representation, i.e. they represent the referent by different ways.
Traditional questions with this regard are these: can T-theories work with intensional contexts, i.e. do theorems of such theory represent only referents of expressions, or as well meanings? The right part of a T-sentence refers to the expression at the left: what is the nature of this reference? Is it a simple indication or this reference can in a way show or display some meaning?
These questions are analyzed in the research literature on more special object domains: counterfactual conditionals; sentences with dispositional predicates; sentences expressing possibilities and causal relations; interrogative and imperative sentences etc. Solutions offered so far can define truthfulness of expression as truthfulness relating to some conceptual structure. In particular, the relativity of truth might be an important precondition, if we talk about conventionality of meaning, consequently keeping our attachment to the idea about conditions of truthfulness of a linguistic sentence as its meaning. On the other hand, such updating is helpful with removing the truth-conditional concept of meaning from under impact of some substantial criticisms - for example, verificationist. According to the latter, the truth-conditional concept of meaning is unsatisfactory, as it does not result in a satisfactory explanation of the phenomenon of understanding a language; any theory of meaning, which is not a theory of understanding or does not give it as a result, does not satisfy the philosophical purpose, for which a theory of meaning is required - and therefore, a theory of meaning should include, besides the theory of reference, also some theory of illocution. These and similar reasons have caused the appearance of a number of "broad" theories of reference, using semantics of possible worlds and other means for giving some socio-cultural dimension to the very idea of reference.
The theory of reference in its traditional (empiricist) variant identified the meaning of an expression with what it indicates, or with referential connection (i.e. did not involve any assumptions about the interpretation of knowledge as a possibility of description). But the inclusion of referentially opaque contexts in the area of consideration of T-theories, which is inevitable at the logic analysis of natural language, has required an essential increasing of expressive opportunities of our formal systems and more flexibility of our account of descriptions. For this purpose a theory of meaning of referential kind should be based on such concept of indication that would allow to express more complex mutual relation between the object of reference and conceptual web, or "web of belief" of the speaker. To have some knowledge of an object, sufficient for its identification, means then to have not simply some information on this object, but some information on the meaning of the term indicating this object [7]. Those are, in particular, "broad" (or "causal") referential theories like rigid designators, indexicals, "speaker's reference" etc. The analysis of such theories shows, that they use different assumptions about representation than "classical" representationism: representation in them appears by and large intermediated. In fact, these and similar theories won't deal with the view, on which the description of an object proceeds from an assumption that this object is by any means given (pre-given). As more complex accounts of description, they proceed from some other assumption, namely: the primary description of a subject can be placed in the context of some system of description - such, that the putting of the indication on the described referent in this system would not harm rules of functioning of this system. Otherwise - as, for example, Davidson pointed it out - such causal theories of reference just would not be theories of meaning, because they (would) look to causal relations between names and objects of which speakers may be ignorant [8].
Within the limits of correspondence truth analysis, we should reason here in the following way: the relation between signs - for example, narrative sentences of indicative mood of an object language - and a reality, which is transcendent to the language, is the relation of descriptiveness. Thus, with regard to the described reality as such, as to the transcendent referent available as a reality, the relation between sentences and what they describe can be set in a concrete situation in (at least) two ways:
Such extended description appears, in the framework of understanding representation as the relation of the description to some reality, to be a description of some other reality, new in respect to the primarily represented one. The constructive attitude is characterized by bringing new elements, whose referential status may be unclarifiable, i.e. these references can not refer to the initially present reality in the same way, that references established in the primary description. Then we should admit, that an attempt of expansion of description, i.e. construction of an explanation, would mean, in the certain sense, doubling of the referent.
Indeed, if we deal with a system of designations, where both descriptive and constructive relations to reality are interconnected - and this reality is supposed to be common, or general for description in the given system of propositions - than we face the question: how this internal consistency can be achieved and kept in a context of various character of correlation with reality inside the given system of propositions? In fact, we describe inside this system two realities taken theoretically for one. We will have to admit then, that an attempt at ontological validation of such conclusion would be rather controversial.
Hence, we have considered traditional views on descriptions as based on the representationist approach to the analysis of linguistic meanings, and have shown problems arising here in connection with the use of the correspondence theory of truth. The alternative, to which we came during this consideration, is this: an account of truth-conditions of linguistic expressions could be developed in terms of the coherence theory of truth.
According to the latter, the level of truthfulness of a statement is defined by its role and place in some conceptual system. The more coherent are our ideas among themselves, the more they are true. The truth of any true proposition consists in its coherence with some specified set of propositions.
The coherence theory differs from the correspondence theory of truth in two essential respects: they give different accounts of the relation between propositions and their truth conditions.
|
Relation of truthfulness |
Truth conditions of propositions |
|
|
Coherence theory |
Coherence |
Consist in other propositions |
|
Correspondence theory |
Correspondence |
Are not propositions, but indeed objective features of the world |
Let's consider both criteria in turn.
The coherentist might want to add some probabilistic connections between other propositions S believes and P. As Laurence BonJour pointed out [9] in connection with a coherence theory of justification, logical consistency is a notoriously weak sort of coherence, and a coherence theorist of truth will no doubt want to expand the kind of coherence that yields truth to include at least probabilistic connections. P will be true for S only if P is logically consistent with the rest of what S believes, and there are interesting probabilistic connections between other propositions S believes and P.
Another common way to deal with the coherence relation is to understand it as some form of entailment, which can be taken as strict logical entailment, or entailment in some looser sense. According to this version, a proposition coheres with a set of propositions iff it is entailed by members of the set. But this is, also, too weak by itself, because this approach will require a further elucidation of the notion "entailment", and so on.
At the same time correspondence theorists might not be bothered with this kind of questions, considering correspondence as a relation sui generis. "Over and above the rhetoric of derision, is there really any principled objections to the view that the all-important concept of correspondence is irreducible, is unanalyzable?" [10] It may well be asked, why then couldn't coherence be a sui generis relation, just as correspondence is?
Since any conceptual analysis must have a foundation, then there are conceptual "atoms" out of which all other concepts are formed and which cannot themselves be analyzed. But since any system has a structure, we can say the same about relations between them. (Therefore, probably, a stronger thesis can be proved: if we find correspondence a relation sui generis, we thus will necessarily have to say the same about coherence.) It was argued by many philosophers that fundamental intentionality involves a number of sui generis relations, the ideas of which are absolutely fundamental but are also not reducible to any other relations. By virtue of its nature - or, precisely speaking, as we are able to identify it as such - the human mind exemplifies certain properties (is in certain states) including properties to correspond and/or to cohere.
Coherentists generally agree that the specified set consists of propositions believed or held to be true. They differ on the questions of who believes the propositions and when. At one extreme, coherence theorists can hold that the specified set of propositions is the largest consistent set of propositions currently believed by actual people (e.g., J.O.Young [11]). According to a moderate position (e.g., Putnam [12]), the specified set consists of those propositions, which will be believed when people like us (with finite cognitive capacities) have reached some limit of inquiry. At the other extreme, coherence theorists can maintain that the specified set contains the propositions, which would be believed by an omniscient being (some early - British idealists' - versions: e.g., Blanshard [13]).
One can still think of the relation of coherence as holding between propositions, but the relevant propositions with which P must cohere are to be defined in terms of their being the actual or hypothetical objects of belief. Abovementioned different versions of a coherence theory of truth might be obtained depending on how one uses belief to restrict the relevant class of propositions with which a proposition must cohere in order to be true. This means that the relevant subclass of propositions can vary from one individual or community to another and it is coherence with an individual's or community's belief system that determines truth with this regard.
So the notion of "specified set of propositions believed to be true" can be considered as still requiring further exposition. Then, if we are keen to keep our arguments about referentially opaque contexts in natural language, we can consider the specified set of propositions in question as the set of all trivially true sentences, which uniquely determines the extension of the concept of truth for all members of the linguistic community. Linguistic community is understood here extensively, as the set of all bearers of the language L.
If we approve such an understanding of the specified set, then the coherentism involves the rejection of representationism even before we adopt constructivism. I will try to prove this thesis in the following way.
Representationism involves acceptance of two principles:
the principle of bivalence (according to which every proposition is either true or false) and
Coherentists reject
the principle of bivalence since it is not the case that for every proposition either it or a contrary, by excluding disjunction, proposition coheres with the specified set, and
the principle of transcendence since, if a proposition coheres with a set of beliefs, it can be known to cohere with the set. If we did not know its truthfulness or falsity, we could not determine its coherence in any way.
Thus, the coherence theory of truth appears to be linked to the coherence theory of justification. One might argue here that the latter does not necessarily involves the former - for example, on the ground that the holder of a coherent set of beliefs might lack a reason to believe his beliefs coherent. This could be a separate topic for discussion, but since my aim here is not to discuss the coherence theory of truth or of justification as such, but their semantic application, I will concentrate on their joint semantic relevance.
One of the basic (if not the main) challenges to coherentism is the following: coherence with a set of beliefs might be the test of truth but truth nevertheless consists in correspondence to objective facts. But this contrargument can be declined: if truth consists in correspondence to objective facts, then coherence with a set of beliefs will not be a test of truth, because there is no guarantee that a perfectly coherent set of beliefs would match objective reality [14].
That is why, if we hold the link between the coherence theory of truth and the coherence theory of justification, we can re-estimate a classical epistemological argument for coherentism. It is based on the view (proper, in particular, to constructivism) that we cannot "get outside" our set of beliefs and compare propositions to objective facts [15]. This argument can be considered as depending on the coherence theory of justification. The argument infers from such a theory that we can only know isolated facts of coherence or of absence of coherence of a certain proposition with the certain set of propositions expressing certain beliefs. We are by no means and we cannot find ourselves in such epistemological position, whence we could conclude, whether this or that proposition corresponds to reality.
This argument is subject to criticism that it can be considered as containing an incorrect implication. From the fact that a proposition cannot be known to correspond to reality we cannot infer that it does not correspond to reality. Even if correspondence theorists admit that we can only know which propositions cohere with our beliefs, they can still hold that truth consists in correspondence. If correspondence theorists adopted this position, they would accept that there may be truths, which cannot be known - for example, that there is some absolute truth, to which we can only come nearer by specification of relative truths, which we know. Alternatively, they can argue as Davidson does it [16], that the coherence of a proposition with a set of beliefs is a good indication that the proposition corresponds to objective facts and that we can know that propositions correspond.
Correspondentists might even argue - and they do - that the coherence theory is not a theory of truth at all [17]. Hence, they proceed from an assumption that they know what the truth is, i.e. they have a definition of truth. And of course, they know what the truth is: according to their views, it is correspondence to facts. Indeed, the coherence theory of truth is not at all a theory of correspondence to facts. But coherentists never said it is.
This difference in meanings of the term itself "truth" may be interpreted as linked to the difference of purposes for which a theory of truth is given. There can be at least two such purposes:
to provide a definition of the conception "is true" as a characteristic of a proposition;
to specify the test-conditions for determining whether or not there is warrant for applying the characterization "is true" for a proposition.
According to Nicholas Rescher, who summarizes this distinction [18], the two questions are strictly non-identical: we can have a criterion or criteria of truthfulness (truth-conditions) of a proposition and still lack a definition, what does it mean for this proposition to be true, and vice versa.
But if we want to be able to use a linguistic unit, then we will need relevant criteria for the success of the usage. That is exactly what the truth-conditional theory of meaning does: it identifies meaning with truth-conditions of a proposition, and this taxonomy is grounded in the conception "meaning as use", identifying meaning of a linguistic unit with the conditions of its use. If we adopt this view, we should say that any talking about meaning-specification, which is not effective for determining rules of application of the sign, is just redundant.
That is why I would not argue that we cannot know, whether some propositions correspond to elements and characteristics of the external (to the description) world: this skeptical conclusion would be a metaphysical statement, which is not really necessary for giving a theory of meaning (as was shown by Kripkenstein). What I am doing here is rather trying to arrange arguments in favor of the irrelevance of such (putative) correspondence for the coherence theory of justification and, accordingly, for the truth-conditional theory of meaning based on the coherence theory of truth. In order to do this, the initial argument for the coherentism must be supplemented. Since we stick to our idea of linguistic community as an ultimate truth-operator, we can supplement the argument as follows.
The correspondence and coherence theories have differing views about the nature of truth conditions. According to the coherence theory, the truth conditions of propositions consist in other propositions. According to the correspondence theory, the truth conditions of propositions are not propositions, but indeed objective features of the world. One way to choose an account of truth (i.e. to define, in which cases this or that account of truth is more adequate) is to pay attention to the process by which propositions are assigned truth conditions. Coherence theorists can argue that the truth conditions of a proposition are the conditions under which speakers make a practice of asserting it ("meaning as use"). This means that speakers can only make a practice of asserting a proposition under conditions, which the speakers are able to recognize as justifying the proposition. Now the supposed inability of speakers to "get outside" of their beliefs becomes significant. It is important, because the conditions under which a proposition coheres with speakers' beliefs are the only truth conditions in a way that they are the only conditions which speakers can recognize as justifying this proposition - and thus as justification of our knowledge of the meaning of this proposition. When the speakers make a practice of asserting the proposition under these conditions, they become the proposition's truth conditions.
Within the suggested aspects of consideration, the recognition of the latter thesis brings us to the adoption of the constructivist approach.
Consider the distinction of representationist and constructional paradigms.
Representationism is taken as epistemological realism, proper to both rationalist and empiricist traditions. It proceeds from the belief that knowledge as such is aimed at acquiring information on some external world, which is transcendent to the learning subject, independent of it and "given" beforehand.
Constructivism (constructionism, to use Goodman's term) is characterized by the rejection of the "given", of any effort to split perception from conceptualization (hence of all such approaches to an observation/theory dichotomy for science), and of the apriorism in favor of a modified coherence view of justification, emphasis on pragmatic considerations in choice of theory etc.
Hence, a constructional system is understood to be an interpreted formal system of definitions and theorems framed in the language of the first-order predicate calculus. The definitions of a constructional system are to be thought of as "real" definitions meeting some definite semantic criterion of accuracy in addition to the usual syntactic criteria imposed on purely formal or "nominal" definitions. Thus, a constructional system is a formalization of some domain of putative knowledge which may be thought of as a set of sentences formulated in presystematic discourse (i.e., generally, discourse of a natural language), some of whose terms are to be appropriately defined in the system using logic plus a special set of terms adopted as primitive in the system (its "extralogical basis"). These primitive descriptions are to be thought of as already having an intended use or interpretation; if it is not obvious, it may be provided by an informal explanation, strictly not part of the system.
Then we can make the following remark about the foundations, on which explanation as a linguistic (textual) characteristic of an object is opposed to its description. The point is that within the constructionist approach what we can distinguish from the description is not a prescription, since such a distinction is rejected: if we suppose that a description allows us to construct an object, then we assume that every description already has a prescriptive value. That is why we can suppose that the analysis of transition from description to explanation is linked to the problem of the relationship between data of representation and theoretical construction and may be considered within its context. It can be done, then, through the distinction of representationist and constructional paradigms, which is not actually identical to the distinction of the language of observation and the language of theory; both in the language of observation and in the language of theory it is possible to allocate both representationist and constructional elements. E.g., according to Schlick, we do not doubt of the facts of geography or history not because we believe them empirically verifiable, but because we know (and do not doubt) the way in which such factual statements are usually made.
For our purposes it is important to stress that not only different fragments of natural language can be used as language of observation and language of theory, but also the same fragment, or all natural language as a whole. Sentences of the language of observation can include some markers, locating meaning in time, space etc., but there are no obstacles for use of the same linguistic means in the sentences of the language of theory. The definition itself of the language of observation as a postulated part of natural language which is devoid of theoretical terms is vulnerable in the sense that the language of observation and the language of theory will nevertheless have the same grammar, i.e. be governed by the same linguistic rules, both synchronical and diachronical. The latter argument covers van Fraassen's objection to this dichotomy, postulating that if we could cleanse our language of theory-laden terms, starting with the recently introduced ones, continuing through "mass" and "impulse" to "element" and so on into the prehistory of language formation, we would end up with nothing useful [19]. So, if expressions like "part" and "A is part of B" are considered in different ways, like belonging to the language of observation or to the language of theory, what happens then to their meanings?
If we consider the set of propositions taken as trivially true by the linguistic community at a moment of time as a set of primitive descriptions ("semantic primitives") of a constructionist system, it will be clear how we introduce in this system new elements, extending descriptiveness (e.g., to "theoretical description", i.e. explanation). Meaning of descriptions would not be reduced thus to their reference, but this fact would still give us no reasons to reject assumptions about meaning as conditions of the truthfulness of a proposition. Then the status of a new element will depend on the nature of his links of coherence with other elements. So we can answer our initial question
how is possible and how is carried out an expansion of descriptiveness which means in fact increasing of knowledge about the world?
in the following way:
increasing of knowledge consists in constructing coherence: establishing of a biggest number of semantic connections of the initial description with the biggest number of semantic primitives of the system of characterization.
The sui generis relation of coherence is perceived through number of semantic links between linguistic units, and these links form an open set. Here we have then a broad semantic approach, allowing to remove the opposition of standard representationist semantics, where the object domain is considered as a set of ontologically homogeneous objects (elements of the given world), and semantics of possible worlds, using reference to ontologically various kinds of objects: to "objects of the real world" and to "objects of the possible world". Accordingly, such approach will allow to model a wider scope of contexts of natural language. For this purpose, as we have seen, it is possible to apply in a truth-conditional theory of meaning not the correspondence concept of truth, but the coherence one.
The question, whether the coherence truthfulness can be used for definition of meaning in the framework of the truth-conditional approach, appears thus to be a question of the possibility of the use of language by some linguistic community. References arising during this use are to be found in the field of coordination of different (individual) pictures of the world, or conceptual schemes of bearers of language. The possibility of the real use of a language may be thus considered as detection of some polysubjectivism in the intersubjectivity, avoiding thus the criticism traditionally addressed to any relativist concept of meaning and consisting in reproaching such concepts in an insufficient granting of referential grounds. The semantic status of the area of coordination of individual pictures of the world - the area which is common for all bearers of a language - appears then open for the exact analysis and clarification.
The enlargement of knowledge as an explanation, from such point of view, will look as follows.
Let us start from the example of the deductive-nomological model, according to which an explanation assumes a description of some initial knowledge about the examined object and description of some additional knowledge of more general character. Conclusions arising in the real process of knowledge can be reduced to a sequence of similar constructions. The explanation, why the statement "Water is H2O" is true, will be then the indication that this statement can be deductively inferred from statements "Substance, which molecule consists of two atoms of hydrogen and one atom of oxygen, is H2O" and "Molecule of water consists of two atoms of hydrogen and one atom of oxygen". Certainly, there can be other ways of inference, but in any case the explanans as a set of statements expressing some knowledge of the examined object domain, and the explanandum as a set of statements expressing some initial knowledge of the examined object, should be connected by the relation of inference of the explanandum from the explanans. The explanation is supposed to consist in deductive or inductive subsumption of the explanandum under a certain generalization - say, a law of nature, like the explanation of the ideal gas law by the molecular-kinetic theory.
Such techniques are supposed to eliminate the so-called paradox of explanation, or paradox of analysis, which consists in the following: to explain, how this or that explanation is made, it is necessary to know the general mechanism of explanation, i.e. to know beforehand what is not known yet and is still examined as an object of explanation. In the deductive-nomological model the paradox appears eliminated, however the estimation of the truthfulness of a statement is as a matter of fact reduced to checking the conformity of this statement to some set of true statements, which is supposed to be given. The truth estimated in such a way can not be correspondence truth; there are still no grounds here to consider the knowledge of the world, obtained in this way, to be true in the sense of the direct correspondence to the world. Thus, the central question here would be a question of choice of semantic primitives, through which the explanation is constructed.
The search of the right set of semantic primitives is one of the most general programs of examination of connection between a thing and its name. Such set is supposed to constitute a certain lingua mentalis, an assortment of universal non-arbitrary "elements of human thought" as a basis of any possible semantic constructions. The methodology of such program would be, probably, to the greatest extent "object-oriented", i.e. will consider the nature of the object of theory - the language. It is clear that the examination of the connection between a thing and its name refers to the deepest foundations of knowledge: even if it were possible to reveal more basic foundations, such revealing, however, ex definitio could not be verbalized. In modern semantic theories the choice of primitives appears to be a kind of metaphysical landmark: either this choice is recognized to be arbitrary - which is associated with the conventionalist approach to meaning, - or, on the contrary, the choice of primitives is granted to be by and large determined. As it would be rather difficult to specify satisfactory reasons, on which the gap between the theory and the empirical fact in semantics should be more, than, say, in physics or chemistry, so far the choice of semantic primitives can be considered arbitrary, from such point of view, not in a bigger degree than the establishment of the list of chemical elements.
However in the latter case it is still not necessary for the researcher to take a representationist position. The lack of strictly epistemological grounds for formalization of empirical data in deductive theories stimulates the quest for more relevant criteria, having pragmatic character. As a theory meets the requirements presented to it (explanation of known facts and prediction of new ones), so far it can be accepted as satisfactory, and the absence or impossibility of substantiation of the theory as a description of some transcendent matter does not reduce its value as a theory.
The traditional account of explanation is, in its most intrinsic characteristic, reductionist: to explain something, from such point of view, is to reduce the unknown to the known. The basic objections here are those: the explanation of a known phenomenon can be done with the help of completely new and unusual theories. The term "explanation" is hence a pragmatic concept, and is, for that reason, relativist: according to Hempel, we can construct some meaningful explanation only in this or that cognitive situation, only for this or that concrete individual - recipient of the information; it is impossible to give some universal explanation, "an explanation in general", which would be valid always and for everybody [20]. (This marked Hempel's move from the deductive model of explanation to the inductive, or statistical one.) The explanation is therefore admitted to be contextually dependent. The traditional reductionism, thus, is identified with the universalism: the treatment of an explanation as of the reduction of the unknown to the known appears to be intending to find out some general and universal criteria of an intelligible theory (for example, causality, possibility of modeling, visualization etc.).
The account of explanation as causality, the most eminent proponent of which is Wesley Salmon, claims that to explain an event is to identify its cause. Trying to resolve Hume's problems regarding causality, Salmon advanced the "statistical relevance" model of explanation, stipulating that the fundamental causal concepts could be explicated in terms of statistical concepts alone. But, being not satisfied with statistical probability alone, he attempted to provide an account of certain causal mechanisms, in particular, causal interactions and causal processes [21]. Salmon took as primitives the notion of a process and that of a spatiotemporal intersection of processes. The aim is to distinguish between processes that are causal and those that are not (causal processes versus pseudo-processes) and to distinguish between those intersections of processes (whether causal or pseudo) that are genuine causal interactions and those that are not. The basic idea is that an intersection of two processes is a causal interaction if both processes are modified in the intersection in ways that persist beyond the point of intersection, even in the absence of further intersections. A process is causal if it is capable of transmitting a mark - that is, if it is capable of entering into a causal interaction.
The very fact that this approach is based on the notion of interactivity (more precisely, of establishing interactivity) correlates clearly with the criterion of relative truth with respect to propositions having a certain definite epistemic warrant. Here one can see how the concept of "is true" can legitimately be viewed as relative truth throughout the area of empirical fact - it is "relative truth once all the possibly relevant returns are in" [22]. Talking about the explanation of an empirical fact in terms of causal interactions of processes, we will have then to identify the point where the mark was left. If we suppose these marks being constitutive (the ontic conception of explanation, which Salmon involves in order to avoid Hume's criticism of causality [23]), then in order to define the crossing point we will have to be able to define previous marks - not ad infinitum, obviously, but for all possibly relevant for the explanation returns.
Another alternative idea of explanation was advanced by Michael Friedman [24] and Philip Kitcher [25]: it is the idea of explanation as unification, according to which it consists in showing that evidently different phenomena can be seen to be fundamentally similar: we increase our understanding of the world to the extent that we can reduce the number of independently acceptable assumptions that are required to explain natural phenomena (observable regularities). To count these assumptions, Friedman offers a concept of a "K-atomic statement," relativized to a knowledge situation K. A statement is K-atomic provided it is not equivalent to two or more generalizations that are independently acceptable in knowledge situation K. A given statement is acceptable independently of another if it is possible to have evidence adequate for the acceptance of the given statement without having evidence adequate to accept the other.
The problem that arises for Friedman's program is that it seems impossible to have any K-atomic statements - at least, any that could plausibly be taken as fundamental laws of nature or any apriori pre-set regulations. Again, implications to the phenomena themselves do not help to create a plausible account of explanation: it is still needing an external justification; but, in its turn, the consistency of independently acceptable assumptions is enhanced by the presence of explanatory relations among them. We can see thus that Hempel's claim for "systematic unification" [26] as an explanatory tool can be interpreted as a claim for coherence.
Within the constructionist approach the claim for coherence in the account of explanation will mean the following. If we take the set of propositions taken as trivially true by the linguistic community at a moment of time to be the set of semantic primitives, or initial descriptions, relating to which the validity of linguistic meanings is defined, it will allow us to avoid the opposition between universalism and relativism in possible accounts of explanation. Indeed, the establishment of a maximum number of semantic connections of an initial description with a maximum number of semantic primitives of the system does not mean in any way a reduction. On the contrary, what we are talking about here is the construction of a new knowledge, the growth of knowledge at the expense of increasing of number of its structural elements. The explanatory force of a theory in such case will depend on the strength of entailment and on the quantity of semantic primitives, with which the explanation establishes links.
Consider, e.g., the "inference to the best explanation" - a trend ascending to C. S. Peirce ("abduction") and made explicit by Gilbert Harman [27]. According to this view, that a theory explains certain phenomena is part of the evidence that leads us to accept it. And this means that the explanation-relation is visible before we believe that the theory is true.
Let us suppose that we have evidence E, and are considering several hypotheses, say H and H'. Then, according to this account, we should infer H rather than H' if H is a better explanation of E than H' is. The criteria applicable here can refer to the statistical theory:
H is a better explanation than H' (ceteris paribus) of E, provided:
(a) P(H) > P(H') - H has higher probability than H';
(b) P(E/ H) > P(E/H') - H bestows higher probability on E than H' does.
This version of the explanation theory would directly involve coherence in its probabilistic account (BonJour [28]) - P will be true for S iff P is logically consistent with the rest of what S believes, and there are purposeful probabilistic connections between other propositions S believes and P. As BonJour puts it, a system of belief is consistent in proportion to its degree of probabilistic consistency. The latter, unlike the logical consistency, is a matter of degree, depending on how many (putative) conflicts the system would contain, and on the degree of improbability involved in each case. The probabilistic account of explanation is subject to criticism that it does not cover facts of low probability, which are thus supposed to be unexplainable, though we understand each of their occurrences equally well. But the suggested account of "explanation to the best coherence" falls out this criticism. For example, such an approach allows us to deal with the "lottery paradox", which arises given a purely probabilistic rule of acceptance, since such a rule would have us infer concerning any ticket in the lottery that the ticket will not win the prize. But we still feel that we lack epistemological grounds for such a conclusion; and indeed, if our inferential reasoning is aimed at making the best explanatory statement as a most coherent (with other true statements) one, the paradox does not arise, since the hypothesis that a particular ticket fails to win the lottery is definitely linked to some true statements, but it is as well linked to other true statements by relations which are logically different but nevertheless still are relations of coherence.
To summarize, we can formulate two arguments about explanation:
(1) nothing is an explanation unless it is true;
(2) one is not warranted in saying 'I have an explanation' unless one is warranted in the assertion 'I have a theory which is acceptable and which explains'.
It is clear that these two arguments differ only from the correspondentist point of view. For the coherentist they are identical.
Such an interpretation allows us to keep basic intuitions according to which explanation relates the unknown to the known, and it allows us to deal with the objection to such a view, according to which explanations may bring some completely new and unusual theories: anyway, this new and odd knowledge will have to be coherent with the set of propositions which are considered to be trivially true by a linguistic community. Otherwise this new knowledge, having no definite meanings, simply will not be able to function, i.e. will not exist as knowledge.
Thus, the discussion of semantic aspect of transition from the description to an explanation appears to be connected with the realization of several dichotomies:
description/explanation;
observation data/theory;
representationism/constructionism;
correspondence theory of truth/coherence theory of truth;
correspondence theory of truth/coherence theory of justification.
I hope that the above study of their mutual relations has given arguments in favor of the non-referential approach to the problem of meaning and, accordingly, in favor of the construction of semantic theory on the properly epistemological bases, which could be a truth-conditional theory of meaning based on the coherence concept of truth or a theory of meaning based on the coherence concept of justification.
______________________________________
1 B. van Fraassen, The Scientific Image, (Oxford, 1980), p. 134.
2 E. Saarinen, 'Quantifier Phrases Are (At Least) Five Ways Ambiguous in Intensional Contexts', in: Ambiguities in Intensional Contexts. Synthese Language Library, v.12, ed. F. Heny, (Dordrecht - Boston - N.Y., 1981).
3 P. F. Strawson, Individuals. (L., 1971), p. 20.
4 D. Davidson, 'Truth and Meaning', in: Inquiries into Truth and Interpretation, (Oxford, 1984), p. 17-37.
5 D. Davidson, 'In Defence of Convention T', in: Inquiries into Truth and Interpretation, (Oxford, 1984), p. 65-75.
6 D. Davidson, 'Introduction', in: Inquiries into Truth and Interpretation, (Oxford, 1984), p. xiv.
7 K. Donnellan, 'Proper Names and Identifying Descriptions', in: Semantics of Natural Language, eds. D. Davidson, G. Harman, (Dordrecht, 1972), pp. 356-79.
8 D. Davidson, 'A Coherence Theory of Truth and Knowledge', in: Truth and Interpretation. Perspectives on the Philosophy of Donald Davidson, ed. E. LePore, (Basil Blackwell, Oxford, 1986), p. 318.
9 L. BonJour, The Structure of Empirical Knowledge, (Cambridge, 1985), pp. 93-100.
10 R. Fumerton, Truth and Correspondence (manuscript).
11 J.O.Young, Global Anti-realism. (Avebury, Aldershot, 1995).
12 H. Putnam, Reason, Truth and History. (Cambridge, 1981).
13 B. Blanshard, The Nature of Thought (L., 1939).
14 J.O.Young, 'Coherence Theory of Truth', in: Stanford Encyclopedia of Philosophy.
15 A.R. White, Truth (N.Y., 1970), pp. 109-122.
16 D. Davidson, 'A Coherence Theory of Truth and Knowledge', in: Truth and Interpretation. Perspectives on the Philosophy of Donald Davidson, ed. E. LePore (Oxford, 1986), pp. 307-19.
17 R. Fumerton, Metaepistemology and Skepticism. (Lanham, Maryland, 1995), p. 142.
18 N. Resher, The Coherence Theory of Truth, (Oxford, 1973), pp. 1-4.
19 B. van Fraassen, The Scientific Image, (Oxford, 1980), p. 14.
20 C.G. Hempel, 'Aspects of Scientific Explanation', in: Aspects of Scientific Explanation and Other Essays in the Philosophy of Science, (N.Y., 1965).
21 W. Salmon, Causality and Explanation, (N.Y. - Oxford, 1998).
22 N. Resher, The Coherence Theory of Truth, p. 181.
23 W. Salmon, 'Scientific Explanation: Three Basic Concepts', in: Causality and Explanation.
24 M. Friedman, Explanation and Scientific Understanding, Journal of Philosophy, 71 (1974), pp. 5-19.
25 P. Kitcher, 'Explanatory Unification and the Causal Structure of the World', in: Scientific Explanation, vol. 13, Minnesota Studies in the Philosophy of Sciences, eds. P. Kitcher and W. Salmon (Minneapolis, 1989), pp. 410-505.
26 C.G. Hempel, Philosophy of Natural Science, (Englewood Cliffs, N.J., 1967), p.83.
27 G. Harman, The Inference to the Best Explanation, Philosophical Review, 74 (1965), pp. 88 - 95; G. Harman, Thought, (Princeton, N.J., 1973), pp. 155-173.
28 L. BonJour, The Structure of Empirical Knowledge, (Cambridge, 1985), pp. 93-100.
Back to Maxim Lebedev's homepage