On the question of the substance of ideas
by David Foss
If ideas are representational with respect to discernible world-objects, then it seems that there must exist at least as many “ideas” as there are conceivable world-objects. This does not, of course, restrict ideas to only those arrangements of matter physically possible; but to the realm of thinkable entire. Furthermore, the content of any given idea, taken as its representational form, must be sufficiently articulate to singularly reflect that world-object (or transworld-object in the case of unworldly or fictional objects) it represents. Such a form, moreover, cannot depend upon the actual existence of its object to adequately differentiate it from distinct [representational] ideas, or adequately characterize it as a determinate component of the thoughts of which it is a part. Such an object, as should be clear, may not really exist.
There have, admittedly, been those who suppose that we cannot possess ideas of things which are not (in reality) simple objects of our world, or some mixture of the simple ideas of them. The suggestion has, on occasion, been that any conception we have of unreal objects is, necessarily, a compound of many simple ideas, each of which really does code to something real. This raises two important questions.
First, what are these “simple” ideas? They cannot be anything like the “idea of a horse (in general)” or the “idea of walking.” For these are clearly complex notions, or conceptions, made up of simpler, and more general concepts. Nor can they be anything like the “idea of that horse,” and so on, for similar reasons (i.e., such ideas are simply too complex to be considered irreducible in some ideational sense).
Second, such an insistence that simple ideas fully map to discernible real objects ignores the fact that the very discernment of real objects, in the sense of “natural kinds” or the individuation of “natural” entities, is itself problematic, and very often dependent upon the ideas we employ to individuate them. There is a sort of vicious circularity here: the ideas are individuated by “natural objects,” which, in turn, are individuated by the ideas we have of them. Examples of this last point abound in the history of science, and I do not think it necessary to consider them here. The point should simply be recognized that, whatever an ideational theory takes as the basis upon which distinct ideas are individuated, it cannot be the material particularity of each idea’s representational extension.
Granting this, and the general point that the number of simple ideas can now expand beyond material fact or possibility, we are left with a somewhat startling result. Leaving aside the matter of individuation, we should notice that such a proliferation of simples is not even restricted by what we take to currently be conceivable. Indeed, if ideas exist, in any sense broader than being present to some consciousness or other, then it is the category of conceivable which is dependent upon them, and not they who are dependent upon the constraints we might impose by the occurrent limits of our imaginative resources. Unfortunately, the loss of these two constraints, occurrent material or conceptual possibility, means that the individuation of candidate simple ideas becomes terribly difficult. We can no longer blindly appeal to material or imaginative indiscernibility as evidence that two idea-tokens instantiate the same idea-type. While this may not be so obviously true introspectively, it is almost undeniable with respect to ideational comparisons between persons. To what degree do I mean by “red” what you mean? If the answer ultimately stipulates that we will mean the same thing if, and only if, “red” for each of us expresses the same idea, then there seems little hope of ever insuring such sameness without appeal to material differentiations or current (accidental) conceptual limits. This is only to say that the appeal to “ideas,” in order to explain “sameness in meaning,” does not actually explain anything if there is no formal (i.e., intelligible) content to those ideas.
Such questions are not intended to refute ideational theories of meaning. Rather, they are intended to cast doubt upon one of the principle merits an ideational theory is thought to have. In particular, one of the most attractive reasons to suppose that language (or meaning) ultimately concerns the expression of proto-linguistic mental entities (ideas), which are themselves only indirectly visible (or known) through the speech acts they give rise to, is that, with “ideas,” we are apt to feel as though we have a unambiguous handle on meaning-components. That is, if meaning ultimately concerns the proprietary mapping of particular phonetic sequences (or other iconic sequences) to particular idea-clusters, then we need only discern the idea-cluster appropriate to any given act to determine its meaningful content.
To consider a particular sort of problem in meaning: in order to determine whether Lee understood what Walter meant by an utterance “p”, we need only confirm that Lee appropriates an idea-cluster by Walter’s utterance of “p”, where Walter intended by “p” to express the idea-cluster . In simple language, Lee understood what Walter said, if and only if, Walter’s utterance incites in Lee the idea Walter was expressing. This seems a rather clean and realistic way in which to model an act of communication. It certainly is a natural way to think about such acts. The ideational model may indeed be less popular than is otherwise might be among philosophers precisely because it seems such a natural way to think about such acts. Common sense has a mysterious way of offending critical minds.
Nevertheless, ideational theories of meaning are currently witnessing something of a revival. The revival is borne, in part, of a general “recognition” that modern efforts to view language in purely behavioristic terms, or (conversely) in purely propositional terms, have largely failed. It is also borne of a resurgence in linguistics of rationalistic a priorism, in the form of transformational grammar, which seeks to distinguish between surface structure (the arrangement of phonetic, word, or material sequences, taken as a linguistic utterance), and deep structure (the generative mechanisms which produce and interpret the structures of meaning underlying novel speech acts). The first point, concerning the general failure of modern empiricist linguistics and descriptivist theories of meaning in philosophy, is admittedly a largely rhetorical claim. Behavioristic linguistics (as one of the more prominent segments of the empiricist school), from the works of Skinner and his disciples to those of early Wittgenstein, have very rarely been entirely behavioristic. That is, the claim that behaviorism refuses to admit any notion of the expressive aspects of language use is popular primarily among its opponents (with a few notable exceptions). Descriptivist approaches to the analysis of meaning, from the work of the Vienna Circle to Russell, frequently does not pretend to offer an analysis of all forms of meaning, and is not entirely antagonistic to ideational addenda. Still, ideas (as meaningful components of linguistic analysis) have been largely ignored by these traditions.
I. Meaning by ideas
So what is an ideational theory? I take as essential to our garden variety ideational theory the claim that sentential meaning is (to a large extent) analyzable in terms of component word meanings. Word meanings (and on occasion, word-cluster meanings) in turn, are analyzable in terms of their proprietary expressive content; that is, in terms of the ideas they conventionally express. Some words express simple ideas. But more often, a word will express a cluster of ideas; or what we might call (following Wayne Davis) conceptions. None of this excludes a very complex picture about utterances.
The expressive content of an utterance may or may not conform precisely to conventional word or sentence meanings. Indeed, a great deal of conversational colour depends both upon there being a fairly clear conventional content to our words and sentences, as well as there being comported by our delivery a certain divergence from the norm. In addition, none of this excludes a very sophisticated picture of conventional comportment. The precise alignment of ideas with words (and their various arrangements) may be fairly rigid, or extremely fluid. An ideational theory need not suppose that ideas map to particular words universally or essentially. The only requirement here is that the conventional content of a word be conventional. That is, the expressive content of a word will be that idea or conception which is dominantly understood (among speakers in the language) by the use of that word. How this dominant understanding is formed or informed, and the relevant senses of word-use in question here, may vary dramatically between ideational theorists, and are not points which particularly characterize ideational theories of meaning. Let us say then, recognizing the degree of simplification here, that an ideational theory will make two important claims: sentences get their meaning (entirely) from the words they contain (compounded by their grammatical arrangement); and words get their meaning from the ideas they conventionally express.
The precise sense of “meaning” in this context is important too, and deserves some elaboration; but I will not attempt a formal definition. Rather, it should suffice to simply state that this is a sense of meaning predicable of symbols directly. It is the sort of meaning we appeal to when we state, “x” means p (of a symbol or icon x). We might say, in the ideational frame, “x” means p iff. an appearance of x represents (or “expresses”) p (where p is an idea or idea-cluster). But even here, I do not wish to state this relationship as an absolute or rigid correspondence. The declarative claim should rather stand as a rough model for the sort of meaning at issue for words, sentences, and the analytic device of an “idea.”
The sorts of meaning not considered central to an ideational semantics, but which form a central concern for other models of word and sentence meaning, include (but are not limited to) sense of “to mean” which refer to intention, or to inferential role at large. An ideational theory seeks to analyze such notions in terms of word meaning. “I meant to drive East,” “A robin means that spring is on the way,” and sentences of the like, include senses of “meaning” which, for the ideational theorist, are either nonlinguistic (in the case the “meaning” of natural signs) or in need of extensive (eliminative) analysis.
In order that ideas may function in this way, in order that they provide a sort of anchor for the meanings of higher order expressions and relations, each idea must be discernible: that is, each idea must be sufficiently articulate to appear to the mind both unique (each idea is exclusively self identical) and nonempty (each idea is cognitively distinguishable from non-identical ideas). Furthermore, each idea must stand in some relation to other ideas (no idea can appear alone) as a constituent of thought. The reasons for this are rather straightforward, but somewhat tedious to rehearse. Let us simply notice that, supposing the idea of a cat to be punctate, this idea can never appear to a mind without being accompanied by some other idea, to which it stands in relation: that cat, brown cat, cat (and the image of a small furry creature), and so on.
I have suggested that the idea of a cat, in the above example, should be punctate, in an important sense. That is, it should depend upon no idea other than itself for its ontological vitality. It is self-evidently, or for the most part self-evidently, an idea of a cat. It is important to notice that this sort of atomism, the manner in which each simple idea is punctate (if indeed it is), is one in which the individuation of each idea does not depend upon the individuation of its representative content.
To see this point, let us suppose that ideas are individuated (metaphysically or ontologically) by the objects they represent. Unfortunately (as I stated above), we find that the objects they represent are individuated (empirically and epistemically) by the ideas that we use to identify them. This is a problem. While not a precise circle, passing as it does between two forms of individuation, the circle becomes vicious when we realize that the certainty with which we can assert the ontological status of a given idea (the validity of asserting the existence of a given idea) is dependent upon the empirical or epistemic grounds for asserting the objective existence of its representational content. That is, the validity of holding that I possess an idea of a horse (i.e., that my idea is of a horse, and not of some other bizarre collection of attributes, images, relations, or what not), where that idea is sufficiently articulate to be individuated from other ideas, is [here] dependent upon empirical or epistemic criteria (normally applied to our acquaintance with that horse, or this cow, in real life). And this is a direct circle, at least in terms of our effort to be justified in holding that my idea of a horse, is an idea of a horse.
So let us say that ideas are not individuated by their representational content. That is, let us suppose that ideas simply are distinct. It is by the aggregation of irreducibly distinct ideas that we acquire different concepts and form novel thoughts.
The matter of each idea being punctate (in this way) is one which shall consume the remainder of this paper. It is normally thought that ideas are precisely the sort of metaphysically punctate components of meaning needed to avoid the spurious consequences of meaning holism. Ideas, as atoms in our semantic universe, are regarded by some as providing the ideal foundation of meaning: from explaining the occurrence of subsentential meaning, to providing a clean semantic model for the emergent science of the mind. But there are some problems with an atomistic semantics, even when borne of an ideational theory of meaning.
There are two obvious realms in which such an ideational theory seems to give us a great deal of help. Semantically, a theory of [punctate] ideas seems to offer a powerful analysis (in its simplicity) of synonymity. And pragmatically, it seems to promise a mental ontology which neuroscience will find congenial to the physics of the brain: it offers a model of thought particularly well suited to material analyses of the mind. Unfortunately, so long as we consider these ideas ontologically (or epistemically) punctate, these two benefits (and their analytic cousins) remain precariously just beyond our reach.
II. Synonymity is but a likeness
Synonymity appears to be a deceptively straightforward relation for an ideational theory. An utterance or expression “p” is synonymous with “q” iff. “p” and “q” express the same idea. This may appear to merely beg the question. (What is it, after all, to “express the same idea”?) And it may appear, as has been suggested frequently by the opponents of such a theory (notably Quine and others), that synonymity has thus been rendered an empty relation. While I sympathize with each of these “objections”, I do not think that they significantly challenge the coherence of such a definition of synonymity. In the first case, we might say that our definition is only part of the story (the remainder being supplied by the ontology of ideas). In the second, we might bite the bullet, and admit, “if no two terms in a language express the very same idea, then there are no proper synonyms in that language.” There does not seem to be anything terribly wrong with such a move. But it does begin to suggest the sort of difficulty I do have with this analysis of synonymity. Strict synonymity is not a very interesting semantic phenomenon. If we tailor our semantics to suit the special problem of explaining when two symbols or signs stand for the same meaning, then it becomes increasingly difficult to explain approximate sameness, or sameness-under-the-circumstances. This is not a problem unique to the ideational approach, but it is particularly acute here, where there are fewer avenues of escape.
Simile, not synonymity, is the true challenge any theory of meaning must face. Too much attention to “identical meaning” (when there is almost no such thing across differs words or sentences), has obscured our ability to explain similarities in meaning. Meaning comes in subtle shades, not in black and white. A theory of meaning must attend to the full palette of a language, and not obsess with the peculiarities of strangely hypothetical phenomena. But, in truth, most approaches to “similar meaning” appeal to an embedded notion of “identical meaning.”
Assignments of “same proposition,” or same meaning, always occurs (in real life anyway) with respect to some specific criterion or purpose. “She meant by ‘The drunk man in the corner,’ the same as ‘John Gunkle’” can be true even if John is not drunk at the time of her utterance, precisely because our assessment of “sameness” may not attend to the full semantic content of her utterance; we may ignore, for the purposes of our interpretation of her speech act, any part of that act which is considered extrinsic to her “point”. Such a phenomenon is well documented in the realm of speaker meaning (most significantly in the dispute between Donnellan and Kripke, on the matter of “attributive/referential use” and “speaker’s/semantic reference”). It will be a matter of some concern for any theory which defines semantic meaning in terms of speaker meanings (even taken aggregately or proprietarily).
As Kripke, in “Speaker’s Reference and Semantic Reference,” makes clear, the problems associated with any theory of speaker’s meaning, given these peculiarities of intention and interpretation, are not mortally catastrophic (or need not be) for most theories of semantic meaning (even where such meaning ultimately refers to speaker meaning). The concern, nevertheless real, is that semantic content frequently transcends the immediate grasp of either the speaker’s or her audience’s awareness: and on such occasions, often will not only fail to obscure the meaning of the utterance, but may in fact facilitate its clarity. We allow for substantial slips of the tongue. We are generous in our interpretations (at least when speaking in non-philosophical company). If we attend too closely to the semantic meaning (understood as the appropriate representational content) of an utterance, it is probable that we shall be guilty of excessive literalism. There is much more to the meaning of the terms and expressions we use [in daily conversation] than we do or should attend to on a regular basis. Given all this, it should be clear that synonymity, if it is the strict relation of sameness in semantic meaning, may be a very rare bird indeed.
Returning to the matter of synonymity then, we should notice that a full “synonymity objection” is not that there are no synonyms or that we cannot know when we have found them, but that such synonymity is not an interesting linguistic phenomenon (if it is one at all). We almost never mean by two distinct terms, words, or sentences exactly the same thing (whether that thing is a proposition, an idea, or an inferential role). Most obviously, they are, after all, different sounds or symbolic arrangements. But a greater difference still, lies in each use: the contexts in which most of our utterances occur, contexts which effect the generative and interpretive principles surrounding those utterances, shift in a myriad of ways, bringing about linguistic applications of striking originality and novelty even where the utterance itself is markedly cliche. These differences amount to something in our interpretation of the utterance; and insofar as such “use” and “interpretation” account for the semantic content of that utterance, it amounts to something in the correct interpretation of the linguistic production (i.e., the semantic content of the utterance). We may dismiss the difference as irrelevant to our current purpose, but this does not make the difference vanish. We are speaking here of the ontological content of utterances, and not merely of the contingent interpretive refinement a given listener may correctly apply.
An obsessive attention to absolute synonymity neglects the sense in which we normally understand synonyms to be “approximately” identical in meaning. If the relation of synonymity exists only where two terms express the very same idea (or have the very same meaning under all proprietary interpretations), then most terms we customarily take to be “synonyms” are simply not synonyms. But then what are they? How are we to make sense of the notion of “expressing approximately the same idea,” or “expressing similar ideas”?
One answer readily presents itself given the existence of ideas: ideas are utterly distinct, discrete, and unrelated to one another. Similarity in meaning only occurs between complex arrangements of many simple ideas. Words represent concepts, and concepts may be more or less identical in meaning depending upon the degree to which they are composed of the same simple ideas.
There is, unfortunately, a significant price to pay for such a move. We must give up the notion that we can possess a simple idea of a dog, or a cat, or whatever, in isolation. That is, we will not be able to think, express, hear, or otherwise cognate an idea as that idea. Simple ideas are not going to be perceivable apart from the clusters in which they occur. More importantly, simple ideas will only be perceived indirectly as part of a sort of constituent ensemble of synonymity relations. We cannot directly perceive, or epistemically detect, a component of meaning which is itself without structure or form. This is indeed a heavy price to pay for insisting upon metaphysically punctate ideas. For now there is no way to distinguish the presence of an idea from the points of similitude between two linguistic constructions.
In fact, this is precisely what such an idea is supposed to be. But why suppose that we need ideas at all here. Various sound-clusters (in normal phonetic communication for example) stand in certain relations with one another, and in certain other relations with a variety of behavioral affects, and it might seem that the positing of ideas at critical intersections of continuity or regularity amounts to little more than a convenient heuristic device. The existence of points of semantic similitude is still safe, so we might still say that ideas “exist”; but it now seems as though we might suppose that it is the existence of the whole fabric of utterances and behavior regulating devices among humans (the existence of the whole language) that grounds the existence of ideas, and not the other way around.
Punctate ideas, in themselves, seem to give us little guidance in the matter of “resonance,” without appeal to extravagant networks of [extrinsic] associations (which then become the authentic bearer’s of semantic ‘content’). That there is such a thing as “resonance” among various clusters of words and meanings cannot be seriously disputed. If such resonance is to be explained by the presence of semantically punctate ideas (appropriately arranged), then it will be difficult, if not impossible, to distinguish between the macro-level semantics brought about by any given idea, and the macro-level semantics brought about by just those extended associative ties the name of that idea signals.
Unfortunately, now the ontology of each idea is threatened. They are like placeholders on the semantic landscape, points of intersection between large clusters of relations (phonetic, imaginative, or whatever), as ontologically secure as the geometer’s points in space or the geographer’s degree of longitude. They become mere “positions” on a semantic grid. Surely such positions exist; but it is far from obvious that their existence is not contingent upon the “surfaces” they “define”. This is troubling when we recall that one of the driving forces behind accepting a ideational theory of meaning is that ideas should be considered ontologically prior to, and capable of explaining (both semantically and metaphysically), the concepts in which they occur.
There are two avenues which an ideational theory can take here. The first is to emphasize the ontological claim: the presence of ideas is needed to explain the phenomenon of [meaningful] thinking. Certainly, something is needed to explain how it is that a human brain, appropriately situated and trained, cannot only produce meaningful utterances for the entertainment of others, but can entertain meaningful thoughts to itself. Modern neuroscience has enough difficulty working out the details of the rather mundane tasks of sense perception, memory, and motor skills. As we step away from the more clearly physical aspects of cognition, to the more logical or “meaningful”, it becomes obvious that we will need a robust model of conceptual life particularly open to material analysis (with a minimum degree of compromise on either side of the mental/physical divide). Ideas appear to supply just the right sort of model.
The second avenue, somewhat less attractive to many a die-hard ideationalist, is to entertain a sort of semantic molecularism; ideas are not entirely punctate (semantically), but enable us to entertain a sort of semantic connectionism short of full semantic holism. Here, however, it will become even more difficult to tell just what an idea is (as a component of meaning), if ideas are going to constitute (at least in part) the bedrock of our semantics.
Let us first look at the possibility that a theory of ideas can give us a firm ontology of meaning, susceptible to a material analysis.
III. Ideas as Brain States
The language of ideas is natively the language of deep structure. Noam Chomsky’s resurrection of “Cartesian Linguistics” is a resurrection of the claim that the generative principles of speech, the principles which organize and produce meanings, are only indirectly observable, and can not be equated with the “surface structure” of a language (the grammar of a given natural language). Ideas are [traditionally] the primitive units of deep structure analysis; much as words have been in surface structure analysis. But we should not be mislead by the parallel.
The advantage, or perhaps the danger, of viewing linguistic phenomena in terms of surface and deep structure is that deep structure need no longer look much like the surface structure. We need not carve up the deep structure (the structure of meaning itself) in terms of words, phrases, and sentences. The notion of a word, of a static meaning-object, is unlikely to find a strict analogue in the world of the brain. Ideas, if they map to any real physiological phenomena, will likely individuate broad regularities in the procedural, massively parallel, mapping of occurrent stimuli and brain “states” to linguistic behavior (either in speech production or interpretation).
This will not be a simple mapping of terms to states (or vice versa). Indeed, there will be nothing “simple” about such a mapping. How we carve up the world of nervous-system regularities seems even more dependent upon our occurrent vocabulary than our carving up the material conditions around us. The isolation of meaning components in the wildly dynamic world of the brain strikes me as not only doomed to failure, but pointless. Different brains will likely exhibit differences in deep structure (if we take this as synonymous with brain function) even where we are typically reluctant to attribute differences in meaning. This is trivially demonstrated by cases of childhood brain damage, where the individual has been denied the use of some region or portion of the brain normally responsible for speech, and as a result has passed these duties to other portions of the brain. We do not suppose that these individuals speak gibberish when they produce speech from regions of the brain normally not responsible for such acts. There is something right in the functionalist’s account of the mind here. It is the procedural regularities in linguistic behavior which enable us to attribute meanings (and ideas) to ourselves and others.
Moreover, if by “deep structure” we mean these regularities, then we have an added burden of making sense of dynamic processes in terms rigid enough to identify (or fix) semantic equivalence classes, while not so rigid as to render static the processes collectively named. We must be careful not to imagine that the classes we identify, by the grouping of various regularities, identify “states” or “moments” of meaning. A neurophysiological regularity (on a scale appropriate to the individuation of components of meaning) is no more a state or “thing” than a C-Sharp. Notice that a C-Sharp is not an entity, in the conventional sense of an “entity.” In order that a C-Sharp be a C-Sharp it must have duration, it must have been produced by the appropriate sort of musical or sonic instrument, and (somewhat controversially) it must be interpreted by the appropriate sort of listening device. If any one of these three elements is missing, a C-Sharp has simply not occurred. Why should we suppose that ideas, or meaning, must be markedly different? I rather like to think of music as “language without words”, and it is here that the resemblance is most striking.
But if ideas are primitives in our neurological universe, if they are the simple elements of our ontology, then they will not look very much like words. They will not be discrete. They will not even be unambiguously simple. And this means that, in our effort to define “ideas” in terms open to material analysis, we have reduced their semantic utility. The problem will have moved from one of mapping brain function to ideational content (where idea/word relations are relatively straightforward); to one of mapping ideas to surface structure, i.e., words and sentences (where there is nothing straightforward about it).
A linguistics with “ideas” will therefore not present a significant advantage over a linguistics without them, if we are differentiating the two categories of theory strictly in terms of ontological vitality. But what of the other avenue named above: that the language of ideas may remain vital to linguistic analysis if we relax the degree to which we consider each idea semantically punctate. To see that this is even possible, we should review the argument that any anatomistic theory (which a theory of non-punctate ideas would have to be) is a holistic theory of meaning. Indeed, we should clarify what a punctate or holistic theory of meaning is, in the context of ideas.
IV. The Anathema of Anatomism
Jerry Fodor and Ernest Lepore’s Holism presents a sustained attack on the coherence of semantic holism (at least in those varieties of holism which have been, of late, popular among philosophers of language). The authors distinguish between atomic properties (those properties capable of being instantiated in one thing) and anatomic properties (those for which, if one thing has it, then at least one nonidentical thing has it also). Much of their assault on [extant varieties of] anatomic semantics is premised by the claim that, in matters of meaning, anatomicity entails holism. That is, if meaning is essentially a relational phenomenon, not only between ideational (or propositional) relata, but between other meaning relations (i.e., if meaning is essentially anatomic), then meaning is essentially holistic (i.e., the meaning of any single expression is dependent upon meaning relations across the whole language). This is not intended by the authors to be a controversial thesis. The term “holism” is frequently applied by the advocates of semantic anatomism, whether it be in the work of Quine, Sellars, the followers of Frege, or of the structuralists in linguistics.
Why should pervasive holism in our semantic theory bother us? As the authors point out, languages rarely seem to behave as though they were holistic. We frequently allow children to know what is meant by, for example, “The red ball is imaginary,” even if they do not know what is meant by, “The ‘Red Menace’ is imaginary.” This is true even if the notions of red are linked (or associated) rather strongly in the popular imaginations of most Americans. Indeed, this is true even where the two connotations of red (the color and the political affiliation) are strongly associated in proprietary usage (i.e., the semantic content of ‘red’ includes the minor “accessory” idea of, or association with, ‘communist’). In other words, we allow children and speakers learning English to “know” the meaning of ‘red’, even when they are not fully acquainted with the term’s range of nuanced meanings in American English.
There are, of course, even stranger formulations of the “consequence” of holism, which entails that to know, believe, understand, or think the meaning of any term or expression in the language, one must know, believe, understand, etc., the language as a whole. Fodor and Lepore offer one such formulation:
... consider the property T* which a belief has iff it expresses a proposition that is the content of some belief of mine. According to the present assumptions, if T* is anatomic, then it is holistic. And if T* is holistic, then (assuming that thoughts are individuated by their propositional contents) it might turn out that nobody has thoughts that are tokens of the same type as my thought about Auntie’s pen [“The pen of my aunt is on the table”] unless he also has thoughts that are tokens of the same type as, as it might be, my thought that the cat is on the mat, my thought that black holes are odd kinds of objects, my thought that some presidents are wimps, or my thought that Salome will never sell in Omaha.
On the face of it, none of this is terribly counter-intuitive (although the authors suggest otherwise). For no one will quite perceive the same nuanced meaning of my utterance “The pen of my aunt is on the table” unless she shares almost all of my beliefs. That is, a listener may catch a lot of the meaning I intend, and indeed, my understand all of the underlying logical and grammatical content of the utterance, as well as most of its conventional portent, so long as she also shares my acquaintance with those logical and grammatical assumptions, as well as a passing awareness of the conventional (or transformational) connotations of the terms; but she will not understand the utterance in its nuanced entirety, as I understand it in the act of speaking; just as the child does not fully “understand” the meaning of ‘red’ (in American English) prior to associating the concept of communism with the color.
Does this mean that no content has been communicated (in the case of a casual listener) or learned (in the case of the child)? Obviously not. Even if holism is true, it would not follow that no content could be communicated without all content being communicated. Only that most content must be communicated (or shared) in order that any be. And this may be true, if we recognize that significant portions of the content are the logical and grammatical rules, as well as the relevant “transformational rules” which make this language the one it is.
But a further confusion clouds the diagnosis here. The authors confuse semantic holism with belief holism. My beliefs may contribute to the substantial nuancing of a particular sentence. But are we thereby entitled to claim that my beliefs are those which determine the meaning of my utterances? I think not. Most of my speech is filled with deferential use of terms and concepts of which I only have a vague acquaintance. I rely, to a large extent, upon the linguistic competence of my audience to supply the full meaning of even mundane terms like “competence,” “audience,” “full,” and so on, which, though I believe myself to understand fairly well, never suppose that I “know” exhaustively. My beliefs only roughly approximate the meanings of terms I use. Even the propositional integrity of my thought depends substantially upon conceptual components which exceed my total comprehension (in the sense of there being occurrences of the ideas, the copula, the subject-predicate form, etc., which have never occurred to me, but which I must take for granted in order that my thought be intelligible even to myself). This, then, is not so much an argument against holism, as it is an argument against confusing belief and semantic holism. Even in those cases in which I think I know what I am saying, there is likely to be more to the possible (valid) interpretations of my utterance than I am aware.
Keeping in mind such a distinction, it is quite plausible that semantic holism may be true even if belief holism is absurd, or merely false. What makes the distinction possible is a reading of semantic content which is not directly correlated to the belief content of an individual speaker, even on the occasion of that speaker’s production of a meaningful expression. This may be achieved in the ideational context by moving semantic content away from the ideas “commonly” or “predominantly” expressed by the use of some term or sign; and toward the ideas “proprietarily” or “customarily” expressed by the use of that term or sign. The added normative and majoritarian components of mapping symbols to meanings are essential for establishing a gap between direct speaker meaning and semantic meaning.
“Holism” seems an exaggerated charge against a semantic theory which merely insists that meaning is not a punctate property. But even if such a charge were appropriate, even if semantic anatomicity entailed semantic holism, the truth of semantic holism is not sufficient to establish some sort of “belief holism”. There is a justifiable reluctance to think that in order for you to understand what I mean, you must share (or at least be acquainted with) all of my beliefs. And if semantic holism necessarily entailed that local meaning (the meaning of a term or phrase) could only be communicated if the speaker’s comprehension of that term (taken as the system of associated terms, ideas, images, etc.) were also communicated, then it would seem that authentic communication almost never occurs.
We rely substantially upon extrapersonal semantic properties when interpreting any utterance. We assume that a speaker is using her or his terms, expressions, and gestures in roughly conventional ways. And although such conventional information ultimately refers back to persons, it is not the individual behavior of each speaker, or the internal belief systems of each individual speaker, which define proprietary content. Indeed, it is not mere belief system content at all; but rather the dominant set of expectations and deferments each speaker has in mind by the utterance of a given expression.
I do not suppose that I fully grasp the meaning of “dog”, even though I may be quite familiar with canine behavior, and consider myself well versed in the ironic or sarcastic uses of the term. I consider it sufficient, in the matter of communicating the idea of a dog, that the speaker and hearer share an appreciation for what is presently important among its various attributes or meanings. Measuring the degree of “shared belief” or shared normative expectations necessary for such communication is notoriously difficult to establish. The authors note, specifically with regard to an inferential role semantics:
... [A further] problem with reconstructing similarity-of-beliefs- entertained by reference to similarity-of-inferences-endorsed is that some inferences have to count for more than others, surely. ... So now we need to know how much the differences between [object]- inferences I endorse and the ones that [another speaker] did count as differences in concept of [that object]. The extent to which this sort of question lacks a principled answer is the extent to which we have no notion of similarity of content that is compatible with a holistic account of belief attribution.
Significantly, however, there is a fundamental premise which should not be taken so much for granted. The individuation of beliefs (or inferences endorsed) themselves depend upon a special sort of widespread linguistic agreement (not “total linguistic agreement”). Once such individuation has been successful, it is almost trivial to claim a relevant notion of “similarity” has been borne out. This individuation appears to largely be the result, not of two speakers possessing isomorphic lists of beliefs-as-propositions, but of the particular utterance or symbol in question being subject to roughly equivalent sets of challenges, interlocutors, or behavioral “inferences.” And this “roughly” is precisely defined as those challenges, interlocutors, or behavioral inferences seen as immediately salient to the situation at hand. I do not care whether my utterance of the word “red” makes you think of apples first or tomatoes, when I use it in the context of describing the color of a traffic-light; The only salient inferences, the only inference-making tendencies relevant to the matter of you and I thinking the same thing by red, are those which you and I both need to have qua the traffic-light. Why is this so fuzzy a thesis? Again, “holism” (if we are to call this sort of thesis one of holism) only requires “broad” agreement, in a situationally localized fashion, and not “total” agreement, in order that “same idea, belief, or inference” make sense.
All of this raises the particular possibility that a view in which “ideas” are less than punctate (semantically) will not be sheer nonsense. One can, heeding the warnings above, entertain a sort of semantic connectionism. Ideas will have a role to play in such a semantics to the degree that they are not considered the particular conceptual mechanism which “rescues” us from the ‘insidious’ consequences of holism. Indeed, the manner in which they are traditionally thought to offer a punctate analysis of semantic phenomena is seriously flawed.
A rather immediate (albeit limited) sense in which this can be seen is in our analysis of the evolution of meaning. That is, semantically punctate ideas give us a rather strange reading of etymological development. The meanings of terms change over time. We often say of particular ideas that they change over time: whether they are the “ideas” we ourselves possess, or the “ideas” conventionally expressed by a given term or linguistic gesture. Unfortunately, there is no easy way to say that an idea has changed, if that idea is semantically punctate.
Niels Bohr’s idea of the atom prior to 1920, and his idea of the atom after 1920, are going to either be the same idea (so how could they mean, or represent, different things) or utterly distinct ideas (so how could they have designated and explained, to Bohr, the same thing). We might read Bohr’s atom as a concept; in which case it might be the associated class of constituent ideas which shifts as the “atom” matures in his understanding. But now it seems that the associations are doing the semantic work. Simply put, as intersections in the meanings of terms shift, are we to suppose that our concepts come to embody new sets of ideas, or that the “ideas” themselves have changed? I see little gained by a blind insistence that any change in meaning involves the selection of a new set of constituent ideas; especially if such “ideas” turn out to be rather vacuous (semantically) in themselves.
Meaning appears, under such conditions, to be a phenomenon of the connections, associations, or mere relations among and between ideas. Ideas, if they are the sub-semantic entities, themselves quite meaningless, which we posit in order to explain the more systematic occurrence of meaning in thought and language, no longer look a great deal like the objects we conventionally call “ideas.” Ordinarily, the term idea (even used technically) applies to the smallest units of meaningful analysis. In a theoretical or common sensical view of language, which considers our spoken utterances or written phrases expressions of implicit mental happenings, an idea will be a simple component of that happening. If ideas are sub-semantic however, then the “happenings” of which they can be simple parts are not meaningful per se, but occur only metaphysically prior to an actual occurrence of meaning.
But what are these ideas then? Are they even part of the deep structure? Deep structure, in Chomsky’s sense, is essentially semantic, even where it is not grammatical in the same manner in which a natural language will be.
Ideas begin to look like a rather arbitrary device if they are positioned prior to the semantic phenomenon they were supposed to explain. Worse yet, my “idea of a pig” will no longer be an idea, properly understood. The content, the very meaning, of this “idea [of a pig]” will be fixed by the fabric of association relations the metaphysical idea is said to hold with respect to a battery of other ideas, and not by any intrinsic property of the idea itself. It will be these relations, and not the presence or absence of this [metaphysical] idea which establishes a synonymity condition between two words, or two speakers. A significant reason for supposing there are such ideas is now substantially undermined. They do not help us resolve the matter of “same meaning”. It simply will not do to claim that “p” and “q” mean the same thing iff “p” and “q” express the same idea.
Ideas, if they are to be relevant to semantic analysis, must have content. They must have meaning. And I do believe ideas to be a powerful semantic tool; i.e., that they should be taken as being meaningful. There are tremendous advantages for linguistic analysis if we consider the form and treatment of utterances as expressive of some generative process, some deeper unspoken “language”, underlying the linguistic behavior of rational agents. Ideas can play a role in our understanding of proto-linguistic phenomena not unlike that played by words in our understanding of natural languages. But we must be careful to not confuse the analytic advantage brought by such a device with proof of the existence of thought-words (they are not precise analogues) nor the existence of punctate “atoms” in our semantics. If ideas correspond to the basic elements of meaning, then it will not be the “idea of a pig” which is associated with the ideas of pink, mud, a curly tail, and so on, but the associations with “pink,” “mud,” “a curly tail,” etc., which is the “idea” of a pig. And this, surely, cannot be confused with a claim about the basic physical elements of thought.
These ideas will neither be “punctate” semantically, nor holistic (in the sense in which a speaker need not know the proprietary associations ad nauseam of a given “idea” to properly use and interpret its expression). Languages are not learned quickly. Nor are they learned one word at a time (even if we think our children first acquire “mama”, before moving on to “maternal parent”). A great deal of behavioral and pseudo-linguistic regularities must be absorbed before the mind begins to correctly assemble linguistic information. Such a context may not seem straightforwardly linguistic in the sense of a broad acquaintance with words, phrases, grammatical constructions, etc., but neither is it the case that (1) individuals first learning a language are considered “competent” speakers, nor that (2) all that counts as deep structure should be a direct part of surface structure. In other words, as I warned above, there is no logical reason to suppose that the structure of thought is restricted to the grammatical structure of any given natural language; behavioral regularities almost certainly account for a significant portion of deep structure “data”, even where there is no direct correlate to such information in the individual’s surface vocabulary.
Only part of one’s semantics may be open to surface structure explication. I may be able to “explain” or “express” only a part (perhaps even a small part) of those associations which make a term, a phrase, or other expression, meaningful. It is these deep, often nonverbal and inexpressible, associative elements which enable “partial understanding” of an expression even where no associated “word” has yet been established. And it is precisely the mechanism of “ideas” in a semantic theory which can clarify and illuminate this sort of proto-linguistic phenomenon, where words and surface grammar are by definition void. It is precisely ideas, then, which can give us a sort of semantic molecularism, half way between the barren slopes of meaning as a punctate relation, and the bizarre jungle of meaning as a holistic relation.