Nick Bostrom
[email protected]
LSE and Stockholm University
July, 1995

Back to Main Page



The state of the art as regards the thesis of indeterminacy of translation is as follows. Very much has been said about it, most of which is based on misunderstandings. No satisfactory formulation of the thesis has been presented. No good argument has been given in favour of the thesis. No good argument has been advanced against it.

In this paper, I attempt to clear up some of the misunderstandings, to provide a satisfactory formulation of the thesis in non-naturalistic terms, to demonstrate how a naturalistic substitute can be derived from this formulation, to refute the best know arguments for and against the thesis, and to show how it relates to the thesis of indeterminacy of reference, the theses of semantic and epistemic holism and to the thesis of underdetermination of theory by data. Finally I argue that there is an interesting sense in which the indeterminacy is a matter of degree, and express my opinion that this degree is probably not very high.



1.1 Introduction

I shall argue later that no adequate formulation of the thesis of indeterminacy of translation has yet been presented. In this introductory section I shall make a brief uncritical presentation of the thesis for readers not acquainted with Quine's philosophy.

The thesis is that divergent translation manuals can be set up between natural languages such that they all are compatible with empirical facts but nevertheless diverge radically from each other in what sentences they prescribe as translations of sentences in the foreign language. Each manual works individually, but they cannot be used in alternation: the fusion of two of these manuals does not in general constitute a manual that is compatible with all empirical facts. The sentences (or anyway many of them) which the divergent manuals correlate to a foreign expression stand in no form of equivalence to each other, however loose.

This thesis should be distinguished from the thesis of indeterminacy of reference which is somehow analogous but is concerned with words instead of sentences. I will say more about the thesis of indeterminacy reference and the way it is related to the thesis of indeterminacy of translation in §3.1 and §2.1.4.

The thesis of indeterminacy of translation is not that it is hard to find out what foreign sentences mean, or that the evidence available to us, finite beings as we are, is always incomplete. It is rather that there isn't anything there to be found: meanings, interlinguistic well-defined meanings, do not exist: there is no fact of the matter as to which meaning a foreign sentence has of the alternatives attributed to it by the rival manuals.

From Quine's writings one gathers that the thesis of indeterminacy of translation is a protest against the uncritical appeal to meanings and analyticity that characterised the logical positivists. Quine speaks of the notion of meaning as a stumbling-block cleared away. The indeterminacy thesis paves the way for Quine's philosophy of science and of mathematics, whose back bone is semantic holism:

Where metaphysics had sought the essence of things, analytical philosophy as of G. E. Moore and after settled for the meanings of words; but still it was as if there were intrinsic meanings to be teased out rather than just fluctuant usage to be averaged out. In later years analyticity served Carnap in his philosophy of mathematics, explaining how mathematics could be meaningful despite lacking empirical content, and why it is necessarily true. However, holism settles both questions without appeal to analyticity. Holism lets mathematics share empirical content where it is applied, and it thus accounts for mathematical necessity by freedom of selection and the maxim of minimum mutilation.1


1.2 The indeterminacy thesis says something that implies that the notion of meaning is irremediably confused

It is clear that Quine takes indeterminacy of translation to have devastating consequences for the family of classic semantic notions. -meaning", "synonymy", "analyticity", "intention", "belief". These are all terms which must not be used as supporting blocks in any scientific theory, at least if they are understood in their traditional sense: as absolute, interlinguistic, discrete entities, the way the logical positivists and others would have it. However, there is to my knowledge no argument in Quine's work from indeterminacy of translation to the impermissibility of that terminology. Therefore it is reasonable to demand of any explication of the indeterminacy thesis that it provides an explicatum that makes this implication obvious. For to Quine it seems obvious; if some formulation of the indeterminacy thesis on analysis turned out to have no close connection whatsoever to the question of the substantiveness of the notions of meaning and analyticity, then it would be evidence that the thesis had been wrongly construed, especially if there exists an explication that provides an interesting proposition that does have such a bearing and in addition fits well into Quine's philosophy and agrees with most of what he has said about indeterminacy of translation. However, I am not very interested about the exegetic part of my claim and I don't care to argue it further. Call the formulation I provide in §1.5 and §1.6 an interpretation, an explication, or an alternative to Quine's thesis, whatever you like.

So the starting point for me is that the indeterminacy thesis says something to the effect that our notion of meaning is irremediably confused. Not quite, though: there could perhaps be other things wrong with that notion beside the large-scale macroindeterminacy Quine wants to draw our attention to. Even if the indeterminacy thesis is false, there is still the possibility that meanings could be indeterminate on a microscopic level. Suppose that no translation manual could assign radically different meanings to foreign sentences from the standard manual and still be correct; i.e. suppose there is what I call macrodeterminacy. Could there not nevertheless be some room for divergence in the fine details, i.e. microindeterminacy? For example, what facts determine that a speaker means that he is adding when he says "I am adding.", rather than that he is performing some freaky quus-function a la Kripke? Or that all occurrences of the sentence "That is a reindeer." have exactly the same meaning, the same inclusiveness as regards borderline instances, rather than just very similar meanings? For there to be a working notion of analyticity, these microproblems with the notion of meaning must be solvable. We want to say that the sentence "There exists a reindeer." is logically equivalent to the sentence "There exists a reindeer."; but this fails if the two tokens of the sentence have different meanings, however small the discrepancy. I think I know how these problems can be solved, under the assumption that Quine's indeterminacy thesis is false; but the manoeuvre is in no way obvious. Therefore it would not seem advisable to define the indeterminacy thesis as the thesis that our notions of meaning, analyticity, etc., do not make sense. If it is indeed equivalent to that proposition, then this is something that should come as a theorem, not a definition. I will not try to give any reasons that they are equivalent, not in this paper. However, what I am looking for as an explication of Quine's thesis is something that says in effect: "There can be no notion of meaning because every purported notion of meaning fails on the macrolevel."

Before we proceed, however, we have to agree upon which sort of meanings we are talking about, for "meaning" is an ambiguous word.


1.3 By "meaning" we understand: truth condition

In one of its senses, "meaning" means linguistic role or usage. Linguistic roles are floating, imprecise and naturalistic; they have no close relation to analyticity, and Quine has nothing against their notion. Indeed, he thinks it is the right substitute for the philosophers' meanings. (See e.g. §23 in Pursuit of Truth)

Is "meaning" to be understood as (possible world) truth conditions or as some kind of structured proposition? It is quite possible (prima facie at least) that there could be indeterminacy in the one sense and determinacy in the other. In fact it seems that a thesis of indeterminacy of structured propositions would be stronger than a thesis of indeterminacy of truth conditions. For, on the one hand, physical facts presumable show fewer manuals to conserve structured propositions than to conserve truth condition, so there would be more candidates to produce the "incoherent sequences" if only conservation of truth conditions (as far as they are revealed in physicalistic facts) were required than if the structure of the expressed proposition had to remain unaltered too. On the other hand, there would perhaps be no new possible "incoherent sequences" in the case of the thesis of indeterminacy of structured propositions that were not incoherent already as regards truth value; for the structure of the proposition that a sentence expresses is presumably naturalistically determined (although there can be doubts as regards sentences containing defined expressions). Anyhow, there is no reason to suppose that the two theses are equivalent.

Intensional structure (or something even more elaborate) rather than truth conditions seems to be what Quine had in mind when he wrote: "The meaning of a sentence of one language is what it shares with its translations in another language."2 Other statements, however, indicates that truth conditions are what is intended, in particular this interpretation seems to be vouched for by the presumption that the thesis should have direct relevance for the fate of the notion of analyticity. One possibility is that Quine does not make this distinction but aims his indeterminacy thesis against all classical semantic notions indiscriminately. This would be unfortunate, though, because, as we saw, the different versions seem to be of different strength. Alternatively, we might think that Quine has not found the notion of "intensional structure" worth bothering about; one is sympathetic to this attitude. But this does not silence the disharmony that arises from Quine's attempt to fusion the indeterminacy thesis with a proposition about practical translation. We return to this in §1.4.1.

There is yet another kind of meaning that should be distinguished from the kind of truth value meaning that the indeterminacy thesis is concerned with. I do not know that there is a special term for this kind of meanings, or indeed that the distinction has been explicitly drawn. I shall call them "thought meanings" and contrast them with "logistic meanings".

Both thought meanings and logistic meanings are truth conditions; the difference is that thought meanings are much more fine grained than logistic meanings. Whereas logistic meanings are known as being what is invoked to explain analyticity and mathematical deductions, it needs to be said what thought meanings are supposed to do. I do not know how to do this briefly, but I will make rough explanation in the following footnote (which you may skip if hurried).3


1.4 Dissatisfying formulations

Quine has on different occasions formulated his doctrine of indeterminacy differently: as a thesis about practical translation, as a thesis about manuals' usefulness, and as a thesis about conservation of stimulus meaning. I will analyse these three formulations in turn and argue that they are all unsatisfactory in that they do not have any direct consequences for the notion of analyticity, as was intended. (§1.4 is a critique over the appropriateness of certain formulations, but as I stated earlier, I make no claim as to whether Quine has formulated an interesting indeterminacy thesis incorrectly or an uninteresting thesis correctly, though I believe the former.)


1.4.1 Ordinary translation

When Quine presents his thesis he makes it seem as if it were a thesis about ordinary translation, but a little reflection suffices to realise that ordinary translation is not what he has in mind. For a translation to be correct in the ordinary sense, it has to obey a multitude of irrelevant constraints like conservation of phrase structure, frequency of use, emotional connotation, cumbersomeness of pronunciation etc., which are to be taken into account when the correctness of a translation is to be determined. When I say "irrelevant", I mean irrelevant for the issue of whether a manual conserves logistic meaning, i.e. truth conditions, or not. For example, would a manual that translated an official document issued by the French government about the costs to society of sex crimes and ways to prevent them be considered correct if it had all technical terms for money and kinds of sexual abuses replaced by corresponding colloquial English expressions in such a way as to make the text ridiculous and obscene? Would a translation of a foreign university-level textbook on mathematics be correct if all propositions were translated into "Three plus eight equals eleven"?

From these examples it is clear that a manual may conserve truth conditions yet be incorrect as a translation in the ordinary sense. So if the incompatibility of Quine's rival manuals consists only in that their fusion would be an incorrect translation manual in the ordinary sense of the word, this does not imply that they are incompatible in the relevant way, i.e. that their fusion does not conserve truth conditions. On the other hand, manuals that conserve truth conditions as far as that is determined by naturalistic criteria, and could candidate to be instances of indeterminacy of translation, will in many cases be ruled out because they are not correct as ordinary translations. They may equate sentences with different emotional connotation, for example. So there is no obvious relationship between indeterminacy of practical translation and indeterminacy of translation of logistic meanings. Thus it is unfortunate to give the thesis the appearance that it is a matter of practical translation.

As an aside, we may note that it seems plausible that there is at least a little indeterminacy of ordinary translation. For often, surely, a translator has two or more alternatives as to how to render a foreign expression that occurs repeatedly in a text; all alternatives may be acceptable, but only if the same expression is used consistently, throughout the text, to replace the nonvarying formulation in the original.


1.4.2 Useful translation

Sometimes Quine speaks as if usefulness were the criterion of manual correctness.4 A manual is said to be correct if it "promotes smoothness of conversation and success in negotiation". This is a property that comes in degrees, and an indeterminacy thesis that employed this notion of correctness would spell something like "There exist two manuals (between two languages) that are both very good at promoting conversation such that their fusion is very bad at doing that." As regards this formulation, let us first notice that when examining the usefulness of a manual, what we primarily look at is not a correlation of sentences, i.e. the abstract object of a set of ordered pairs, but rather the tokens which manifest this abstract object.

Consider a computer program that translates German into English. Suppose that it yields exactly the same output as a skilled human translator using the standard manual, except that it prefixes a "It is not the case that " to every sentence. Such a program, if you could invent it now, would make you a millionaire. It is certainly very useful, since it need only be supplemented by another simple program, or an uneducated secretary, that removes the prefix, in order to render a cheap and elegant translation of any document whatsoever.

This example shows that if the thesis is not to be trivialised (as Massey's manuals would do, if they were admitted as correct; see § then the utility condition must be construed so as to exclude the "It is not the case that"-manual from being useful. What sophistry could avail us here? Perhaps we might argue that in the example, what was really being used and found useful was not the "It is not the case"-manual but the standard manual, because the standard manual was what defined the way in which the speakers' dispositions were being correlated. In passing by, before we follow this suggestion to see where it leads us, let us mention that many irrelevant contingent factors are powerful determinators of a manual's utility. Just to name one: the standard manual has the great privilege that every other manual would force us to relearn. Such factors should in any case not be allowed to play a role in deciding which truth values a sentence has. (See §2.2.2.)

Now to correlating dispositions. This idea is interesting also apart from the attempt to block the reductio of the utility criterion. The idea is that what determines the correctness of a translation manual is the way it correlates dispositions. What we would like to say is that it is the conservation of thoughts or intentions associated with sentences that make manuals correct, but we must look for a kosher substitute. The naturalistic counterpart to intention is some kind of dispositions. But not any notion of disposition will do. People may have different dispositions and mean the same thing. And there are linguistic dispositions that concern features other than meaning, e.g. spelling etc.

A large part of Quine's work has investigated into exactly which sorts of language behaviour and dispositions might determine meanings, to the extent they are determined, and how. Let us therefore move on to that part and conclude that formulating the indeterminacy thesis in terms of manuals' usefulness is not a good idea.


1.4.3 Concrete examination of evidence available to translator

I think that the preceding two interpretations are not fair to Quine. His indeterminacy thesis is quite intricate, and it does not have the character of a logical or philosophical proposition, but is more like a rationalised description of the principles involved in actual translation and language learning.

In Word & Object as in later works, Quine presents his thesis by first describing what sort of evidence a radical translator has to go on and then proclaiming that this evidence is not enough to fix a unique correct manual. Hereby he sets forth a thesis which consists of two parts: (1) The claim that all relevant evidence has been described; (2) The claim that there are divergent manuals which are compatible with the described evidence.

If we interpret "divergent" as "differing in truth conditions", then this formulation would be quite acceptable (modulo the vagueness as to how divergent the manuals have to be) if the description of the relevant sorts of evidence were sufficiently precise. But Quine's description of the evidence available is not sufficiently precise. Quine does not make it clear what, exactly, counts as evidence for a manual's correctness; to the extent to which this is undefined, to that extent is the indeterminacy thesis vague.

Without going into details, let us quickly review what Quine has done to delimit the evidence that determines correctness. Translation of observation sentences is generally determined, because we all have an "uncanny knack for empathising another's perceptual situation"5:

The linguist notes the native's utterance of "Gavagai" where he, in the native's position, might have said "Rabbit". So he tries bandying "Gavagai" on occasions that would have prompted "Rabbit", and looks to natives for approval. Encouraged, he tentatively adopts "Rabbit" as translation.6

Translation of truth functional constructions of observation sentences are also relatively safe:

Our linguist then goes on tentatively identifying and translating observation sentences. Some of them are perhaps compounded of others of them, in ways hinting of our logical particles "and", "or", "but", "not". By collating the situations that command the native's assent to the compounds with the situations that command assent to the components, and similarly for dissent, the linguist gets a plausible line on such connectives.7

And then? What can the linguist do with the other sentences?

He can keep a record of these unconstrued sentences and dissect them. Some of the segments will have occurred also in the already construed observation sentences. He will treat them as words, and try pairing them off with English expressions in ways suggested by those observation sentences. Such are what I have called analytical hypotheses.8

The crucial word here is "suggested"...

Let us consider, then, what constraints our radical translator can bring to bear to help guide his conjectures. Continuity is helpful: successive utterances may be expected to have some bearing on one another. --- He will favour translations that ascribe beliefs to the native that stand to reason or are consonant with the native's observed way of life. But he will not cultivate these values at the cost of unduly complicating the structure to be ascribed to the native's grammar and semantics, for this again would be bad psychology; the language must have been simple enough for acquisition by the natives9

Considerations of the sort we have been surveying are all that the radical translator has to go on. This is not because the meanings of sentences are elusive or inscrutable; it is because there is nothing to them, beyond what these fumbling procedures can come up with. Nor is there hope even of codifying these procedures and then defining what counts as translation by citing the procedures; for the procedures involve weighing incommensurable values. How much grotesqueness may we allow to the native's beliefs, for instance, in order to avoid how much grotesqueness in his grammar or semantics?10

The last two sentences here are curious -this talk about "weighing incommensurable values". The idea, one could think, is that some manuals are better in some ways and other in other ways, so that what we need is not a single concept of "absolutely meaning conserving manual" but rather a family of concepts of "manual conserving those and those features of the expressions". It is not evident how this, even if true, would make it impossible for us to define a notion of manual correctness in terms of the cited procedures. We could have, and indeed we have, several notions of correctness of translation: translations true to the letter, translations true to the author's intention, translations that yields texts that have an effect on the intended audience that is similar to the effect the original text had on its target group, etc. So this sort of pragmatic diversification can't be what the passage is controverting.

What, on the other hand, could be thought to really make impossible the use of Quine's description of the field linguist's procedures to define a reasonably clear notion of manual correctness is that it might seem too vague. It does not specify, for instance, how much grotesqueness we may allow to the native's beliefs in order to avoid how much grotesqueness in his grammar or semantics; or does it? Well in fact it does, though not as explicitly as one could wish. The cue is the word "psychology": we should ascribe to the natives exactly those beliefs which it is psychologically plausible that they have11; we should ascribe to them a semantics which assigns those meanings to their expressions which it is plausible that the expressions have. -And what evidence should count? -All. -And how should it be weighted? -In accordance with the rules of rationality and our concept of meaning. In the next subsection I will explain what I mean by this. In any case, unless one takes on a mentalistic approach of this sort, it is not well defined by the cited procedures what evidence should determine a manual's correctness and in what ways. Therefore the indeterminacy thesis is not well defined either, in Quine's presentation.

Let's dwell a little on this point. Consider the questions: "Why on earth should a manual be respected for making successive utterances have some bearing on each other? Where do all of these criteria come from? Why may we use some criteria but not others in determining the correctness of a manual?" I do not know that Quine has an answer to these interesting questions.

What we want is thus a criterion for which criteria to apply. My explication of the thesis in the next section will give such a criterion and allow us answer those embarrassing questions in a satisfactory way. My approach is natural and straight-forward, as we shall see; it was hinted at above when I spoke of psychological plausibility.


1.5 A better formulation of the indeterminacy thesis

The morale of §1.4 is that Quine's naturalistic formulations of the indeterminacy thesis do not express what was intended. In this section a non-naturalistic formulation will be suggested which is claimed to be more adequate. In the next section it will be shown how to derive a naturalistic formulation which is roughly equivalent to the mentalistic one, sufficiently close to equivalent to serve Quine's purposes.

When Quine formulates his philosophy and in particular his doctrine of indeterminacy of translation he takes heed lest he should use any of those notions he dismisses, e.g. "meaning", "analyticity", "possible world" etc. Not that he never uses any of these words; but he avoids using them except where it is obvious how to replace them with Quine-acceptable terms or where merely a colloquial hint is intended and nothing is strictly claimed. This is as it should be, for it would be dubious moral to demand the exclusion of certain words from the philosophical vocabulary and go on using these very words as building blocks in formulating ones doctrines. However, for those not persuaded of the impermissibility of the terminology, no reason exists for not employing it in their own philosophy, even when that philosophy deals with the possibility of indeterminacy of translation. In fact, the issue of indeterminacy can be made clearer if access to the classical semantic vocabulary is granted. So long as the words "meaning", "synonymy" etc. are used with care, they can be utilised also in communicating with convinced Quineans. For Quine does not claim them to be totally incomprehensible: some things said by sentences involving these words are quite substantial; only those myriads of sharp distinctions required for upholding a working notion of analyticity are not to be had. But in order to state something like Quine's thesis, no such fine-grained distinctions need be made. That the following non-naturalistic formulation could be Quine-acceptable is supported by the fact that it can be transformed into a naturalistic formulation, as we shall see in the next section.

The claim that lies at the bottom of Quine's pronouncements is, I think, the following:

Between two natural languages manuals can be set up such that (1) they differ as to which logistic meanings have the expressions in the second language that the manuals yield as translations of given expressions in the first language; and (2) they are correct in the sense that if only the totality of naturalistic data is taken into account then they qualify as falling under our notion of a logistic meaning conserving manual.

This explication appeals to an unexplained "notion of logistic meaning conserving manual", so it is not very informative, it does not take us far. But it does have the merit of getting things right from the start; if everyone had taken this formulation as their starting point when specifying the indeterminacy thesis, many an absurd argument for or against the thesis would not have been expressed. We will see examples in §2.

In the present context it is useful to think of a notion in the following way. A notion is the mental equivalent to a definition. The notion of a bear, for example, is a cluster of fairly consistent beliefs of the form "x is a bear iff x has the property A to the degree a and the property B to the degree b; x is a bear iff x has the property B to the degree b' and C to degree c; x is a bear iff..." The notion is thus a set of connected beliefs about the membership criteria to "bear", and "bear" may either be taken as the word "bear" or as nameless mental "button" (which, when pressed, activates all the beliefs in the notion of bear, and other beliefs in addition). Just as a purported definition can be inconsistent, so can also a purported notion be inconsistent: we may then call it a confusion if the inconsistency is serious enough (i.e. if sufficiently great adjustments in the cluster of beliefs are necessary to make it consistent).

Now, one belief which is certainly an important part of our notion of logistic meaning conserving manual is that sameness of logistic meaning is a transitive relation. So if it could be shown that the indeterminacy thesis, as stated above, is true, then, under the assumption that naturalism in linguistics can be shown to be mandatory, as Quine thinks, it would follow that our notion of logistic meaning is inconsistent; if there is much indeterminacy, then it is seriously inconsistent and our purported notion of logistic meaning amounts to a confusion. Thus, in our formulation, the indeterminacy thesis has the devastating consequences for the purported notion of analyticity which Quine assumes it has.


1.6 A naturalistic indeterminacy thesis

What are we to say about the mentalistic formulation of the indeterminacy thesis if it is true? Is it then nonsensical or confused?! Well, it would be a matter of degree. Anyway, we may avoid such problems by opting for a naturalistic formulation instead. This alternative should be especially attractive to people who, like Quine, believes in indeterminacy of translation.

The naturalistic version I am about to propose is coarser than the mentalistic one, but if Quine is right in thinking he has discovered a phenomenon which is so abundant, he does not have to tinker with subtleties and borderline cases: it suffices for him to draw a broad distinction in naturalistic terms: the rival manuals he has in mind should fall far within the boundary of empirical admissibility but far outside the border of equivalence of logistic meanings. The following formulation illustrates what I have in mind:

Between two natural languages manuals can be set up such that competent, well-informed speakers of English will confidently assent to (1) the sentence "The purported notion of logistic meaning conserving manual contains a belief according to which these manuals qualify as falling under it, if only naturalistic data are taken into account." and also to (2) the sentence "But the purported notion of meaning conserving manual also contains a belief according to which the fusion of these two manuals does not qualify as falling under it."

Here the word "confidently" is to be understood in such way as to rule out small fluctuations and confusions on part of the "competent, well-informed" speakers. It looks as if we tried to smuggle in the forbidden notion of meaning, in the form of an appeal to competent speaker's linguistic intuitions, but the point is that we could let "confidently" be a very restrictive property, since all borderline cases, which would have to be correctly decided upon for a notion of synonymy to be definable in terms of "confidently", all these borderline cases can be disregarded by our notion of "confidently", because the manuals in question are such that the relevant sentences are very confidently assented to by competent language users. In other words: the divergent manuals Quine imagines should be so obviously correct and incompatible that one would need only a small fraction of the discerning abilities of the notions of "logistic meaning conserving manual, as concerns naturalistic facts" and "manuals differing in their assignments of logistic meaning" to determine that they were correct; this fraction can easily be provided by a naturalistic formulation.,p> This naturalistic formulation serves the purpose of explaining how the indeterminacy thesis could be expressed if it is impermissible in general to use the notions "meaning" and "part of our concept" etc. My formulation is an alternative to those of Quine, which we found to be inadequate. Quine has criticised Naess's semantic method of seeking answers to questions about synonymy through distributing a questionnaire to ordinary language users:

And now the test suggested is that we ask the natives the very question which we do not understand ourselves: the question for which we ourselves are seeking a test. We are moving in an oddly warped circle.12

Does the same critique apply to my suggestion? To some extent it does. However, one must bear in mind that what I am seeking here is not a definition of synonymy or a general criterion of what is part of the meaning of an expression: I am looking at a particular case and I'm only interested in certain broad features of the term that the natives are supposed to have an opinion on. I do not care whether their opinion is prejudiced or well-grounded. What is necessary, however, is that the right sort of explanation precede the query; else the subjects will not count as well-informed. One could for example choose a number of intelligent individuals and quote some passages from classical logical positivist writings; then present to them an alternative translation manual between two languages they know. Finally one put the specified sentences to them, asking for their assent or dissent, allowing them time to reflect before delivering their answer. A procedure along this line could surely be stated in naturalistic term, and if the non-naturalistic formulation of §1.5 is adequate, then we have good reason to believe that such a naturalistic explication would also be extensionally equivalent. (We have good reason to believe this even if we do not yet have good reason to believe that the thesis is false or to believe that it is true.) I call the technique (cheat-technique) involved in my naturalistic formulation the method of pragmatic assent.13

I said that to some extent Quine's critique over Naess' method also applies to my suggestion. This is to the extent to which: speakers would have difficulty to understand the question, different speakers would answer differently, and their answers would tend to vary with details in the presentation preceding the query. I suggest that this extent is not so wide as to prevent my naturalistic formulation from being an acceptable surrogate for the non-naturalistic formulation of the indeterminacy thesis. I do not think that there is a practically pervious way of providing a better naturalistic formulation other than along the line indicated. I also think, however, that except where it is necessary to do otherwise, it is better to use the non-naturalistic formulation.

In the remainder of this paper I take the "indeterminacy thesis" to be understood in the way I have just explained.



Many arguments have been advanced for or against the indeterminacy thesis. In this chapter I shall review some of them, the most important ones, and I shall try to show that all of them are flawed. Moreover, several of these arguments betray serious misunderstandings of the thesis, and I shall point out as we go along how such confusions would have been avoided had the formulation I suggested been adopted.


2.1 Refutation of some arguments in favour of the thesis
2.1.1 An insufficient argument for the indeterminacy

One argument for the indeterminacy thesis takes the following form. (1) The only naturalistic criteria for what truth conditions a sentence has, are through those of its empirical consequences which a subject can test by a short series of observations (a sort of verificationism). (2) But these are not sufficient. (Something like this argument can be read into Word And Object)

The reason for (1) is that: (1.1) Only the observation consequences will effect the speakers´ dispositions to assent to or dissent from a sentence when faced with a stimulation within the "modulus"14; and (1.2) Such effects are the only relevant ones because our only source of information is the speakers´ dispositions to react to stimuli, and (1.2.1) Only stimulus of a short interval can be used since there is no guarantee that the meaning of a sentence for a speaker will not change during a long period of stimulation.

Reasons for (2) could be (2.1) full-scale examples, but such may be tedious to construct; or (2.2) indications or proofs that there must be such examples.

We could call (1.2.1) an argument that the noise level is too high. It is obviously a matter of degree how effective a physicalistic explication can be made for abstracting changes of logistic meaning for a speaker from changes of his other beliefs.

Briefly, the reply to this is that we are still waiting for (2.1) or (2.2). In particular one should be keenly aware of the possibility that the truth conditions of a theoretical sentence could be derived from the meanings of its words, while the words derive their meanings from their relations to words that occur in observation sentences. The inner parts of language may thus be uniquely determined, as to their meaning, by their interconnections to other inner parts and ultimately to those parts that lay on the surface of language and are immediately determined by their connections to external stimulation.


2.1.2 Putative examples of divergent manuals

One way of arguing for the indeterminacy thesis is to present examples of correct divergent translation manuals (2.1). Let us briefly review some of the most important suggestions for such examples.

* Massey´s manuals15

Massey has devised three manuals, the "dualizing", the "contradictory" and the "Cretan", that conflict radically with the standard manual but which he nevertheless claims to be correct. When we would say that a speaker asserts S, the dual manual holds him to deny the dual to S16; the contradictory manual holds that he denies the negation of S; the Cretan manual that he asserts the dual to S but that he lies whenever we would say he speaks the truth.

The correctness of these manuals is of the same sort as the correctness of a physical theory that says that bodies dropped near the surface of the earth fall upwards, and prescribes that we by "upwards" should understand: downwards. Massey´s manuals would prove indeterminacy only if it could be shown that "assent" and "dissent" are not reasonably clear terms anchored in some physicalistic facts. Quine's case would be weak indeed, if it depended upon such a lemma.17

From the point of view of my §1.5-interpretation of the thesis, this reply is straightforward. In order to instantiate indeterminacy of translation, Massey's manuals would have to satisfy the second clause of the definition, i.e. they would have to qualify as falling under our notion of a logistic meaning conserving manual if only the totality of naturalistic data is taken into account. But our notion of meaning is surely such that it would be psychologically extremely unlikely that everybody in a language community went around constantly telling lies to one another. (We have to involve psychological theory since from a meaning assignment alone, no observational consequences about the behaviour of the speakers can be derived.) So the only way for the proposed manuals nevertheless to satisfy the second clause is through the notion of lying, or the notions of assent versus dissent, to have no anchoring at all in naturalistic facts. This assumption is then easily seen as gratuitous.

* Duality in projective geometry (Frege)

In one axiomatization of geometry one can interchange the words "plane" and "point" throughout the axioms without changing their truth values. As Kirk has correctly pointed out18, this is only because the terms "plane" and "point" are uninterpreted in that axiomatization. It has not been shown, for example, that it is correct to suppose that someone says he sees a "plane" when he is looking at the figure of a point.

* Poincare´s example of underdetermination of theories of physical space19

The example is that a theory asserting that physical space is spherical and finite and that bodies shrink uniformly as they approach the borders could be specified so as to be empirically equivalent to a theory that says that space is infinite and bodies do not thus change their size. Yet, the finite space would have a centre, a singular point, which the infinite space would lack. So the two theories would be logically incompatible, assuming that all notions they involve are thoroughly meaningful. The suggestion is that divergent but empirically correct manuals could be established between the terms of these two theories. But if this is to suffice to establish indeterminacy of translation of ordinary language, those of the theories´ notions that are also used in other contexts would have to be translated compatibly with these other usages; it is not obvious that one could do so. And if, instead, the theories´ notions are disconnected from ordinary language, then great care have to be taken that they do not get confused with associations belonging to the terms' usage in ordinary language (as was the case with "plane" and "point" in §

It is pertinent to issue a general warning against mathematical examples intended to prove counterintuitive propositions of nonformal concepts: they tend to benefit from the ambiguity between an intuitive notion and its formal explication.


2.1.3 An argument based on a false principle of belief attribution

It has been stressed by Davidson, and Quine also, that assignment of meanings to sentences is closely intermingled with assignment of beliefs to speakers.

What beliefs should we ascribe to a person? Some philosophers say we should apply the principle of charity, others suggest that those beliefs should be ascribed to a subject which it would be most rational for him to have. The truth, however, is that we should ascribe exactly those beliefs to a subject which it is psychologically plausible that that subject has under the circumstances in question. It is a task of the psychologist to find out which beliefs a person holds. Another question is what we should mean by "believe"; to answer that question the philosopher and the psychologist may work together in an investigation of what concept of believing we most need in our understanding of the world. The principles of charity and of rationality are only rules of thumb that perhaps in many cases approximate what is psychologically most plausible; whenever those principles yield verdicts that conflict with what we think reasonable, we should simply disregard them, because they derive their authority solely from being approximations to psychological plausibility.

To adopt the principle of rationality as one's codex of belief attribution can influence one's evaluation of the evidence in favour of the indeterminacy thesis. One might be misled to reason like this: "Our theories are underdetermined by all possible data, i.e. there are several equally simple but mutually incompatible theories that equally well explain our observations; therefore we have no rational reason to prefer one of these theories to another; therefore we may correctly be assumed to hold any of these theories (according to the principle of rationality); thus there is no fact of the matter as to which one of several mutually inconsistent beliefs we hold, and so it is likely that there will be indeterminacy as to which opinions we express and what propositions we assert."


2.1.4 Pressing from above and from below

We come now to Quine's own arguments. He says that there are two ways of pressing the doctrine of indeterminacy of translation: The "pressing from above" is the argument that:

the same old empirical slack, the old indeterminacy between physical theories, recurs in second intention. Insofar as the truth of a physical theory is underdetermined by observables, the translation of the foreigner's physical theory is underdetermined by translation of his observation sentences. If our physical theory can vary though all possible observations be fixed, then our translation of his physical theory can vary though our translations of all possible observation reports on his part be fixed. Our translation of his observation sentences no more fixes our translation of his physical theory than our own possible observations fix our own physical theory.20

There are three reasons why the pressing from above need not necessitate acceptance of the indeterminacy thesis. First, it needs the support of the underdetermination thesis, which is controversial. From the assumption that a physical theory is underdetermined by all possible observation, a solid verificationist would draw the conclusion that not all terms in the physical theory are thoroughly meaningful; he would rather welcome Quine's argument as a justification for his verificationism in this case.

Second, even if the truth values of some theoretical sentences are not determined by an assignment of truth values to all observation sentences, indeterminacy for those theoretical sentences still doesn't follow. For instance, the two theories which are equally justified by all observations might have different structures: they might employ unisomorphic machineries of abstract entities to produce their observation consequences. It is not implausible that the structure of the theories are reflected in their formulations in such a way that this may be used to determine which theory the speakers entertain.

Moreover (third reason), there might be psychological considerations as to which one of several beliefs, equivalent as to which expectations of observations they generate, that a given speaker has in mind and intends to express.

By "pressing from below", Quine means pressing whatever arguments for indeterminacy of translation can be based on the inscrutability of terms. The well-known "Gavagai"-example was intended as an example of such inscrutability. Quine tells us that his readers "have responded with suggestions of how, with the help of screens or other devices, we might hope to give the native informant an inkling of the desired distinctions and so settle the reference."21 But

The most we can hope from the screens and kindred aids, then, is an indirect hint as to which of various analytical hypotheses regarding pronouns, identity, plurals, etc. might in the end work out most naturally. When this kind of hint is available, should we say that the supposed multiplicity of choices was not in fact open after all? Or should we say that the choice is open but that we have found a practical consideration that will help us in choosing? The issue is palpably unreal, and the doctrine of the indeterminacy of translation depends in no way upon it.22

"Palpably unreal"? I would say that it is a question of whether an explication of "meaning" involving such considerations would be adequate. The indeterminacy thesis "depends in no way upon it"? If the meaning of terms were fixed, then the only way for the meaning of sentences not to be fixed would be through indeterminacy of the contribution to logistic meaning by modes of composition. But it scarcely seems plausible that the indeterminacy of translation could derive solely from the absence of any fact of the matter as to which meaning a mode of composition generates out of given meanings of the constituents. There are relatively few basic modes of composition, and these can often be easily studied when they operate upon terms whose meanings generate sentences whose meanings can be determined by a field linguist theory. In fact, Quine does not mean that the indeterminacy thesis in no way depends upon the inscrutability thesis, for later on the same page he continues:

The gavagai example had only this indirect bearing on indeterminacy of translation of sentences: one could imagine with some plausibility that some lengthy nonobservational sentences containing gavagai could be found which would go into English in materially different ways according as gavagai was equated with one or another of the terms 'rabbit', 'rabbit stage', etc.

The pressure is not compelling. Even if we grant the indeterminacy of reference, this is still no reason to accept indeterminacy of logistic meanings. The only argument in this passage from the former to the latter is the statement that "one could imagine with some plausibility that ...".

I am aware that these refutations seem very hasty and maybe trivial. But then again: so are the arguments they refute. Adopting my interpretation of the thesis would seem to lessen the temptation of a simple proof or disproof. The thesis in my formulation looks like something substantial, controversial and important. Maybe it seems counterintuitive, but as we shall now see it is not so easy to prove that it is false.


2.2 Refutation of some arguments against the thesis
2.2.1 How the indeterminacy could go unnoticed.

Prima facie, we have little reason to suspect any considerable flexibility in our translations; and we need an explanation of how there could be such without our noticing it (before Quine pointed it out).

Consider two languages with only a few sentences each, of the kind that very small children may have. Here all sentences are high on observability, and means of variation and construction are very sparse: there is no indeterminacy of translation. If we add a single term, there will probably still be no indeterminacy. We can go on adding term after term, and for some time there will be only one acceptable translation manual, since it will take a while before the domain gets big enough to allow those "compensatory moves" that will be necessary to adjust a rival manual. But according to Quine, there comes a time when the structure of the theoretical sentences will possess not only one but two or more "stable states" (i.e. be such that several radically divergent but empirically adequate translation manuals could be set up between them). In practice, the translation will remain in its original state -that state is, after all, stable; but there will be other stable states as well, and it is a matter only of historical contingency that the translation ended up in that stable state. We could start up a completely new interpretation, and if we started it off from a certain other domain, it would end up in a stable state different from the first one. (This is the crucial fact overlooked by Robert Kirk in his refutation of the indeterminacy thesis.)

Thus translation of sentences would be relative to stable states, i.e. translation manuals. To paraphrase Quine, it is not the translation of a single sentence but only translation of a language as a whole that faces the tribunal of experience. One reason why indeterminacy was not easy to discover, then, is that a developing translation tends to remain in its stable state, and that all translations in practical use between a given pair of languages are, in fact, developments from a single Urübersetzung between those languages, which may perhaps have evolved as the languages themselves developed out of a common ancestor. "Radical translation is a rare achievement, and it is not going to be undertaken successfully twice for the same language."23

A second factor that may account for the absence of divergent manuals, and which I think more potent than prejudice, is that there are other pragmatic considerations, not essential to synonymy, that decide which translation we adopt, beside the consideration that the traditional manual is the best since it does not force us to relearn. For example, words that are used frequently should be translated into words that are frequently used.

Since my formulation speaks of logistic meanings, not usage or linguistic roles, and since it does not appeal to the usefulness of manuals, it would seem that it minimises the risk for improper considerations of this sort to infect our judgement about whether there is indeterminacy or not.


2.2.2 Not considerations about language learning

Another way, which does not look promising, of explicating correctness of translation is through considerations about language learning. "If an expression e in language A is rendered as an expression e' in B, then e and e' should be learned at approximately the same stage in the typical language user's language acquisition process."

First, it is not clear that this move would expel the indeterminacy. Second, I think the language acquisition process is one of those lower level features of language behaviour that is idealised away in the higher levels of description, where the notions of "correct translation" and "equivalent sentences" belong. A sentence has the same logistic meaning independently of whether it is generally learned in the ordinary way or through a $10 quick-learning cassette course or through the consummation of a fancy "language pill".,p> This would be yet another example of how we could go astray through a lack of understanding of exactly which criteria are relevant.


2.2.3 Not straightforward pragmatic considerations

Given only Quine's formulation of the indeterminacy thesis, one could be tempted to interpret "empirically correct manual" as referring only to those manuals that would really be a good choice for a translator. Or at least, one could define a notion of correctness to denote only the pragmatically best translation manual, and then one could regard this as an explication of "correctness" that would escape the indeterminacy and which we could be happy with.

However, as we saw in §1.2, the term "correct" as used here is tied to the notion of logistic meaning, and so any explication of "correctness" must really be an explication of "logistic meaning conservation" if it is to serve in a formulation of the indeterminacy. But simply looking to the manual's usefulness is not justified from the meaning conservation point of view. For a typical pragmatic reason for the preferability of one manual over another is that by some historical contingency we have become accustomed to the former but not the latter; which is irrelevant to the issue of which expressions in the languages that really are synonymous with which.

There is, of course, the possibility to formulate an indeterminacy thesis in term of usefulness, as we saw in §1.4.2. In that case one should probably settle for something more sophisticated than common usefulness; e.g.: "tends to be a useful manual in those and those respects".


2.2.4 Bennet's argument

A naturally occurring argument against the indeterminacy thesis is set forth in the following quotation from J. Bennett:

The 'reverberations across the fabric' [of that system of interconnected sentences which constitutes our language and theory] have their own detectable pattern; and the two translations of [the tribal sentence] S cannot both be secure unless E1 and E2 [two nonequivalent English translations of S] both reverberate in precisely the similar ways. For example, if any experiences would render problematic the acceptance of E1 but not that of E2, then there could be behavioral evidence favoring the translation of S by one of them rather than by the other. Or if E1 were relevant to some theoretical issue, while E2 did not have an exactly analogous relevance to an exactly analogous theoretical issue, then again there could be a basis for preferring one translation. In short, every single fact about E1 must be mirrored by a fact about E2. By any reasonable standard, therefore, the two sentences are synonymous, and so the indeterminacy of translation thesis is false.24

"reverberations across the fabric" sounds physicalistic; "relevant to some theoretical issue" sounds as if it presupposes a classical notion of interlinguistic meaning. This is the crux. The argument is that for E1 and E2 to be correct translations of S, they have to "reverberate in precisely the similar ways" as S. If by "reverberate in precisely the similar ways" you mean: play exactly the same linguistic role (i.e. being used in exactly the same way) then what your requirement is naturalistic, but so strong that no sentence can be correctly translated by any sentence, except itself. That would clearly be unreasonable. If, on the other hand, you mean that the two sentences must have the same theoretical implications, then you are assuming what you set out to prove, namely that sentences have a well defined set of truth conditions associated with them in a naturalistically scrutable manner. In either case the argument fails. There is no other obvious interpretation of "reverberate in precisely the similar ways" than these two. One hope that no one working with my explication of the thesis would produce an argument like this one. I submit, though, that it might sound rather convincing if one is not clear over what the indeterminacy thesis really says.


2.2.5 Robert Kirk's attempt to prove determinacy

Robert Kirk claims to have refuted Quine´s thesis of indeterminacy of translation25. I will here discuss the core of Kirk´s argument which is set forth in his book Translation Determined, chapter 11, where he intends to demonstrate domestic determinacy, i. e. that no two translation manuals from one natural language to the same language can satisfy all Quine-acceptable constrains and yet be incompatible with each other. This, together with the claim (which Kirk argues for earlier in the book) that indeterminacy of translation implies domestic determinacy, has as a consequence the negation of Quine´s indeterminacy thesis.

Kirk asks us to imagine a "twin-language" to English, 'Martian', which is so defined that 'all relevant facts about it about behavioural dispositions, brain states, and whatever else Quine will accept as facts are exactly like relevant facts about English, except that each Martian morpheme is the reverse of its m-correlate [where 'm' is the 'morpheme-to-morpheme correlation under which Martian and English are translatable.'26]'27. The word corresponding to 'bird' is 'drib', and so forth. The argument then proceeds by an attempt to justify the claim that there is a possible procedure of learning a natural language which can be regarded as progressing in small minimal steps, 'minimal in the sense that the increment constituted by a given step cannot be regarded as consisting in its turn of the successive acquisition of semantically relevant increments'28. Kirk goes on to describe one such route for learning English and one for learning Martian. Emma and Marcia are two children learning to speak their native languages, English and Martian, respectively; and they are exactly similar in all their dispositions etc., except that Emma is disposed to utter 'bird' whenever Marcia is disposed to utter 'drib' and so on. As a starting point Kirk takes the stage at which Emma and Marcia each have an active vocabulary of about 200 basic expressions, such as two year children can have, and some simple construction means whereby they can form sentences such as 'Adopti book' (=I´ve dropped it: the book) or 'Ere-e-is, bird' (=There it is: a bird). Kirk believes that Quineans would concede that Emma´s and Marcia´s utterances are 'too simple, and too closely keyed to publicly checkable circumstances, to leave room for the sorts of compensating adjustments that would be required for the construction, within these rather limited language fragments, of permissible rivals to the ultra-simple morph-for-morph manuals that gives such perfect results here.'29 In order to expand the determinacy of these fragments, Kirk first considers what would happen if Emma and Marcia had such phrases at their command as "We say that '-----' when ....." and 'The new predicate applies to -----'. In that case Emma and Marcia would be able to extend their languages by definitions and explanations of new expressions, e.g. colour words or 'atom', and this would involve no indeterminacy that was not already in the language, since 'she could always replace any sentence in which they occurred by a sentence in which they were merely mentioned'30. Kirk goes on to argue:

obviously there is no reason why colour words or 'atom' should not have been introduced before the languages included that simple semantic vocabulary. Nor could the determinacy or indeterminacy of translation of sentences including colour words or 'atom' depend on whether or not those words happened to have been introduced with the aid of semantic words. For the question whether translation between a given pair of languages is subject to the indeterminacy is to be settled by the actual character of those languages at whatever stage they may have arrived, regardless of the route by which that stage was reached.31

So the simple fragments of Emma´s and Marcia´s can be extended in such a way that no new indeterminacy is incurred; and since there was no indeterminacy in the translation between the original fragments, by induction there can be no indeterminacy in the translation between the whole languages either, thus no domestic indeterminacy at all.

The part of Kirk's argument just explained covers 21 pages in his book; clearly, he is in great pain to make his argument watertight. There seems, however, to be a silent consensus that he has failed (I've never heard anyone mention his proof). The first objection that comes to ones' mind is that there might be some unnoticed build-up of looseness of fit that eventually allows room for the horse to turn in its box. Kirk has anticipated this and prepared a reply:

given that assumption [that English can be acquired by a succession of small 'steps'] the objection would only work if we could do exactly the same sort of thing for a non-standard rival to T [the standard translation] as my argument does for T. For clearly, if a rival to T is permissible at all, there must be exactly the same possibility of building up this rival manual by successive increments as there is for the case of T itself. I believe my argument rules out this possibility because at each stage Tn supplies exact versions - - -, while no exact rival versions are permissible.32

I think that Quine would agree that if we could supply a definition of every theoretical sentence in terms of logical constructions of observation sentences (if an complete Aufbau could be successfully erected) then nothing would stand in the way for indeterminacy of translation to be explicated away straight forward. It is an integral part of Quine's doctrine that a complete Aufbau is impossible. Theoretical expressions are not generally reducible to logical constructions of observation sentences. If one wants to squeeze out the empirical content of theoretical sentences, one has to make them relative to the theory formulations in which they occur. One could then define the meaning of a theoretical sentence as the set of all ordered pairs whose first component is a set of sentences and whose second component is the set of all observation sentences the sentences in the first component implies if taken in conjunction with the sentence whose meaning is being defined; or something like that. This would dispose of indeterminacy of translation but only at the price of making theoretical sentences untranslatable.

It is therefore a flop in Kirk's argument that it presupposes that more advanced language games can be defined in the observational vocabulary of a two-year-old. Ideally, she may be able to explain her new linguistic tools, but that is not enough: for the argument to hold, she would have to be able to define them. An explanation leaves it to the other girl to form some conception of the new word or sentence. There is no reason why there should not be different conceptions, intuitively speaking, which would enable her to communicate successfully with her extraterrestrial sister.

This difference in conceptions would not show up until their languages had developed to the stage at which alternative ways of talking about the same things begin to be feasible. It is therefore possible that the requirement that manuals have to be a closure of some manual between these simple language fragments would indeed prevent indeterminacy. But this is, in effect, just to impose constraints of language learning; I argued in §2.2.2 that this would be impermissible.



3.1 Indeterminacy of reference and indeterminacy of logistic word meaning

A logistic meaning is to a sentence what x is to a word.

What is x? Not the denotation of a word, for what corresponds to that is the truth value of a sentence. We can know the logistic meaning of a sentence without knowing its truth value, and we can know x, the logistic meaning of a word, without knowing what it denotes. We do not know which things are blue although we understand the predicate. Just as the logistic meaning of a sentence is its truth conditions, i.e. a function from possible worlds to truth values, so the logistic meaning of a word (or formula) should be taken as its denotation conditions, i.e. a function from possible worlds to denotations. The logistic meaning of "blue" is a function that takes out the set of objects that are blue, given a complete totality of facts.

A analogue to the thesis of indeterminacy of logistic meanings (by which we have hitherto meant: logistic sentence meanings) now comes spontaneously to one's mind: the thesis of indeterminacy of logistic word meanings. It is to be formulated in exactly the same way as we defined the thesis of sentence meaning indeterminacy in §1, just disambiguate that formulation in the obvious way.

Ontological relativity, or Quine's thesis of indeterminacy of reference, is something else. It differs on two accounts. First, it concerns the actual denotation, not the denotation conditions, of words. Second, and more importantly, it seems to be disconnected from our intuitive notion of reference. The thesis is that there are divergent acceptable ways of assigning denotations to predicates. But here "acceptable" means simply: conserving truth value of all sentences.33

I will return to Quine's ontological relativity in a moment, but first I wish to speak of my own thesis of indeterminacy of word meaning. In this case, as opposed to the one with the thesis of indeterminacy of translation, I am not suggesting that my formulation is an explication of Quine's thesis; it is simply another thesis.

First we make some distinctions. We note that, just as was the case with the indeterminacy of logistic meaning, indeterminacy of reference may come in degree. We may say that there are theses of amounts of indeterminacy of reference: how many and how wild can the correct "proxy functions", i.e. permutations, be? There is also the possibility of microindeterminacy of reference, and this problem is as pressing as the corresponding problem for logistic meanings. How does it come about that two ostensions of the same bear pick out exactly the same object rather than just almost completely overlapping spacetime regions? I think this problem can be solved in the same way as the problem of microindeterminacy of logistic meaning, but it should be distinguished from theses of various amount of macroindeterminacy of reference.

Further, we must determine whether or not it is permissible to invoke intuitions as criteria for whether an assignment of denotations is correct. This is obviously a very essential issue.

To begin with, it must be understood that to appeal to intuitive notions of this and that is not necessarily to beg the question from Quine. There are three regulations to observe however, if such an appeal is going to be successful. (i) If you argue that something falls under a concept, then you must give naturalistic data according to which it does so. This is because of Quine's naturalism. (ii) You must recognise that the fact that x falls under the purported concept Y does not imply that x does not fall under the purported concept not-Y. This is because a purported concept need not be a concept: it may be inconsistent. If e.g. the term "meaning" leads to confusion, this is the state of affairs one should expect. This regulation holds generally, not only in a debate with Quineans. (iii) You should be prepared to meet a demand to translate your jargon of notions and concepts into naturalistic terms. Regulation (i) said that, given a concept, only naturalistic facts should determine whether something falls under it; regulation (iii) says that only naturalistic facts should determine what the "concept" is in the first place. This regulation is lifted if it is supposed that semantic eliminativism is mistaken. Even if that is not supposed, (iii) could still be suspended. One would than have to reckon that what is said may be partially meaningless, but this need not prevent that some useful points could be made.

As long as (i)-(iii) are respected, nothing prohibits a thesis or an argument that appeals to intuitions from be fully acceptable from a Quinean point to of view. The thesis of indeterminacy of translation is an example, at least under my §1-explication.

So there is no obstacle to formulating a thesis of indeterminacy of word meaning in analogy to the thesis of indeterminacy of logistic sentence meaning we presented in §1. What we require of an assignment of word meanings to predicates is that it conserves word meaning in the intuitive sense as far as this is naturalistically determinable.

Back to Quine's ontological relativity.

We may now ask what may be the philosophical significance of the thesis of indeterminacy of reference, interpreted so as to be proved by the presentation of any proxyfunction. Does it show (a) that there are incompatible theories of reference, all of which are equally adequate, albeit we have happened to choose one particular theory by opting for our actual notion of reference? Or does it show (b) that there is no fact of the matter as to which sense of "reference" we are using?

I think it does not show (b) because in order to prove that the purported notion of "reference" is merely purported, it is not enough to show that there can be divergent denotation assignments conserving the truth values of all sentences; one would also have to show that conserving truth value of all sentences is sufficient for an assignment to be correct in the intuitive sense, as far as this is naturalistically scrutable. That has not been shown. There is no obvious reason why there may not be determining criteria hooked directly onto words.

I think it does not show (a) either. (a) would in effect be a case of underdetermination of theory by data, underdetermination of the countless possible theories of reference. I will say more about the relation between indeterminacy and underdetermination in §3.4, but it should be clear in any case that two theories which have exactly the same structure and are straightforwardly intertranslatable should not count as examples of nonequivalent theories. That would make the underdetermination thesis trivial and uninteresting: it would amount to nothing more than the truism that the same word (sign) can be used in different senses. Now, this is precisely what we have here: two theories with exactly the same structure, straightforwardly intertranslatable via the proxy function. In other words: proxyfunctions do not prove that there are empirically equivalent nonequivalent theories, because the theories to which they give rise are equivalent.

Nor does the thesis of indeterminacy of reference have any obvious bearing on the thesis of indeterminacy of translation. As regards my thesis of indeterminacy of logistic word meanings, on the other hand, there exists such a connection. If we assume, as seems reasonable, that the meaning34 of a sentence is uniquely determined by the sequence of its words and the meanings of these words, then indeterminacy of translation implies indeterminacy of word meanings. (Since the order of words in a sentence is surely scrutable and since that order plus the meaning of the words would determine the sentence's meaning, it is evident that that if the sentence's meaning is not determined, this can only be so because the meaning of the words is not determined.) This assumption is reasonable because the way in which word order contributes to sentence meaning seems to be naturalistically scrutable. There is a limited number of construction rules, and they can often be tested in cases where they are applied to sentences whose meanings are naturalistically scrutable to yield sentences whose meanings are directly naturalistically scrutable. In those cases the effect of the operation stands out clearly, and one can readily generalise to more complicated cases.

The converse is not obviously true; indeterminacy of word meaning could cancel out so that sentence meanings were still determined.


3.2 Is it a matter of degree?

What the indeterminacy thesis is intended to bring out is not so much that there is a considerable zone of fuzziness and arbitrariness in the determination of the meaning of many a sentence, but rather that the issue whether a certain sentence can be assigned a certain meaning is dependent upon which meanings are assigned to other sentences. This is the point of semantic holism. A corresponding proposition holds for the thesis of indeterminacy of reference.

At the same time one feels that any nontrivial indeterminacy must be a matter degree -just how confused is the notion of meaning. Especially my explication brings this conclusion near at hand: it must be a question of more or less, not of yes or no. But the holistic scenario seems to indicate that indeterminacy tends to come in blocks, so that it is a discrete phenomenon after all. How are we to resolve this conflict? There are at least two natural ways of specifying a degree of indeterminacy.

First, there is the question of which expressions can be divergently translated. Is it only highly theoretical sentences or does the indeterminacy reach down to common standing sentences or even some observation sentences? Or is it even limited to particularly obscure statements in philosophy or psychology and other soft sciences? This is clearly one relevant variable; any exact thesis of indeterminacy of translation would have to specify its value, i.e. make clear exactly which set of sentences is submitted to indeterminacy.

Second, we may ask: how drastically do we have to change our notion of meaning in order to cure it from inconsistency. This is a second variable that determines exactly what the indeterminacy thesis says. Quine suggests that we have to make meanings relative to translation manuals. That would surely be a very drastic modification: so drastic that it is better to say we have another concept, not a new version of the original one. But it could also happen that we would need only to make some small adjustment in order to expel the indeterminacy, perhaps regard some expressions as ambiguous in unexpected ways.

If we regard our purported notion of meaning as fixed, then these two variables are not independent. An indeterminacy thesis would be specified as soon as we fixed the value of one of them, since they are related via the purported notion of "meaning". Given that we know this purported notion and the set of expressions which can be divergently translated, we should in principle have at least a rough idea of what modifications would be necessary to eliminate the indeterminacy. And given that we know what changes are required, we could figure out which translations they would render incorrect and so which expressions were struck by indeterminacy before the changes.

So our intuition that indeterminacy must come in degrees and the appearance the its being a discrete phenomenon can be reconciled. If the indeterminacy is such that it can be removed by making some minor changes to our notion of meaning, then we might say that there really isn't any indeterminacy, we have just been somewhat confused about what "meaning" means. If, on the other hand, a complete transformation of our purported notion is required, we would instead admit that there were no such notion at all. The outcome is thus discrete, though the underlying phenomenon comes on a continuos scale. There will, of course, be vagueness as to exactly how much confusion a concept can be said to accommodate: when it ceases to be real and becomes merely "purported".


3.3 Vagueness, epistemic holism and indeterminacy

Suppose you open a philosophical journal at random and read one sentence in the middle of a section. The sentence is grammatically well-formed, but is it true or false? For many sentences, you would not know this not only because your philosophical insight is imperfect, but also because it would be very difficult to know what the author meant by that sentence without having read the rest of the text.

The typical situation for a philosophical argument is that you cannot determine whether it is valid or at which step it goes wrong, until you know what it is aiming at, what it is supposed to prove. Only that determines the exact content of every intermediate stage of the proof.35

Is this a symptom of holism? Yes, but not necessarily semantic holism, of the strong sort that is claimed by the indeterminacy thesis. Let us suppose that the indeterminacy thesis is false. The way I would then prefer to think about the phenomenon I illustrated, epistemological holism, is as follows. At least in philosophy and other soft sciences, many sentences seems to be highly vague, or ambiguous, and there are two equivalent ways of describing this situation.

One way is to make the meaning of such sentences relative to contexts. Theoretical sentences do not have a private fond of observational consequences they can call their own but obtain empirical significance only when taken in conjunction with other sentences. This means semantic holism, but not in the strong sense, for there is another, equivalent, description which does not relativise meanings.

This other way is to insist that each sentence has a meaning on its own. We must then take the meanings of theoretical sentences to be quite diffuse and "spread out", so to speak; but that need not deter us. Given the holistic description, it is trivial to transform it into an particularistic one. Simply define the meaning of a theoretical sentence, as, e.g., a function from contexts or theories (taken to be equivalence classes of sets of sentences under the synonymy relation) to more definite meanings. This obviously presupposes that there is no indeterminacy of translation, since otherwise the synonymy predicate will not be well defined. But if the indeterminacy thesis is false, the road seems paved to an intuitively plausible meaning assignment which does not make reference to particular theories or contexts.

Thus the thesis of epistemic holism, while plausible in the moderate form advocated above, does not imply that there is indeterminacy of translation or semantic holism in the strong sense. The implication does seem to obtain in the opposite direction. If there is indeterminacy of translation, then we would certainly not expect that soft science sentences could generally be verified or falsified one at a time, without considering them as part of a theory; for if that were possible, they would succumb to a determinate translation by the same clause that except observation sentences from the indeterminacy. What the thesis of epistemic holism amounts to is something like the claim that stimulus meaning falls short of being a reconstruction of logistic meaning. That is a necessary but not sufficient condition for indeterminacy of translation.


3.4 Underdetermination and indeterminacy

The thesis of underdetermination of theory by data, maintained by Quine, is that there are logically incompatible but empirically equivalent theories. If one may use the term "logistic meaning" then this boils down to saying that there are contingent nonempirical facts. (By empirical we mean: in principle possible to verify or falsify by experiment.)36 The underdetermination theory is thus equivalent to the negation of some from of verificationism, under the assumption that it makes sense to talk about such things as logistic meanings.

But suppose now that Quine is right about indeterminacy of translation. The problem is that the underdetermination thesis seems to be nonsensical if neither "empirical content" nor "logical incompatibility" make sense! Whether we say: "there are logically incompatible but empirically equivalent theories" or we say: "there are contingent nonempirical facts", we are talking nonsense if the indeterminacy thesis is true. (I here refer to the thesis I formulated in §1.5 or §1.6; but the same holds for Quine's indeterminacy thesis if I am right in claiming that it should be construed in such a way as to imply that it does not make sense in general to talk about interlinguistic, absolute synonymy, analyticity etc.) So how are we to understand the underdetermination from a Quinean perspective?

As far as I can see, what remains of the underdetermination thesis after the elimination of the terminology that presupposes the coherence of the notion of meanings is the claim that there can be two theories exploiting very dissimilar abstract machineries but fulfilling the same explanatory and predictive purpose. "Very dissimilar" and "the same" are to be taken in some vague intuitive sense. If worst comes to worst, we can always fall back on the method of pragmatic assent to define some rough material equivalent.

In order for such a thesis to be really interesting, it would be necessary to make more precise what we mean by "very dissimilar": whether it is true or false certainly hinges on what we are prepared to count as very dissimilar.37

Underdetermination does not by itself imply indeterminacy of translation; that we saw in §2.1.4. Now we have just seen that indeterminacy of translation would make it necessary to reformulate the underdetermination thesis. Whether indeterminacy of translation would bring with it underdetermination depends upon how different are the various rival translations, whether they would count as "very dissimilar". It also depends upon whether they would fulfil the same explanatory and predictive purposes: typically, perhaps, the rival manuals would each have a distinct set of virtues, making them suitable for different objectives, so that they would not constitute an instance of underdetermination.


3.5 My opinion

I do not think that the notion of meaning is so confused that we should cease using it; indeterminacy of translation is at worst of a modest degree. Meanings will do more good than harm, especially if we are aware that in particular among the more abstract parts of discourse there might be some arbitrariness in the assignment of logistic meanings to sentences. Quine's reasonings may serve to alert us of this condition.

I am not in sympathy with those philosophers who use arguments based on the theory of classical semantics, its mentalistic side, to downplay the philosophical significance recent developments in neuroscience. I think that just as Carnap's philosophy was in some sense succeeded by Quine's, so may Quine's find its transfigured continuation in neurophilosophy and AI. Quine's verdict on Paul Churchland's latest book "an outstanding philosophical achievement"38 indicates that il Maestro shares this view, that the naturalisation of epistemology and semantics, and even ethics and aesthetics, are to be pursued in terms of neural networks and other concepts from the converging lines of research in the cognitive and brain sciences and AI.

Your comments are welcome: [email protected]

Back to Main Page



Bennett, J. Linguistic Behaviour

Churchland, Paul. The Engine of Reason, The Seat of the Soul MIT Press, 1995

Kirk, R. Translation Determined
------------'Interpretation, Idealisation, and Norms.', Phil. Stud., 70(2), 213-223 (1993).

Massey, G. J. "Indeterminacy, Inscrutability, and Ontological Relativity", 1978. In "Studies in Ontology", Am. Phil. Quat. Monograph series, No. 12, pp. 43-55

Quine, W. V. From Stimulus to Science Harvard University Press, 1995
------------ "Methodological Reflections on Current Linguistic Theory", Synthese, 21, pp. 386-398, 1970
------------ "On the Reasons for Indeterminacy of Translation", J. of Phil. 1970
------------ "Ontological Relativity", J. of Phil., LXV, 7 (April 4, 1968)
------------ Pursuit of Truth Revised Edition
------------ The Roots of Reference
------------ "Three Indeterminacies", in Perspectives on Quine, edt. Robert B. Barrett and Roger F. Gibson
------------Word and Object


1 ibid., pp. 55-56

2 Pursuit of truth, p. 37

3 First we make meanings relative to interpretations. Everything means anything, but according to different interpretations. This pencil here has the meaning "It will rain tomorrow." according to an interpretation that assigns that meaning to it; according to other interpretations it means other things. An interpretation is just a function from objects to truth conditions.
Then we can have theories about these interpretations. These theories may contain a predicate "Correct interpretation(x)", but these predicates are relative to theories. (I speak of theories here in the sense of "systems of doctrine or belief" rather than in the logician's sense, "set of sentences".) So in general we have the predicates "Correct interpretation in the sense of theory #1 (x)", " Correct interpretation in the sense of theory #2 (x)", etc.
One of the most important fields for which such a theory containing a correct-interpretation predicate is suitable is the field of human behavior, or many aspects thereof. It is obvious that we need several theories of human behavior, or rather several levels of explanation which may be integrated and reduced in an overall theory. (The reason, of course, is that the amount and kind of information available, and the detailedness of the predictions needed, and the amount and type of computational or intellectual power at hand, varies strongly from occasion to occasion. A theory that is good when we want to understand the details of the co-ordination between eye and hand would be bad when we try to find out in daily life whether someone is likely to respond aggressively to an obtrusion, and vice versa.3 )
One straightforward theory of humans is that they are perfectly rational beings, infinitely intelligent, acting so as to maximize expected utility. To give full content to this theory one has to specify (1) what a persons preferences are, (2) what basic beliefs about data he has, and (3) a canon of rationality. The latter canon could be taken as a law of induction plus logic. Then to predict what his behavior will be, one simply calculates, for each possible action, the probability according to his canon of rationality and the basic beliefs that it should have a certain outcome and multiply this probability with the value he attaches to it. The prediction is then that he will perform that action for which this product is greatest.
Is this theory true or false? That cannot be determined until we have made clear what we mean by all the terms in our formulation of it. It is trivial that there is some assignment of preferences, basic beliefs, and canon of rationality that would make the theory yield the right predictions. But if the theory should be of any use, we must know something about which of these assignments we should make. Here is a thing to notice. The theory, or the part of it which can be explicitly formulated, need not say exactly what assignment should be made: that may be left to the judgement and intuition of its users. As long as the theory says something that may be helpful to its users in making this assignment, it may be of value.
So before we can say whether the theory is true or false we have to make some specifications. Let us decide that "preferences" and "canon of rationality" and "basic beliefs" should be understood in the pretheoretic, intuitive sense. The more diffuse this sense, the weaker the theory; but it will not be wholly devoid of might as long as "preferences" and "canon of rationality" are not wholly nonsensical: which they aren't. Now, the theory, as we have formulated it is certainly a poor theory for human beings, (but perhaps not so bad for a chess computer). Many would say that the theory is decidedly false for humans; but I hold that there is one merely moderately stretched sense of "basic beliefs" which makes the theory true, albeit weak.
Let us disregard for the moment that part of the canon of rationality which consists of a law of induction and concentrate on what is derivable from pure logic. Suppose a persons' preferences are specified as ordered pairs of a sentence and a number, the number being a measure of the value the person attaches to the state of affairs expressed by the sentence. If we then took the basic beliefs to be exactly mirrored in the atomic sentences the person would assent to, the above theory would be empirically false, if those atomic sentences are understood in the usual way, i.e. when we regard them as carriers of logistic meaning. For people do not possess unlimited intellectual powers, and they often act irrationally, even if we only consider their conscious, deliberate actions. On the other hand, we could stipulate that the atomic sentences the person believes in should be assigned meanings such that the theory holds true about his conscious, deliberate actions. Here we come to the thought meanings. The interpretation of the sentences in which a person believes and disbelieves that makes him perfectly rational is the interpretation which assigns (his) thought meanings to the sentences. The thought meaning of a sentence varies strongly from individual to individual and from one minute to the next. The thought meaning of a sentence for a speaker at a point of time is the meaning of that sentence in that person's idiolect at that time.

4 e.g. ibid., pp. 46-47

5 ibid., p. 42

6 ibid., p. 42

7 ibid., p. 45

8 ibid., p. 45

9 ibid., p. 46

10 ibid., p. 47

11 I was pleased to find that Quine would use this very term on a video-taped interview, produced by Rudolf Fara. Quine said: "I do not believe in the charity principle; I think that beliefs should be assigned on the basis of psychological plausibility.." (I quote from memory, so it might not be verbatim.)

12 Methodological Reflections on Current Linguistic Theories, Synthese, 21, pp. 386-398, 1970

13 In analogy to what Quine has called "semantic assent". In semantic assent you simply switch from using a sentence to mentioning it. In pragmatic assent you switch from claiming a proposition to claiming something about peoples' reaction to the sentence supposed to express that proposition. I think that the method of pragmatic assent deserves an investigation of its own, but I will not undertake it in this paper. Pragmatic assent has been used before, in Turing's well-know test for conscious intelligence, for instance; though I do not know that it has been recognised as a general philosophical method. 14 The "stimulus meaning" of a sentence for a speaker is the ordered pair whose first component is the class of all stimulations of an endurance not longer than the time interval called the modulus that would prompt his assent, and whose second component is the set of those stimulations within the modulus that would prompt his dissent.

15 G. J. Massey, 1978, "Indeterminacy, Inscrutability, and Ontological Relativity".

16 The dual to a sentence is obtained by replacing all predicates with their inverses (i.e. new predicates true of all and only those objects of which the original predicates were false), all conjunctors by their duals (i.e. "and" by "or"; "or" by "and" etc.; negation remains unaltered), and existential quantifiers by universal quantifiers and vice versa.

17 Of course, Quine does not think it does. (See The Roots of Reference, §12, for some naturalistic criteria for assent and dissent.)

18 Translation Determined, pp. 178-182

19 See, e.g., Quine´s "Three Indeterminacies" in Perspectives on Quine, p.13 20 "On the reasons for indeterminacy of translation", p.183

21 ibid., p. 181

22 ibid., p. 182

23 Quine, Pursuit of Truth, pp. 50-51

24 J. Bennett, Linguistic Behaviour, p. 262

25 He maintains his claim in an article published in 1993: 'Interpretation, Idealisation, and Norms.', Phil. Stud., 70(2), 213-223.

26 R. Kirk, Translation Determined, 214.

27 ibid., pp. 214-215.

28 ibid., p. 219.

29 ibid., p. 221.

30 ibid., p. 225.

31 ibid., p. 227.

32 ibid., p. 235.

33 See e.g. Quine, From Stimulus to Science, p. 72

34 For simplicity I here regard sentences with scope ambiguity or with an otherwise ambiguous surface syntax (e.g. "Every boy in the class is in love with a girl who is older.") as nonambiguous , having as their meanings the disjunctions of the meanings of all their possible disambiguations. Since indeterminacy due to scope ambiguity is not what Quine has in mind and has never been taken as a serious threat against the notion of analyticity, it may be expedient to work with meanings that are defined so as to spare us qualms about that sort of ambiguity.

35 I am therefore slightly pessimistic about attempts to provide strict proofs in philosophy; that a philosophical reasoning is very explicit and progresses in minuscule small steps -like Kirk's "proof" of determinacy of translation- tends, in my view, often to make it take longer time to see it through. But who can tell? -quite possibly philosophy will profit in the long run from what looks like unfruitful formalism and a tedious roundabout way of dealing with things.

36 Why? Well, because of the following argument, which we may also take as a contextual definition of what we mean by the critical terms.
Assume that T1 and T2 are logically incompatible. Then there is a sentence S, such that T1 implies S and T2 implies ¬S. Since T1 and T2 are empirically equivalent, S must be nonempirical. S must also be contingent. This is because otherwise S would be either contradictory or necessary. Then one of the theories, T1 say, would be contradictory. But a contradictory theory implies every sentence, in particular B&¬B, for some observation sentence B. B&¬B is empirical, since it can be empirically falsified by verifying B or verifying ¬B. However, T2 implies ¬(B&¬B), so T1 and T2 would disagree about an empirical sentence, contradicting the premise that T1 and T2 are empirically equivalent. Hence S is contingent and nonempirical. This means that ¬S is contingent and nonempirical. Either S or ¬S is true. Therefore there is a contingent nonempirical fact.
Conversely, assume that there is a contingent, nonempirical fact, claimed to obtain by the sentence S. Let T1={S} and T2={¬S}. Then T1 and T2 are logically incompatible. Remains to show that T1 and T2 are empirically equivalent. Assume, for a contradiction, the opposite, i.e. that there is an empirical sentence S* such that T1 implies S* and T2 implies ¬S*. But then S* is equivalent to S, so S would be empirical too, contrary to the premise. It follows that T1 and T2 are empirically equivalent and logically incompatible. Thus the indeterminacy thesis, under this interpretation, is equivalent to the proposition that there is a nonempirical (contingent) fact.

37 More generally, I think holism in any form will always be a matter of degree, and we should not expect that any fruitful discussion of holism is possible without making some effort to give a quantitative specification of what degrees we are talking about. Even more generally, I think that philosophers should try harder to quantify their claims. We are no doubt a long way from treating philosophical problems fruitfully with mathematical formulae and numerical measures: it might not even be possible; but there are other, more modest ways of giving some rough indication of how much there is of something. At the very least, we could state that something is a matter of degree when it is so, and then try to explain the relevant dimensions along which this degree is in principle to be specified. This means starting from a typical philosophical problem and deliberately working our way down towards the empirical science. The opposite movement in philosophy is to staple abstraction upon abstraction, in a scholastic manner, and it too may have some justification; but at the present I think it would be good to have more of the downward inclination.

38 Paul Churchland, The Engine of Reason, back cover.


Nick Bostrom's Home Page