Chapter 6
Translational Theories of Meaning
6.1 Introduction
In chapter 2 I suggested that there are at least three important traditions in twentieth-century theorizing about meaning: the intention-based, the causal, and the compositional-semantic traditions, as I called them. These traditions, I maintain, are not mutually exclusive and they intertwine in various ways. The present chapter is essentially about a, historically speaking, somewhat less important tradition, but a conceptually no less compelling one that also can be seen as intertwinable with the others. I will call this tradition the translational tradition in the theory of meaning. In this tradition, the basic idea is that meaning is to be explained in terms of relations of speakers to translation schemes of some sort which provide translations of public-language sentences into Mentalese. Let me begin by telling something of this tradition's recent history before I discuss the details of a theory of meaning which belongs primarily to it.
First a little of the prehistory. Quine is probably a central figure in the prehistory of the translational tradition that I have in mind now. For he seems to take the task of understanding the utterances of others in terms of an interpreter translating those utterances into the interpreter's personal idiolect. Quine speaks of the construction of translation manuals to aid in the radical translation situation, but he seems to expect that the work of interpreting in less radical situations is more-or-less the same:
Thinking in terms of radical translation of exotic languages has helped make factors vivid, but the main lesson to be derived concerns the empirical slack in our own beliefs.<1>
Quine, however, doesn't take the "target" language of the translations that make forinterpretations to be Mentalese. So I hesitate to see Quine as aiming at a translational theory of meaning in the present sense.<2> Still, the idea in Quine seems to be that language-processing can be understood in terms of possession somehow of a translation manual that takes you from utterances in the language of another to expressions in a language that you have some sort of pre-interpretational understanding of and vice-versa. And that is a view not far from the translational theorist's.
Another important factor in the prehistory of the translational tradition is the history of the sort of 'semantic' theory that began with Katz and Fodor's paper "The Structure of a Semantic Theory".<3> The scare quotes around 'semantic' are there, in fact, because today the Katz/Fodor-style 'semantics' is generally understood to be less a semantic theory, that is, less of a theory that says what expressions mean or denote, than a theory that translates natural-language expressions into a special symbolism of what Katz and Fodor called semantic markers which can be used to determine certain of the semantic features of the original expressions.<4> I'll return to this point in just a bit. But first I will say a little about what a Katz/Fodor-style theory is supposed to do.
The special symbolism of semantic markers is sometimes referred to today as "Markerese". The idea of Katz/Fodor-style theories, then, is that given the Markerese translations of natural-language expressions, certain semantic features of the natural-language expressions are supposed to be determinable. For example, in principle, it is supposed to be possible to determine whether a natural-language sentence is analytic or not, or whether it has a meaning at all, given just its Markerese translation. And given the Markerese translations of two separate sentences it is supposed to be possible todetermine whether the sentences are synonymous or not. The semantic competence of natural-language users is, according to Katz and Fodor (1963), supposed to be characterizable in terms of some such a translation scheme along with definitions in terms of Markerese expressions alone of a certain class of semantic features and relations, for example, analyticity, meaningfulness, meaninglessness, synonymy, etc.<5>
Beginning, as far as I can tell, with Bruce Vermazen's review, Vermazen (1967), a popular criticism of the Katz/Fodor-style theories began to circulate in philosophical literature. This criticism has it, more-or-less, that whether or not they could ever succeed in being able to predict the extensions of all those notions in the class of semantic features and relations with which it explicitly deals, such theories do not try to say what the meanings of expressions are; and saying what the meanings of expressions are is what a theory called a "semantic" theory should do: you can know the Markerese translation of any sentence you like and all the things Katz and Fodor say that you can figure out on the basis of this translation, but all that wouldn't entail that you knew what the meaning of the sentence is. So the criticism goes and so it got repeated in various forms in quite a number of places.<6> And for many philosophers, it seems, this sort of criticism settled the matter with respect to theories of meaning that relied on translation and so such theories were largely dropped from discussion in philosophical literature through most of the 70's and 80's.
One final aspect of the prehistory of the translational tradition is probably good to mention and this aspect will bring us right into the actual history I am aiming for.<7> Computer scientists often talk in terms of the 'semantics' of higher-level programming'languages' - languages like C, PROLOG, etc. - as being given by the 'translators' that take programs written in a higher-level language into a computer's executable 'machine-language'.<8> This talk by computer scientists was originally, perhaps, metaphorical, but George Miller and Philip Johnson-Laird suggested that human language processing could be understood in just such terms.<9> In fact, they took the computer metaphor seriously enough, it seems, as to even suggest that the semantics of natural-languages should be spelled out in procedural terms:
In our view ... [the] broader psychological theory of linguistic competence must be formulated in procedural terms.<10>
Fodor dropped the proceduralism of Miller and Laird-Johnson and suggested that natural-language understanding could be accounted for by a theory which involves translation of natural-language sentences into the LOT:<11>
...[I]t is characteristic of the organization of general purpose digital computers that they do not communicate in the languages in which they compute and they do not compute in the languages in which they communicate. The usual situation is that information gets into and out of the computational code via the operation of compiling systems which are, in effect, translation algorithms for the programming languages that the machine 'understands'. ...[I]f the view of communication that I have been commending is true, then these remarks hold, in some detail, for the mechanisms whereby human beings exchange information via natural languages. To all intents and purposes such mechanisms constitute 'compilers' which allow the speaker/hearer to translate from formulae in the computational code to wave forms and back again.<12>
Fodor also argued that the Vermazen-style arguments against Katz/Fodor-style theories were irrelevant to this claim about natural-language processing. It might be true that knowing the M-translation alone wouldn't tell you the meaning of the sentence it translated, but if the M-translation were processed in the appropriate way this wouldn'tmatter since understanding itself is constituted by processing of M-sentences according to the LOT theory. Understanding a public-language sentence, on the LOT view Fodor has argued for, is a matter of standing in a certain computational relation to an M-sentence token which is the result translation from the public-language sentence. This understanding is constituted, we might say, not by interpreting an M-sentence, but by processing it.<13>
Thus, we might characterize the Fodor view more generally in the following way. For a population to speak a language is for the members of the population to stand in some sort of cognitive relation to a translation manual that takes sentences of the language and provides their M-translations. This is the central thesis, roughly put, of the translational tradition that I am trying to call attention to here. A bit later in this chapter I will give a slightly more rigorous gloss of this thesis, but for now this will do.
The question of Mentalese expression-meaning now becomes prominent since, essentially, if a translational account of public-language expression meaning is correct, then public-language expression-meaning really depends on Mentalese expression-meaning. I will discuss this question below as well.
I don't know whether anybody picked up on the theme of translational theories of meaning between Fodor (1975) and Schiffer's recent paper "Actual-Language Relations" (Schiffer (1993)).<14> Schiffer's attraction to a translational theory can be understood in terms of his desire to do without compositional-semantics in the explication of the notion of expression-meaning. Schiffer provides what appears a somewhat more rigorous statement of the translational theory of meaning suggested by Fodor in Fodor (1975). But, in order to make his version of the theory more generally acceptable, he tries to do away with the empirical aspects of Fodor's LOT hypothesis by suggesting that there is a sense in which the LOT hypothesis is either trivially true or merely a useful but harmless metaphor.<15> I believe, as I argued in chapter 3 above, that Schiffer is right about the LOT hypothesis and that it can, indeed, be given a reading in which it is trivially true, or, at least, as trivially true as the supervenience of propositional-attitude properties on physical properties.
But I believe that Schiffer's translational theory of the actual-language relation is untenable and will argue so in this chapter. First, I will discuss Schiffer's theory and some of the motivations for its details. Then I will state what appears to be a rather serious counterexample to it and some further problems for it. I will discuss why I think that it is unlikely that Schiffer's theory is reparable given my counterexample. And after all of that is done, I will mention another important counterexample to Schiffer's theory that works wholly independently of the main one I discuss in this chapter.
6.2 Schiffer's Theory of the Actual-Language Relation
Before providing his theory of the actual-language relation in his paper "Actual-Language Relations", Schiffer defines two technical notions that he employs in the theory. These are the notion of a practice in [a population] P of meaning in [a language] L and the notion of an L-determining translator.
Schiffer's definition of the first of these notions is as follows: for any population P and language L,[P][T]here is a practice in P of meaning in L iff often when a member of P means q, for arbitrary q, he does so by uttering some sentence that means q in L.<16>
The notion of an L-determining translator is defined by Schiffer as follows:
[L]An L-determining translator is a language-processing mechanism that determines a mapping of each sentence of L onto a Mentalese sentence that means in Mentalese what the L sentence means in L.<17>
Schiffer's states his theory of the actual-language relation, then, as follows:
[T]A language L is used by a population P iff there is a practice in P of meaning in L and the processing of L utterances proceeds via an L-determining translator.<18>
I would like to give names to the two conjuncts on the right of the biconditional of [T]. I will call the first conjunct, "there is a practice in P of meaning in L", the employment-clause, and the second conjunct, "the processing of L utterances proceeds via an L-determining translator", the whole-language-clause. Let me discuss these in turn.
I will assume for the sake of argument throughout that the employment-clause states<19> something that is a necessary condition for a population to use a language. It is useful to see why, by itself, however, the employment-clause does not state a sufficient condition for a population to use a language.
Suppose that there is a practice in a population P* of meaning in a language L* and
that there is no language L*' that P* uses such that L*'L*. According to the definition [P]
this means, closely enough, that often when a member of P* wants to communicate some
proposition to someone, he or she will utter an L* sentence that means . But even if
there is a practice in P* of meaning in L*, there will still only be a finite number of L*
sentences actually uttered by members of P*. Consider the finite set Q*=df{
The whole-language-clause is intended by Schiffer to pin down what the employment-clause cannot. I will now discuss Schiffer's motivation for the whole-language-clause of [T].
Consider two members, June and Bingo, of some population P*. Suppose June utters some sentence in Bingo's presence which means among the members of P* the proposition that Metallica rules. It seems more-or-less harmless to suppose that Bingo understands June just in case Bingo comes to have the belief that June said that Metallicarules. But if Bingo thinks in Mentalese, then her belief that June said that Metallica rules will be realized by the tokening in her belief-box of an M-sentence that means in M that June said that Metallica rules. The M-sentence that means that June said that Metallica rules will have to contain a part that means that Metallica rules. That part, since it means the same thing as the sentence that June uttered, namely, that Metallica rules, can be said to be an M-translation of . Thus, somehow a translation of the public-language sentence into M can be said to have been achieved in the process of Bingo's coming to understand June's utterance of .
The picture of language understanding presented in this tale would appear to be generally correct given the LOT hypothesis, so that, more-or-less, for a hearer to understand what a speaker said in uttering a sentence that means a proposition among the members of the population to which they belong is for the hearer to have tokened in his or her belief-box an M-sentence that means that the speaker said, or meant, . Further, let us suppose that for any proposition , any M-sentence that means in M that some so-and-so said must contain an M-sentence that means . Then it would seem that a necessary condition, assuming the truth of the LOT hypothesis, for a hearer to understand what a speaker said in uttering a sentence that means a proposition among the members of the population to which they belong is that a sentence that means in M be tokened in the hearer.<21> So language understanding would seem to require translation into M if the LOT hypothesis is true.
But even if translation into Mentalese is necessary for understanding the sentences that actually get uttered in a population, it surely cannot be necessary for every sentenceof a language to be actually translated into Mentalese in order for the language to be used by a population. For infinite languages this would be impossible and if actual translation were required, no population could speak an infinite language.
Schiffer's insight at this point is that if there is a device that, in some relevant sense, determines a mapping from public-language sentences onto their M-translations, even if it is never used for actually translating most public-language sentences into M, then such a device almost surely must be what determines that a population uses the particular language it uses. Thus the whole-language-clause of his theory [T] of the actual-language relation. An L-determining translator for a language L, by definition, determines a mapping of L sentences onto their M-translations. Such a mapping, it would seem, is sufficient to pick out the language L because each M-sentence is uniquely tied to the proposition that is its meaning in M via the belief-making property it has. Thus, adding to the employment-clause the whole-language clause might seem, prima facie at least, to provide a necessary and sufficient condition for a population to use a language.
Now, supposing that Schiffer is right about [T] - I will argue just below that, in fact, he is quite wrong -, then the meaning of public-language expressions is determined by the meaning of their M-translations. And one might object to [T] if there was nothing to say about how M-expressions have their meanings. To say that an M-sentence means a proposition , it will be recalled, is to say that the M-sentence has a physical property - what I have been calling its belief-making property - such that having that property and being tokened in the belief-box is metaphysically sufficient for believing . So to explain how M-expressions have their meaning is to explain how the belief-making properties ofM-sentences realize the beliefs that they do. For Schiffer, this question is answered by the provision of a correct compositional-supervenience theory for M.<22> For the meaning of every M-expression is uniquely determined by a correct compositional supervenience theory for M.
I am willing to accept Schiffer's theory meaning for M, though the matter has been debated.<23> So I don't think there are serious objections to Schiffer's account of meaning due to a lacuna in that account with respect to issues of Mentalese expression-meaning. But, in fact, I believe that Schiffer's account of the actual-language relation is badly off the mark for other reasons.
I will now show why Schiffer's whole-language-clause, as it is stated in [T] fails to provide, along with the employment-clause, a sufficient condition for a population's using a language. Afterwards I will argue that there is no obvious way to repair [T] and that for interesting reasons this reflects poorly on the whole project of giving a translational theory of meaning.
6.3 A Counterexample to Schiffer's Actual-Language Relation
The counterexample I will now present is pretty straightforward. The basic idea is that an L-determining translator for a language L can be a proper part of an L'-determining translator for a language L' which is not identical to L. Processing of L'-sentences by a population P' will proceed via the L-determining translator since it proceeds via an L'-determining translator and the L-determining translator is a proper part of this L'-determining translator. That's the rough idea. I will now provide a little detailand show precisely how this sort of scheme can count as a good counterexample against Schiffer's theory [T].
Consider a population P1 that uses the language L1. Now suppose the following four things:
[C](i) the set S1=df{
(ii) there is a language L2, different from L1, such that there is a non-empty set S2 such that (a) S2L1L2, (b) S2S1, (c) there is a practice in P1 of meaning in S2, and (d) there are no L2 utterances by members of P1 that are not tokenings of S2 sentences;
(iii) in the head of each member of P1 is an L1-determining translator which is used in processing L1 utterances;
(iv) the L1-determining translators inside the heads of members of P1 each contain an L2-determining translator as well as a mechanism that determines the mapping fZ which is such that, for any , if is the value of the mapping determined by the L2-determining translator for argument , then fZ() is the M sentence that means in M what means in L1.
The situation described by these suppositions represents a counter-example to Schiffer's theory since, though it follows from these suppositions both that there is a practice in P1 of meaning in L2 and that processing of L2 sentences proceeds via an L2-determining translator, P1 does not use L2. Let me show why all this follows from [C].
Note first that the set S2 is a subset of L2: since every member of S2 is a member of the intersection of L1 and L2, and since every member of the intersection of L1 and L2 is be a member of L2, then every member of S2 must be a member of L2. Also, it follows from supposition (ii), along with the definition [P], that often when a member of P1 means some proposition , she or he utters a sentence that means in S2 to do so. Since every member of S2 is a member of L2, then often when a member of P1 means some proposition , she or he utters a sentence that means in L2 to do so. Thus, it willclearly follow that there is a practice in P1 of meaning in L2.
Supposition (iii) has it that processing of L1 sentences proceeds in P1 by way of an L1-determining translator. But since processing by way of an L1-determining translator proceeds, according to supposition (iv), by way of an L2-determining translator, processing of L1 sentences proceeds in P1 by way of an L2-determining translator. S2 is a subset of L1, so processing of S2 sentences proceeds in P1 by way of an L2-determining translator. And finally, since the only L2 sentences ever encountered by P1 members are S2 sentences, it follows that processing of L2 utterances by members of P1 proceeds by way of an L2-determining translator.
So it follows from the story supposed above that there is a practice in P1 of meaning in L2 and that processing of L2 utterances by P1 members proceeds via an L2-determining translator. According to Schiffer's theory [T], it should follow that P1 uses L2. But P1 does not use L2. So Schiffer's theory is false.
This counter-example plays on the simple idea that [T] does not say enough about the role of the L-determining translators in utterance processing. Roughly, as long as an L-determining translator is implicated in some way in utterance processing - it doesn't actually matter how it is implicated - the whole-language clause of [T] is satisfied. Even though, for a particular L, an L-determining translator in some sense does help "nail down", as Schiffer says, the entire language L, still the nailing-down done by an L-determining translator is beside the point when the translator is placed in certain processes like the one described by my supposition (iv) above. There is, presumably, a right sort of way that a language can be nailed down, and Schiffer has not succeeded in telling uswhat that way is.
It might at first appear that there is an easy remedy of Schiffer's theory in the face of my counter-example. Roughly put, just have the whole-language clause disallow an L-determining translator that is a proper part of some other translator. But such a requirement, supposing it could be stated with greater clarity, would be too strong. Consider a population P2 that uses language L1. Let L3 be a language radically different from L1. The members of P2, we can suppose, all have L1-determining translators in their heads and these are proper parts of L3-determining translators. But in processing of L1 utterances, let's suppose, only the L1-determining translator is used somehow. This situation, I believe, shows quite clearly that it is not necessary that a translator used to determine a language never be a proper part of some other translator which determines a different language.
The trick is to say something more than merely that an L-determining translator is used in processing of L-utterances: the necessary and sufficient conditions for its meaning-making use, as we might perhaps call it, need to be specified. But there are other problems still for Schiffer's [T] to which I now turn.
6.4 Hell-Determining Translators
I will now convert the counter-example presented in 6.2.1 into a sort of nightmare for Schiffer's theory. Then I will show that a rather serious objection to Schiffer's theory [T], independent of the sort of counterexample given in section 6.3, can be raised.
Schiffer tells us in [L] that, for a language L, an L-determining translator is "amechanism that determines a mapping of each sentence of L onto a Mentalese sentence that means in Mentalese what the L sentence means in L." But what counts as a mechanism that determines a specific mapping?
Any mechanism with inputs and outputs describable in physical or topic-neutral terms will determine what systems-engineers call a transfer function. A transfer function for a mechanism describes in physical or topic-neutral terms what output state will occur for the mechanism given some input.<25> In mechanisms used for doing the sorts of things that are commonly described as "computing functions", e.g., calculators or computers, it is not the inputs and outputs described is physical or topic-neutral terms that really determines what function we will say the mechanism computes, but both interpretations assigned to the input and output states and all sorts of complex idealizations that we make of the input and output states of such machines.<26> But such a heuristic/intentional notion of determination of a function is not something that Schiffer can rely on in his theory. If a function is going to be determined by a machine, by a mechanism, in Schiffer's theory, then there will have to be some good sense to be made of the notion of determination here in non-intentional terms.
Ignoring the interpretations that people place on input/output states of machines as well as other apparently intentional aspects of the attributions of functions to machines, it would seem the only sober view that one could entertain is that a function, or mapping, is realized, or determined, by a mechanism if and only if it is isomorphic to the transfer function of the mechanism.<27>
Using this notion of determination of a mapping to interpret Schiffer's [L], it willfollow that, for any language L, any mechanism with a transfer function isomorphic to
a translation-function, as I will call such things, fL from L-sentences to their M-translations will determine fL as well as all of the infinity of translation-functions
isomorphic with fL.<28> In, my counterexample [C] of section 6.3 above, I supposed that an
L2-determining translator was employed in utterance processing by P1 members. Let be
the infinite set of translation-functions isomorphic with the translation-function from L2
into M. And let ' be that infinite subset of such that each member of ' has as a subset
the finite translation-function from S2 sentences to their M-translations. For each function
f in ' there will correspond a unique language L such that L={
An important thing to note is that even if my counterexample [C] were somehow avoided, there would still be a somewhat-hellish consequence of Schiffer's theory [T] given the present interpretation of the notion of determination of a function by a mechanism. For consider again the L1-using population P1. Suppose that f1 is the translation-function for L1. There will be infinitely many languages different from L1 butwith translation-functions isomorphic to f1 . Let 1 be the set of these languages. Any L1-determining translator will also be an L-determining translator for each language L in 1 on current assumptions. And that will mean that each member of 1 will be a language used by P1 according to [T] even though the only member of 1 used by P1, by hypothesis, is L1.
The somewhat-hellish consequence, actually, constitutes a second serious issue for Schiffer's theory since it is independent of the problem raised in section 6.3 above. The somewhat-hellish consequence can be drawn from Schiffer's theory given the present reading of the notion of the determination of a function by a mechanism. The hellish-consequence can be blocked, that is, by somehow blocking the counterexample of section 6.3, but not the somewhat-hellish consequence. To block the somewhat-hellish consequence a reading of the notion of the determination of a function by a mechanism that is more fine-grained than the present isomorphism-reading, but that uses no intentional notions, will have to be supplied. I turn to this possibility now.
6.5 Determining a Function
There is a way that the hellish and somewhat-hellish consequences of section 6.4 might be avoided, but it would require giving up the thought that an L-determining translator was a mechanism in any ordinary sense of that notion and it will open itself to further counterexamples in part because of this.
An L-determining translator for a language L can be taken to be an M-expression - tokened in some "subdoxastic" attitude-box - that expresses exactly one function, viz.,the mapping which takes each L-sentence onto the M-sentence that means in M what the L-sentence mean in L. An expression that denotes a function is not generally considered a sentence, so I will suppose that an L-determining translator is an M-sentence such that for some translation-function f, the sentence logically entails for each L-sentence that the M-translation of is f(). And I will say that an L-determining translator, so construed, is an M-sentence that expresses a translation-theory for L.<29> An L-determining translator will now determine exactly one function by virtue of its having the specific belief-making property that it has.<30> So understood, an L-determining translator doesn't look like a mechanism exactly anymore. But, we can take Schiffer's talk of an L-determining translator as a mechanism as merely a sort of metaphor. Problems arise, however, because of this relaxing on the notion of mechanism here and I will get to these in a moment. For now I would just like it to be pretty clear that we do seem forced to the present reading by the arguments of section 6.4.
Two things should be noted immediately. First, Schiffer's theory now looks something like Chomsky's theory of the actual-language relation which I discussed in chapter 4. The only difference is that Schiffer has us standing in a cognitive relation of some sort to a finite definition of translation-function where Chomsky had us standing in a cognitive relation of some sort to a finite definition of a language, that is, to a CMT. Schiffer's theory has over Chomsky's, perhaps, that it is far more plausible that there could be a finite definition of a translation-function than of a language. But, just as it was seen that Chomsky needs to be clearer about the nature of the cognitive relation he takes language-users to stand in to CMTs, so Schiffer needs to be clearer about the attitude-boxthat his L-determining translators are thought to occupy. And I will return to this point in just a bit.
A second thing that should be pointed out is that Schiffer's theory now requires language-users to have propositional attitudes about neural states somehow since now an L-determining translator is an M-sentence that talks about such things. An L-determining translator, that is, specifies M-translations of public-language sentences. So it must use expressions that refer to M-sentences. But M-sentences are supposed to be neural states of some sort. So, if an L-determining translator is in some subdoxastic attitude-box of a language user, then that language user must be having a propositional attitude about all sorts of complicated neurological things. This might seem rather implausible, even on the assumption that this knowledge is unconscious. It seems prima facie pretty unintuitive to believe that people generally have some such knowledge, not to mention how unintuitive it is to suggest that people more than a couple of hundred years ago had unconscious knowledge of their own neural states. Of course, a defender of the view might try to explain how this could be by saying that language users don't need to think of neural states as neural states, that is, they could have some other mode-of-presentation of these states. That's fine as far as it goes, but to make such a story plausible more would have to be said. Though I do think this is a problem for Schiffer's theory, I am willing for the remainder of my discussion, at least, to ignore it.<31>
It should be obvious that construing L-determining translators in the present way, that is, as M-sentences that state a translation-theory for a language, though it apparently doesn't help at all with respect to the counterexample of section 6.3, saves Schiffer's [T]from the hellish and somewhat-hellish consequences. Isomorphism is not the issue. The translation-function determined by an L-determining translator will be unique because the belief-making property of the L-determining translator establishes only a single interpretation for it. So, we have apparently calmed the worries represented by the hellish and somewhat-hellish consequences.
But new problems will now arise. If L-determining translators are construed as M-sentences determining translation-functions, then utterance processing via an L-determining translator can no longer be seen as a necessary condition for using a language L. There are two counterexamples to the necessity of this condition which I will now discuss.
1. M-sentences by themselves don't do anything. They just sit in attitude-boxes where thought processes can make use of them. So, by itself, an L-determining translator, on the latest construal of that notion, can hardly be understood to be a mechanism in an ordinary sense. Mechanisms do things, M-sentences don't.
The following analogy may be helpful. You may have a book perfectly describing how to translate Finnish into Urdu, but by itself the book doesn't do anything. If the book, however, is given to someone who knows how to read it and make use of what is read, it can help to make accurate translations of Finnish sentences into Urdu.
An L-determining translator doesn't actually translate, but it describes, so to speak, what the results of translation will be for any input. Actual translation, if there is to be any, must be carried out by some mechanism or process that takes as input L-sentence tokens and produces as output M-sentence tokens. Let's call any such mechanism a -processor. Notice that a -processor for a finite language LF need not use an LF-determining translator, i.e., it need not make use of an M-sentence that describes a function from each LF-sentence to its M-translation: such a -processor could be hard-wired to produce the proper output for any given input independently of any M-sentence whatsoever.<32> A population that used such a -processor in an appropriate way in LF-utterance processing could be said to use LF. So clearly, to construe an L-determining translator in the way we have in the present section of this paper, will force [T] to state something unnecessary for language use. So unless we can find another way of construing the notion of an L-determining translator, we are faced with a dilemma: either hell-determining translators, that is, L-determining translators on the original isomorphism-construal discussed in section 6.4, or the failure of [T] to state a necessary condition for language use.
But, in fact, the dilemma can be avoided with a definition like the following, inelegant though it may be:
[L']for any item x and language L, x is an L-determining translator iff either (a) x is an M-sentence that expresses a translation-theory for L, or (b) L is a finite language and x a -processor hard-wired such that, for any L-sentence , if a token of occurs at x's input, then a token of an M-translation of occurs at x's output.
2. There is yet another counterexample to the necessity of processing via an L-determining translator understood as an M-sentence stating a translation-function. For suppose that some language enjoyed a CMT. This is surely possible for a finite language where a CMT could be merely a list of all the meaning facts of that language. But perhaps even infinite language can turn out to enjoy CMTs as well. Whatever thecircumstance, if a language enjoys a CMT, it is perfectly conceivable that users of that language can process utterances by some sort of knowledge of a CMT. But in that case neither [L] nor [L'] are broad enough to fit this case.
One tricky matter here is that Schiffer seems originally to have intended to capture just such a case with his notion of an L-determining translator as given in [L]:
The translator can take more than one form. It can use an internally represented grammar of L, la Chomsky; or it can eschew such a grammar altogether and use an internally represented translation manual la Harvey.<33> Perhaps there are other forms the translator can take.<34>
But, it is hard to know how to be entirely charitable to Schiffer here. It just doesn't look possible to read his definition of an L-determining translator in such a way that all that comes out true. He needs to be clearer about what he takes a mechanism to be and what he takes the determination of a function by a mechanism to be.
In any event, I don't believe that Schiffer can state a single simple condition that can meet all possible cases here. Of course, the following definition of the notion of an L-determining translator is possible:
[L'']for any item x and language L, x is an L-determining translator iff either (a) x is an M-sentence that expresses a translation-theory for L, or (b) L is a finite language and x a -processor hard-wired such that, for any L-sentence , if a token of occurs at x's input, then a token of an M-translation of occurs at x's output, or (c) x is an M-sentence that expresses a CMT for L.
It's not pretty, but it seems to avoid both the counterexamples I have raised here. What's sad is that this is just the sort of thing Schiffer seems to have wanted to capture with an elegant generalization. But also, it is not entirely clear that there is not some other way that has been overlooked by way of which utterance processing can go on. Soit is not entirely clear that [T] states a necessary condition for language use given the definition [L''] of the notion of an L-determining translator.
6.6 Repairing Schiffer's Theory [T]?
In section 6.3 I showed that Schiffer's theory [T] fails to state a sufficient condition for language use. In section 6.4 I showed a horrible consequence of this failure and I presented an independent argument against the sufficiency of [T]. The independent argument drew what I called the somewhat-hellish consequence from [T] and in section 6.5 I suggested a way that Schiffer's theory could avoid this consequence. But the price for avoiding the somewhat-hellish consequence was that [T] no longer stated a necessary condition for language use. The central issue here was Schiffer's definition [L]. I was able to patch up [L] as [L''] and avoid the two specific counterexamples to the necessity of [T] that I suggested, but I can think of no argument that demonstrates the necessity of [T] given [L'']. Still, I can't think of any obvious counterexamples to the necessity of [T] given [L''], so I am willing now to suppose tentatively at least that, given [L''], Schiffer's [T] does state a necessary condition for language use.
But what about the counterexample from section 6.3? For, whether a specific L-determining translator is a hard-wired L-determining -processor, an M-sentence expressing a translation-theory for L, or an M-sentence expressing a CMT for L, it still must be embedded in an utterance-processing system, and [T] seems not to care enough about the nature of this embedding. So the counterexample of section 6.3 seems to retain its force against [T], in spite of the change from [L] to [L''] and the supposition that thischange avoids the hellish and somewhat-hellish consequences as well as the statement of anything unnecessary by [T].
Can Schiffer's theory be saved from the counterexample of section 6.3?
I don't think so. I'll make some general comments about the sort of thing that I take Schiffer to be trying to do and then I will say why I don't think that he will be able to remedy [T] to avoid the sort of counterexample that [C] represents.
As I see things, Schiffer's insight really is that if we think in M, then the question of what nails down the language L that is used by a population P is equivalent to the question of what relation RP is such that P uses L just in case both there is a practice of in P of meaning in L and RP(P,fL), where fL is the translation-function which takes L-sentences into their M-translations.<35> We can call RP the actual-translator relation.<36> To give a non-trivial specification of RP is, clearly, to give the actual-language relation since every translation-function determines a unique language given the semantics of M. And we can say that the goal of a translational theory of meaning is to specify the actual-translator relation.
Schiffer's talk about L-determining translators being used in utterance processing is his attempt to specify the actual-translator relation. His suggestion, more specifically, is that there is some sort of thing - he calls it an L-determining translator - that is responsible for making sure that the correct translation-function gets determined, and then there is some sort of relation to this thing that is responsible for determining whether a person uses the language picked out by the translation-function determined by the L-determining translator. But Schiffer isn't terribly clear about either what an L-determiningtranslator is or what relation we are supposed to stand in to such a thing. Even if [L''] does better in saying what an L-determining translator is than Schiffer's [L] did, still, we don't know at all what relation we are supposed to stand in to such a thing in order to use a language.
Suppose that [L''] tells us what an L-determining translator is. A relation R must be specified such that for any population P, language L, and L-determining translator DL, P uses L just in case both there is a practice in P of meaning in L and R(P,DL). Since there are three separate sorts of things that can count as L-determining translators, this relation R will have to state the necessary and sufficient conditions for the use of each so that a language gets spoken. Thus, the form of R will have to be something like the following:
R(P,DL) just in case either (1) DL is an L-determining -processor and C1, or (2) DL is an M-sentence stating a CMT for L and C2, or (3) DL is an M-sentence stating a translation-function for L and C3.
To state the condition C2 would, in essence, be to complete the job that Chomsky left incomplete in his attempt to provide a theory of the actual-language relation. C2 will have to state the necessary and sufficient conditions for the use of a CMT such that this use constitutes language use in the appropriate sense. This surely is not a small task.
The statement of the condition C3, will be equally difficult and for reasons that it is somewhat easier to see than in the case C2. For C3 is the condition necessary and sufficient for the use of a translation manual such that this use constitutes language use in the appropriate sense. To believe that such a set of necessary and sufficient conditions can be stated is more or less like believing that we could provide a non-trivial set ofnecessary and sufficient conditions for the perfect-translating use of a book that specified the translation-functions from Finnish into Urdu and vice-versa. I suppose that if an Urdu speaker had such a book and that if that book were written in Urdu, there is some way that it can be used for her or him to get along in Finnish. But, if Finnish and Urdu are infinitely large languages, then I can't imagine what a set of necessary and sufficient conditions for the perfect use of this book could be such that the Urdu speaker having it would be a Finnish speaker. I am extremely sceptical that there is such a set of necessary and sufficient conditions.
It should be clear that the problems with specifying what might be called the perfect- translating use of a translation manual will be analogous to the problems of specifying what might be called the perfect interpreting use of a CMT. I believe that to specify either of these notions adequately will be equivalent in important respects to solving the problem of specifying the necessary and sufficient conditions for someone to be a Finnish speaker given mastery of Urdu and the above described translation manual. And I believe that this latter problem is impossible to solve if Finnish and Urdu are both infinite languages. Thus, I am extremely sceptical that a theory too much like Schiffer's theory [T] can be ever be true.
6.7 A Further Problem for Schiffer's Theory
Besides the above, Schiffer's theory seems unable to avoid the counterexample Loar raised to his own theory in Loar (1981) and that I discussed in chapter 5 above. Consider the case where, let's suppose, were any English speaker to utter 'Apollo set theevening sky ablaze', they would be understood as meaning that the sunset is intensely beautiful this evening. Presumably, that is, English speakers would have a device that translated 'Apollo set the evening sky ablaze' with the M-sentence 'the sunset is intensely beautiful this evening' (pretend that is an M-sentence). Nothing in Schiffer's characterization of an L-determining translator would exclude considering the process, whatever it is, that produced this translation from being an English'-determining translator where English' is just like English except that it pairs 'Apollo set the evening sky ablaze' with the proposition that the sunset is intensely beautiful this evening. Thus, English speakers actually speak English', on the present assumptions, according to Schiffer's theory. Thus, this is another objection to the sufficiency of Schiffer's conditions for a population to use a language. This problem will be important to the discussion in the next chapter.
6.8 Summary
Schiffer's theory has many problems. I have suggested some rather ad hoc solutions for some of these problems with the proposal of [L''] to replace Schiffer's [L], but the problem represented by the counterexample in [C] of section 6.3, I don't believe can be solved and I have given at least a rough indication of why I think this is so. Schiffer's theory, independently of all of that, also fails to accommodate the sort of case that Loar found to be a counterexample to his own theory in Loar (1981) and that I discussed in chapter 5.