"The thesis is then this: manuals for translating one language into another can be set up in divergent ways, all compatible with the totality of speech dispositions, yet incompatible with one another. In countless places they will diverge in giving, as their respective translations of a sentence of the one language, sentences of the other language which stand to each other in no plausible sort of equivalence however loose." (Quine, WO, p. 27)
While preserving internal consistency and correlation to empirical events, numerous mutually incompatible "translation manuals," or systems for translating any given utterance, can be assembled for the language in question. No one translation manual is better than any other in the sense of translating utterances; all do this equally well.
Quine's concern here is not with translation of nuance and subtlety, though he would, of course, say that these too bear no one determinate translation. Quine's thesis, however, suggests much grosser differences in translation manuals than tone and nuance alone. Second, the theory is not about epistemology, but ontology. That is, it does not state that the true meaning of an utterance is always undetermined by the limited number of utterances we have heard, but rather that meaning is undetermined even in the face of all the evidence. Clearly, the latter is a much stronger statement, and it is what we will be concerned with here.
This theory, finally does not describe what a linguist would actually experience when doing radical translation. Indeed, Quine would agree that there often is, among the range of possible translations, one which is clearly the simplest and, for the linguist's purpose, the best. Further, in such a case, this translation is the one that should be used, not because it is the "correct" one, but because it reduces the cognitive gap between the aliens and ourselves and makes possible our thinking of the aliens as relatively like ourselves, thereby facilitating communication.
That a more simple-seeming translation exists is not evidence per se that this translation is correct, and makes no assertions about how the aliens are or are not like us. However, it is easier for us to think about their motivations and desires if these seem to be like ours. This "principle of charity" (Quine, in ibid.) means that in reality linguists will, in many cases, come up with similar translations. But despite the pragmatic use of such safeguards, it remains possible in theory, says Quine, that many radically different translations of different utterances could also be workable.
What, then does Quine's theory purport to encompass? Perhaps its scope will become more apparent as we trace his framing of the argument. Consider, says Quine, a field linguist engaged in a project of radical translation. Radical translation Quine describes as "translation of the language of a hitherto untouched people" (WO, p. 28); by this he means that the linguist has no interpreter and is dealing with a language wholly unrelated to the his. Additionally, the culture differs wholly from that of the linguist, so that the assembly of the manual is not "aided by traditional equations that have evolved in step with a shared culture" (ibid.).
Quine considers how the linguist would go about learning the language radically. He suggests that sentences given in response to an externally noticeable stimulus may be the linguist's best starting point. "The utterances first and most surely translated in such a case are ones keyed to present events that are conspicuous to the linguist and his informant." (ibid., p. 29). For example, if a rabbit runs by and a native says, "Gavagai," the linguist may guess the utterance to mean "Rabbit." Quine then considers how the linguist might verify this guessed translation, deriving the words for "yes" and "no" and then trying out the question "Gavagai?" in various situations in the presence and absence of rabbits and rabbit-like stimuli.
Confirmation that "Gavagai" may, in fact, mean "rabbit" comes when the linguist's queries are affirmed or denied when expected, in the obvious presence and absence of rabbits. Granted, the situation is somewhat more complicated than this rather Augustinian picture draws. Quine considers cases which could lead the linguist to error, such as a native's knowledge of a rabbit-fly, which hovers over rabbits invisible to the observer, or a native's misunderstanding the meaning of his query, "Gavagai?" in a specific context ("The native may dissent from 'Gavagai' in plain sight of the rabbit's ears, because the rabbit is in no position for shooting" (ibid., p. 39)). Despite such interference, however, the linguist is eventually able to establish with some degree of certainty the "stimulus meaning" of "Gavagai."
A word about stimulus meanings is necessary here, because they are central to Quine's indeterminacy argument. Quine defines affirmative and negative stimulus meanings; the former he defines, "for a given speaker, as the class of all stimulations that would prompt his assent" (ibid., p. 32) and latter similarly for dissent. Stimulus meanings provide a basis for Quine's conception of how two things can be equated; he states that "the imagined equating of 'Gavagai' and 'Rabbit' can now be stated thus: they have the same stimulus meaning," (ibid., p. 33), that is, that the same set of stimuli would prompt agreement to both "Rabbit?" and "Gavagai?".
The stipulation in the definition of "for a given speaker" is of course problematic for all situations of translation from a non-native language to one's native tongue. Presumably, under this definition, the linguist would find the two terms to be equivalent only after assessing the stimulus meaning of "Gavagai" for the native, and comparing it to the linguist's own stimulus meaning for "rabbit," assuming (as it appears to be) that stimulus meanings are in fact empirically verifiable. Indeed, such a procedure would probably much resemble Quine's linguist asking the native, "Gavagai?" under a variety of circumstances to elicit "yes" or "no" answers, as described before. However, such an exercise would yield knowledge only of that native's idiolect as compared to the linguist's idiolect.
This fact raises an interesting problem which Quine fails to address concerning stimulus meanings. In his essay, "Epistemology Naturalized," written about the same time as Word and Object (1965 vs. 1964), Quine gives an extended discussion of observation sentences. Here, he states that
"an observation sentence is one on which all speakers of the language give the same verdict when given the same concurent stimulation. To put the point negatively, an observation sentence is one that is not sensitive to differences in past experience within the speech community." (Quine, Ontological Relativity (OR), p. 87)
This assertion about observation sentences demarcates them from other sentences according to a criterion of agreement within a community. Does this, then, mean that non-observation sentences are not agreed upon by the speech community? If we take this definition to lay out sufficient conditions for S to be an observation sentence (i.e. if S is not sensitive to differnces in past experience within the speech community, then it is an observation sentence), then this conclusion is necessarily true: If S is not an observation sentence, then it is sensitive to differences in past experience, etc.
But if this is the case, then aside from observation sentences (whatever these may be), the linguist's careful assessment of a native's stimulus meanings for words will yield a detailed set of tentative equivalences for one speaker and one speaker alone. If we are to take Quine's assertion regarding observation sentences seriously, then a search for community consensus on stimulus meanings of sentences will be futile for all but the most simple and concrete of utterances.
In fact, the situation is worse yet, for what an observation sentence is, is itself in question. In the same essay, Quine asserts that "what counts as an observation sentence varies with the width of community considered" (OR, p. 88), depending on how much outside knowledge the observer brings to the experience. While he clains "we can also always get an absolute standard [of what counts as an observation sentence] by taking in all speakers of the language, or most," (ibid.), he does not elaborate on how this is to be done. It seems to me that taking the least common denominator, eliminating anything that is, to some member of the community (excluding "deviants" such as madmen and the blind), not an observation sentence, is the only way to do this that does not lead to the bizarre and contradictory situation of some speakers lacking the information needed to make an observation sentence. If this is our method, then though some sentences will remain, it appears that a remarkable number will no longer qualify as observation sentences, and will be sensitive to past experiences of the speaker.
Returning to the main argument, we see that this assertion has some important implications. In finding stimulus meanings for native utterances, the linguist may expect that all but the most concrete, "brute fact"-based of any given native's utterances may have idiosyncratic stimulus meanings. That is, this narrow class of least-common-denominator observation sentences aside, there will be no population-wide consensus on the stimulus meaning of any word.
Is "Gavagai" an observation sentence? Within some very specific bounds, I think it still is. However, consider these bounds. "Gavagai" is only an observation sentence when given under conditions that every speaker of the language (excluding "deviants") would agree upon. This rules out such cases as inferring the presence of rabbits from rabbit-flies, and cases in which prior knowledge of a rabbit warren, for example, would lead a speaker to interpret stirring grass and a flash of color as a rabbit. In fact, if we are to take the "least common denominator" stipulation to heart, as I think we must, then the stimulus meaning of the observation sentence "Gavagai" is probably something like "clear sight of a rabbit under good lighting for a long enough time to distinguish it from other similar objects," even if the stimulus meaning of the utterance for any one speaker probably allows more latitude.
For the purposes of the indeterminacy of translation argument, this appears to me to mean that the linguist can assemble a list of tentative equivalences between the utterances of any given native and the linguist's native language, but that, given Quine's assertions on the subject, there is little reason to expect population-wide consensus on the stimulus meanings, and therefore on the equivalents in the linguist's language, in cases aside from observation sentences.
Before resuming Quine's argument where we paused to consider stimulus meaning, one more brief observation: Though the preceding pages paint a gloomy picture about agreement even between native speakers of the same language, the fact that significant communication and discourse can and does take place between people every day should be some indication that either Quine (anyhow, my interpretation of Quine) is more pessimistic than necessary about actual consensus among native speakers, or that people understand one another based on information aside from matching up stimulus meanings. If the latter is true, this may be a case in which Quine's "web of belief," discussed in "Two Dogmas of Empiricism", comes into play.
Let us assume our linguist has succeded in constructing a tentative manual of translation based, for the sake of simplicity in the light of the preceding discussion, on the utterances and speech behavior of one native. This done, says Quine,
"insofar as the concept of stimulus meaning may be said to constitute in some strained sense a meaning concept, ... it would seem to constitute a meaning concept for general terms like 'red' and 'rabbit.' This, however, is a mistake. Stimulus synonymy of the occasion sentences 'Gavagai' and 'Rabbit' does not even guarantee that 'gavagai' and 'rabbit' are coextensive terms, terms true of the same things."
Even after constructing a manual according to the narrow rules defined so far, the linguist still cannot be assured that the apparently equivalent words point to the same things in the world. Consider, says Quine, all the extensions we can apply to "Gavagai." Consider, in addition to the sentence (0) "There is a rabbit," four more sentences with identical stimulus meanings:
The only grounds, says Quine, on which we can assume the first, simple translation (0), is that we take "for granted that the native is enough like us to have a brief general term for rabbits and no brief general term for rabbit stages or parts." There is, however, no feasible "critical experiment" one could perform to distinguish between them on the grounds of stimulus meanings, for "point to an integral part of a rabbit and you have pointed again to the remaining four sorts of things." More sophisticated means of distinguishing between the five (or more) possibilities, such as establishing how many "Gavagai" are observed when one rabbit is present and the like, are not feasible here, says Quine, because all such tests presuppose a sophisticated facility with the language, whereas our linguist has at the moment definite knowledge only of "yes" and "no."
By the same argument, though sentences such as "That F is the same as the one we saw yesterday" would distinguish between the cases ("That stage in the history of a rabbit is the same one we saw yesterday" is clearly not coherent), they can only be distinguished if the linguist knows the translation of "is the same one," for example. If this were unknown, nothing prevents a translation of "That stage in the history of a rabbit is part of the same animal as the one I saw yesterday." (ibid.).
This, therefore, is the argument for Quine's theory of the indeterminacy of radical translation. Utterances, even if their stimulus meanings are known, can be translated in more than one way that leads to a coherent and consistent system. Because translations are derived solely from utterances' stimulus meanings, several (sometimes very different) translations with identical stimulus meanings can be found for any given utterance in a language. Attempts to differentiate between these multiple translations through questions about identity and number will fail if the translator does not already have some facility with the language, something specifically assumed not to be the case in a situation of radical translation.
It is evident that this formulation leaves many questions open, as well as a number of objections which could be raised. The important question of the status of observations sentences in Quine's theory, which has implications both for the theory and for his philosophy of science, I leave for later discussion. Here, I will consider one class of objections with which Quine's theory will have to deal.
As formulated above, Quine's theory considers the possibility of distinguishing between translations via further questions such as "How many 'Gavagai' are over there?" Quine rejects this possibility, reasonably enough, on the grounds that a linguist at this stage would not know the words (or translations) for "how many," and other identity and quantity terms. However, let us carry the idea a few steps further. Is it possible that only one translation manual could exist which, across the entire language, gave an internally consistent set of translations in keeping with stimulus meanings?
The argument is intuitively appealing. Language feels complex enough that we ought to be able to build a mutually supporting network of meaning out of it. Indeed, it seems that this "gut feeling" about language is one important reason why Quine's counterintuitive argument has met so much controversy.
We can arrive at the idea of a complex, mutually supportive system like this by several different routes. It is not difficult to think of simplified systems for which, even if known information is incomplete, as it is here, the results are determinate. To draw an example from a grade school puzzle, if Rose, Iris and Violet are each wearing a flower, and none is wearing the flower bearing her name, and Rose is not wearing a violet, then the system is determined: with this sparse information we know the outcome for each variable (person) in the game. If this system is to provide a good analogy, two criteria must be met: Natural language must work by determinate rules in the same way as logic problems do; and there must be enough starting information available to reach a stage where the remainder of the "game" can be determined, leading to one correct translation. (Note that if the second clue were not there, we would have two possible solutions, equally correct).
We can come to this point by different reasoning: Consider the example mentioned above, "That rabbit is the same as the one we saw yesterday" vs. "That stage in the history of a rabbit is part of the same animal as the one I saw yesterday." To distinguish between these, we need to know the translations for "is the same" and "is part of the same animal," and which is which. While this is not possible based on what the linguist knows at this point, could it be that a further piece of information, preferably linked to an easily distinguished stimulus, would make the distinction? Perhaps we could test the truth-value by inserting another word for "Gavagai," to yield, for example, the putative translations, "That book is the same as the one we saw yesterday" vs. "That stage in the history of a book is part of the same animal as the one we saw yesterday." Trying these new sentences out on a native speaker under the apprortiate conditions (i.e. in the presence or absence of a book seen yesterday) might point out which is correct, and which meaningless.
There are a number of replies to be made to this objection that the language as a whole provides adequate information to yield one and only one correct translation manual. The first, I think, continues in Quine's vein: How are we to know the word for "book"? By the arguments above, the word might mean "book," "stage in the history of a book," etc. Suppose the phrase to be translated, in addition to "is the same" and "is part of the same animal," could also be "is part of the same entity." We then would be creating one of at least two coherent and consistent sentences, either "That book is the same as the one we saw yesterday" or "That stage in the history of a book is part of the same entity as the one we saw yesterday," again with no way of knowing which. That the regress will continue unless we can find one "hard" translation to begin with is evident. Stepping back has not helped, for without at least one firm piece of information, it would appear we have nowhere to begin.
Based on this argument, it seems that the intuitively appealing idea that we can find one "correct" translation manual because language makes a complex network of semantic and logical associations which only permits one such translation is not correct after all. It appears as if either language fails to work like a logic game (i.e. is not self-contradictory, has rules of logic and syntax for how to combine words, etc.) or the linguist lacks sufficient starting information about the language being studied. The first reply seems to address the latter failing (that is, the linguist begins with insufficient knowledge and so knows no unequivocal definitions, but could if given a "seed" of information, i.e. one or a few definite translations), while the second deals, in an admittedly weak sense, with the former.
Naturally, there are many more sophisticated arguments against Quine's theory of the indeterminacy of radical translation; I have chosen to discuss this one because its intuitive impact is so great and because, as mentioned, it seems to me a major reason for the opposition and difficulty of comprehension Quine's theory has faced.