lunes, 19 de diciembre de 2011

Blackbox Realism: On Quine and the Indeterminacy of Translation


On Quine and the Indeterminacy of Translation

      In this paper I seek to develop some considerations surrounding Quine's thesis for the indeterminacy of translation. As presented in his canonical Word and Object (1964), the thesis states that, for any pair of languages, different incompatible theories of translation, all adequate to the relevant available empirical facts, may be proposed. Thus, Quine seeks to undermine the idea that translation between two languages implies achieving congruence of meaning between them, if the latter is understood as entailing synonymy qua sameness of reference. In doing so, he casts doubts upon the co-dependent traditional notions of meaning, reference, and synonymy. This is supported through what Quine calls the underdetermination of translation by data, which states that the set of empirical facts rendered available for a translator can accept of many possible, incompatible meanings, all of them being adequate to the observable data. There are many ways in which words or expressions fit the facts in spite of being ontologically or referentially ambiguous. Therefore, for a given theory of translation, that theory's conditions for success do not depend on mapping synonymous expressions between the two languages. The translator cannot infer synonymy from the congruence observed in identical behavioral responses to stimuli, when contrasting expressions between languages. Rather, the job of the translator is to preserve the standard relations between given linguistic expressions and behavioral conditions for assent or dissent, relative to observation instances. Quine frames these views in his theory of 'radical translation', supported in the underdetermination of translation by data.
     I shall follow Scott Soames' reconstruction of Quine's argument for the indeterminacy of translation (IOT) by clarifying the relationship between three central concepts: stimulus meaning, observation sentences, and occasion sentences[1]. In the first section, I follow and expand on Soames' presentation of Quine in seeking a formal explanation of the abovementioned three interrelated concepts. I shall then propose that Quine's arguments for indeterminacy fail to account for non-observational criteria relevant for translation, which cannot be captured by the restriction of empirical data to stimulus meaning and observation sentences. In particular, I shall argue that it is possible to propose additional criteria for any two hypothetical alternative compatible theories of translation, which are supposed to share their stimulus meaning and so be equally adequate translations by Quine’s criteria. Although these considerations do not belie the argument for indeterminacy, they do set constraints to the scope of what the latter solicits us to conclude about what can be of empirical relevance for the task of translation. More specifically, it will allow us to consider the possibility that non-truth-functional linguistic expressions can be relevant for translation. I conclude that while Quine's theory of radical translation should be read as an argument for indeterminacy, it should not be read as a comprehensive theory of translation applicable outside of strict cases where the only relevant empirical data is limited to linguistic expressions that function as perceptual reports.
      I then briefly show some problems that arise for the thesis of indeterminacy, by considering it in conjunction with Quine's pragmatic endorsement of behaviorism and physicalism. I focus on the behaviorist disavowal of representational contents, the physicalist pragmatic endorsement of the truth of physics, and the thesis on ontological relativity according to which to be is to be the value of a variable. I suggest that the conjunction of these three theses create problems for Quine’s deflationary scientific realism, and also for the thesis of the indeterminacy of translation. In particular, I argue that his behaviorism can be read in two senses: an epistemological and an ontological one. In conjunction with his other two positions, both options leave the plausibility of the inscrutability of reference in demand of further justification for its claims.
I - The Indeterminacy of Translation
      Quine's focus on theories of translation is contingent upon his intention to advance a general point about language. His two main philosophical targets comprise at least two prevalent orientations in philosophy of mind and language from his time:
1) The logical atomist theory of meaning - which preserves Frege's idea that meaning is primarily a matter of a series of relations between sentential expressions and the items of reference correlative to those expressions. The main idea that the logical atomists adapted from the Fregean view was that, through propositional analysis, an eventual theory of language would yield an understanding of meaning that would allow the logician to neatly separate meaningful sentences from meaningless ones, by isolating those expressions that are apt for empirical verification from those which aren’t. In turn, discerning which sentences are verifiable and thus meaningful would rely on knowing which expressions succeeded at denoting an item of reference in the world. This is an empiricist and verificationist thesis.
2) The sense-datum theories of mind - according to which sense-data, in some form, determines the basic content for all further propositional attitudes, i.e. experience and so knowledge originates in our registration of sensorial stimuli. This (roughly neo-Aristotelian) position entails that a fundamental layer of mental content underlies all propositional attitudes, and so that sensibilia anchors linguistic acts on an analytically available referential frame.
       In relation to the logical atomists and philosophy of language, Quine preserves the verificationist constraint according to which the meaning of a sentence is determined by its conditions for verification or refutation. He also advocates the thesis that sentences are the basic constituents of meaningful expressions: “I follow Frege in deeming sentences the primary vehicles of meaning.”[2]At the same time, he rejects that an analysis of language can clarify meaning by explaining the congruity between expressions and referents, because it turns out that conditions for verification render items of reference ambiguous.  Rather than to seek an elucidation of meaning in terms of word-world representational mappings, Quine proposes instead that language is mostly a matter of social convention, i.e. of dispositions to react to linguistic stimuli by responding in accordance to behaviorally conditioned responses within given socio-cultural contexts, or as cases of what Brandom (1998) calls reliable differential responsive dispositions. The opening line from Word and Object thus reads: "Language is a social act... in acquiring it we have to depend entirely on intersubjectively available cues as to what to say and when." Quine's argument constraining the available empirical data for translation theories to stimulus responses turns out to be the fundamental feature of his views on translation, and the inscrutability of reference that the latter evinces.
     In relation to the sense-datum theorists and the philosophy of mind, Quine endorses the idea that linguistic acts begin in experience, and that the meaning we assign to words is contingent on our capacities for perceptual discrimination. That is, again, part of what is at stake in Quine’s epistemic restriction of translation to observation sentences. However, he rejects the idea of a ubiquitous layer of sense-content which could serve to render linguistic reference unambiguous, i.e. sensibilia has no determinate content which unequivocally and transparently fixes an item of reference across a community of speakers[3]. Alex Orenstein explains in that regard that “the conjecture of indeterminacy is that there is no reason to think, given the empiricism/behaviorism involved in translation and its ontological underpinnings, that translation is determinate [about reference].” (Orenstein, 2002, pp. 144) Rather, communication proceeds while leaving the item(s) of reference inscrutable through the available evidence. In this regard, Soames condenses the two central claims that Quine seeks to advance by using theories of translation: "i) that the class of all possible data for such a theory radically underdetermines the claims about meaning that it makes, and ii) that this indeterminacy could not be resolved even if we had access to all physical facts."[4] (Soames, 2003: Pg. 226) The former follows given the ontological and referential indeterminacy left undecided after observing linguistic behavior. The latter follows from the fact that adequacy to the (natural) facts that constitute the world admit of semantically incompatible expressions, i.e. expressions with divergent meanings can be adequate to the facts. Quine thinks the inscrutability of reference sets epistemological constrains about which ontological commitments other linguistic subjects target in their use of language, since ambiguity of reference persists. However, this does not entail that every possible theory or set of statements are on equal footing before reality: Quine also endorses the physicalist thesis according to which physics describes the totality of facts that structures the world. I shall try to show why Quine’s physicalism becomes difficult to reconcile with behaviorism below. It is important to understand, however, that Quine’s argument for the underdetermination of translation by data in fact depends on the realist thesis that a) there exists an ontological structure composed of true facts which is the world, b) that those facts are specifically those advanced by current bona fide physics. Let us explain how exactly Quine argues for the underdetermination thesis in this context.
      Given the verificationist constraint, all scientific theories, including theories of translation, must be tested against plausible observational data. Quine's argument will consist in trying to delineate what the relevant data amounts to for theories of translation specifically. It will turn out that the set of data available for the translator underdetermines the choice for translation theory, i.e. more than one theory may fit the data equally as well. From this, it follows that a theory of translation must include in its conditions for success a severe and fundamental epistemic constraint. That is, the underdetermination of translation by data entails the indeterminacy of translation. Following Soames (2005), let us propose a definition of these two central ideas[5]:
1) (The Underdetermination of Translation by Data) (UTD)
      Let L1 and L2 be arbitrary languages, and let D be the set of all observational truths (known and unknown) relevant to translation from one to the other. For any theory of translation T for L1 and L1, compatible with D, there is a theory T', incompatible with T, that is compatible with, and equally well supported by, D.
2) (The Indeterminacy of Translation) (IOT)
     Translation is not determined by the set N of all empirical truths, known or unknown. For any pair of languages and a given theory of translation T for those languages, there are alternative theories of translation, incompatible with T, that accord equally well with N and so that are just as adequate to the facts. There is no objective matter of fact on which they disagree, and no objective sense in which one is true and the other is not.
        The plausibility of the inference from IOT to UTD supervenes on whether Quine's argument for what constitutes relevant data is reasonable to uphold. We may provisionally anticipate that it is at least not intuitively obvious what a) observational truths amount to precisely in Quine's account, and b) why we should restrain ourselves to a consideration of such facts. The latter is motivated by the empiricist and verificationist constraint that theories should be tested against the backdrop of evidentially salient data. For translation theories, the task is to correlate expressions between different languages or dialects. The set of correlations that map expressions in one language to another is what constitutes a translation manual or translation theory. The basic constituents of such a theory will then yield statements of the form:
“Word or sentence s1 in L1 means the same as word or sentence s2 in L2.”
    However, as we have anticipated, attempts to anchor translation on reference-synonymy are undermined by the data. To see why this is the case we must ask what ‘behavioral data’ amounts to in the production of a translation manual. The hypothesis advanced by Quine is that this data involves the observation of occasions where the foreigner uses specific expressions in his native language, where the translator compares them to situations where we use expressions in our language. As we surmised above apropos the endorsement of the Fregean thesis, Quine advocates the restriction of meaning to sentential expressions, but we should remark that the latter may include one-word interrogative utterances, and holophrastically construed sentences such as ‘Here-is-a-rabbit!’ or ‘This-is-red!’ The anchoring of such linguistic behavior in non-verbal, observational stimuli provides the observational data that Quine deems essential for translation, i.e. for what he calls stimulus meaning. Coming to know the relation between conditions for assent or dissent relative to expressions in given occasions constitutes the basis for a translation manual. As Orenstein puts it: “Quine’s linguist offers a hypothesis equating two such sentences (one the native’s and another the linguist’s) and checks it against a native speaker’s assenting or dissenting to the native sentence in the presence of some non-verbal stimuli.” (Orenstein 2002, pp. 134) Without further ado, let us introduce the three essential definitions at work in Quine’s account, following and expanding on Soames presentation (Soames, 2005, pp 254-255):
(Stimulus Meaning - SM)
     The stimulus meaning of a sentence S (for a speaker at a given time t) is a pair of classes- the class of situations which would prompt the speaker to assent to S if queried (the affirmative stimulus meaning of S), and the class of situations which would prompt the speaker to dissent from S if queried (the negative stimulus meaning of S).
Occasion Sentences - OCS)
     S is an occasion sentence for a speaker if and only if the speaker's assent to, or dissent from, S depends in part on what the speaker is observing.
(Observation Sentences - OBS)
     S is an observation sentence in a language L if and only if i) S is an occasion sentence for speakers of L, and ii) the stimulus meaning of S varies trivially from one speaker of L to another.
      The third criterion tries to render the interference of background assumptions ineffectual, to make variability in stimulus response minimal, i.e. it excludes judgments in which collateral-information affects stimulus responses. Quine’s central idea is then that the empirical constraint to observation sentences, as defined by stable stimulus meaning across occasion sentences, yield data that does not sufficiently support reference-synonymy. The latter is therefore not part of the experimental material or the goal of translation manuals. The translator can at best achieve stimulus-synonymy, while reference-synonymy would require that one may unambiguously assert sameness of meaning, which entails sameness of reference. This impossibility is precisely what Quine’s argument is designed to prove:  “The recovery of a man’s current language from his currently observed responses is the task of the linguist [or translator] who, unaided by an interpreter, is out to penetrate and translate a language hitherto unknown.  All the objective data he has to go on are the forces that he sees impinging on the natives’ surfaces and the observable behavior, vocal and otherwise, of the native. Such data evinces “meanings” only of the most objectively empirical or stimulus-linked variety.”  (Quine, WO: pp. 28-9)
    As a result of the constraint to stimulus-synonymy, the empirical prediction for theories of translation will hold generally that:
(Empirical Prediction of Translation Theories)
    Translation of observation sentences must preserve stimulus meaning. If a translation theory states that an observation sentence S1 in L1 means the same as S2 in L2, then the theory predicts that S1 and S2 have the same stimulus meanings in their respective linguistic communities.
     With this in mind we might stipulate a revision of our earlier general formulation, for the theorematic statements advanced by translation theories, as follows:
“Word or sentence s1 is in L1 is stimulus-synonymous to word or sentence s2 in L2.”
   At this juncture, we should reiterate that the relevant data described in Quine’s argument is of two kinds: the observational data gathered in stimulus responses, and the totality of physical facts which constitute the world, and relative to which the underdetermination by observational data occurs. Whereas the former is available, relevant and tractable for the translator, the latter constitutes the factual background that can be equally adequate to different and incompatible translation manuals, and so to different sets of stimulus responses. The idea is then that a) observational data is referentially ambiguous and so that different translations are equally supported by such data, and b) the facts of physics do not help resolve this ambiguity.
     Quine’s proposed thought-experiment for radical translation asks us to imagine the task of translating a hypothetical native language called Jungle, and which bears little in common to English. Having already stipulated that the translator is capable of discerning the appropriate gestures/expressions in Jungle for assent or dissent, to determine occasions for negative and positive stimulus meaning in the native’s language, Quine considers a specific example in a translator’s attempt to translate the native expression Gavagai! He stipulates that we could find out that the natives will assent and dissent to the one-word interrogative Gavagai? in the same situations that we are disposed to assent and dissent to the one-word interrogative sentence Rabbit? On this basis, the translator might be tempted to conclude that both Gavagai and Rabbit are referent-synonymous. Such a hypothetical translator argues as follows:
1)      For any pair of expressions S1 in L1, and S2 in L2, it is possible to empirically determine that both expressions are synonymous.
2)      If two expressions are synonymous, then they have the same meaning.
3)      If two expressions have the same meaning, then they must have the same referents.
4)      Therefore, if two expressions are synonymous, then they have the same referents.
5)      Therefore, it is possible to empirically determine for any pair of expressions in distinct languages, that they have the same referent.
     As we have suggested above, Quine’s contention against this argument is to disambiguate between stimulus-synonymy and reference-synonymy. Although Quine agrees in that the traditional notion of meaning advanced by the tradition is construed in terms of reference-synonymy and implies it, his point is that translation can at best warrant establishing stimulus-synonymy, and from the latter reference-synonymy doesn’t follow.  It follows that premise (1) fails if not qualified to read ‘stimulus-synonymous’, given the empirical constraint set by stimulus meaning. Furthermore, premise (2) is also only sensible to uphold if one qualifies it to mean ‘referent-synonymy’, and so the inference to (3) requires such a qualification.
       At this point, it might seem as if Quine is oscillating between two possible ideas: is he trying to dispense of the notion of meaning altogether by showing that synonymy simpliter is evidentially undermined? Or is he trying to redefine meaning so that the coinage of stimulus-synonymy can be said to achieve a behavioral account of meaning? This question lies outside the scope of this paper, but it should just be remarked that in any case synonymy and meaning, as construed by the tradition’s focus on referential relations, remain the target of Quine’s argument.
    The second qualification that we must note at this point is that, given the constriction to stimulus meaning and observation sentences, the set of expressions that can be sorted out using the abovementioned procedure is limited. Since the sentential expressions which yield stimulus responses depend by definition on observation instances, they do not comprise sentences whose determination is intractable by such means. For example, expressions like Rabbits share genetic material with Hares or Columbus discovered America will not work under such circumstances (Soames 2005, pp. 229). This raises a question for Quine, as well as for translation tout court. First, in excluding such sentences, haven’t we obviated what is an obvious and crucial part of the task undertaken by real translators everywhere? Quine himself does not address why such sentences are not subject to semantic analysis at a loss for evidential support, which renders the hypothetical nature of radical translation seem less realistic in scope. However, Quine might insist that sentences in a language must be, in the last instance, tethered to plausible direct or indirect knowledge of stimulus responses relative to observation, or at the very least, perceptual instances. One cannot do without being able to correlate expressions to some sort of perceptual stimulus, since it is through the latter that all communication functions to anchor language on the world. That such information might occur indirectly (through third-party testimony, recordings of some sort, or otherwise) does little to change this fundamental constraint.
        To consider why such an argument is persuasive for Quine consider the following radical example: suppose that one finds what appears to be an Alien tabula with indecipherable inscriptions presumably coming from a community of Alien speakers, equipped with similar perceptual capacities to ours. We have no diagrammatic representation of these symbols anywhere, and we lack any contact or knowledge about how any of these expressions might relate to the situations that we might register through perception in specific occasions. It seems reasonable to suggest that it would then prove to be utterly impossible to translate or understand anything about this manual; the bare minimum required to understand a language is to know how certain experientially available situations trigger stimulus responses in the right instances, i.e. to understand how they relate to the world. Without any idea of which situations correspond to which expressions in the native language we couldn’t even be capable of proposing candidates from our language to serve for stimulus synonymy, let alone reference-synonymy. Thus, the paucity of criteria offered in radical translation is meant precisely to illustrate a minimal set of conditions for a translation theory, and not an exhaustive delineation of all possible cases. This response, however, still begs the question about what it is precisely about non-observational sentences that renders them evidentially trivial or empty; are there other forms of evidence besides those rendered in SM or OBS? If so, why are these secondary or trivial for translation? Could these serve to overcome the underdetermination of translation by data as restricted to stimulus responses?[6] We shall return to this issue below, but for now let us return to reference.
     So far we haven’t addressed how the underdetermination of data rests on the inscrutability of reference. Although Quine himself reminds us to keep these two theses separate, it is clear that the former is meant to be supported by the truth of the latter[7]. The question finally amounts to asking why stimulus-synonymy fails to entail reference-synonymy.  Assume a pair of sentences in two different languages with identical SM: s1 in L1 and s2 in L2. Suppose that the two expressions in question are the English Lo, a Rabbit! and the Junglese Gavagai! If both expressions are stimulus-synonymous it follows, by definition of SM, that those circumstances on which a speaker of English would assent/dissent to the one-word interrogative Rabbit? are the same as those in which a speaker of Junglese would assent-dissent to the one-word interrogative Gavagai? From this it might be tempting for the translator to conclude that Gavagai refers to rabbits, and so that Gavagai and Rabbit are not just stimulus-synonymous, but referent-synonymous, i.e. that they have the same meaning where the latter entails sameness of reference.
        However, is this really established by the data? Quine remarks that the native Gavagai could just as easily refer to an undetached-rabbit-part, a temporal rabbit-stage, the form of Rabbithood, and who knows what else. Each of those possibilities remains adequate to the stimulus at hand. Thus, Quine argues that “Given that that a native sentence says that a so-and-so is present, and given that the sentence is true when and only when a rabbit is present, it by means follows that the so-and-so are rabbits.”[8] In other words, while it might be perfectly true that Gavagai? is assented to by the native speakers in the exact same situations that Rabbit? is assented to by English speakers, the former’s reference remains indeterminate. Since the referents of both expressions could be dissimilar, it follows that sameness of meaning is not deducible from the evidence. This thesis is also called the thesis for relative neutrality, i.e. the data is not partial to one possibility rather than the other. All of these states correspond to the physical world, in which rabbit, rabbit-stage, and rabbit-part stimulations all yield the same stimulus response in the same real situation. Thus, commitment to one hypothesis rather than another requires supplementary commitments called ‘analytical hypotheses’. For example, in order to ask ‘Is this rabbit the same as that?’ the translator must decide on how to translate articles, pronouns, identity predicates, among other things. Translating into Jungle requires us to reach beyond what SM renders available (Orenstein, 2005, pp. 135). The selection of a translation manual involves the choice of such divergent sets of possible translations, all of which are equally supported by the data. This allows us to understand how the notion of incompatibility between theories is cashed out in terms of the failure of reference-synonymy. The following reconstruction offers the basic position[9]:
(Incompatibility Between Translation Theories)
      Let T1 be a translation theory containing statement (i), and let T2 be a translation theory containing statement (ii). The union of T1, T2, and a set which includes following premises is inconsistent:
a) Rabbits are not undetached spatial rabbit parts, undetached spatial rabbit parts are not stages or rabbits, and rabbits are not temporal stages of rabbits.
b) 'Rabbit' (as we use it now) refers to an object if and only if it is a rabbit, the same with every respective expression.
c) If two words refer to different things then they don't mean the same thing.
d) If a word of a phrase w means the same as a word of phrase x, and w means the same as a word or phrase y, then x means the same as y. (transitivity)
      This reconstruction obviously rests upon the reading, suggested above, that Quine would be first and foremost be looking to displace the notion of meaning altogether, rather than redefine it.  An alternative reading, which sees Quine as relaxing the notion of meaning, could simply qualify premises (c) and (d) to read ‘reference-meaning’ or ‘reference-synonymy’, thereby leaving it open that two expressions with different referents could nevertheless be said to enjoy sameness of meaning in the sense of ‘stimulus-meaning’. In any case, the point outlined above is that given a translation theory T1 that contains a statement (i), and an alternative theory T2 containing a statement (ii), the union of T1, T2 and the set containing (a-d) is logically inconsistent.  With this in place, we might offer a formal reconstruction of the argument for IOT, as follows:
(A Possible Reconstruction of the Argument)
1) For any instance of radical translation one must base oneself on observation sentences, and the stimulus meaning for the expressions proper to that community.
2) The meaning of a sentence or word is only intelligible in relation to the entirety of the sentences-words which compose a language.
3) For every theory of translation T1 which maps statements of the form s1 <> L1 = s2 <> L2 on the basis of observation sentences, there is a possible alternative theory T2 which has s3 <> L1 = s2 <> L2, and in which s1 and s3 refer to different objects.
4) If two terms refer to different things they have different meanings; synonymy implies sameness of reference.
5) Therefore, s1 and s3 are not synonymous.
6) Therefore, T1 and T2 are semantically incompatible, even though they both correspond equally as well to the observation statements and so to the data available by stimulus meaning.
7) Therefore, in theory, more than one possible translation, all equally adequate to the data rendered available by stimulus-meaning and to the totality of physical facts, and yet incompatible with regards to their reference-meaning, are possible.
8) Therefore, the data gathered by stimulus-meaning underdetermines any claims to semantic synonymy, insofar as the latter entails sameness of reference, i.e. UDT entails relative neutrality.
         At this point, one might wonder whether the thesis for the inscrutability of reference is over-hastily drawn. One could argue that what Quine takes to be an irremediable ambiguity resulting from the relative neutrality obtaining from surveying observation sentences might be nevertheless resolved upon further questioning. Specifically, if one could learn to ask the native more specific questions about the identity and determinate content of what he/she is speaking about, it is not clear why the ambiguity should persist. Quine in fact considers such a possibility, but thinks that nevertheless the inscrutability of reference persists:
     "It will perhaps be countered that there is no essential difficulty in spotting judgments of identity on the part of the jungle native, or even of a speechless animal. This is true enough for qualitative identity, better called resemblance. In an organism's susceptibility to the conditioning of responses we have plentiful criteria for his standards of resemblance of stimulations. But what is relevant to the preceding reflections is numerical identity. Two pointings may be pointings to a numerically identical rabbit, to numerically distinct rabbit parts, and to numerically distinct rabbit stages, the inscrutability lies not in resemblance, but in the anatomy of sentences. We could equate a native expression with any of the disparate English terms 'rabbit', 'rabbit stage', 'undetached rabbit part', etc, and still by compensatorily juggling the translation of numerical identity and associated particles preserve the conformity to stimulus meanings of occasion sentences."
(Quine, Word and Object, pp. 52-54)
         What Quine attempts to argue for in this rather cryptic passage is the following: the question about whether the native refers to the same object(s) as us depends on what, if anything, the native understands that is equivalent to our understanding of identity notions such as: is the very same thing as, or is identical with. But behavioral evidence does not decide this question. Imagine that the word squiggle is hypothesized as a candidate for expressing the notion of identity. One could utter Gavagai squiggle Gavagai? with the belief that if the native assents to my utterance he must be referring to a Rabbit as opposed to an undetached spatial part of a rabbit, or a temporal rabbit stage, etc. This would seem to follow because whereas the rabbit is one thing that remains identical from a time t1 to t2, a rabbit stage isn't. The same might hold for spatial parts, unless the rabbit was immobile. Of course, other candidates for reference might still persist, but the point would be that nevertheless there are ways to narrow down the ambiguity, and so that inscrutability may be resolved with sufficient effort.
            And yet, as far as the evidence goes, squiggle could mean identical with, is an undetached spatial part of the same extended whole as, is a temporal stage of the same enduring complex as, and various other iterations. Thus, for all we know, while the native could be assenting to the underlying belief in unified Rabbithood, he could still be referring to stages or parts, and a variety of iterations like the ones above. With this in place one might begin to wonder about the empirical constraint to stimulus meaning, which Quine thinks sets minimal conditions for translation, as we surmised above.
      The question would be whether the scope of data captured by the Quinean examples of radical translation truly captures the breadth of data relevant and available to the translator. This amounts to asking about whether stimulus meaning exhausts the relevant data for translation. In order to see why this is problematic for Quine, imagine the following scenario. It is at least not a priori ruled out that two expressions in a given language might be stimulus-synonymous. For the argument’s sake let us hypothesize that the interrogatives in Spanish Sombrilla? and Paraguas? are taken to be stimulus-synonymous. Furthermore, assume that a native Spanish speaker, looking to find the appropriate translation for these cases, soon discovers that the stimulus meaning for the English Umbrella? is stimulus-synonymous to both Sombrilla and Paraguas. According to Quine’s criteria, Umbrella could be translated accordingly by both terms, and there are no further grounds to decide upon this issue. Yet it is plainly obvious that such concerns by far delimit the scope of Quine’s thought experiment in a way that obviates empirical considerations taken by translators everywhere. For example, consider that the translator is deliberating on the abovementioned example in the process of translating a piece of poetry. Although determination of stimulus-meaning is part of the labor at stake, the translator must also, in deciding whether to use Sombrilla or Paraguas, pay attention to other salient factors, i.e. for example, the phonetic structure of the poem. It might be that translating Sombrilla for Umbrella allows the translator to preserve the rhyme-structure of the work, which would be destroyed otherwise, or preserve in it a case of alliteration in conjunction with other choices, in the process of conveying a particular idea, etc. Analogous examples could be given to show that stylistic decisions, while based on salient data available for the translator, are irreducible to stimulus-meaning. Although these considerations are not arguments against Quine’s thesis of UTD or IOR, they do provide a reminder of the restricted scope of his experiments.
    However, it turns out that the issue of competing expressions with stimulus-synonymy has more substantial consequences. To see why, it suffices to consider non-referent-synonymous expressions in two distinct languages, which would be assented / dissented to in all possible circumstances. These kinds of expressions are labeled by Soames stimulus-analytic sentences (Soames 2005, pp 234). For example, consider the English expression There have been dogs. It seems plausible enough to suggest that this is a sentence whose assent to or dissent from would bear little to no variance between speakers of the language, and further that assent to it would be ubiquitous across all instances, i.e. it is stimulus-analytic. The problem then emerges when we consider what we would take to be an obvious falsehood in the native language that is also stimulus-analytic. For example, ubiquitous assent to the interrogative Katamerai? in Jungle might actually mean This is the work of the omnipresent Sun-God! The problem is that, restricting ourselves to stimulus-meaning, we could fail to distinguish universally believed English truths from universally believed native falsehoods, and vice versa.
“This raises the possibility that two sentences might differ in meaning, even though utterances of them would generally convey the same information (and hence prompt the same assents or dissents) owing to the fact that utterances of one of the sentences generally would implicate a proposition that was part of the meaning of the other. In such cases, the difference in meaning between the two sentences would be all but invisible to Quine’s radical translator, and Quine’s constraints on the empirical adequacy of translation would allow the sentences to be assimilated to one another.” (Ibid, pp. 235)
     At this juncture, the weight of Quine’s thought experiment turns on the constriction to stimulus-meaning and observation sentences. The problems raised by considering stimulus-analytic sentences show not only that such a restriction might miss some relevant data for the task of translation, but that it can often lead to indistinguishing between semantically incompatible notions. Now, one could argue that this, far from being a limitation in Quine’s account, is actually a result of it: that he is precisely revealing the impossibility of inferring semantic determinacy from available data. However, this again turns on the assumption that we agree in that the constriction of data to stimulus-meaning does not obviate other relevant forms of evidence. As we have suggested above, stylistic decisions in translation include such considerations. The question of stimulus-analytic sentences which prompt expressions that are semantically incompatible, but functionally equivalent, suggests that assent or dissent is motivated by factors which exceed the strict truth-functional conditions of the sentential contents. In particular, conventional and conversational implicatures, and other Gricean categories come to mind. These are precisely the kind of background factors Quine looks to set aside in his notion of observation sentences, but how much it leaves out raises questions. As Soames puts it: “At some point, one must recognize the severe limitations he has imposed on himself, as well as the tentative and approximate character of his adequacy conditions on translation.” (Ibid. pp. 235) Additional evidential factors that he takes could amplify the scope of observational evidence include: situational features like the introduction of a word into a language (verbal definitions vs. ostensive illustrations), situations where individuals acquire competency in the use of the word (explanations vs. examples), spontaneous use of words without prompt (English speakers find it natural to use rabbit rather than rabbit-stage). There are all plausible criteria which could affect the translator’s job, and which amplify the scope of evidence relevantly. Nevertheless, it is not obvious that such considerations would alter the basic conclusion about indeterminacy.
     II – Behaviorism, Physicalism, Ontological Relativity
   At this stage, we might decide to call into question the reliance on observational evidence entailed in Quine’s endorsement of behaviorism. The latter allows him to proscribe from the relevant evidence considerations about beliefs, intentions, and all cognitive states; the contents of wishes and desires, and motivational states of the speakers; the contents of perceptual experiences and the relation of the latter to their environment, etc. All of these factors are not tractable through the salient behavioral evidence yielded by stimulus responses. 
     Yet as several commentators have noted, behaviorism is problematic. It is clear that in other domains non-observational facts come into play for supporting our theories and hypotheses, and Quine’s restriction to observationally salient data in stimulus responses leaves such factors out of the picture. Among others, Burge (2010) and Soames (2005) have pointed out that, in order for the indeterminacy argument to be compelling, Quine would need to provide an independent argument for behaviorism, which he does not do. Yet it is not clear how departure from behaviorism would help overcome the indeterminacy argued for by Quine, say, by a hypothetical inclusion of any of the abovementioned items into the set of relevant data. Some commentators, like Soames (2005), argue that aggregating facts about the neurophysiology of individuals does not intuitively help refute the indeterminacy thesis. Specifically, he doubts that the IOR may be resolved through facts about the anatomy of the subject, or information about the content of mental states. Let us pause on this issue for a moment.
      Soames calls to question the exclusion of these factors, insofar as the restriction to stimulus responses and observational evidence is supported by an endorsement of behaviorism that is itself not argued for. I just add to Soames that without a proper explication and justification for his behaviorism, precisely what kind of conclusions we should draw about the abovementioned items of data is up in the air. One possible reading is that Quine endorses an ontological behaviorism, that seeks to destroy the notion of representational content altogether, and so to claim not only that reference is ambiguous because of an epistemic limitation in observers, but that there is no determinate relation between words, mental states, and things that could clarify the concept of reference in terms of representational content[10]. Sometimes Soames seems to read Quine in this direction, for example, when he claims that, for Quine, “[T]he ordinary notions of meaning and reference are rejected as illegitimate pre-scientific concepts that have no place in a scientifically respectable description of the world... The problem in Quine’s view is not that we are in danger of forever remaining ignorant of the facts about meaning and reference. The problem is that there are no genuine facts to be ignorant off” (Soames 2005, pp 226) How are we to position this thesis alongside Quine’s conjoint commitment to metaphysical physicalism and semantic holism?
     Recall that Quine (1969) endorses ontological relativity, understood as the position according to which existence is to be the value of a variable, i.e. an argument for an existentially quantified true statement[11]. This is part of Quine’s inferentialist commitment which follows from the claim that empirical content undeterdetermines the choice of theory. An ontological disavowal of representational contents under this reading would entail that there are no true facts about reference to fulfill this role, i.e. one doesn’t quantify over ‘referents’, if by the latter we mean something that is supposed to inhere outside our ‘referring’ expressions. 
      Another way to make the same point is to say that reference is nonsensical except by taking it relatively to a background theory, never ‘absolutely’ or with regard to anything ‘external’, i.e. Quine’s position is that of a deflationary realism in which the objectivity of the world is thinly construed in terms of the holistic network of inferential relations that hold between expressions in a given language, bereft of all transcendent skyhooks. In that regard, Quine seeks to exemplify through the indeterminacy of translation and the inscrutability of reference the extent into which ontology is relative to background theoretical assumptions which are socially convened upon and so holistic in nature. As FØllesdal (1995) claims, “Quine, more than any other philosopher, has made us see the far reaching implications of the public nature of language.”[12]
     This reading creates a problem when attempting to reconcile behaviorism with Quine’s simultaneous endorsement of physicalism. If the notions of representational content and of reference have no value; why and how are scientific statements any different than any other, if at all? What grants scientific description, and our understanding of it, a genuine prerogative in adequately describing the determinate structure of the world, without ambiguity? That is, how is it that physics and only physics yields ‘true facts’ about what it existential quantifies? McDowell (1996), similarly to Quine, has claimed that deflationary standards for truth and ontology show that there is nothing more to the reality of beings or properties besides there being true statements about them. However, he thinks that there are many statements which we take to be true, and which are not intuitively part of natural-scientific discourse. Notably for McDowell, value-statements can be also said to be true or false, and do not transparently follow from truths about physics, i.e. they are not logical consequences of physics. It is even less obvious that mundane truths formulated in common language can be shown to be entailed by physics. In this regard, Soames (2005) reasonably claims that to believe that statements such as A car exists, or I own a blue car could be shown to be logical consequences of physical truths is, for the moment, an utterly speculative thesis which exceeds any evidential support made available by current physics. He proposes thus to read the determination of all truths by physics as one of mere (metaphysical) supervenience, as opposed to strict logical consequences. Whereas to construe a given fact q as a logical consequence of a physical fact p requires that q be deducible from p, supervenience only implies the weaker claim that a given entity q could not exist without p. Yet even with a weakened sense of supervenience, it is not clear where from the authority of physics derives, nor the specific relation it bears to other (presumed) truth-functional sentences. These controversies show that even in denying the valence of the concept of reference, problems remain with respect to the endorsement of physicalism. If Quine wants to deny that the notion of reference plays any part in current bona fide physics, and so that the latter need not remain tethered to it, he still needs to explain how it is that scientific discourse describes a mind independent world. Such an explanation would need to be epistemologically explanatory in showing how one can disambiguate between our possible descriptions of the world and the world itself. But a deflationary account which constraints realism to existential quantification seems difficult to reconcile with such a demand since, like McDowell suggests, all kinds of statements appear to be truth-apt which do not seem to be about physics, and which are not construed as consequences of the latter. And it is not clear that in the weakened sense of metaphysical/material supervenience, statements existentially affirming the valence of cars and other middle sized items can be taken to be true on virtue of lower level facts about non-observables, 'real patterns' (Dennett; Ladyman and Ross), or anything else. An account of the relationship between the semantic content of conventional statements and those of physics would be in order, even if some have argued that a physicalism construed around supervenience is empirically supported by current science[13]. On any account, ontological relativity coupled to ontological behaviorism would then run into problems when explaining the relationship between mind and world, beyond the constraints of language and social convention, in order to assert the determinability of a mind-independent world required by any realist account.  It is important to see that ontological behaviorism becomes of a piece with the inferentialist core, in entailing that there is nothing more to representational contents than propositional contents, while insisting that grasping the latter has nothing to do with finding out how concepts match some item in extra-experiential reality, and so nothing to do with reference. The representational ‘content’ of a proposition is taken to be nothing but the inferential relations into which it enters.
      As we saw above, standards for truth and falsity are socially convened upon, rather than anchored on a non-conceptual world by necessity[14]. At the same time, philosophy endows science the prerogative before the real. Thus the ‘totality of true facts’ which make up the world and which physics describes are simultaneously taken to be a matter of a rational pragmatics subordinated to social convention, and a matter of grasping correctly a determinately structured mind-independent reality. These two aspirations in Quine’s realism are not so easy to reconcile. On the one hand, behaviorism entails that there is nothing more to grasping a concept than to grasp its conditions for use, and the latter is simply to understand the inferential relations the concept can enter into, including which counterfactuals it can support in specified contexts. But this severely restricts Quine’s capacity to explain how physics in particular can relate to a mind independent world, since the truths of physical statements are, like those proper to any other discursive register, subjected to pragmatic convention. As Quine himself claims, the distinction between physical objects and other objects ends up being one of degree rather than kind: “As an empiricist, I continue to think of the conceptual scheme of science as a tool, ultimately, for predicting future experience in the light of past experience [. . .] The myth of physical objects is epistemologically superior to most in that it has proved more efficacious than other myths as a device for working a manage-able structure into the flux of experience.”As Brassier (2008) argues, this ultimately subordinates Quine’s realism to the pragmatic wager between assorted ideological operations and interests, in principle incapable of drawing a relation between competing discourses in terms of how they correctly relate to objects and facts in the world, independent of pragmatic strictures. At a loss for a delineation of the relation between concepts and objects, instrumentalism lurks to usurp Quine’s alleged commitment to scientific realism. Worse still, it might be argued that the failure to distinguish how thought describes being conflates the two, resulting in a kind of idealism. This is precisely what Badiou (2007) proposes, claiming that without explaining the distinction between fact and form, and so leaving the relation between the actual world and formal propositions obscure, one folds the two together. As a result, representation is conceived as a feedback loop, and epistemology becomes incapable of discerning its discursive register from what it purportedly describes. Brassier (2008) writes: “Thus in a surprising empiricist mimesis of the serpent of absolute knowledge swallowing its own tail, naturalized epistemology seeks to construct a virtuous circle wherein congruence between fact and form is explained through the loop whereby representation is grounded in fact and fact is accounted for by representation.[15]” Given that representational is itself accounted in terms of semantic content, the problem becomes clearer. Weakening representational contents to propositional inferential relations whose truths are convened upon socially finally amounts to disintegrating the link to the empirical world that the naturalist claims science is capable of describing. This is to deny that representational contents may bear any ontological status, and to accept that our scientific realism be reduced to the pragmatic prescription of physicalism.
     Furthermore, Quine would owe us an explanation of what the endeavors of those special sciences that make use of the notion of representational content in terms of referential relations are talking about, as part of those sciences in which the ‘intentional stance’ remains.  Burge (2010) for example describes how perceptual psychology examines mechanisms for pre-linguistic objective individuation carried out by our sensory organs, so that representation is causally anchored in relation to environmental stimuli. Are we to simply render such talk derivative, unscientific in tenor, or its postulates epiphenomenal and thus with no proper ontological valence?  We shall say more about Burge below, but we should underline that it is clear that Quine would need of an additional argument to disavow representational content simpliter. Since it is the eviction of the latter which confines objective individuation to linguistic behavior, it follows that without such an argument, the annihilation of the referential relation between words and things is not transparently supported by the demands of a realist physicalism.
      In response to this, Quine could claim that physicalism can be rendered compatible with the wholesale destitution of representational content, along the lines suggested above: for example, if one endorses a form of physicalist eliminativism, in which relations between mind and world may be reduced to phenomena proper to fundamental physics. Quine (1953) advocated this position, as conceived through the distinction between the scientific descriptions of basic physical structure (illata), and posterior theoretical re-descriptions of the former (abstracta). The basic idea then was that the former substructure supports the supervenience of higher-order structure back into basic physical relations between elementary particles. As Ladyman and Ross (2009) point out, however, this view rests on the ‘Democritean faith’, refuted by contemporary science, that eventually it will be possible to decompose everything into elementary particles and relations among them, through fundamental physics. They argue that current physics is ambivalent with regard to the ontological status of unobservables, and so that reductionism is not motivated by present science. In addition, this ‘faith’ wouldn’t be easily palliated even if science did motivate such a reduction, since, given Quine’s endorsement of ontological relativity, the endowment granted to physics is in principle a pragmatic matter, and so a position that is prescribed rather than explained.  Thus, even if reductionism within science were sound, it would still be controversial whether only physical phenomena should be granted ontological status.
        On any account, it is not clear how physicalism and behaviorism are rendered compatible if Quine endorses an ontological behaviorism. A complete disavowal of representational contents and reference seems to entail the liquidation of the epistemological relation between mind and world, and so of knowledge understood (at least partly) as the objective representation of reality. Here it is not only translation and so knowledge of what others mean when they speak that leaves us in a blackbox with respect to other languages and cultures. More generally, the relationship between thought, language, and the world is rendered obscure. Without a clarification of how representation admits purchase on being, it is not even clear how any theoretical posits are capable of gaining traction on phenomena. We might conclude, therefore, that without an epistemological account of how thought represents the world it is difficult to find motivation for Quine’s endorsement of scientism.
        Alternatively, one may suggest that Quine advocates a form of epistemological behaviorism. Under such a reading, there may be facts about reference that, for contingent limitations, we are just incapable of knowing. Thus, even if there are genuine representational relations between mind and world, and if words and things stand in referential relations too, the problem is that salient evidence undermines our capacity to know of them. Representational contents would be said to exist, insofar as mental states and perceptual occasions are not merely linguistically individuated posits or conventionally accepted ‘true statements’, but determinate items in relation to the external world. The problem here would be fundamentally epistemological, inasmuch as we must accept ignorance on ontological commitments, mental states, and the representational contents that the latter bear, when confronted by the limitations of observational evidence. This is a more moderate reading of Quine’s position, but does it fare any better? Even on this reading, it remains unclear why enquiry into the representational contents of our mental states should prove fruitless, i.e. incapable of resolving the IOR. Soames (2005) seems amenable to Quine’s skepticism, and on this regard he comes closer to reading Quine as an epistemological behaviorist: “We can no more read off the contents of a person’s words from physiological claims about neurons than we can read off the contents of his words from statements about the noises he makes in certain environments.” (Ibid, pp, 246) But why should this follow? Why should we accept that an enquiry into the internal constitution of mental states and their contents should be incapable of informing us about the ontological commitments of others, and the items to which their expressions refer, seeing that at least some of them do refer, after all?  On this account, Burge (2010) in particular has raised a fundamental point of contention against Quine’s behaviorist externalism by proposing an epistemological externalism of his own, where objective representational content begins in fundamental instances of perceptual individuation causally relating environmental stimuli to our sensory organs. Advocating study into the science of perceptual psychology, such a position could entail, Burge suggests, that primitive conditions for spatial individuation constrain higher-order linguistic reference. As a result, we would be capable of successfully resolving the IOR through the amplification of our admissible data, since the former now appears artificially restrictive by excluding precisely the kind of empirical insight that would allow the relevant disambiguation for the translator. Burge agues as follows:
    “[Quine] is right in that it does not follow from utterances that occur when and only when rabbits or rabbit facsimiles occur that the utterer mentions rabbits. But he just assumes that the only relevant evidential consideration is the history of black-box utterances in the presence of rabbits. He thinks that if this evidence does not warrant unique attribution of a referent or a meaning, such attribution is gratuitous… Quine does not confront the natural view that the semantics of language is initially determined by perception. He does not consider how perceptual representational content- hence perceptual singular reference and perceptual attribution- are established.” (Burge, 2010, pp. 214-215)
    On this account, which we might deem ‘neo-Aristotelian’ in spirit, the relationship between words and things would be antedated by a more primitive relation between mental states and the world. Since the latter is rendered unambiguously determinate, relative to our perceptual faculties in relation to environmental stimuli, and tractable through scientific investigation, Burge thinks that Quine would have to admit that the supplementation of data by such information could restrict IOR, if not eliminate it. Although I am inclined to agree with Quine in that it is not clear that such an investigation could resolve the IOR, at least, he cannot dismiss the import of representational content by endorsing behaviorism without further argument.

    A second possibility once again brings us back to eliminativism, but this time one which would preserve the notion of knowledge of the external world while dispensing of representational contents. For example, an epistemological account tethered to neurophysiology could explain how the interaction between the environment and the brain does not make appeal to anything like representational contents, even if it describes a robust interaction between a sapient being and its exteriority. If Quine wants to take the eliminativist route in this regard, and claim that knowledge of the world is to be cashed out in terms other than representational contents, other questions are left pending[16]. As we saw above,
Soames (2005) convincingly argues that physicalism in Quine must be understood in terms of metaphysical supervenience[17]. Yet Kim (2010) for instance argues that functionally individuated psychological properties which retain intentional content must be ‘functionalized’ redescriptions of those physical properties with which they are metaphysically identical, or else the local supervenience of the psychological on the physical fails. This failure entails dualism, epiphenomenalism about mental properties, or the view that a singular event can have two causes, a mental and a physical one[18]. Without reinforcing how precisely psychological-kinds which are said to have representational content supervene on non-representational physical terms, it becomes unclear how the former can be brushed off from having any epistemological import in favor of behaviorism. On any account, denying that representational content may disambiguate about ontological commitments, and so overcome the epistemic limitations yielded by observational data, leaves in the dark the relation between physics and the ‘special sciences’, and specifically perceptual psychology. Moreover, it would still not be clear how construing the epistemic relation between mind and world in non-representational
 interactions between organism and environment couldn't help resolve the indeterminacy thesis. Nevertheless, we should always remember that Quine transfers the burden of proof by challenging his opponents to find a supplementary fact about the world which would render the inscrutability of reference ineffectual. Burge’s contention is that in particular the science of perceptual psychology can rise to this challenge. Other philosophers of mind advocate alternative candidates to fulfill this promise. Whether they do or don’t exceeds the scope of this paper.
      In addition, some of the problems that appeared by reading Quine as an ontological behaviorist, reemerge in reading him as an epistemological behaviorist. With regards to the endorsement of physicalism, it is unclear how scientific discourse qua communicational practice, subject to the behaviorist constraint to observational data, relates to other forms of communication and discourse. Is there no inscrutability with regards to the facts of physics, like when, for example, I think to understand what a scientist tells me when he describes the behavior of an electron or some other non-observable cause? Are theoretical posits intractable to behavioral evidence equally susceptible to the inscrutability of reference? The latter is a question which brings to mind Quine’s endorsement of the real ontological status of abstract objects, and so the obscure relation between the latter and observable, concrete physical particulars[19]. If the inscrutability of reference holds even at the sub-observational level, so that even the statements advanced by physics leave reference indeterminate, then Quine’s endorsement of physicalism would render the consensus in scientific community potentially indifferent to the true facts of the matter. But this seems problematic, since it is said that physics describes the world as it is, and given semantic incompatibility between theories, extending indeterminacy to physics would beg the question about whose physics adequately describes the world, and how do we know this? Postulating an ‘ideal physics’ doesn’t quite help, since the question repeats itself in having to clarify the relation between our actual physics, and the ideal physics in exception to the pitfalls of the former. If the UTD applies to physics, then subjecting the epistemic purchase of actual physics to stimulus meaning would seem make Quine’s behaviorism conducive to a kind of radical agnosticism about reference for all intersubjective communication, and not just cases of ‘radical translation’. However, then it becomes difficult to see how such a view could amount to a proper scientific realism, since all of a sudden ontological ambiguity affects even intra-linguistic or intra-theoretical communication. Quine on this reading seems one short step from endorsing full-blown ontological relativism about the world, and not just ontological relativity about discourse, as a result of the behaviorist restriction to stimulus meaning, and the epistemic gulf opened by the inscrutability of reference. 
    If, on the other hand, IOR doesn’t hold at the sub-perceptual level, then it is not clear why the inscrutability yielded through communication in non-scientific description couldn´t be clarified through determinate, fundamental physical facts.  Soames (2005), for example, seems to think that physicalism would entail the disambiguation of inscrutability if we understand by the former the thesis that everything that is true must supervene on physical truths.

         “Whatever any of us means by rabbit, it is natural to suppose that our meaning what we do depends ultimately on physical facts. For example, we may ask whether a physically identical twin- someone (in a physically identical possible world-state) whose utterances, behavior, brain states, causal and historical relations to the environment, and interactions with other speakers…completely and exactly match mine- could mean by ‘rabbit’ what I actually mean by, say ‘undetached rabbit part’. It seems to me that the answer to this question must be ‘No’, for the very same reasons that physicalism itself seems acceptable on this interpretation.” (Soames, 2005, pp. 251)
      Although Soames separates Quine’s behaviorism from his physicalism and focuses on the latter, I think that, on the contrary, consideration of the former yields reasons to doubt the plausibility of endorsing the latter. This follows even if Soames is right in saying that it is not clear why the IOR persists under such a physicalism. If one additionally casts doubts upon the notion of representational contents, it seems on the other hand difficult to understand how physics stands in relation to the IOR. Since it is physics that provides knowledge of the facts underlying the possibility of there being equally adequate but incompatible theories of translation and so incompatible ontological commitments, Quine cannot obviate the task of describing how physics yields determinate knowledge of a mind-independent world, relative to which sentential utterances are rendered ontologically ambiguous or at the very least epistemologically foreclosed.
         Finally, if Quine does not completely eliminate the epistemic purchase of representational contents in tethering the latter to physics, he either needs to explain how such contents don’t resolve the inscrutability of reference, or else claim how they constitute an exception. That is, if physics relates to the world beyond the constraints of convention, and it yields knowledge of such a world, how does it escape the inscrutability of reference, if at all? The former option leads into the kinds of problems we suggested above, making it problematic to discern whose physics is to be championed. The latter option of claiming physics is in exception, renders scientific description and communication metaphysically determined and unambiguous, while leaving other descriptive registers irremediably affected by the IOR, and so subject to the UTD. Thus, Quine could claim that physicalism supports IOT and the UTD insofar as all non-physical linguistic utterances remain referentially ambiguous. At the same time, he can claim that physics, and only physics, gives communicable knowledge about the real, and is not subject to IOR. Finally, the ontological ambiguity presented by IOR would be said to be supported by the plurality of possible incompatible theories while admitting of determinate physical facts described unambiguously by science. As we have seen, given the endorsement of ontological relativity, this prerogative granted to science cannot but be prescriptive in nature. Whether this constitutes a sufficient ground for a full blown philosophical realism is an open question, although I remain skeptical for reasons I have been arguing. If Quine wants to restrict realism to knowledge of unobservables or abstract principles not registered through SM, and so not subject to the UTD, he would need to explain why a supplementation of data by fundamental physical facts couldn’t clarify the entirety of statements that supervene on physical facts. A successful resolution to the quandaries brought by Quine’s behaviorism, physicalism, and ontological relativity would entail showing show that the UTD and IOR hold, while not sacrificing the valence of physical knowledge. Even more, it would need to render the UTD relative to the data yielded by physics in particular. This would be to simultaneously endorse the following two conditions without tension:

All genuine truths (facts) are supervene on physical truths.

(The underdetermination of translation by physics)
 Translation is not determined by the set of all physical truths (facts), known and unknown. For any pair of languages and theories of translation, incompatible with T, that accord equally well with all physical truths.

     As we have seen, however, this is precisely what Quine’s simultaneous endorsement of behaviorism and physicalism renders problematic. Whether these can be reconciled or not lies outside the scope of the present study, and is reserved for another occasion.
·         Badiou, Alain, The Concept of Model, translated by Z.L Fraser and Tzuchien Tho, re: press, 2007.
·         Brassier, Ray. Badiou’s Materialist Epistemology of Mathematics, in Angelaki 10, 2005.
·         Brandom, Robert, Making it Explicit, Harvard University Press, 1998.
·         Burge, Tyler, The Origins of Objectivity, Oxford Press, 2010.
·         Kim, Jaegwon, Essays in the Metaphysics of Mind, Oxford University Press,  2010.
·         Ladyman, James & Ross, Don, Everything Must Go, Oxford University Press, 2009.
·         McDowell, Mind and World, Harvard University Press, 1996.
·         Orenstein, Alex, W.V Quine, Princeton University Press, 2002.
·         Quine, W.V. Whither Physical Objects?, from R. S. Cohen, P. K. Feyarabend and M. W.
Wartofsky (eds.) Essays in Memory of Imre Lakatos, Boston: D. Reidel Publishing Company, 1976.
·         Quine, W.V. Theories and Things, London: Belknap Press, 1981.
·         Quine, W.V. Naturalism; Or, Living Within One's Means. Dialectica 49, 1995.
·         Quine, W.V, Word and Object, MIT Press, 1964.
·         Quine, W.V, Ontological Relativity, Columbia University Press, 1977.
·         Soames, Scott, Philosophical Analysis in the 20th Century: Volume 2, Princeton University Press. 2005. 
·         Follesdal, Dagfinn, In what Sense is Language Public?, in On Quine, P. Lombardi and M.Santambogia (eds), Cambridge: Cambridge University Press, 1995.

[1] These three terms will be defined technically below.
[2] Quine, W.V, Reply to Anthony, in Knowledge, Language, and Logic, pp. 419.
[3] In this regard, it should become obvious that Quine’s position is continuous with Davidson, and more crucially Wilfrid Sellars, in claiming that perception is a conceptual achievement, insofar as it involves the exertion of judgment in order to yield determinate content. Perception is not self-evidential, and brute sensation is epistemically inert, i.e. sensibilia do not constitute a ubiquitous layer of pre-conceptual determinate content, available for analysis. This will turn out to be a crucial point of contention for thinkers such a Burge (2010).  See Sellars Empiricism and the Philosophy of Mind (1956)
[4] We shall not be concerned in this paper in assessing whether Soames' reading of Quine does the latter justice. We shall simply assume, for the moment, that his reconstructions are cogent and consistent with Quine's views.
[5] SOAMES, Scott, Philosophical Analysis in the 20th Century: Volume 2, Princeton University Press,  pp. 227
[6] Another set of questions concerns the privilege accorded to observation in Quine’s account. Is the priority given to sight and observation merely arbitrary, so that similar example could in principle follow from experiments constrained to other sense-capacities? If not, then Quine would need a separate argument to show why only observation can yield the kind of data needed for translation.
[7] Quine, In Pursuit of Truth, Harvard University Press, pp 47-48
[8] QUINE, W.V, Speaking of Objects, in Quintessence, 91-91.
[9] Soames, 2005, pp. 240.
[10] Of course, Quine could disavow reference without necessarily disavowing ‘representational contents’, i.e. if one somehow construes the latter as high-order iterations of physical facts for which the concept of reference would have no use. I consider this possibility below. We should just keep in mind that here we use representational contents, in the sense in which the latter are determined by a relation to the world, and thus that sentences are tethered to such contents.
[11] I owe this formulation to Peter Wolfendale, who in his excellent (unpublished) Essay on Transcendental Realism provides a brief but elegant account of Quine (and other’s) positions.
[12] FOLLESDAL, Dagfinn, “In what Sense is Language Public?”, in On Quine, P. Lombardi and M.Santambogia (eds), Cambridge: Cambridge University Press, 1995.
[13] See Ladyman & Ross (2007), Chapter 4.
[14] In this regard, Quine famously advocates an extensionalist approach to ontology that reduces reference to being the extension of a concept. According to Quine, intensional determinants are what remains of Aristotelian essences, and are logically opaque instead. Reference in the sense of a relation to a mind independent reality would thus remain tethered to the latter notion, which exceeds the formal transparency of a purely extensional ontology.
[15] Brassier, 2008, pp. 139.
[16] This would not be strictly analogous to the earlier ontological behaviorism, since it could entail the position that internal mental states do relate to the world objectively, but do not require postulating anything like representational contents, or psychological kinds.
[17] Soames, 2005, pp. 256-257.
[18] Marras (2000) shows that this argument generalizes over the other special sciences as well.
[19] Quine, W. 1964. Ontological Reduction and the World of Numbers, The Journal of Philosophy 61: 209-1