- Subjective, Intersubjective, Objective by Donald Davidson
Oxford, 237 pp, £30.00, September 2002, ISBN 0 19 823753 7
I do wish Donald Davidson would write a book. I mean, a proper book with a beginning, a middle and an end, in contrast to the collections of papers of which the present volume is an instance. My wishing so is not invidious. These bite-sized essays, each a mere fifteen or twenty pages long, often impress one as serious philosophical achievements even when they are read piecemeal, as they were written. But when they’re read together, one sees (what I, at least, hadn’t fully realised) that Davidson is a kind of bird that’s become rare almost to extinction: a systematic philosopher. That’s to say that he holds to a small number of very general principles the application of which, he claims, resolves a heterogeneous bushel of philosophical puzzles. The puzzles range from, for example, whether animals believe things (apparently they don’t), to whether the concept of the self is irreducible (apparently it is), to how many people it takes to think about a light bulb (apparently it takes two; see below). A lot of the fun of reading these papers is seeing how an exiguous collection of commitments plays out in so many different domains.
But, though it’s fun, it’s also frustrating. Davidson’s format often doesn’t give him room to make clear exactly what his general principles amount to; and the arguments from the principles to the conclusions that they’re alleged to underwrite tend to be elusive. One thinks perhaps that one sees how they’re supposed to go; but then one thinks that perhaps one doesn’t. Much gnashing of teeth. Much tearing of hair. Lacking the book that I wish Davidson would write, the best I can do is trace a strand through the web of arguments that he weaves. For all that it’s a main strand, there is much more going on in these papers than this way of proceeding will let me attend to. So be it.
Perhaps Davidson’s primary commitment is to the thesis that languages are intrinsically public objects. ‘If Wittgenstein held that language is necessarily social, then the central thesis of this essay is Wittgensteinian.’ And again: ‘Language is in its nature . . . intersubjective; what someone else’s words mean on a given occasion is always something that we can in principle learn from public clues.’ And yet again: ‘The theory of truth we must presume lies in available facts about how speakers use the language. When I say available, I mean publicly available – available not only in principle, but available in fact to anyone who is capable of understanding the speaker or speakers of the language.’ What with one thing and another, I guess Davidson thinks that there aren’t – can’t be – any private languages. ‘And, given the close connections between thought and language, analogous remarks go for the contents of the [propositional] attitudes.’ Just as the meaning of utterances must depend entirely on facts that aren’t privy to the speaker, so, too, the content of thoughts must depend entirely on facts that aren’t privy to the thinker. Only what hearers are able to interpret could be a language; only what second persons can discern can be the content of a thought. The pending questions are: Why does Davidson believe all that? Is he right to believe it? And, if he’s right, what follows from it?
To begin with, the inherent publicity of language (or thought) must amount to more than the inherent publicity of communication. It’s just a truism that it takes two to communicate, as Davidson is fully aware. His idea is that whatever it is actually used for, it’s intrinsic to something being a language that it’s the sort of thing that could be a medium of communication. Likewise, however reticent the thinker in a given case, it’s intrinsic to something being a thought that its content could be publicly manifest. Davidson’s shorthand way of saying this is that the contents of utterances and of thoughts must be available to someone who is in the epistemic position of a ‘radical interpreter’. I’m not entirely clear what epistemic position Davidson thinks radical interpreters are in, but the basic intuition is that a language has to be the kind of thing that such as you or I could learn to translate.
Only a radically interpretable language could be a vehicle of communication; that seems plausible enough. But it doesn’t quite follow that languages are, per se, radically interpretable. For, why couldn’t a language that isn’t used for communication be useful for something else? Consider inferring, which is, I suppose, something that creatures with minds do all the time. The premises and conclusion of an inference must be expressed in some way or other. The usual way of expressing them is to utter (or think) a linguistic formula that means them. Presumably, you infer from P to Q by saying (or thinking) something along the lines of ‘P, therefore Q’. Well, it certainly seems that you could use a language to express the inferences you want to evaluate but not for anything else. And it seems not to be obvious that such a language would have to be accessible from, as it were, ‘the outside’; that it would have to be a language that a radical interpreter could learn. If Davidson thinks that isn’t really imaginable, he has to say why. I’m about to turn to this, but first a detour. I want to make clear why he supposes the issue of interpretability matters so.
Here’s how I think Davidson thinks the geography goes. It would be very interesting to show that languages have to be interpretable because the conditions for interpretability are not trivial. Interpretation is possible only on the assumption that the informant is possessed of at least a minimal degree of logical consistency, and that most of what he says is true. But the second assumption is incompatible with scepticism, and (since rationality is a normative notion) the first implies that the constraints on interpretation can’t be formulated in the value-neutral vocabulary of the physical sciences. To show that what is said (or thought) must be interpretable is thus to dispense, at a single stroke, with both scepticism and reductionism. ‘It follows from the nature of correct interpretation that an interpersonal standard of consistency and correspondence to the facts applies to both the speaker and the speaker’s interpreter.’ Davidson thinks he can do all that work on the plausible assumption that publicly available facts determine the content of language and thought.
By the modest standards that philosophers are accustomed to, that would be a very big return on the investment. In the good old days, philosophers used to draw grand pictures as a matter of course; they often supposed they could prove, by some relatively simple argument, that nothing moves, or that time is unreal, or that there isn’t anything mental, or that there isn’t anything except what’s mental, or that there is or isn’t freedom of the will, or that God exists, or that numbers do, or that God doesn’t exist or that numbers don’t; and so forth. As it turned out, nobody ever did prove any of these things, and most of us have by now become more modest in our ambitions. Davidson, however, writes philosophy like Wagner wrote operas: nothing less than everything is ever at stake. There has to be a place in philosophers’ heaven for people who have ideas for really stunning arguments which, in the event, don’t quite work.
I don’t think that languages have to be interpretable, nor do I think that if languages have to be interpretable, that would defeat scepticism and reductionism. I’ll go through this in reverse order.
The reason that interpretability doesn’t preclude scepticism (or, mutatis mutandis, reductionism) is one that Davidson is, in fact, entirely aware of. I simply can’t understand why it doesn’t worry him more.
It is an artefact of the interpreter’s correct interpretation of a person’s speech and attitudes that there is a large degree of truth and consistency in the thought and speech of an agent. But this is truth and consistency by the interpreter’s standards. Why couldn’t it happen that speaker and interpreter understand one another on the basis of shared but erroneous beliefs?
(My emphasis.) Why indeed?
Peter observes that Paul says ‘mouse’ mostly when there are (what Peter takes to be) mice on view. So Peter interprets Paul as meaning mouse when he utters ‘mouse’. Fine so far, but it seems entirely compatible with this scenario that Peter and Paul have both misidentified the (as psychologists say) ‘distal stimulus’. Most of the time, what Peter and Paul both take to be mice are distal dust balls lurking untidily beneath the distal furniture. (My cat often makes the same mistake.) It looks like what’s going on is precisely that ‘speaker and interpreter understand one another on the basis of shared but erroneous beliefs.’ But if that is what’s going on, Paul’s being correctly interpretable as having a mouse-thought by no means shows that there are mice. What worries sceptics is that everybody might be wrong together. Paul got it wrong, but Peter did, too, and in the same way; so Peter doesn’t notice Paul’s mistake. Maybe something shows that that couldn’t happen; but, so far, Davidson seems to have left open the possibility of universal folie à deux.
Here’s what Davidson says: error ‘cannot be the rule. For imagine . . . an interpreter who is omniscient about the world . . . The omniscient interpreter, using the same method as the fallible interpreter, finds the fallible speaker largely consistent and correct. By his own standards, of course, but since these are objectively correct, the fallible speaker is seen to be largely correct and consistent by objective standards.’ I don’t get it. An omniscient interpreter can’t adopt the policy of making the speaker’s utterances true by his (the interpreter’s) lights; for, if he does, he’ll mistranslate whatever the speaker says that’s false. An omniscient interpreter could exercise universal charity only towards a speaker who is also omniscient. Not, I imagine, a situation frequently encountered.
So much for whether Davidson’s inference from the intrinsic interpretability of language and thought to the refutation of scepticism is valid. What about his claim that language and thought are intrinsically interpretable? That they are would be big news all by itself, however the implications for scepticism or reductionism play out. There are, as far as I can tell, two closely related, mutually compatible arguments that Davidson has on offer. Both are familiar, so I’ll sketch them very briefly. I’ll then turn to a third and more novel approach that Davidson has recently taken to elucidating the relations between interpretability, content and the rejection of scepticism. And then we’ll be through.
First, there is ‘the private language argument’ properly so called. The private language argument is philosophy’s unicorn; it’s widely believed to be at large somewhere in the pages of Wittgenstein’s Philosophical Investigations, and there are some who claim to have glimpsed it in a glass, darkly. But nobody has had a clear view of the thing, and there isn’t general agreement as to either its form or its substance. Here, anyhow, is Davidson’s way of putting it:
The central argument against private languages is that, unless a language is shared, there is no way to distinguish between using the language correctly and using it incorrectly; only communication with another can supply an objective check. If only communication can provide a check on the correct use of words, only communication can supply a standard of objectivity in other domains.
So, then: if there’s no interpretability, there’s no communication; if there’s no communication, there’s no ‘objective check’ on using the language right; and if there’s no objective check on using the language right, then there is no difference between using it right and using it wrong. But what you can use any way you please is ipso facto not a language.
If that is the private language argument, it appears on the face of it to be hopeless. The reason is well known. It requires an inference from an epistemological premise to an ontological conclusion (it moves from ‘there’s no check on the difference’ to ‘there’s no difference’). Such inferences are warranted only if some sort of verificationism is assumed, and verificationism isn’t true. Perhaps there is a way to make a private language argument run without the verificationism; many have thought there must be. But Davidson offers no such formulation, and I, for one, am thoroughly bored with the whole business.
Here’s the second kind of argument Davidson thinks might underwrite the interpretability requirement; it depends on a certain way of understanding first language acquisition.
What we learn first is to associate what in the end turn out to be one-word sentences (‘Mama’, ‘No’, ‘Dog’, ‘Blue’) with situations, events, objects and their features. Soon the child learns the magical power of making sounds adults find appropriate and hence reward. These are only preliminaries to fully fledged talk and thought . . . [but] these primitive relations between two people in the presence of stimuli from a shared world contain the kernel of ostensive learning.
I take this to be more or less the line of thought found in Dewey, Skinner and other behaviourist psychologists fifty years ago: having a language (or a concept) requires making learned discriminative responses, the acquisition of which depends on socially mediated reinforcement. The learnability of a language thus implies its interpretability: since learner and teacher must both have epistemic access to the discriminative stimulus, they must both inhabit the same, public world.
I won’t spend long on this. First, the psycholinguistics it assumes is archaic, not to say prehistoric; see any modern text. Second, a sceptic could rightly complain that, precisely because it takes other minds and a public world for granted, endorsing this sort of story about language learning begs all the claims that he’s sceptical about. Third, and most to the point, the usual candidate for a language that is used to infer but not to communicate is ‘Mentalese’, the so-called language of thought in which, according to some theories of cognition, higher mental processes transpire. But everybody who thinks that there’s a mental language (since, by the way, William of Ockham) takes for granted that it must be innate. If so, then it’s not an objection to Mentalese that learned languages are ipso facto interpretable. For the last couple of hundred years, Anglo-American theorising about mind and language has routinely begged the question against Nativism. Maybe it’s time to kick the habit.
We’ve been considering Davidson-style arguments that language and thought are ipso facto interpretable. I don’t think they work; it would be astounding if they did. According to Davidson, an interpretation of a language is (roughly) a hypothesis about the truth conditions of its sentences. Securing such a theory requires empirical inferences from facts about speakers and hearers to conclusions about the semantics of their language. Accordingly, if it’s a priori that languages are interpretable, then it’s a priori that such facts are available from the epistemic position of a radical interpreter; that is, it’s a priori that they are available to us. But surely that would be unprecedented? Surely, empirical inferences don’t work that way? Imagine a philosopher of science who claims that the data required to select the correct astrophysical account of (say) the origin of the Universe must ipso facto be accessible to radical astronomers. Wouldn’t we sort of wonder why that should be so? Surely, we’d think, it’s possible in principle that the crucial facts are just too far away, or too long ago, or too big, or too small, or too deeply hidden, for us to get at them? Presumably God didn’t make the astrophysical universe to suit the convenience of astronomers, radical or otherwise; so perhaps we’re not in a position to find out how the Universe works. Or perhaps our position is all right, but we’re just not smart enough. Well, if God isn’t prepared to guarantee the success of empirical theories in astrophysics, why on earth should we suppose that He guarantees the success of empirical theories in semantics?
To recap: Davidson wants an argument that most of what we say (or believe) must be true. So far, the candidates we’ve considered have all been inferences from the assumed a priori interpretability of language and thought, and they look to be neither valid nor sound. Of late, however, Davidson has been considering a rather different sort of line. We should have a look at this in closing.
These days, a lot of philosophers claim that the content of utterances (or thoughts) is determined, in part, by their causal history. This view is called ‘externalism’ in semantics; I won’t bother you with the details except insofar as they impinge directly on our current concerns. Well, Davidson seems increasingly to be tempted by the possibility of running his antisceptical argument (not from the interpretability assumption but) with externalism as the crucial premise. The following serves to give the feel of it.
The relation between belief and truth is not simple. But that it exists, and is central to meaning and truth, can be seen from the simplest cases, and their role in understanding other cases. The simplest cases are those where a sentence such as ‘That’s a book’ or ‘This is yellow’ is caused to be held true by the conspicuous presence of books or yellow things. This is evidence that these sentences are true when books or yellow things are present; the reasoning is that what determines the meaning of such sentences is what routinely makes them true.
One may feel a twinge of nostalgia for ‘paradigm case arguments’ and other such follies of one’s youth.
I don’t think this kind of argument works very well either. Sceptics are worried that maybe there’s nothing out there. So, if refuting sceptics is the goal, it seems that an externalist semantics begs the very thing that worries them. But never mind. What I want to emphasise is that, if externalism does prove that most of our beliefs are true, it apparently does so without any assumptions about radical interpretation. That might be, from Davidson’s point of view, rather an embarrassment. Over the years, he’s invested a lot in the thesis that radical interpretability is the fact about thought and language on which Realism about their content turns. But externalism seems to make all that superfluous. Where is the radical interpreter, now that we don’t need him?
Here is what I take to be Davidson’s new twist. Actually, he thinks, the interpreter does still have a job to do, even on the externalist assumption that the content of a thought is constituted by its cause. For, whenever the world causes a thought, there will of course be a chain of events leading from the one to the other; most of which will not be relevant to the determination of content, even according to externalism. (The mouse reflects the light; the light travels to the retina; the retinal excitation stimulates the brain; and so forth.) A semantic theory must therefore find a principled way of deciding which link(s) in such chains are meaning-constitutive. This is where Davidson thinks the interpreter earns his wages. The name of the game is ‘triangulation’. The meaning-constitutive event is the link where the causal chain from the world to a thought intersects the causal chain from the world to an interpretation of the thought. ‘Until the triangle is completed connecting two creatures, and each creature with common features of the world, there can be no answer to the question whether a creature, in discriminating between stimuli, is discriminating between stimuli at the sensory surfaces or somewhere further out or further in . . . It takes two points of view to give a location to the cause of a thought.’ That’s why it takes two to think about a light bulb.
If that’s true, it’s very surprising. For one thing, the problem of distinguishing ‘the’ cause from other links in a causal chain is perfectly general; it seems to have nothing in particular to do with linguistic or cognitive content. The usual assumption is that ‘the’ cause of an event is at the point where certain hypotheticals converge. Thus, it’s plausible that the actual and possible causal chains that project back from a mouse-thought generally intersect at a mouse. Starting at a mouse is just what they’re likely to have in common in, as Davidson says, the simplest cases. But if that’s so, the need to triangulate on an interpreter disappears. What makes a thought a mouse-thought is that it is, or would be, caused by a mouse in all the nearby possible worlds.
I don’t know if that sort of story will work, but I do think Davidson has a dilemma. If it works, no interpreter is needed; if it doesn’t, supplying one won’t help. Having the causal line from the mouse to me cross the causal line from the mouse to my interpreter at the mouse isn’t sufficient for the content of either of our thoughts to be mouse. (He might be thinking pest and I might be thinking furry thing.) What’s further required is that the lines would continue to cross at the mouse in the (relevant) counterfactual cases: he would have thought mouse even if it hadn’t been a pest; I would have thought mouse even if it hadn’t been furry. But if my thought is invariant across all the (relevant) causal chains so long as they preserve the mouse, and if externalism is true, then that should all by itself entail that I’m thinking mouse. In which case, who cares what my interpreter is thinking? I suspect that triangulation is a sort of make-work programme for radical interpreters whom semantic externalism would otherwise make redundant.
I’ve discussed Davidson on the relation between content and interpretation at some length, because the present collection of papers is a reliable introduction to his views on this topic, and his views have been very influential. But, here at the close, I want to step back a pace or two and say what I think is really going on.
A lot of the last century of the philosophy of language and the philosophy of mind has been driven by attempts to find a bullet-proof argument against scepticism. The preferred tactic has been Pragmatist, practically without exception. (Pragmatism, to stick with the triangulation metaphor, is located a little past Peirce and Dewey, at the intersection of Wittgenstein and Quine.) The basic idea is to opt for an epistemic account of truth; hence for an a priori connection between what’s true and what’s believed to be. Some philosophers, for example, have managed to convince themselves that the truth is what (i.e. whatever) everyone will come to believe ‘eventually’. We should all live so long.
That sort of account doesn’t work because, of course, truth isn’t an epistemic notion. What’s true depends on how the world is, not on how we think it is. Actually, and very refreshingly, Davidson is aware of this. He therefore proposes to take the notion of truth as primitive; a fortiori, as not susceptible of epistemic reconstruction. That is surely a step in the right direction, but then he spoils it. Having wisely avoided an epistemic construal of truth, he disappointingly embraces an epistemic construal of truth conditions. Roughly, according to Davidson, the truth condition of a sentence is whatever a radical interpreter would come to believe that it is, given relevant facts about what causes the sentence to be uttered. And believing is an epistemic state.
Well, the jig’s up. We can’t deduce our epistemology from our semantics or vice versa. The world is prior to the mind, so what we believe is one thing, but whether what we believe is so is quite another. If that’s scepticism, then scepticism is true.