Sensory Empiricist Epistemology - A Moral Risk

This has probably been remarked on by others, at least indirectly, but it's worth spelling out:

At the heart of the 'internal / external', 'subjective / objective', 'senses / science' conundrum lurks an issue of good faith. When we cannot understand what someone (A) is saying - when what they are saying does not 'make sense' to us - we might represent this to another person (B) as a hypothesis about A's competence or honesty as a language user.

Within the empiricist tradition, there are many statements (e.g. first-person observation statements) where this failure, or this dishonesty, must translate into a specific issue with A's sensory apparatus: either it isn't working properly (they don't see what's there) or they are lying (they see, but pretend not to). We know (externally, demonstrably) that the linguistic interaction is going wrong, and we claim (problematically, unresolvably) that this results from a lack of facility or from moral failure.

This easy projection of the external communication breakdown onto the internal space of a potential interlocutor has some ugly consequences. Among mental health practitioners, there is considerable disagreement about whether certain conditions can be treated as personal deficiencies or whether they can only be understood contextually - whether, for example, the family, the society, or the therapist/client relationship need to be considered along with the capacities of the patient or client. Traditionally, people have been diagnosed as mad on the grounds of their social interactions, but treated as though the incomprehensible aspects of these are the product of personal perceptual and moral defects.

If we look for epistemological fundamentals in the possibility of a shared language, we are forced to shift the focus from the individual to the interaction - from personal 'failure' to relationships. From this perspective, we are also forced to recognise that diagnosing someone as having an 'individual defect' must always be provisional and can never have a 'moral' dimension. Once we have put someone outside the class of interlocutors (or potential interlocutors?) our judgments about their intentional states cannot be tested against their participation in the conversation. We cannot know their reasons, and so cannot judge them.

This doesn't mean that we cannot protect ourselves from their actions, of course. It does mean that we cannot pretend that we can know that this is 'for their own good'.

We also cannot legitimately invite them to consult their own internal condition to explain our behaviour towards them; to consult their empirical equipment or their conscience to find the roots of their 'madness'.

Not only do we have no shared language with which to do this, but we have no legitimating empiricist mythology with which to comfort ourselves.

What is the world really like?

I'm tempted, obviously, to give an oblique, Austinian, kind of answer to this question, but I think it  needs taken seriously. Perhaps Austin was a bit presumptuous and condescending, for all that he was broadly right.

So: I was lying in bed last night, half asleep, thinking about what the world was like, and having some meta-thoughts about what that question meant. I might share these thoughts with you, or I might not. If I did, it would be through this medium of 'textual encoding' that I'm using now. Is this like the medium of picture encoding that I would use to send you a photograph? (Is that, itself, a 'transparent' medium?)

We might say that (a) there is a way that the world is, (b) that I might have thoughts (correct or incorrect) about the way the world is, and (c) that I might (honestly or dishonestly) represent to you what I claimed to be thinking. If my thoughts are correct, and I represent them honestly, then I would be telling the truth.

The problem with this picture (as a sketch for an epistemology) is that stages (a) and (b) are more or less opaque until they have been articulated - a stage (c) process. And yet, somehow,  it provides an intuitively compelling model for empirical truth-telling.

The problems are overwhelming, of course. The way the world is can only be present to my cognition if I'm thinking about it, and I can't bring either the presence or the thought into this conversation with you without articulating them.

So what about the intuitively compelling model?

The Monadology seemed very idiosyncratic to me when I was first introduced to it, as, I'm sure, it does to many who think the problem it addresses can (implausibly) be solved in some other way, but Leibniz' picture of solipsistic universes kept mysteriously synchronized with one another poetically captures the human condition, if some of its elaborations are less convincing.

The synchronicity is, of course, a projection of our shared language. How this reflects back to the interior is a central focus of therapeutic practice, which draws on both the articulable and on the inarticulable (which can only be 'shown' or 'discovered') in the relationship between the therapist and the client - or, as one might more modestly claim, between one human being and another.

The solipsistic condition cannot, of course, be represented (although we can have a name for it - 'beetle', perhaps ...).

The therapeutic dilemma is that the attempt to explore this is both essential and inexplicable. I am quite sure that the things that go on in my 'mind', as I might say to you, are profoundly engaged with what I say and how my behaviour is represented in the world. I can, however, neither show you this nor tell it to you in a way that sidesteps the need to presuppose semantic congruence.

And neither of us should imagine that we can agree about what words we might privately be reciting to ourselves. Words such as: "I wonder if we are making any sense to one another at all."

Far less imagine that we can discover this particular thought in one another without accompanying words.

More on Stroud and Privileged Statements

Consider:

(a) The world allows us to talk about it in the way that we do.

(b) The world is the way that we say that it is.

They appear to mean different things but, taken completely seriously and literally, they are methodologically equivalent.

If meaning = truth conditions, in other words, they do mean the same thing.

So what about metaphysical anxiety?

To construct the kind of TA that Stroud specifies, we need to be able to project from the preconditions of scepticism to a refutation of the sceptical position. He offers the possible starting point of 'privileged statements' which must be true in any possible language, but then reflects that there is a world independent of language that may not be limited by our particular grammatical constraints. He forgets, as I've observed, that we need to know what that phrase 'a world independent of language' means before we can consider his objection, and this ultimately comes down to use (in so far as it can be accounted for at all).

His instinct that projecting from language to a metaphysical reality doesn't get us very far is correct, though. We would only be establishing 'metaphysical realism' at the expense of reducing it to a grammatical requirement - we certainly couldn't then point to it as a support for the validity of our linguistic computations. We would be achieving a metaphysical result at the expense of rendering it pointless.

We might deal with a sceptic who expresses doubts about physical reality this way, but only by showing that physical reality has no role to play in any foundational epistemology.

A sceptic of the possibility of any foundational epistemology can be dealt with more directly, of course, by pointing to the grammatical preconditions of the possibility of asking sceptical questions. We cannot treat someone as a competent interlocutor without attributing reliable knowledge states to them  - 'Do you know you are asking a sceptical question?' can have only one answer, if the sceptic is engaged in the conversation.

In both cases - of physical reality and of knowledge states - the 'metaphysical projections' can be exhaustively accounted for in grammatical terms. Going back to (a) and (b) above - the world allows us to talk as though both exist, and we cannot methodologically distinguish this case from the case in which they 'really do'.

Language and Humanity

What is language?

Or, what is 'language'? Perhaps, in an Austinian frame, 'How do we use the word "language"?

Except that I don't want to duck what Barry Stroud would have called 'metaphysical anxieties', the acknowledgement of which does not, of course, commit me to any metaphysical conclusions.

So. We can start with some potential 'root-like' statements:

"A language is what we use to talk to one another."

It is partly the collection of behaviours, technologies, objects, and expressions of capacities, which are necessary to the sense of 'We are having a conversation'. This is a bit messy, of course. And I say 'partly' because none of these things could count on their own - it is the way they are used that make them the components of a language. If I look at an ancient carving that has patterns of dots and lines on it, I might hypothesise that it is an inscription in a contemporary language, but until I can translate it I cannot be sure. And, of course, without a native speaker to test this hypothesis with I cannot deal with a certain kind of sceptical uncertainty about my conclusion. In the language I am using now, I can ask the question whether I have got the translation wrong in some subtle but catastrophic way. In a conversation with a native speaker in the language itself, I cannot explore this possibility without (unintelligibly) expressing doubt about whether we are having a conversation at all.

(Another consequence of the Quine/Davidson radical translation problem is that we can't firmly draw a line between language and non-language on the grounds of these components.)

The string of characters you see on this page only comprises an example of language use because you can interpret them, and believe they were written by someone in a certain complicated intentional state, and with certain objectives. If you discovered they were a surprising output from a random process you might be puzzled and disturbed, but you would have to acknowledge that your initial supposition about their linguistic nature was a mistake.

Ursula le Guin had this about right.

(Do we need a word here? Why don't we stop at 'we can talk to one another' or 'we can communicate'? We can't if we want the ancient inscription question to be intelligible; if we want to ask: 'Is what I am looking at a language?')

Perhaps another root:

"A language is anything we can intelligibly translate into the language we are using now."

"Can intelligibly translate" must imply "Could have a conversation with native users of the language", for Davidsonian (at least) reasons. (Think of "I have successfully translated Martian, and have found that nothing Martians say in it makes any sense.")

Like the other concepts which generate open question paradoxes, 'language' is both a useful/necessary concept and also one that can only be given a recursive account of. But the problem seems bigger here: we are left with something like 'medium of communication', which depends only on the possibility of communication and is completely unspecific about the nature of the medium.

What about grammar, spelling, constructive meaning etc.? Surely these things are required to sustain intelligibility? After all, it was at least a pattern of lines and dots I was hypothesizing about, not a pictorial representation or a geometric construction.

But le Guin is right here as well: the existence of these clues guides and does not determine. We can only say 'we find it easier to come up with a translation hypothesis where these are present'. We cannot say (without circularity) 'These are essential to something being translatable'.

Chomsky makes a mistake here: He thought he had found a deep grammar, some abstract characteristic of all languages, but in fact he only found our cognitive limitations. There are only some kinds of organisations of objects and technologies which we have the capacity to translate, or the inclination to attempt to translate. This does not have the consequence that there are languages we cannot translate - which would be incoherent (le Guin's story is satire) - but it does have the consequence that 'language' is something human beings have the capacity to produce. To imagine that we can translate Martian or Dolphin is to imagine Martians and Dolphins having quite parochially human characteristics - to imagine them as being like quasi-human-beings dressed up in funny suits. As Wittgenstein speculated about lions: if they could talk we could not understand them. Or, in other words, that they cannot talk is part of what makes our conception of what counts as language intelligible. A Dolphin at a keyboard is not substantially less puzzling than a random text generator appearing to produce a 'theory of language'.

(In case this sounds 'human-centric', I would ask you to consider what happens when we do have instances of communication with animals - the range of what can be 'said' is quite narrow, even where it is important to us. I love my cat, and believe that she knows how to 'tell' me she's hungry, but I can't distinguish between real linguistic intention and my empathic interpretation of her behaviour. And she certainly doesn't try to do philosophy ...

We really haven't done a lot better with dolphins than with domestic cats, despite many efforts. And, of course, Martians don't exist.)

Where I'm headed, whether you think this is the only possible destination or not, is somewhere like this:

'Language' is a concept we can only give a recursive account of - by pointing to some undeniable examples (e.g. what we are using to have this conversation), and showing how other examples depend on these (via translation, implication, etc.).

Also, talking to one another is a profoundly human capacity. Language is, without qualification, human language. The thing that we have been able to construct that we call language has certain structural features because that is how human cognition works. We can do grammar and spelling, but we can't do the integrated phase transition gyrations of the globular group minds of Gliese 667Cc, because we have no idea what that means.

We might ask, of course, why we can do grammar and spelling (and all the other stuff that goes with linguistic capacity), and we have found partial answers. We will only ever find partial answers, however, for two reasons that occur to me just now:

(1) The question is hard to articulate without confusion - if it is about what kind of biological machinery a human being has in order that it can produce language, we find ourselves in a self-referential loop.

(2) We can't give an account of linguistic capacity in terms of an individual. The sense of an individual having linguistic capacity can only be given in terms of the individual's membership of a language-using community.

(2) has a consequence similar to (1), but also the further consequence that no capacity or incapacity on the part of an individual member of the community would trump the community's judgement of their status as a language user. While we might refer to behavioural or biological facts about an individual, these would never be conclusive (for Goodman/Kripke reasons), and could not stand against the community's judgement. The only constraint on the intelligibility and correctness of this judgement would be whether making it rendered the members of the community unintelligible to one another (in which case no intelligible judgement, rather a false intelligible judgement, would have been made).

And of course, such a judgement would have to be internal to the community. A judgement from outside to the effect that the community clung to unintelligible beliefs is not distinguishable from a judgement that the translation schema of the outside observer was defective (Martians and nonsense again).

(At least that is the only intelligible judgement of this circumstance that can be made in our community ...)

So, if you're with me so far, here is another good thing we can do:

The structure of language (Chomsky, constructive theories of meaning, fascination with computation etc.) has been a distraction. We project this structure onto 'the world', and 'the world' allows us to do this, so the distraction has taken a terrible grip on the imagination of theorists. What has received less attention is exactly how human beings use, or can use, this structure. The mistake here is like imagining that enumerating the logical states that a piece of computer hardware can, in principle, be in, is more important than exploring the logical states that a working computer (doing human-useful tasks ...) might actually explore.

(It's easy to see this from some of the themes in this blog. For instance, it is very clear that a kind of anxiety - perhaps Stroud's metaphysical anxiety - blocks our recognition of the consequences that the various paradoxes I have taken seriously - "Moore's"; Open Question; Goodman/Kripke - have for theorising about truth, knowledge, meaning, and language. These consequences are not only not blocked by the structural and computational features that have been the focus of theorists, but they are also consequentially tied to them - both as a way of giving an account of their 'validity' and because they are the only foundational account that they will allow. The law of non-contradiction prevents us from discarding the consequences of this train of thought since other possibilities are incoherent; and it is also explained by the approach, since it, itself, can be thought of as a conclusion drawn from linguistic experiments in contradiction. Despite being the 'Sherlock Holmes' solution, some deep prejudices about the relationship between language and reality - prejudices that have nothing to do with any structural or semantic features of any actual or possible language - stand in the way of understanding it. Just as they seem to have stood in the way of Stroud's recognising the profound importance of his 'privileged statements'...)

A potentially tragic consequence of this diad - fascination with computational structure and blindness to its human application - is that we imagine we can construct talking machines without taking some important features of the human nature of language processing into account. We have picked out structure, parsing, some aspects of perception, computation, and computational prediction as the most important elements of 'intelligence' without really asking ourselves critical questions about whether these fully capture it. (Turing's infamous test side-steps the problem by cancelling out the 'humanity' term - he lets a human being be the ultimate judge.)

('Neural networks' which model some aspects of the way the human brain functions produce results in ways that cannot be ultimately validated. We can observe their spooky accuracy, but we have no way of predicting whether they might fail in some crucial test. We have sacrificed knowing how they work to the production of a simulacrum, demonstrations of whose 'reliability' depend on quite a shaky metaphor.)

Our capacity to construct machines that mimic these computational structures (permitted by a 'world' that allows us to talk about it in the way that we do) has progressed to a point where the ambiguities avoided by Turing's sleight of hand cannot be ignored. When these machines 'talk' to one another, they develop 'languages' that we do not understand. At the same time, some very ordinary human capacities seem to defy computational tractability. We do not understand our own 'intelligence' in a way that allows us to safely attribute it to our constructions.

When I talk to a human being, I have the comfort of knowing what human beings are like. This comfort does not depend upon a ground upwards demonstration of their humanity. It depends only on my ability to recognise them, and the relative absence of  'difficult cases' (although I'm sure we all know one or two ...). I might also be reassured by my knowledge that millions of years of evolution (with a good deal of attendant death and misery) have produced the biological and social behaviour that I now depend upon.

I don't have any of these comforts when I 'talk' to a machine. Cats and dolphins, if not Martians, seem a lot less alien.