Monday, 20 October 2014

Understanding, States and Correctness

This post considers the negative account of understanding in §§138-242. By “negative account” I mean Wittgenstein's criticism of understanding as a state which governs our use of language. It is not, however, a straightforward description of that criticism as it develops in the Investigations. Instead, I've reordered the arguments in an attempt to bring out their key features – for the more I consider them the more it seems to me that Wittgenstein makes the same (or similar) points over and over again. Or perhaps it would be more accurate to say that the points he makes about a specific type of state turn out to be applicable across the board (and this, of course, is not a coincidence).

My next post will outline Wittgenstein's positive account of understanding – ie, his description of the role the concept plays in our lives and the circumstances within which it operates. Hopefully that won't take me ten months to write.


Let's start with the interlocutor's remark in §138:

But we understand the meaning of a word when we hear or say it; we grasp the meaning at a stroke, and what we grasp in this way is surely something different from the 'use' which is extended in time!

Why does this represent a challenge to Wittgenstein's claim that the meaning of a word is it's use in the language (§43)? Well, consider the following exchange between Lee and Jones:
Lee: I bought a piccolo today.
Jones: What does 'piccolo' mean?
Lee: This is a piccolo:
[Lee takes a piccolo out of his bag and shows Jones.]

At once the penny drops. Jones understands the meaning of the word “piccolo” at a stroke; she has grasped not merely what Lee meant in his specific remark, but how the word is to be used in a vast (indeed infinite) range of possible applications. But if that is so then how can meaning be use? Jones didn't study the use of the word in order to grasp its meaning – indeed her understanding seems prior to the use she will now go on to make of it. We might put it more strongly: how could a study of use (ie, what has happened in a relatively tiny number of past occasions) possibly determine how a word will function in an infinity of potential future cases?

Such thoughts can be built upon in a number of ways (as we shall see), but four points stand out:

  1. The above account presents understanding as an entity of some kind. As such, it readily suggests that understanding is a state. Jones was in one state before she understood, and acquiring the “entity” moved her into a new state: the state of understanding.
  2. The State governs our use of words. We use them as we do because we're in a particular state.
  3. The state enables us to use a word correctly. It is because Jones has acquired the entity that she can now respond to “piccolo” correctly when someone else uses it, and will also use it correctly herself. So the state must provide a standard of correctness against which usage can be assessed. Inevitably, this raises the issue of rules, since using a word correctly simply is using it in accordance with the relevant rule. So what is grasped at the moment of understanding would seem to be the rule itself (in some form or other). The state of understanding requires the possession of, and correcting functioning of, a rule.
  4. Achieving understanding seems to require a mental act of comprehension; the “entity” didn't merely fall into Jones – she grasped it. She not only saw the entity, she understood its meaning. She was conscious of the fact that she could now go on to use the word correctly.
Taken together, these points outline a foundational theory of understanding. But of course we still need to fill in the details: what sort of state are we talking about? In what sense does it govern use? And how does it provide a standard of correctness? As we shall see, the heart of Wittgenstein's objection to this whole approach lies in an unbearable tension arising out of the answers to these last two questions.

Physical States

We'll consider physical states first, if only because they seem the least mysterious option on offer. Here we are talking about “a state of an apparatus” (§149) such that it produces a given output for a given input. In the case of humans, the apparatus in question is held to be the brain, but it's worth noting that the theory doesn't rule out the existence of understanding in inorganic structures such as computers. For us, however, the process runs as follows: when we hear a word, the brain receives it as an input (or stimulus) which causes various neurological reactions (or processing), resulting in an output (or response). Clearly this account offers an explanation as to how the state governs use: it is a causal process. But how does it budget for correctness? Well, the state represents the rule for the use of the word. The rule is encoded into the brain's neurological structure just as the rules encoded into an iPhone's circuits allow Siri to answer our questions. The correct set of rules, properly encoded, results in the correct response for a particular verbal stimulus. Obviously our current understanding of the brain is too coarse to enable us to identify the detailed workings of these rules, but part of the theory's force lies in the thought that they must be there, or else understanding would seem to lie outside the realm of natural laws. It would be not simply mysterious but akin to magic (cf §158).

The first problem to note with this account is that it means my own understanding is hidden from me. I have no access to the rule I need to follow in order to use a word correctly (cf §153). So at best I can only infer that I understand a word – or we might say that I interpret my current state to be one of understanding. Likewise, when I use a word I can only infer that I'm using it correctly; I interpret its meaning in such-and-such a way. But from what do I make these inferences? What are my interpretations based upon? Consider a normal case of interpretation: I'm presented with the sentence “The cit purred and licked her paws”. Here I can easily infer that “cit” means “cat” (perhaps it's a typo). But the situation we're now presented with is one in which all our words need to be interpreted. So I'm not just inferring the meaning of “cit”; I'm also inferring the meaning of “inferring”. (And “meaning”. And “word”. And “and”.) It is difficult to make any sense of the phrase “I infer that by 'infer' I mean infer”. Worse still, the meaning of any interpretation I come up with will itself need to be interpreted, and so will the interpretation of my interpretation. And so on.

In fact, this same objection holds against the basic idea that the rules of language are encoded in our brains. Given that we have a language which is not a code, it's a simple enough job to decode a sequence like “Gl yv li mlg gl yv”. Likewise, it is because we have language in uncoded form that we can built coded versions of its rules into iPhones and PCs. But if all language everywhere is a code then the process of decoding can never end – and that amounts to saying that there's no such thing as a code in the first place. And no such thing as language, either. The concepts of interpretation, inference and encoding only make sense within a framework where there are some things which are not interpreted, inferred or encoded. The former are parasitic upon the latter. If you try to make them the fundamental basis for language then language itself collapses in a heap.

It's important to realise that the basic problem lies with the need for a standard of correctness. The physical state hides the rules of language, but you cannot follow a rule if you don't know what it is. And while it is possible to act in accordance with a rule without knowing that it exists, this can only happen within a broader practice of rule-following. Because the physical state makes following a rule impossible, it also makes acting in accordance with a rule impossible. And that, at bottom, is why notions such as interpretation can get no purchase here; we are attempting to act in accordance with a rule when there is no such thing as following it. In such a situation, the very notion of a rule disappears.

What this comes down to is that brain processes do not amount to rules. They may well operate according to causal laws, but there is a categorical distinction between rules and laws of nature. You don't need to know the laws of gravity in order to fall at 9.8 metres per second per second. And there is no such thing as correctly (or incorrectly) following that law. A causal process might do what we hoped or disappoint us; it might operate as predicted or might surprise us. But it cannot make a mistake. Therefore a causal process cannot provide a standard of correctness for usage, and therefore it cannot provide a foundational account of understanding.

Function and Behaviour

But it might be objected that we're needlessly over-complicating things. There's a much simpler way to allow for correctness in relation to physical states and processes. All that's required is for the outputs to be properly calibrated – that is, given the same input, the individual states reliably produce the same output. From this point of view it doesn't even matter if the states in question are completely different; after all, mobile phones can be constructed in numerous different ways but that doesn't bother us. It's the function that counts.

Interestingly, this approach leaves us in some doubt as to the precise criteria for specifying understanding. Our starting point was that understanding is a state; find the state and you've found the thing itself. But switching the focus towards function muddies the waters. Now it seems that understanding is a state plus the output (ie, behaviour). And since the state can take any number of different forms it seems to be the output which is doing all the heavy lifting here. The requisite state is merely whichever one happens to be in place when correctly calibrated outputs are produced. Doesn't it thereby drop out of the picture so far as defining understanding is concerned? Can't we say that understanding is simply correct behaviour? Or is there something important about the fact that the behaviour is produced – ie, that it's linked to inputs via the state? Should we take an inclusive approach and say that understanding is something which emerges out of the process as a whole, so that the total “system” includes all the linguistic interactions between people?

However, the real problem with this account is not which part to designate as the essence of understanding. The whole approach fails because, again, it cannot provide a standard of correctness. To say that we can generate a standard of correctness by properly calibrating outputs simply presupposes a standard – ie, we're assuming that the meaning of “properly calibrated” has already been settled. But that's precisely what we're supposed to be explaining! It's no help to say an output is right if it accords with everyone else's, because we've yet to fix what counts as “being in accordance with” in any given case. Here it's tempting to say that they're in accordance if they're the same – but this takes us in a circle, because whether two things count as “the same” depends on whether they are both in accordance with a particular rule. So we're defining “accords with” in terms of “the same” and “the same” in terms of “accords with”!

This brings out the significance of Wittgenstein's remarks on sameness which are dotted throughout §§138-242 (see §208, §§215-216 and §§224-227). It is tempting to suppose that the sameness of any two objects is an intrinsic feature of the world – something we can simply “read off” by regarding the objects themselves. From this point of view, a rule would be a description of the feature in question, and whether or not it was a correct rule would depend on how accurately it described the feature. So, for example, the rule “This = 'red'” would successfully correlate the word “red” with red objects because such objects were intrinsically the same as . The sample merely described redness and the rest followed as a matter of course. But, as we're already seen when discussing ostensive definition, this doesn't work as a foundational account because without a pre-established context the definition achieves nothing at all (cf §§28-31). In this case, the context is that the box is a sample of a particular colour and that we already know what counts as “being the same colour”. In different circumstances the rule would have had an entirely different effect. Likewise, if someone buys two copies of the morning paper has he bought two things once or one thing twice? That is, are we to classify the two papers as “the same” or “different”? The answer, of course, depends on why we're asking; it cannot be divorced from the activity in which the question arises. What a rule achieves depends on the context in which it is applied; it is woven into the way we live, the way we interact with things (with colours, for example). It does not function as a description.

But the function/behaviour approach implicitly assumes that rules describe “sameness” as an intrinsic feature of the world, and therefore, in themselves, provide a context-independent standard of correctness. All you need is the rule; the rest takes care of itself. But if you do not specify in advance what is to count as “being the same” (ie, “being in accordance with”) then any response can be classified as “the same” – or “different” – according to some formulation or other. It's easy to overlook this point because the examples of “sameness” used are extremely typical, and so we readily imagine a context in which they would be correct. But it is therefore we who are providing the standard of correctness. And this brings us back to our earlier point: far from showing how the input/process/output model gives rise to correctness, it simply presupposes what it is claiming to explain.

Mental States

Since the mechanistic causality of physical states presents an insurmountable difficulty, might not mental states prove more accommodating? Certainly there is something appealing about this idea. The mental realm offers something less rigid, something stranger – it is (we suppose) a nebulous realm whose workings we don't quite understand and yet seem capable of near miraculous results. This connects with the thought that, at best, a mechanical system can merely manipulate dead signs; you need the mind to breath life into the process and turn it into meaningful language (cf, the Blue Book, p3-4). And, in a way, such a transformation simply represents the move from mere causal “happenings” to correct or incorrect use. For unless our use of words can be right or wrong they cannot be meaningful – indeed, they do not even count as words. They are just bleats or scratches on a page.

Of course, in proposing mental states as a way forward we must be careful to avoid the difficulties we've already encountered. So, for example, such states must not operate in a causal or quasi-causal manner; a mental mechanism is no more helpful than a physical one. Likewise, a hidden state (or process) will be no use to us; that would simply re-introduce the absurdity of language as constant interpretation. Yet we still need to account for the way in which the state governs behaviour and produces a standard of correctness against which such behaviour can be judged.

Taken together, these requirements point us towards the linked notions of guidance (as opposed to causation) and characteristic experiences (as opposed to hidden processes). This picture can be fleshed out in a number of ways, but roughly speaking it hinges on the idea that we are directly aware of our understanding; we feel it. And this feeling guides us in our use of words. It suggests one use rather than another.

As usual, the devil is in the details. For example, is the feeling we get understanding itself? – in which case, we only understand a word while we're experiencing that feeling, which seems absurd. Or is it merely an indication that we understand? – in which case our actual understanding again seems hidden from us. Moreover, is there even such a thing as a characteristic experience of understanding? Do you have a characteristic experience every single time you hear the word “dog” or “the”?

Wittgenstein spends a considerable amount of time undermining the plausibility of the whole account, but it seems to me that it falls on two basic points he makes regarding correctness. First of all, feeling you understand isn't the same as understanding. Being sure you're right isn't the same as being right. Thinking you're following a rule isn't the same as following it (§202). If it was then there would be no way of settling disputes over rules, and therefore no standard of correctness for their use. Secondly, insofar as the experience guides me, it again fails to provide a standard of correctness, because “if it can guide me right, it can also guide me wrong” (§213). Guidance is not enough; I still need a means of evaluating it. And, of course, that cannot take the form of further guidance, because then I would need a means of evaluating the guidance about the guidance.

We've ended up back at the familiar regress, and this should alert us to the fact that “guidance” here is playing a similar role to inference and interpretation in earlier accounts. Guidance suggests an application of a rule, and that is akin to being offered an interpretation. But (we might object) can't guidance tell us what to do? Obviously it can, but then it's either up to us whether we do as we're told or we have no choice in the matter. In the former case, we need a standard of correctness by which to decide if it is right to obey. In the latter case, there's no question of a standard because we're again in a rigidly determined system. Our discussion repeatedly brings us back to this dilemma: either the state governs use in a rigid way, in which case there cannot be right or wrong usage; or else it leaves the final decision up to us, in which case we have no firm basis for a decision since that's precisely what the state is supposed to be providing.


But doesn't this introduce the role of the mental in the wrong way? Perhaps all this talk of characteristic experiences is besides the point. After all, item (iv) on our list spoke of mental acts rather than a passive awareness of sensations – isn't that what we need to add in order to provide a standard of correctness? This is what the interlocutor is getting at in §186 when he says, “The right step is the one that is in accordance with the order – as it was meant.” And it's a suggestion that brings us back to the difference between dead signs and living language. When I say something I don't just respond to a stimulus like a bell ringing at the push of a button; I talk about things and I mean what I say. And this meaning is (it seems) a conscious act which, inter alia, specifies how you should respond to my words. If you do so correctly (ie, as I meant it) then you've understood. Understanding, therefore, is the mental act of grasping the intention behind the word.

Hence (it seems) it is the mental act of meaning which supplies a standard of correctness and transforms dead signs into living language. But what do we actually mean here by “mental act”? What sort of act are we talking about? The answer seems to be something like picturing to ourselves what we mean by our words. As Wittgenstein (discussing “red”) rather scathingly puts it, “It is as if, when I uttered the word, I cast a sidelong glance at my own colour impression, as it were, in order to say to myself: I know all right what I mean by the word” (§274).

The most obvious objection to this account is that it doesn't seem to fit the facts. In the flow of a normal conversation we're not aware of performing acts of meaning relating to individual words or even groups of words. Perhaps (we might say) the acts go by too quickly to be noticed, or maybe they're performed subconsciously. But this just brings us back (yet again) to the problem of hidden criteria. If meaning is provided by associating a word with a sample and I'm not aware of making this association then how do I know what I mean by my words? The sample is supposed to be the rule that I follow, but, as we've seen, I cannot follow a rule of which I am unaware.

Even setting aside the issue of hiddenness, however, could the mental act achieve what we require of it? When the interlocutor introduces the subject of intention in §186, Wittgenstein immediately brushes it aside:
But that is just what is in question: what, at any stage, does follow from that sentence. Or, again, what at any stage we are to call “being in accordance” with it.

And consider in this context §239:
How is he to know what colour he is to pick out when he hears “red”? – Quite simple: he is to take the colour whose image occurs to him when he hears the word. – But how is he to know which colour it is 'whose image occurs to him'? Is a further criterion needed for that?

In other words, simply associating a word with a mental sample (or rule) isn't enough – it has to be the correct sample. Yet again, the process we posited in order to inject correctness into language itself requires a prior standard in order to make it work.

Let's pause a moment and consider the act of intending a bit more closely. I said it was a kind of picturing, but what might that involve? Here are three suggestions:
  1. I picture the object I mean by the word.
  2. I picture the use to which the word should be put.
  3. I picture both the object and its use as complementary parts of a whole.
How might the act described in (i) work? It seems to be a case of projecting the pictured object as a sample. If the picture is grasped by the hearer, she will use it to identify the appropriate objects in the world. But this just takes us back to where we were regarding the calibration of outputs. Which objects count as the appropriate ones? We are again trying to use a rule as a description, and so its status as a standard of correctness is either missing or assumed. Options (ii) and (iii) seek to rectify things by picturing the rule's application (it should be borne in mind that the “picture” need not be a literal one). But this amounts to providing a second rule for the application of the first – so the question now arises as to the application of the new rule. The regress has reappeared. (We are, in fact, right back in the difficulty described in §§139-141 and §146. See Understanding Part 1: Pictures for more on this.) Moreover, these aren't just problems for the person hearing the words; they apply to the speaker as well. Neither party has a standard of correctness and so neither can tell the meaning of what's said. (§504: “But if someone says, 'How am I to know what he means – I see only his signs?', then I say, 'How is he to know what he means, he too has only his signs?'”)

Still, it might be objected that we're not doing justice to the mental act. Of course it's not just about inwardly looking at a sign while uttering a word. How could anyone think that that would be enough? It's about meaning that the picture (or sample or rule) ought to be applied in a particular way. So this “meaning” is something that happens alongside paying attention to the picture. But what is this something extra? Wittgenstein brings out the strangeness of the claim in §332: “Utter a sentence, and think it; utter it with understanding. – And now don't utter it, and just do what you accompanied it with when you uttered it with understanding!” And again in §510: “Try to do the following: say 'It's cold here', and mean 'It's warm here'. Can you do it? – And what are you doing as you do it? And is there only one way of doing it?” The point, of course, is that whatever we end up trying is utterly unlike anything we'd be prepared to call “an act of meaning”. The very notion of meaning or intending as a substantive act begins to appear hollow.

At this point we are flirting with arguments which are given their fullest expression in the sections on private language (§§243-315), so I'm reluctant to delve into too much detail in this post. But I hope I've said enough at least to indicate how the interlocutor's account of intention struggles to fulfil its mission.

Logical Compulsion

Maybe, however, we're still looking at the mind in the wrong way. We've been talking about psychological laws and processes, and these are inevitably analogous to physical ones – so it's only to be expected that the same problems would emerge in both accounts. But isn't the important thing about the mind the fact that it is the medium by which we apprehend the underlying logic of things? We need a non-causal process which nonetheless does more than simply guide us; doesn't the notion of logical compulsion provide us with just this feature? For the force of logic is imperative yet not causal. Consider a typical logical statement such as “-(P˄-P)”. You cannot have both “P” and “not-P” at the same time. We are (somehow) forced to acknowledge the statement's truth – we cannot help but see that it is correct, that it must be correct. And yet this is not a matter of causation; if anything, the compulsion runs deeper than that. It is not founded upon the contingent laws of nature but upon the necessary structure of the world. It provides the framework within which such contingent laws can exist.

Tracing through the implications of this for language can take many forms, but the basic idea is that somehow logic locks words into a descriptive relationship with the world. Let's return for a moment to the distinction between language and dead signs. Seeing how the latter can become the former means answering a simple question: how do signs work? How is it that scratches on a page or emitted sounds actually mean something? How is it that they describe the world? Well, in the case of the Tractatus, it's a matter of logical structure. Names (which correspond to simples) are combined in a way that matches the logical structure of what they depict. They mirror the essence of a possible state of affairs. It is vital here that what is mirrored is a logical structure. That is what gives the connection its peculiar depth. Logic, we might say, acts as a kind of force which binds the two together in a depicting/depicted relationship. It adds an imperative quality to what otherwise would merely be a contingent likeness. So given that a particular mark or sound is being used as a sign, you must see that it represents such-and-such a state of affairs. Of course, this account doesn't explain anything. As an answer to “How do signs depict?” it amounts to replying “They just do”. And the Tractatus, at least, is audaciously upfront about this: you cannot say how the sign depicts. It shows that it does so by exhibiting the requisite logical form. This might seem unsatisfactory, but it must be the case because (the argument goes) it is a condition of the possibility of language itself.

The problems with this account are frustratingly familiar. First, it presents us with a sign which is a description of a possible state of affairs and yet which has the imperative quality of a rule. As such, it implicitly rests upon the idea of intrinsic sameness, and so it presupposes its standard of correctness. It says, in effect, that because the sign and state of affairs share a structure we cannot help but see that they're the same. But to say they share a structure is just another way of saying they're the same, and “they're the same because they're the same” says nothing. We still need to know what counts as “sharing a structure”.

It might be objected here that we've overlooked the sublime nature of logic; it is precisely because we're dealing with a logical force rather than an empirical one that this problem doesn't arise. Logic just is the realm in which the notion of intrinsic sameness holds good. Aside from the fact that this makes logic look uncomfortably like magic, it runs headlong into our other perennial problem: if logic creates an unbreakable bond between the sign and what it signifies then how is a mistake possible? Yet again we have produced an account of language which is too rigid to allow for the notion of correctness. And, as before, if we attempt to soften things – by an appeal to the guiding inner voice of “intuition”, for example – then we're left without a foundation for our judgements. Each guiding voice needs another to underwrite its advice.

We can express this as a problem arising from the fact that we're treating the sign as both a description and a rule. If you treat a description as a rule then its connection with what it describes is too rigid to allow for mistakes. But if you treat a rule as a description then it loses its imperative force and always stands in need of something further to justify its function. Rules and descriptions are conceptually distinct; running them together produces only chaos.


We have traced the idea of understanding as a state through various permutations, and each time we've been confronted by a similar set of difficulties, all of which centre upon the need to account for correctness. This suggests that the problem does not stem from choosing the wrong type of state, or misrepresenting its workings. It is far more fundamental than that. So where have we gone wrong?

Well, in a sense we have been misled by the very question we're trying to answer, viz: what is understanding? It's one of those questions that, according to Wittgenstein, produces in us a mental cramp. “We feel that we can't point to anything in reply to them and yet ought to point to something” (Blue Book, p1). He then remarks “We are up against one of the greatest sources of philosophical bewilderment: a substantive makes us look for a thing that corresponds to it”. And that's precisely our position regarding understanding; a form of expression (“we grasp the meaning at a stroke”) tempted us into treating it as a kind of entity. Of course, we weren't sure what type of entity we were dealing with, so we began to produce various hypotheses concerning its nature: it must be a physical state, or a functional state, a mental state, and so on. But the one constant amongst all this was that we were investigating a thing, and that by itself locked us into a particular way of considering it. Understanding, whatever its precise form, was a discrete entity which could be considered in isolation from its surroundings. It was context-independent. And insofar as it exhibited influence over other things, such as behaviour, it would do so in what was essentially a rigid, mechanical relationship.

And because understanding is bound up with grasping and following rules, those rules were themselves treated as things – mechanistic structures processing inputs in the brain or the mind or a platonic realm of logical compulsion. At the same time, however, these rules were assumed to function as descriptions, so that the computation was a matter of comparing a rule with a context-independent reality (either the reality of things being “thus and so” or the reality of the output which commonly followed from the relevant input). This act of comparison could only get off the ground if it was assumed that “sameness” or “being in accord with” was an intrinsic feature of the world, so that a standard of correctness was automatically provided along with the rule itself. That is, given a particular rule, it could not possibly be seen as anything other than a picture of such-and-such, and so there could be no doubt as to whether it was being correctly applied on any specific occasion.

Unravelling this illusion begins at §28 with the remark that “an ostensive definition can be variously interpreted in any case”. The ramifications of this simple observation echo throughout the Investigations in a variety of associated contexts; see, for example, §§85-87, §139, §§162-164 and §186. But the underlying point remains constant: a rule (or sample or picture) does not come with its method of application built into it. Imagining that it does so amounts to sublimating the concept; it turns the rule into something occult and utterly mysterious. And once this illusion is dispelled the whole mythology which has been built on top of it collapses. A rule cannot function as a description and therefore cannot state a context-independent fact (indeed, there are no context-independent facts). A rule is not itself a thing, so understanding the meaning of a word cannot be a matter of possessing this thing. Therefore understanding cannot be a state, for that was just another way of saying that understanding meant possessing (in some form or other) a rule. This is why normative explanations cannot be reduced to causes or rigid processes; they represent categorically different types of explanation, and the attempt to run them together only yields confusion (it's an example of what Wittgenstein calls “the crossing of different pictures” in §191).

And that, in turn, is why the switch from causation to something less rigid (guidance, interpretation, inference, etc) gets us nowhere. It introduces an alien type of explanation into the causal account. At the same time, it strips these new notions from their customary framework where they function alongside other, more direct practices. But it is only within this wider framework that they make sense; guidance is grounded in practices which are not themselves guidance, and the same goes for interpretation and inference (cf §1: “Explanations come to an end somewhere”). When we illicitly import them into the causal account, we expect them to provide a foundation for understanding, and this is precisely what they cannot do. Cut free from their own foundations they produce a regress whereby meaning itself vanishes into thin air: each interpretation requires a further interpretation and any response can be brought into accord with any rule according to some interpretation or other. In this situation, language collapses and with it go even our most basic concepts.

All this amounts to a damning indictment of the idea that we can found our concepts in a context-free description of reality – that we can start with “brute facts” about the world and show how features such as “understanding”, “knowing” and “meaning” arise out of them. (Think of the Tractatus: the facts are just there, and the philosopher's job is to explain how language reflects them.) The temptation to take this approach is built into the initial question: What is understanding? It doesn't express puzzlement about this or that aspect of understanding; it is not a question which arises out of a concrete difficulty. So it is unlike (eg) “What are the physical processes which underpin understanding?”. It is a question without a context and therefore invites a context-free answer. It is an example of the “engine idling” (§132), and the response it tempts us to make doesn't just fail to provide a convincing answer; nor does it merely destroy the concept it is trying to explain; it makes concepts themselves impossible – including the concept of brute facts. It is a response which brings itself into disrepute.

The context-free question is, we might say, a paradigmatically philosophical question. If, therefore, there's no legitimate way of answering it, that amounts to saying there's no such thing as philosophy itself. For Wittgenstein, however, the correct response is to describe the various contexts in which the relevant concept operates – the complex role it plays in our lives. So the answer is neither context-free nor an attempt to produce a single description covering all possible circumstances. And it is this approach which forms the basis of his positive account of understanding.


  1. I really appreciate your blog; WIttgenstein's difficult, and discovering your blog as made it much more interesting and digestible. Thank you for all your hard work at sussing out the Investigation!

    1. Thanks very much, Philo! Comment much appreciated.

  2. Welcome back, Phillip, and thanks for the post. Your posts always help as I read PI - and reread it and reread it ...

    Note: I've bolded words from the intentional idiom to indicate that I use them reluctantly since I consider most ill-defined but hard to avoid

    I find the concept of correct use confusing as presented in the post. I think treating its application to speaker and hearer separately helps. The speaker may be faced with correct use of a word in several ways. One is simply using the appropriate word for the intended purpose, eg, not using "piccolo" to refer to an object that is actually named "clarinet". Second is correct grammatical use, which is rule-based - although I suspect in much casual discourse those rules don't play an active role). But IMO, more important is structuring an utterance that uses the word and is such that there is reasonable expectation of achieving the purpose of the utterance, viz, a specific reaction by the hearer. Correct use in this sense is a complex function of speaker, hearer, and environment. The utterance is used correctly if the hearer's reaction is the one sought by the speaker. (What I mean by "the hearer understands the utterance".) But I don't see correct use in that sense (nor in the first sense) as rule based; it's learned through experience (presumably in that complex relationships between sensory neuronal stimulation and motor neuronal activation are formed).
    I'm less clear on how correct use applies to a person in the role of hearer. Of course, each person at one time or another may play either role, but with respect to any specific utterance a person is either speaker or hearer. And as hearer, the person's task is to understand the utterance. As in the piccolo example, an utterance may aid the hearer in correct future use of a word as speaker. But even in that example the objective of the speaker's utterance is only to name the object so that the hearer - Jones - can use the name in future utterances. Jones understands Lee's utterance if in future utterances she correctly refers to an object by using piccolo instead of pointing at the object or describing it. But correct use - eg, "a piccolo is a high pitched flute" must already be in place. (What I take to be the point of PI §31 - and in fact of your observation that "her understanding seems prior to the use she will now go on to make of [the word]".) So, it isn't clear to me how correct use applies to a person's role as hearer.

    "because Jones has acquired the entity ... she can now respond to 'piccolo' correctly when someone else uses it"

    I'm going to jump on this statement because I consider the underlying issue especially important since my position on it is counter to the usual concept of understand. IMO, only the speaker knows what reaction to an utterance is sought. In simple cases - "please pass the salt" - the common reaction will almost always be the one sought. But not always, in which case the speaker's use of words in an utterance may nonetheless be correct in that the hearer's reaction is the one sought - ie, the hearer understands the utterance - but I don't see in what relevant sense that makes the hearer's reaction necessarily "correct". If Lee had produced the piccolo and maliciously said instead "This is a clarinet", that Jones proceeds to use that word in referring to piccolos would indicate that she understood Lee's utterance (in my sense of that word), but her use would be decidedly incorrect.

    I have numerous comments on other issues, but this is enough for now.

    1. Hi Charles. Good to hear from you again.

      I'll just take your last para, since it seems to sum up the problems with what I take to be your approach.

      You say the hearer can never know what the speaker wants. So the hearer always has to guess what the speaker wants. So the hearer does not directly understand what the speaker says, but has to interpret it. And that includes the hearer's parents when they taught the hearer to speak. So the hearer doesn't directly understand her own words and has to guess what she means. But how is the hearer to do this? On what is she to base her guess? If the speaker says "pass the salt" she passes the salt and the speaker says "thank you" - but "thank you" could be deceptive, as could nods, smiles, etc. So when the hearer becomes a speaker and says "pass the salt" she doesn't know what her own words mean - she's only guessing. But, again, what does she base her guess on? And how does she know what she wants?

      Basically, I think you're misusing the term "know" here. You seem to be claiming that so long as there's a possibility of being wrong then someone can't know. But (a) that's not how the word is used, and (b) if it was, no-one would ever know anything.

    2. Although I don't see the matter as obviously relevant to your post, I think a discussion of the use of "know" in an epistemological context should specify the theory of knowledge being applied. I subscribe to a theory based on Sellars' "Empiricism and Philosophy of Mind" with some Davidson additions. In the context of that theory and the model I described in my previous comment, I have the following problems with your paragraphs about "know":

      - I neither said, nor intended to imply, that "the hearer can never know what the speaker wants". Davidson makes what seems to me a compelling argument in his essay "Three Varieties of Knowledge" that we can to some extent know the mind of another - indeed must know a good bit about the mind of a fellow language speaker in order for communication to work. Although I see on rereading my comment that it wasn't clear, the example about passing the salt was intended to be the usual case when (in effect, although not literally) the hearer "knows the mind" of the speaker. But as I said in the comment and as you note, sometimes the hearer's utterance may have the devious purpose of causing the hearer to react in an unusual (possibly inappropriate) way, and in general that purpose will be transparent to a hearer.

      - In my model, a hearer doesn't guess or interpret, only reacts - as I described in my previous comment. Davidson says a lot about interpretation, but he doesn't seem to make major incursions into the head, so perhaps that's OK for his purposes.

      - Parents initially train a young child to associate words with sensory stimulation and later to react to the stimulation. In Sellars' sense of the word, there's no “knowing” going on in the child until about age four. If that's correct, the training scenario for younger children is in general irrelevant to a discussion that assumes fully capable language users.

      - In my reaction-based concept of “understand”, "the hearer doesn't directly understand her own words" makes no sense. "her words" would only apply when she is the speaker, in which case I don't see how "understand" (in my sense above) applies.

      - Fallibility is assumed in the theories of knowledge I assume, so I certainly wouldn't (and didn't) claim that "so long as there's a possibility of being wrong then someone can't know".

      OTOH, once discussion moves from the realm of application of the intentional idiom to inside the head (where I find my model useful), I suppose I do believe that nobody “knows" anything since I consider that word inapplicable.

    3. Charles,

      You said "only the speaker knows what reaction to an utterance is sought" and I took this to mean that the hearer never knows what the speaker means. If that's unreasonable of me, show me why.

      In your subsequent post you said "(in effect, although not literally) the hearer "knows the mind" of the speaker". Please explain to me what you mean by that because, I must admit, in its present form it makes no sense to me.

      And, please, there's no need for "Davidson said...". Who cares? I'm putting forward the views of Wittgenstein, but I wouldn't dare suggest those views are correct because Wittgenstein said them. Again: who cares? We're two intelligent human beings with views and (we hope) good reasons for our views. Let's discuss them as such..

  3. Hi Phil, great to see you posting again - there is so much in this post that it is hard to comment on! One quibble and one personal hobby horse. First, the quibble. I think the piccolo example at the start is a bit dangerous, since it does seem to suggest (in contrast to what you later say) that seeing the word a concept relates to is the easiest/quickest way of understanding what it means. I know this is the last thing you would want to suggest, but unless it is a joke or a deliberate irony it seems a bit confusing. Second, the hobby horse. I think one of the most puzzling things here is the link between understanding/meaning and a specific point in time. Showing that I understood (or explaining what I meant) can be an extended process and can be going ok and then get stuck (you write a poem; i can to understand it; you nod delightedly for the first five minutes of my explanation and then I say something that makes you say: No, no you haven't understood it at all). Similarly, if I say: "I need to go to the Bank" and you say: "The underground station?" I will reply: "Of course, not" and there are an indefinite number of such questions ("Did you mean the Barclay Bank in Newquay?") to which I will have answers. So it does then seem amazing that in a split second I do something that has a limitless number of consequences which I could spend the rest of my life spelling out! How can all this be achieved in a moment unless there is some kind of amazing magic to the mind?

  4. Hello Paul - thanks, as always, for the comment.

    On the quibble, I was trying to set out the interlocutor's approach in its most persuasive form, so the account intentionally skates over various problems and inconsistencies. That whole bit should be read with scare quotes round it - and I probably should've made that clearer.

    On the hobbyhorse, this is the "crossing all the bridges before you get to them" idea, isn't it? And the truth is that I simply didn't manage to fit in a discussion of that aspect. I kept trying to, but somehow the appropriate opening never seemed to appear (in fact, I hint at it in the piccolo part and then never return to it). That's a failing on my part, of course; as you say, it's one of the most puzzling aspects of understanding.

    With a bit of luck, however, I'll be able to cover it in my explanation of W's positive account. It crops up naturally (I think) when discussing regularity, context and typical/atypical usage.

  5. "one of the most puzzling things here is the link between understanding/meaning and a specific point in time"

    Indeed, Paul, but since it's the interlocutor who asserts that having a picture "come before the mind" amounts to understanding an uttered word, aren't we to assume that the interlocutor is probably wrong?

    As I suggested above, it seems necessary to distinguish (in the case of objects) between name and use. What comes before the mind in an instant upon hearing an uttered name may be a mental image of an object that bears that name, but in what sense does that amount to understanding the utterance? Once the picture is "before the mind", what determines how the hearer is to respond? Your examples of understanding suggest - correctly IMO - that understanding is much more complex, involving context and often requiring elapsed time.

    Thinking of understanding an utterance in terms of "correct" (in the sense of being consistent with the speaker's purpose in producing the utterance) context-dependent response is compatible with that view. Determining what to say or do in response to a passage from a poem, or which object "Bank" names, seems clearly a context-dependent process, not a flash of instantaneous unanchored insight.

    1. "Determining what to say or do in response to a passage from a poem, or which object "Bank" names, seems clearly a context-dependent process, not a flash of instantaneous unanchored insight."

      Context certainly comes into - but the question is, in what way should we describe its significance? Is it something we check when making sure we're saying (or hearing) the right thing? I think the answer is: much of the time, no.

      And it's a similar story with a word like "bank". Much of the time we do not assure ourselves that the word is being used in sense x rather than sense y. In that respect it's not like reading a poem where interpretation is a much more common part of the activity. But in some circumstances even an ordinary word like "bank" can need interpretation.

      And, to stress again, W's point is that if we did interpret every single word or phrase then how could the process ever end?

  6. "Is [context] something we check when making sure we're saying ... the right thing? "

    Depends on how "checking" context is envisioned as fitting into the overall processing of speech.

    We learn the uses of a word and how to respond to those uses in various contexts. Eg, we may learn one response to "I'm going to the bank" when the speaker is leaving a suburban home to run errands in the town center and a different response when the speaker is leaving a summer home to go fishing in an adjacent river. One can look at the processing of an utterance as matching representations (presumably neural) of the utterance and the present context (an "input pattern") against a set of learned "response patterns", each comprising a representation of an utterance, a context in which a response to that utterance was learned, and the learned response. In which case context isn't something "checked" separately, it's an integral part of the process. (A greatly simplified model of the presumably complex process.)

    So, I would change the question to be something like:

    Is the role of context in determining what we say in response to an utterance something of which we are aware?

    which I too would answer "much of the time, no". In the case of familiar simple utterances, the matching seems to be effectively instantaneous and transparent. Only in cases for which finding a match that is close enough (in some sense) to warrant activating the associated response does the search process take an amount of time which makes us aware of it.

  7. Charles,

    You need to re-word your amended question slightly; at the moment it seems a bit ambiguous to me. Do you want to insert "What" at the start or change "in" to "that of"?

    1. Actually, re-reading your subsequent para, it's clear you mean the latter, so disregard my above comment. (This all very ironic, given the subject matter!)

      Now, I don't see how putting things in terms of matching representations within a context gets out of our difficulty - in fact, it seems to be a different way of stating the same position that was objected to in the first place.

      What counts as matching? Isn't this presupposing a standard of correctness (ie, what counts as "the same")? Where has that standard come from?

      And once the utterance and representation are matched with each other, what counts as matching them to the context? Don't we get the same problem there?

    2. Indeed, exchanges on complex topics like these seem to be case studies in how difficult "understanding" is when discussion moves beyond quotidian chit-chat.

      Contrary to what it may seem, in my comments I try to keep somewhat close to what I take to be the Wittgensteinian tack of addressing how language is actually used. But for my own purposes, I assume a somewhat detailed model of how this all works at the neural level, a model that I think is consistent with the thinking of some neurophysicians - at least as I interpret their (more or less popularized) writing.

      So, for better or worse here's the model I assume. The basic entity is a "learned" (via training and brain plasticity) neural structure comprising at least three functional components: one that in some way "encodes" the neural activity consequent to an experience of sensory stimulation, another that "encodes" contextual information, and a third that "encodes" the motor neuronal stimuli that implement a reaction to the encoded input. Such a neural structure is what I have in mind when I refer to a "context-dependent behavioral disposition".

      Because of my background in electronics, I tend to think of such a structure as analogous to a so-called "matched filter" - essentially a resonant circuit that responds to a range of inputs but responds maximally to one specific input. Ie, the "matching" is not binary - match "correctly" or not at all - but is a matter of degree.

      The reaction component of a disposition is executed only if the matched filter's response exceeds a threshold determined by individual experience, in particular experienced rewards and punishments consequent to various threshold settings. So, the setting of a threshold isn't rule-based and there's no "correct" setting for everyone. (Which is why I focused on the idea of "correct use" in my first comment - I'm unclear how to interpret that term.)

      I separate out context (often overlooked) only to emphasize its importance - I actually assume it's implemented as part of the neural representation of sensory stimulation. Ie, the processing isn't a two-step process - first process sensory input, then factor in context - but rather is integrated in the "matched filter".

      Note that this model addresses a common objection to simple "stimulus-response" models, viz, it isn't realistic to assume a stored (stimulus,response) pair for every possible stimulus. I can now flesh out "match that is close enough" from my last comment: it means "results in a matched filter response that exceeds the threshold". Ie, not every possible stimulus must have it's own matched filter in order to elicit a reaction.

      No doubt much more than anyone cares to know about my thinking, but I don't know how else to respond to those questions.

    3. Charles,

      Yes, I can see that that's the model you go by when approaching this issue. But I've spent quite some time outlining as clearly as I could what I consider to be serious (indeed, deadly) problems when that model is used as a foundational account of understanding. And I'm sorry, but so far all you've done is reiterate the model rather than addressing the problems. You've not answered my points, you've ignored them.

    4. Well, I certainly didn't intend to ignore any criticisms of the model. Your post starts with a discussion of PI §138 and the interlocutor's claim that "we grasp the meaning [of a word] at a stroke". Since there is neither context nor speaker purpose involved in that idea, it's irrelevant to my model which includes both.

      A major issue in the post is "correct use". My first comment addressed my confusion on that point and I eagerly await a reply.

      Your post addresses the idea inferred from "grasping" that understanding is "an entity", a state. I agree that it isn't and my model obviously doesn't assume that. Of course, a disposition is a state, but that state isn't itself "understanding". It's a latent action, and it's the action itself that I see one component of understanding. Another component is speaker intent, so the section of the post on intent is relevant to the model. Unfortunately that section is written in terms of several concepts that I are problematic for me. I've already noted my confusion about correctness. And I consider mental picturing a red herring since I am skeptical of its causal efficacy. I intended to address that in due time.

      "Sameness" is addressed in my model in that a matched filter response to an input is a measure of how "close" (in the sense described) the present input is to the input to which the filter is matched.

      Another idea that you discuss is that a person must in some sense be aware of her having understood. I am skeptical about introspection in general, so there is nothing like that in the model.

      Another issue is following rules. I'm not sure what's involved in that idea, so I have nothing to say about it so far.

      I don't know what "mental states" are, so there are none explicitly in my model and I have nothing to say about that section.

      Although I appreciate your considerable effort in trying to pull all this together into a coherent package, as I reread the relevant sections of PI it seems to me that each one warrants some independent discussion, and doing so would ease us into the bigger picture. Attacking it all at once seems quite ambitious.

      A comment that addressed every point in the post would be at least as long as the post - assuming it could even be written. So, saying nothing about any given point shouldn't be interpreted as ignoring the point.

    5. Charles,

      OK, to get back to your original point about correctness. You suggest separating out speaker and hearer. The speaker makes sounds that he estimates have the best chance of producing the desired outcome. The hearer has to understand the utterance - and it is not entirely clear to you what that actually amounts to.

      Having re-read your first comment I'm not entirely clear about what's troubling you, so you may need to come back on that one. Nonetheless, a few remarks:

      Regarding the speaker, is estimating likely outcomes a consistent feature of speaking? It certainly happens some of the time - for example, when speaking to someone with limited English we'll try not to use long words, unusual phrases, etc. We will think quite carefully about what we say. But it seems to me that much of the time we do no such thing. We just speak. And, in the absence of any evidence to the contrary, we assume the other person understands us. "Evidence to the contrary" includes: a wrong reaction by the listener; the listener telling us that she didn't understand; the listener looking a bit puzzled or confused; our prior knowledge that the listener only has basic English skills (and so might only be pretending to understand in order to avoid embarrassment); etc.

      However, if you make estimation a ubiquitous element of speaking then it seems to me we're back with problem of interpretation. I can never be sure you understood me, even if you do what I want (it might be a fluke) - and I can't be sure you didn't understand me if you don't do what I want (maybe you're simply disobeying me). Moreover, I can't be sure I'm using words correctly, since they were taught to me by people who, likewise, couldn't be sure I'd ever properly grasped their meaning. My best guess is that they meant x and their best guess is that I understood a word whose meaning they too can only guess at.

      There has to be a use of language which doesn't involve estimating or interpreting or guessing or else the whole thing fails to get off the ground.

      And, of course, the same goes for the listener. In the case of Jones, she was brought up short by an unfamiliar word. She asked its meaning and Lee gave an explanation. That is: Lee explained the rule he was following in using the word. And because the rule was a fairly simply one, and because Jones already spoke the language fluently, he could expect that that was (more or less) all that was needed. If he'd had reason to suppose that Jones didn't know the meaning of "flute" or even "musical instrument" then a more basic, more extensive explanation would've been necessary. But because Jones is fluent in English she can easily grasp the new rule. So she (a) understands what Lee meant when he said he had a piccolo, and (b) will probably understand the word in future (unless she forgets). From the point of view of their conversation, (a) is all that Lee is really concerned about, but it is an empirical fact about fluent language speakers that they get the hang of simple new words relatively easily. (Actually, in this case Lee's explanation is rather basic, and it is no means obviously true that Jones will be able to use the word correctly in future. Can she, for example, tell the difference between a piccolo and a penny whistle? In my post I said she'd be able to because I was writing from the point of view of the theory.)

      I'll stop there to give you a chance to respond. And thank you, as always, for your patience in discussing these things.

  8. No patience required. Some statements in PI seem compatible with my view, some not, and I'm grateful for help figuring out which are LW's, which are the interlocutor's, which I'm misinterpreting, etc.

    The speaker makes sounds that he estimates ...

    In order to simplify a process that is quite complex in general, I initially have in mind very simple sentences in very simple contexts and assume that a hearer's reaction to such sentences is essentially automatic. Therefore, I avoid words like "estimate" which suggest a much more complex process.

    is estimating likely outcomes a consistent feature of speaking?

    So, I'd reword this as:

    Is achieving an anticipated purpose a consistent feature of uttering a sentence?

    No. Some utterances are just a matter of form. A casual "Good morning, how are you today?" at the office normally isn't uttered to produce anything more than "Fine, thanks" - a response the purpose of which, in turn, is typically to terminate the exchange.

    it seems to me that much of the time we ... just speak.

    Much speech certainly seems to be content-free, but that isn't inconsistent with most speech having a purpose: to fill an awkward pause, to impress, to (mis)inform, etc And the purpose of some speech may not be to elicit an immediate overt reaction, but rather to result in latent action in the form of a new disposition in the hearer.

    in the absence of any evidence to the contrary, we assume the other person understands us

    I don't see how we can make that assumption without specifying what evidence can confirm it. (My suggested interpretation of "understanding" does so.) In any event, what we do in the absence of confirming evidence presumably depends on context. If an anticipated reaction is important but not forthcoming, we normally wouldn't assume understanding and just hope for the best; we'd follow up in some way.

    I can never be sure you understood me, even if you do what I want

    In my interpretation of "understand", if my anticipated reaction to an utterance occurs, that is precisely the criterion for success. However, since I assume the possibility of latent reactions, I do agree with the following:

    Even in the absence of the anticipated hearer reaction, an utterance may have been understood in that a disposition to produce that reaction in the future may have been created. But only some future triggering of that disposition can provide evidence of that understanding.

    I can't be sure you didn't understand me if you don't do what I want (maybe you're simply disobeying me).

    With my interpretation, failure to react as anticipated is failure to "understand". Disobeying, lying, et al, are possible explanations for that failure, but to assert that the utterance is nonetheless "understood" suggests some other interpretation. And that is just the danger I see in using the intentional idiom in discussions such as this. If we're exploring what might constitute some word from that idiom (here, "understanding"), using the word as it's used in casual conversation can easily mislead since the word may not be sufficiently well-defined.

    There has to be a use of language which doesn't involve estimating or interpreting or guessing or else the whole thing fails to get off the ground.

    I don't know if it's by design, but this sounds like a statement of the holism of linguistic interpretation, ie, that there is, so to speak, a "core" of utterances each of which is assumed (individually and defeasibly) to be correctly interpreted and on which interpretations of other utterances is built. Since my gurus are all fans of holism, I naturally defer to them on the point and agree.

    A much briefer discussion of the hearer will follow.

  9. By now it will come as no surprise that I have a different take on the example of the piccolo. As I said before, it seems roughly equivalent to the chess king example in PI §31, but because I prefer the analogy of tools for words (because IMO both are used for a purpose!), I'll create yet another example scenario.

    Suppose young Tom has been apprenticed to mute carpenter Cal and learns the trade entirely by following Cal's example. Ie, he knows the use of every tool but not its name. Cal takes on a partner, who can speak and refers to tools by name. When Tom says he doesn't know what a "saw" is, the new partner points at one. How are we to describe what changes Tom undergoes when that happens?

    He obviously doesn't learn the use of the tools - that's already in place. He does learn to associate a name with each tool. So, one question seems to be whether he also learns the use of the names. The message I get from PI §31 is that the answer is "no". If Tom is otherwise fluent in the language, he presumably has discussed his work with others (ie, he participates in a language game) but has had to use descriptions in referring to tools: "I injured myself today with the tool I use to cut boards into two pieces". Then learning the name of a tool amounts to learning an abbreviation of a description. In sentences that use the description, Tom can now substitute the name for the description. And the ability to produce those sentences was already in place.

    Whether he also acquires the ability to "understand" those sentences is unclear (at least in my interpretation of the word). Are the dispositions that caused reactions to sentences that use the description still in place? Or in some sense "modified" to accept the names? Or are new dispositions that trigger on the names created? Who knows.

    Note that I don't address "rules". That's because I don't "understand" that word in this context. (Ie, I can't react as you presumably intend - I'm missing the necessary dispositions.) Eg, you say "Lee explained the rule he was following in using the word." In the piccolo example as described, I see him as simply teaching Jones an association between name and object. How does that amount to "explaining a rule"?