250
The Essential Davidson
if we were to assume that no question can arise concerning a speaker's interpretation of his own words. But of course it can, since what his words mean depends in part on the clues to interpretation he has given the interpreter, or other evidence that he justifiably believes the interpreter has. The speaker can be wrong about what his own words mean. This is one of the reasons first person authority is not infallible. But the possibility of error does not eliminate the asymmetry. The asymmetry rests on the fact that the interpreter must, while the speaker doesn't, rely on what, if it were made explicit, would be a difficult inference in interpreting the speaker.
Neither speaker nor hearer knows in a special or mysterious way what the speaker's words mean; and both can be wrong. But there is a difference. The speaker, after bending whatever knowledge and craft he can to the task of saying what his words mean, cannot improve on the following sort of statement: 'My utterance of "Wagner died happy" is true if and only if Wagner died happy'. An interpreter has no reason to assume this will be his best way of stating the truth conditions of the speaker's utterance.
The best way to appreciate this difference is by imagining a situation in which two people who speak unrelated languages, and are ignorant of each other's languages, are left alone to learn to communicate. Deciphering a new language is not like learning a first language, for a true beginner has neither the reasoning power nor the stock of concepts the participants in the imagined situation have to draw on. This does not, however, affect the point I wish to stress, since what my imagined interpreter can treat consciously as evidence is exacdy what conditions die first learner to be a language user. Let one of the imagined pair speak and the other try to understand. It will not matter whether the speaker speaks his 'native' tongue, since his past social situation is irrelevant. (I assume the speaker has no interest in training the hearer to cope with die speaker's original speech community.) The best the speaker can do is to be interpretable, that is, to use a finite supply of distinguishable sounds applied consistendy to objects and situations he believes are apparent to his hearer. Obviously the speaker may fail in this project from time to time; in that case we can say if we please that he does not know what his words mean. But it is equally obvious that the interpreter has nothing to go on but the pattern of sounds the speaker exhibits in conjunction with further events (including, of course, further actions on the part of both speaker and interpreter). It makes no sense in this situation to wonder whether the speaker is generally getting things wrong. His behaviour may simply not be interpretable. But if it is, then what his words mean is (generally) what he intends them to mean. Since the 'language' he is speaking has no other hearers, the idea of the speaker misusing his language has no application. There is a presumption—an unavoidable presumption built into the nature of interpretation—that the speaker usually knows what he means. So there is a presumption that if he knows that he holds a sentence true, he knows what he believes.
15
A Nice Derangement of Epitaphs
Goodman Ace wrote radio sitcoms. According to Mark Singer, Ace often talked the way he wrote:
Rather than take for granite that Ace talks straight, a listener must be on guard for an occasional entre nous and me... or a long face no see. In a roustabout way, he will maneuver undl he selects the ideal phrase for the situation, hitting the nail right on the thumb. The careful conversationalist might try to mix it up with him in a baffle of wits. In quest of this pinochle of success, I have often wrecked my brain for a clowning achievement, but Ace's chickens always come home to roast. From time to time, Ace will, in a jersksome way, monotonise the conversation with witticisms too humorous to mention. It's high noon someone beat him at his own game, but I have never done it; cross my eyes and hope to die, he always wins thumbs down.1
I quote at length because philosophers have tended to neglect or play down the sort of language-use this passage illustrates. For example, Jonathan Bennett writes,
I doubt if I have ever been present when a speaker did something like shouting 'Water!' as a warning of fire, knowing what 'Water!' means and knowing that his hearers also knew, but. thinking that they would expect him to give to 'Water!' the normal meaning of 'Fire!'2
Bennett adds that, 'Although such things could happen, they seldom do.' I think such things happen all the time; in fact, if the conditions are generalized in a natural way, the phenomenon is ubiquitous.
Singer's examples are special in several ways. A malapropism does not have to be amusing or surprising. It does not have to be based on a cliche, and of course it does not have to be intentional. There need be no play on words, no hint of deliberate pun. We may smile at someone who says 'Lead the way and we'll precede', or, with Archie Bunker, 'We need a few laughs to break up the monogamy', because he has said something that, given the usual meanings of the words, is ridiculous or fun. But the humour is adventitious.
1   The New Yorker, 4 April 1977, p. 56. Reprinted by permission, 1977, The New Yorker Magazine, Inc.
2  Jonathan Bennett, Linguistic Behaviour, Cambridge, 1976, p. 186. Donald Davidson, 1985.
252
The Essential Davidson
Ace's malaprops generally make some sort of sense when the words are taken in the standard way, as in 'Familiarity breeds attempt', or 'We're all cremated equal', but this is not essential ('the pinochle of success'). What is interesting is the fact that in all these cases the hearer has no trouble understanding the speaker in the way the speaker intends.
It is easy enough to explain this feat on the hearer's part: the hearer realizes that the 'standard' interpretation cannot be the intended interpretation; through ignorance, inadvertence, or design the speaker has used a word similar in sound to the word that would have 'correcdy' expressed his meaning. The absurdity or inappropriateness of what the speaker would have meant had his words been taken in die 'standard' way alerts the hearer to trickery or error; the similarity in sound tips him off to the right interpretation. Of course there are many other ways the hearer might catch on; similarity of sound is not essential to the mal-aprop. Nor for that matter does the general case require that the speaker use a real word: most of 'The Jabberwock' is intelligible on first hearing.
It seems unimportant, so far as understanding is concerned, who makes a mistake, or whether there is one. When I first read Singer's piece on Goodman Ace, I thought that the word 'malaprop', though the name of Sheridan's character, was not a common noun that could be used in place of 'malapropism'. It turned out to be my mistake. Not that it mattered: I knew what Singer meant, even though I was in error about the word; I would have taken his meaning in the same way if he had been in error instead of me. We could both have been wrong and things would have gone as smoothly.
This talk of error or mistake is not mysterious nor open to philosophical suspicions. I was wrong about what a good dictionary would say, or what would be found by polling a pod of experts whose taste or training I trust. But error or mistake of this kind, with its associated notion of correct usage, is not philosophically interesting. We want a deeper notion of what words, when spoken in context, mean; and like the shallow notion of correct usage, we want the deep concept to distinguish between what a speaker, on a given occasion, means, and what his words mean. The widespread existence of malapropisms and their kin threatens the distinction, since here the intended meaning seems to take over from the standard meaning.           «
I take for granted, however, that nothing should be allowed to obliterate or even blur the distinction between speaker's meaning and literal meaning. In order to preserve the distinction we must, I shall argue, modify certain commonly accepted views about what it is to 'know a language', or about what a natural language is. In particular, we must pry apart what is literal in language from what is conventional or established.
Here is a preliminary stab at characterizing what I have been calling literal meaning. The term is too incrusted with philosophical and other extras to do much work, so let me call what I am interested in first meaning. The concept applies to words and sentences as uttered by a particular speaker on a particular
A Nice Derangement of Epitaphs                               253
occasion. But if the occasion, the speaker, and the audience are 'normal' or 'standard' (in a sense not to be further explained here), then the first meaning of an utterance will be what should be found by consulting a dictionary based on actual usage (such as Webster's Third). Roughly speaking, first meaning comes first in the order of interpretation. We have no chance of explaining the image in the following lines, for example, unless we know what 'foison' meant in Shakespeare's day:
Speak of the spring and foison of the year, The one doth shadow of your beauty show, The other as your bounty doth appear...3
Litde here is to be taken literally, but unless we know the literal, or first, meaning of the words we do not grasp and cannot explain the image.
But 'the order of interpretation' is not at all clear. For there are cases where we may first guess at the image and so puzzle out che first meaning. This might happen with the word 'tires' in the same sonnet:
On Helen's cheek all art of beauty set, And you in Grecian tires are painted new.
And of course it often happens that we can descry the literal meaning of a word or phrase by first appreciating what the speaker was getting at.
A better way to distinguish first meaning is through the intentions of the speaker. The intentions with which an act is performed are usually unambiguously ordered by the relation of means to ends (where this relation may or may not be causal). Thus the poet wants (let us say) to praise the beauty and generosity of his patron. He does this by using images that say the person addressed takes on every good aspect to be found in nature or in man or woman. This he does in turn by using the word 'tire' to mean 'attire' and the word 'foison' to mean 'harvest'. The order established here by 'by' can be reversed by using the phrase 'in order to'. In the 'in order to' sequence, first meaning is the first meaning referred to. ('With the intention of' with 'ing' added to the verb does as well.)
Suppose Diogenes utters the words 'I would have you stand from between me and the sun' (or their Greek equivalent) with the intention of uttering words that will be interpreted by Alexander as true if and only if Diogenes would have him stand from between Diogenes and the sun, and this with the intention of getting Alexander to move from between him and the sun, and this with the intention of leaving a good anecdote to posterity. Of course these are not the only intentions involved; there will also be the Gricean intentions to achieve certain of these ends through Alexander's recognition of some of the intentions involved. Diogenes' intention to be interpreted in a certain way requires such a self-referring intention,
Shakespeare, Sonnet 53.
254
The Essential Davidson
as does his intention to ask Alexander to move. In general, the first intention in the sequence to require this feature specifies the first meaning.
Because a speaker necessarily intends first meaning to be grasped by his audience, and it is grasped if communication succeeds, we lose nothing in the investigation of first meaning if we concentrate on the knowledge or ability a hearer must have if he is to interpret a speaker. What the speaker knows must correspond to something the interpreter knows if the speaker is to be understood, since if the speaker is understood he has been interpreted as he intended to be interpreted. The abilities of the speaker that go beyond what is required of an interpreter—invention and motor control—do not concern me here.
Nothing said so far limits first meaning to language; what has been characterised is (roughly) Grice's non-natural meaning, which applies to any sign or signal with an intended interpretation. What should be added if we want to restrict first meaning to linguistic meaning? The usual answer would, I think, be that in the case of language the hearer shares a complex system or theory with the speaker, a system which makes possible the articulation of logical relations between utterances, and explains the ability to interpret novel utterances in an organized way.
This answer has been suggested, in one form or another, by many philosophers and linguists, and I assume it must in some sense be right. The difficulty lies in getting clear about what this sense is. The particular difficulty with which I am concerned in this paper (for there are plenty of others) can be brought out by stating three plausible principles concerning first meaning in language: we may label them by saying they require that first meaning be systematic, shared, and prepared.
(1)  First meaning is systematic. A competent speaker or interpreter is able to interpret utterances, his own or those of others, on the basis of the semantic properties of the parts, or words, in the utterance, and the structure of the utterance. For this to be possible, there must be systematic relations between the meanings of utterances.
(2)  First meanings are shared. For speaker and interpreter to communicate successfully and regularly, they must share a method of interpretation of the sort described in (1).
(3)  First meanings are governed by learned conventions or regularities. The systematic knowledge or competence of the speaker or interpreter is learned in advance of occasions of interpretation and is conventional in character.
Probably no one doubts that there are difficulties with these conditions. Ambiguity is an example: often the 'same' word has more than one semantic role, and so the interpretation of utterances in which it occurs is not uniquely fixed by the features of the interpreter's competence so far mentioned. Yet, though the
A Nice Derangement of Epitaphs                            255
verbal and other features of the context of utterance often determine a correct interpretation, it is not easy or perhaps even possible to specify clear rules for disambiguation. There are many more questions about what is required of the competent interpreter. It does not seem plausible that there is a strict rule fixing the occasions on which we should attach significance to the order in which conjoined sentences appear in a conjunction: the difference between 'They got married and had a child' and 'They had a child and got married'. Interpreters certainly can make these distinctions. But part of the burden of diis paper is that much that they can do ought not to count as part of their basic linguistic competence. The contrast in what is meant or implied by the use of 'but' instead of 'and' seems to me another matter, since no amount of common sense unaccompanied by linguistic lore would enable an interpreter to figure it out.
Paul Grice has done more than anyone else to bring these problems to our attention and to help sort them out. In particular, he has shown why it is essential to distinguish between the literal meaning (perhaps what I am calling first meaning) of words and what is often implied (or implicated) by someone who uses those words. He has explored the general principles behind our ability to figure out such implicatures, and these principles must, of course, be known to speakers who expect to be taken up on them. Whether knowledge of these principles ought to be included in the description of linguistic competence may not have to be settled: on the one hand they are things a clever person could often figure out without previous training or exposure and they are things we could get along without. On the other hand they represent a kind of skill we expect of an interpreter and without which communication would be gready impoverished.
I dip into these matters only to distinguish them from the problem raised by malapropisms and the like. The problems touched on in the last two paragraphs all concern the ability to interpret words and constructions of the kind covered by our conditions (l)-(3); the questions have been what is required for such interpretation, and to what extent various competencies should be considered linguistic. Malapropisms introduce expressions not covered by prior learning, or familiar expressions which cannot be interpreted by any of the abilities so far discussed. Malapropisms fall into a different category, one that may include such things as our ability to perceive a well-formed sentence when the actual utterance was incomplete or grammatically garbled, our ability to interpret words we have never heard before, to correct slips of the tongue, or to cope with new idiolects. These phenomena threaten standard descriptions of linguistic competence (including descriptions for which I am responsible).
How should we understand or modify (l)-(3) to accommodate malapropisms? Principle (1) requires a competent interpreter to be prepared to interpret utterances of sentences he or she has never heard uttered before. This is possible because the interpreter can learn the semantic role of each of a finite number of words or phrases and can learn the semantic consequences of a finite number of modes of composition. This is enough to account for the ability to interpret
256
The Essential Davidson
utterances of novel sentences. And since the modes of composition can be iterated, diere is no clear upper limit to the number of sentences utterances of which can be interpreted. The interpreter thus has a system for interpreting what he hears or says. You might think of this system as a machine which, when fed an arbitrary utterance (and certain parameters provided by the circumstances of the utterance), produces an interpretation. One model for such a machine is a theory of truth, more or less along the lines of a Tarski truth definition. It provides a recursive characterisation of the truth conditions of all possible utterances of the speaker, and it does this through an analysis of utterances in terms of sentences made up from the finite vocabulary and the finite stock of modes of composition. I have frequendy argued that command of such a theory would suffice for interpretation.4 Here however there is no reason to be concerned with the details of the theory that can adequately model the ability of an interpreter. All that matters in the present discussion is that the theory has a finite base and is recursive, and these are features on which most philosophers and linguists agree.
To say that an explicit theory for interpreting a speaker is a model of the interpreter's linguistic competence is not to suggest that the interpreter knows any such theory. It is possible, of course, that most interpreters could be brought to acknowledge that they know some of the axioms of a theory of truth; for example, that a conjunction is true if and only if each of the conjuncts is true. And perhaps they also know theorems of the form 'An utterance of the sentence "There is life on Mars" is true if and only if there is life on Mars at the time of the utterance.' On the other hand, no pne now has explicit knowledge of a fully satisfactory theory for interpreting the speakers of any natural language.
In any case, claims about what would constitute a satisfactory theory are not, as I said, claims about the propositional Icnowledge of an interpreter, nor are they claims about the details of the inner workings of some part of the brain. They are rather claims about what must be said to give a satisfactory description of the competence of the interpreter. We cannot describe what an interpreter can do except by appeal to a recursive theory of a certain sort. It does not add anything to this thesis to say that if the theory does correcdy describe the competence of an interpreter, some mechanism in the interpreter must correspond to the theory.
Principle (2) says that for communication to succeed, a systematic method of interpretation must be shared. (I shall henceforth assume there is no harm in calling such a method a theory, as if the interpreter were using the theory we use to describe his competence.) The sharing comes to this: the interpreter uses his theory to understand the speaker; the speaker uses the same (or an equivalent) theory to guide his speech. For the speaker, it is a theory about how the interpreter will interpret him. Obviously this principle does not demand that speaker and interpreter speak the same language. It is an enormous convenience that
4 See the essays on radical interpretation in my Inquiries into Truth and Interpretation, Oxford University Press, 1984.
A Nice Derangement of Epitaphs                            257
many people speak in similar ways, and therefore can be interpreted in more or less the same way. But in principle communication does not demand that any two people speak the same language. What must be shared is the interpreter's and the speaker's understanding of the speaker's words.
For reasons that will emerge, I do not think that principles (1) and (2) are incompatible with the existence of malapropisms; it is only when they are combined with principle (3) that there is trouble. Before discussing principle (3) direcdy, however, I want to introduce an apparent diversion.
The perplexing issue that 1 want to discuss can be separated off from some related matters by considering a distinction made by Keith Donnellan, and something he said in its defence. Donnellan famously distinguished between two uses of definite descriptions. The referential'use is illustrated as follows: Jones says 'Smith's murderer is insane', meaning that a certain man, whom he (Jones) takes to have murdered Smith, is insane. Donnellan says that even if die man that Jones believes to have murdered Smith did not murder Smith, Jones has referred to the man he had in mind; and if that man is insane, Jones has said something true. The same sentence may be used attributively by someone who wants to assert that the murderer of Smith, whoever he may be, is insane. In this case, the speaker does not say something true if no one murdered Smith, nor has the speaker referred to anyone.
In reply, Alfred MacKay objected that Donnellan Shared Humpty Dumpty's theory of meaning: ' "When / use a word", Humpty Dumpty said,... "it means just what I choose it to mean."' In the conversation that went before, he had used the word 'glory' to mean 'a nice knockdown argument'. Donnellan, in answer, explains that intentions are connected with expectations and that you cannot intend to accomplish something by a certain means unless you believe or expect that the means will, or at least could, lead to the desired outcome. A speaker cannot, therefore, intend to mean something by what he says unless he believes his audience will interpret his words as he intends (the Gricean circle). Donnellan says,   .
If I were to end this reply to MacKay with the sentence 'There's glory for you' I would be guilty of arrogance and, no doubt, of overestimating the strength of what I have said, but given the background I do not think I could be accused of saying something unintelligible. I would be understood, and would I not have meant by 'glory' 'a nice knockdown argument'?5
I like this reply, and I accept Donnellan's original distinction between two uses of descriptions (there are many more than two). But apparendy I disagree with some view of Donnellan's, because unlike him I see almost no connection between the answer to MacKay's objection and the remarks on reference. The
5 Keith Donnellan, 'Putting Humpty Dumpty Together Again', The Philosophical Review, 77 (1968), p. 213. Alfred MacKay's artide, 'Mr Donnellan and Humpty Dumpty on Referring', appeared in the same issue of The Philosophical Review, pp. 197—202.
258
The Essential Davidson
reason is this. MacKay says you cannot change what words mean (and so their reference if that is relevant) merely by intending to; the answer is that this is true, but you can change the meaning provided you believe (and perhaps are justified in believing) that the interpreter has adequate clues for the new interpretation. You may deliberately provide those clues, as Donnellan did for his final 'There's glory for you'.
The trouble is that Donnellan's original distinction had nothing to do with words changing their meaning or reference. If, in the referential use, Jones refers to someone who did not murder Smith by using the description 'Smith's murderer', the reference is none the less achieved byway of the normal meanings of the words. The words therefore must have their usual reference. All that is needed, if we are to accept this way of describing the situation, is a firm sense of the difference between what words mean or refer to and what speakers mean or refer to. Jones may have referred to someone else by using words that referred to Smith's murderer; this is something he may have done in ignorance or deliberately. Similarly for Donnellan's claim that Jones has said something true when he says 'Smith's murderer is insane', provided the man he believes (erroneously) to have murdered Smith is insane. Jones has said something true by using a sentence that is false. This ís done intentionally all the time, for example in irony or metaphor. A coherent theory could not allow that under the circumstances Jones' sentence was true; nor would Jones think so if he knew the facts. Jones' belief about who murdered Smith cannot change the truth of the sentence he uses (and for the same reason cannot change the reference of the words in the sentence).
Humpty Dumpty is out of it. He cannot mean what he says he means because he knows that 'There's glory for you' cannot be interpreted by Alice as meaning 'There's a nice knockdown argument for you'. We know he knows this because Alice says 'I don't know what you mean by "glory"', and Humpty Dumpty retorts, 'Of course you don't—til I tell you'. It is Mrs Malaprop and Donnellan who interest me; Mrs Malaprop because she gets away with it without even trying or knowing, and Donnellan because he gets away with it on purpose.
Here is what I mean by 'getting away with it': the interpreter comes to the occasion of utterance armed with a-theory that tells him (or so he believes) what an arbitrary utterance of the speaker means. The speaker then says something with the intention that it will be interpreted in a certain way, and the expectation that it will be so interpreted. In fact this way is not provided for by the interpreter's theory. But the speaker is nevertheless understood; the interpreter adjusts his theory so that it yields the speaker's intended interpretation. The speaker has 'gotten away with it'. The speaker may or may not (Donnellan, Mrs Malaprop) know that he has got away with anything; the interpreter may or may not know that the speaker intended to get away with anything. What is common to the cases is that the speaker expects to be, and is, interpreted as the speaker intended although the interpreter did not have a correct theory in advance.
A Nice Derangement of Epitaphs
259
We do not need bizarre anecdotes or wonderlands to make the point. We all get away with it all the time; understanding the speech of others depends on it. Take proper names. In small, isolated groups everyone may know the names everyone else knows, and so have ready in advance of a speech encounter a theory that will, without correction, cope with the names to be employed. But even this semantic paradise will be destroyed by each new nickname, visitor, or birth. If a taboo bans a name, a speaker's theory is wrong until he learns of this fact; similarly if an outrigger canoe is christened.
There is not, so far as I can see, any theory of names that gets around the problem. If some definite description gives the meaning of a name, an interpreter still must somehow add to his theory the fact that the name new to him is to be matched with the appropriate description. If understanding a name is to give some weight to an adequate number of descriptions true of the object named, it is even more evident that adding a name to one's way of interpreting a speaker depends on no rule clearly stated in advance. The various theories that discover an essential demonstrative element in names do provide at least a partial rule for adding new names. But the addition is still an addition to the method of interpretation—what we may think of as the interpreter's view of the current language of the speaker. Finding a demonstrative element in names, or for that matter in mass nouns or words for natural kinds, does not reduce these words to pure demonstratives; that is why a new word in any of these categories requires a change in the interpreter's theory, and therefore a change in our description of his understanding of the speaker.
Mrs Malaprop and Donnellan make the case general. There is no word or construction that cannot be converted to a new use by an ingenious or ignorant speaker. And such conversion, while easier to explain because it involves mere substitution, is not the only kind. Sheer invention is equally possible, and we can be as good at interpreting it (say in Joyce or Lewis Carroll) as we are at interpreting the errors or twists of substitution. From the point of view of an ultimate explanation of how new concepts are acquired, learning to interpret a word that expresses a concept we do not already have is a far deeper and more interesting phenomenon than explaining the ability to use a word new to us for an old concept. But both require a change in one's way of interpreting the speech of another, or in speaking to someone who has the use of the word.
The contrast between acquiring a new concept or meaning along with a new word and merely acquiring a new word for an old concept would be salient if I were concerned with the infinitely difficult problem of how a first language is learned. By comparison, my problem is simple. I want to know how people who already have a language (whatever exacdy that means) manage to apply their skill or knowledge to actual cases of interpretation. All the things I assume an interpreter knows or can do depend on his having a mature set of concepts, and being at home with the business of linguistic communication. My problem is to
260
The Essential Davidson
describe what is involved in the idea of 'having a language' or of being at home with the business of linguistic communication.
Here is a highly simplified and idealised proposal about what goes on. An interpreter has, at any moment of a speech transaction, what I persist in calling a theory. (I call it a theory, as remarked before, only because a description of the interpreter's competence requires a recursive account.) I assume that the interpreter's theory has been adjusted to the evidence so far available to him: knowledge of the character, dress, role, sex, of the speaker, and whatever else has been gained by observing the speaker's behaviour, linguistic or otherwise. As the speaker speaks his piece the interpreter alters his theory, entering hypotheses about new names, altering the interpretation of familiar predicates, and revising past interpretations of particular utterances in the light of new evidence.
Some of what goes on may be described as improving die method of interpretation as the evidential base enlarges. But much is not like that. When Donnellan ends his reply to MacKay by saying 'There's glory for you', not only he, but his words, are correcdy interpreted as meaning There's a nice knockdown argument for you'. That's how he intends us to interpret his words, and we know this, since we have, and he knows we have, and we know he knows we have (etc.), the background needed to provide the interpretation. But up to a certain point (before MacKay came on the scene) this interpretation of an earlier utterance by Donnellan of the same words would have been wrong. To put this diffěrendy: the theory we actually use to interpret an utterance is geared to the occasion. We may decide later we could have done better by the occasion, but this does not mean (necessarily) that we now have a better theory for the next occasion. The reason for this is, as we have seen, perfecdy obvious: a speaker may provide us with information relevant to interpreting an utterance in the course of making the utterance.
Let us look at the process from the speaker's side. The speaker wants to be understood, so he intends to speak in such a way that he will be interpreted in a certain way. In order to judge how he will be interpreted, he forms, or uses, a picture of the interpreter's readiness to interpret along certain lines. Central to this picture is what the speaker believes is the starting theory of interpretation the interpreter has for him. The speaker does not necessarily speak in such a way as to prompt the interpreter to apply this prior theory; he may deliberately dispose the interpreter to modify his prior theory. But the speaker's view of the interpreter's prior theory is not irrelevant to what he says, nor to what he means by his words; it is an important part of what he has to go on if he wants to be understood.
I have distinguished what I have been calling the prior theory from what I shall henceforth call the passing theory. For the hearer, the prior theory expresses how he is prepared in advance to interpret an utterance of the speaker, while the passing theory is how he does interpret the utterance. For the speaker, the prior
A Nice Derangement of Epitaphs                            261
theory is what he believes the interpreter's prior theory to be, while his passing theory is the theory he intends the interpreter to use.
I am now in a position to state a problem that arises if we accept the distinction between the prior and the passing theory and also accept the account of linguistic competence given by principles (l)-(2). According to that account, each interpreter (and this includes speakers, since speakers must be interpreters) comes to a successful linguistic exchange prepared with a 'theory' which constitutes his basic linguistic competence, and which he shares with those with whom he communicates. Because each party has such a shared theory and knows that others share his theory, and knows that others know he knows (etc.), some would say that the knowledge or abilities that constitute the theory may be called conventions.
I think that the distinction between the prior and the passing theory, if taken seriously, undermines this commonly accepted account of linguistic competence and communication. Here is why. What must be shared for communication to succeed is the passing theory. For the passing theory is the one the interpreter actually uses to interpret an utterance, and it is the theory the speaker intends the interpreter to use. Only if these coincide is understanding complete. (Of course, there are degrees of success in communication; much may be right although something is wrong. This matter of degree is irrelevant to my argument.)
The passing theory is where, accident aside, agreement is greatest. As speaker and interpreter talk, their prior theories become more alike; so do their passing theories. The asymptote of agreement and understanding is when passing theories coincide. But the passing theory cannot in general correspond to an interpreter's linguistic competence. Not only does it have its changing list of proper names and gerrymandered vocabulary, but it includes every successful— i.e. correcdy interpreted—use of any other word or phrase, no matter how far out of the ordinary. Every deviation from ordinary usage, as long as it is agreed on for the moment (knowingly deviant, or not, on one, or both, sides), is in the passing theory as a feature of what the words mean on that occasion. Such meanings, transient though they may be, are literal; they are what I have called first meanings. A passing theory is not a theory of what anyone (except perhaps a philosopher) would call an actual natural language. 'Mastery' of such a language would be useless, since knowing a passing theory is only knowing how to interpret a particular utterance on a particular occasion. Nor could such a language, if we want to call it that, be said to have been learned, or to be governed by conventions. Of course things previously learned were essential to arriving at the passing theory, but what was learned could not have been the passing theory.
Why should a passing theory be called a theory at all? For the sort of theory we have in mind is, in its formal structure, suited to be the theory for an entire language, even though its expected field of application is vanishingly small. The answer is that when a word or phrase temporarily or locally takes over the role of some other word or phrase (as treated in a prior theory, perhaps), the entire burden ofthat role, with all its implications for logical relations to other words, phrases,
262
The Essential Davidson
and sentences, must be carried along by the passing theory. Someone who grasps the face that Mrs Malaprop means 'epithet' when she says 'epitaph' must give 'epithet' all the powers 'epitaph' has for many other people. Only a full recursive theory can do justice to these powers. These remarks do not depend on supposing Mrs Malaprop will always make this 'mistake'; once is enough to summon up a passing theory assigning a new role to 'epitaph'.
An interpreter's prior theory has a better chance of describing what we might think of as a natural language, particularly a prior theory brought to a first conversation. The less we know about the speaker, assuming we know he belongs to our language community, the more nearly our prior theory will simply be the theory we expect someone who hears our unguarded speech to use. If we ask for a cup of coffee, direct a taxi driver, or order a crate of lemons, we may know so little about our intended interpreter that we can do no better than to assume that he will interpret our speech along what we take to be standard lines. But all this is relative. In fact we always have the interpreter in mind; there is no such thing as how we expect, in the abstract, to be interpreted. We inhibit our higher vocabulary, or encourage it, depending on the most general considerations, and we cannot fail to have premonitions as to which of the proper names we know are apt to be correcdy understood.
In any case, my point is this: most of the tíme prior theories will not be shared, and there is no reason why they should be. Certainly it is not a condition of successful communication that prior theories be shared: consider the malaprop from ignorance. Mrs Malaprop's theory, prior and passing, is that 'A nice derangement of epitaphs' means a nice arrangement of epithets. An interpreter who, as we say, knows English, but does not know the verbal habits of Mrs Malaprop, has a prior theory according to which 'A nice derangement of epitaphs' means a nice derangement of epitaphs; but his passing theory agrees with that of Mrs Malaprop if he understands her words.
It is quite clear that in general the prior theory is neither shared by speaker and interpreter nor is it what we would normally call a language. For the prior theory has in it all the features special to the idiolect of the speaker that the interpreter is in a position to take into account before the utterance begins. One way to appreciate the difference between the prior theory and our ordinary idea of a person's language is to reflect on the fact that an interpreter must be expected to have quite different prior theories for different speakers—not as different, usually, as his passing theories; but these are matters that depend on how well the interpreter knows his speaker.
Neither the prior theory nor the passing theory describes what we would call the language a person knows, and neither theory characterizes a speaker's or interpreter's linguistic competence. Is there any theory that would do better?
Perhaps it will be said that what is essential to the mastery of a language is not knowledge of any particular vocabulary, or even detailed grammar, much less knowledge of what any speaker is apt to succeed in making his words and
A Nice Derangement of Epitaphs
263
sentences mean. What is essential is a basic framework of categories and rules, a sense of the way English (or any) grammars may be constructed, plus a skeleton list of interpreted words for fitting into the basic framework. If I put all this vaguely, it is only because I want to consider a large number of actual or possible proposals in one fell swoop; for I think they all fail to resolve our problem. They fail for the same reasons the more complete and specific prior theories fail: none of them satisfies the demand for a description of an ability that speaker and interpreter share and that is adequate to interpretation.
First, any general framework, whether conceived as a grammar for English, or a rule for accepting grammars, or a basic grammar plus rules for modifying or extending it—any such general framework, by virtue of the features that make it general, will by itself be insufficient for interpreting particular utterances. The general framework or theory, whatever it is, may be a key ingredient in what is needed for interpretation, but it can't be all that is needed since it fails to provide the interpretation of particular words and sentences as uttered by a particular speaker. In this respect it is like a prior theory, only worse because it is less complete.
Second, the framework theory must be expected to be different for different speakers. The more general and abstract it is, the more difference there can be without it mattering to communication. The theoretical possibility of such divergence is obvious; but once one tries to imagine a framework rich enough to serve its purpose, it is clear that such differences must also be actual. It is impossible to give examples, of course, until it is decided what to count in the framework: a sufficiendy explicit framework could be discredited by a single malapropism. There is some evidence of a more impressive sort that internal grammars do differ among speakers of 'the same language'. James McCawley reports that recent work by Haber shows
... that there is appreciable variation as to what rules of plural formation different speakers have, the variation being manifested in such things as the handling of novel words that an investigator has presented his subjects with, in the context of a task that will force them to use the word in the plural... Haber suggests that her subjects, rather than having a uniformly applicable process of plural formation, each have a 'core' system, which covers a wide range of cases, but not necessarily everything, plus strategies... for handling cases that are not covered by the 'core' system... Haber's data suggest that speakers of what are to the minutest details 'the same dialect' often have acquired grammars that differ in far more respects than their speech differs in.6
I have been trying to throw doubt on how clear the idea of 'speaking the same dialect' is, but here we may assume that it at least implies the frequent sharing of
passing theories.
6 James McCawley, 'Some Ideas Noc to Live By', Die Neuem Sprachen, 75 (1976), p. 157. These results are disputed by those who believe the relevant underlying rules and structures are prewired. My point obviously docs not depend on the example, or die level at which deviations are empirically possible.
264
The Essential Davidson
Bringing in grammars, theories, or frameworks more general than, and prior to, prior theories just emphasises the problem I originally presented in terms of the contrast between prior theories and passing theories. Stated more broadly now, the problem is this: what interpreter and speaker share, to the extent that communication succeeds, is not learned and so is not a language governed by rules or conventions known to speaker and interpreter in advance; but what the speaker and interpreter know in advance is not (necessarily) shared, and so is not a language governed by shared rules or conventions. What is shared is, as before, the passing theory; what is given in advance is the prior theory, or anything on which it may in turn be based.
What I have been leaving out of account up to now is what Haber calls a 'strategy', which is a nice word for the mysterious process by which a speaker or hearer uses what he knows in advance plus present data to produce a passing theory. What two people need, if they are to understand one another through speech, is the ability to converge on passing theories from utterance to utterance. Their starting points, however far back we want to take them, will usually be very different—as different as the ways in which they acquired their linguistic skills. So also, then, will the strategies and strategems that bring about convergence differ.
Perhaps we can give content to the idea of two people 'having the same language' by saying that they tend to converge on passing theories; degree or relative frequency of convergence would then be a measure of similarity of language. What use can we find, however, for the concept of a language? We could hold that any theory on which a speaker and interpreter converge is a language; but then there would be a new language for every unexpected turn in the conversation, and languages could not be learned and no one would want to master most of them.
We just made a sort of sense of the idea of two people 'having the same language', though we could not explain what a language is. It is easy to see that the idea of'knowing' a language will be in the same trouble, as will the project of characterising the abilities or capacities a person must have if he commands a language. But we might try to say in what a person's ability to interpret or speak to another person consists: it is the ability that permits him to construct a correct, that is, convergent, passing theory for speech transactions with that person. Again, the concept allows of degrees of application.
This characterization of linguistic ability is so nearly circular that it cannot be wrong: it comes to saying that the ability to communicate by speech consists in the ability to make oneself understood, and to understand. It is only when we look at the structure of this ability that we realize how far we have drifted from standard ideas of language mastery. For we have discovered no learnable common core of consistent behaviour, no shared grammar or rules, no portable interpreting machine set to grind out the meaning of an arbitrary utterance. We may say that linguistic ability is the ability to converge on a passing theory from
A Nice Derangement of Epitaphs                             265
time to time—this is what I have suggested, and I have no better proposal. But if we do say this, then we should realize that we have abandoned not only the ordinary notion of a language, but we have erased the boundary between knowing a language and knowing our way around in the world generally. For there are no rules for arriving at passing theories, no rules in any strict sense, as opposed to rough maxims and methodological generalities. A passing theory really is like a dieory at least in this, that it is derived by wit, luck, and wisdom from a private vocabulary and grammar, knowledge of the ways people get their point across, and rules of thumb for figuring out what deviations from the dictionary are most likely. There is no more chance of regularizing, or teaching, this process than there is of regularizing or teaching the process of creating new theories to cope with new data in any field—for that is what this process involves. The problem we have been grappling with depends on the assumption that communication by speech requires that speaker and interpreter have learned or somehow acquired a common method or theory of interpretation—as being able to operate on the basis of shared conventions, rules, or regularities. The problem arose when we realized that no method or theory fills this bill. The solution to the problem is clear. In linguistic communication nothing corresponds to a linguistic competence as often described: that is, as summarized by principles (1)—(3). The solution is to give up the principles. Principles (1) and (2) survive when understood in rather unusual ways, but principle (3) cannot stand, and it is unclear what can take its place. I conclude that there is no such thing as a language, not if a language is anything like what many philosophers and linguists have supposed. There is therefore no such thing to be learned, mastered, or born with. We must give up the idea of a clearly defined shared structure which language-users acquire and then apply to cases. And we should try again to say how convention in any important sense is involved in language; or, as I think, we should give up the attempt to illuminate how we communicate by appeal to conventions.